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"
78 #define CONSUMERD_FILE "lttng-consumerd"
81 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
82 static int tracing_group_name_override
;
83 static char *opt_pidfile
;
84 static int opt_sig_parent
;
85 static int opt_verbose_consumer
;
86 static int opt_daemon
, opt_background
;
87 static int opt_no_kernel
;
88 static char *opt_load_session_path
;
89 static pid_t ppid
; /* Parent PID for --sig-parent option */
90 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
92 static int lockfile_fd
= -1;
94 /* Set to 1 when a SIGUSR1 signal is received. */
95 static int recv_child_signal
;
98 * Consumer daemon specific control data. Every value not initialized here is
99 * set to 0 by the static definition.
101 static struct consumer_data kconsumer_data
= {
102 .type
= LTTNG_CONSUMER_KERNEL
,
103 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
104 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
107 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
108 .lock
= PTHREAD_MUTEX_INITIALIZER
,
109 .cond
= PTHREAD_COND_INITIALIZER
,
110 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
112 static struct consumer_data ustconsumer64_data
= {
113 .type
= LTTNG_CONSUMER64_UST
,
114 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
115 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
118 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
119 .lock
= PTHREAD_MUTEX_INITIALIZER
,
120 .cond
= PTHREAD_COND_INITIALIZER
,
121 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
123 static struct consumer_data ustconsumer32_data
= {
124 .type
= LTTNG_CONSUMER32_UST
,
125 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
126 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
129 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
130 .lock
= PTHREAD_MUTEX_INITIALIZER
,
131 .cond
= PTHREAD_COND_INITIALIZER
,
132 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
135 /* Command line options */
136 static const struct option long_options
[] = {
137 { "client-sock", required_argument
, 0, 'c' },
138 { "apps-sock", required_argument
, 0, 'a' },
139 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
140 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
141 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
142 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
143 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
144 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
145 { "consumerd32-path", required_argument
, 0, '\0' },
146 { "consumerd32-libdir", required_argument
, 0, '\0' },
147 { "consumerd64-path", required_argument
, 0, '\0' },
148 { "consumerd64-libdir", required_argument
, 0, '\0' },
149 { "daemonize", no_argument
, 0, 'd' },
150 { "background", no_argument
, 0, 'b' },
151 { "sig-parent", no_argument
, 0, 'S' },
152 { "help", no_argument
, 0, 'h' },
153 { "group", required_argument
, 0, 'g' },
154 { "version", no_argument
, 0, 'V' },
155 { "quiet", no_argument
, 0, 'q' },
156 { "verbose", no_argument
, 0, 'v' },
157 { "verbose-consumer", no_argument
, 0, '\0' },
158 { "no-kernel", no_argument
, 0, '\0' },
159 { "pidfile", required_argument
, 0, 'p' },
160 { "agent-tcp-port", required_argument
, 0, '\0' },
161 { "config", required_argument
, 0, 'f' },
162 { "load", required_argument
, 0, 'l' },
163 { "kmod-probes", required_argument
, 0, '\0' },
164 { "extra-kmod-probes", required_argument
, 0, '\0' },
168 /* Command line options to ignore from configuration file */
169 static const char *config_ignore_options
[] = { "help", "version", "config" };
171 /* Shared between threads */
172 static int dispatch_thread_exit
;
174 /* Global application Unix socket path */
175 static char apps_unix_sock_path
[PATH_MAX
];
176 /* Global client Unix socket path */
177 static char client_unix_sock_path
[PATH_MAX
];
178 /* global wait shm path for UST */
179 static char wait_shm_path
[PATH_MAX
];
180 /* Global health check unix path */
181 static char health_unix_sock_path
[PATH_MAX
];
183 /* Sockets and FDs */
184 static int client_sock
= -1;
185 static int apps_sock
= -1;
186 int kernel_tracer_fd
= -1;
187 static int kernel_poll_pipe
[2] = { -1, -1 };
190 * Quit pipe for all threads. This permits a single cancellation point
191 * for all threads when receiving an event on the pipe.
193 static int thread_quit_pipe
[2] = { -1, -1 };
194 static int ht_cleanup_quit_pipe
[2] = { -1, -1 };
197 * This pipe is used to inform the thread managing application communication
198 * that a command is queued and ready to be processed.
200 static int apps_cmd_pipe
[2] = { -1, -1 };
202 int apps_cmd_notify_pipe
[2] = { -1, -1 };
204 /* Pthread, Mutexes and Semaphores */
205 static pthread_t apps_thread
;
206 static pthread_t apps_notify_thread
;
207 static pthread_t reg_apps_thread
;
208 static pthread_t client_thread
;
209 static pthread_t kernel_thread
;
210 static pthread_t dispatch_thread
;
211 static pthread_t health_thread
;
212 static pthread_t ht_cleanup_thread
;
213 static pthread_t agent_reg_thread
;
214 static pthread_t load_session_thread
;
217 * UST registration command queue. This queue is tied with a futex and uses a N
218 * wakers / 1 waiter implemented and detailed in futex.c/.h
220 * The thread_registration_apps and thread_dispatch_ust_registration uses this
221 * queue along with the wait/wake scheme. The thread_manage_apps receives down
222 * the line new application socket and monitors it for any I/O error or clean
223 * close that triggers an unregistration of the application.
225 static struct ust_cmd_queue ust_cmd_queue
;
228 * Pointer initialized before thread creation.
230 * This points to the tracing session list containing the session count and a
231 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
232 * MUST NOT be taken if you call a public function in session.c.
234 * The lock is nested inside the structure: session_list_ptr->lock. Please use
235 * session_lock_list and session_unlock_list for lock acquisition.
237 static struct ltt_session_list
*session_list_ptr
;
239 int ust_consumerd64_fd
= -1;
240 int ust_consumerd32_fd
= -1;
242 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
243 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
244 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
245 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
246 static int consumerd32_bin_override
;
247 static int consumerd64_bin_override
;
248 static int consumerd32_libdir_override
;
249 static int consumerd64_libdir_override
;
251 static const char *module_proc_lttng
= "/proc/lttng";
254 * Consumer daemon state which is changed when spawning it, killing it or in
255 * case of a fatal error.
257 enum consumerd_state
{
258 CONSUMER_STARTED
= 1,
259 CONSUMER_STOPPED
= 2,
264 * This consumer daemon state is used to validate if a client command will be
265 * able to reach the consumer. If not, the client is informed. For instance,
266 * doing a "lttng start" when the consumer state is set to ERROR will return an
267 * error to the client.
269 * The following example shows a possible race condition of this scheme:
271 * consumer thread error happens
273 * client cmd checks state -> still OK
274 * consumer thread exit, sets error
275 * client cmd try to talk to consumer
278 * However, since the consumer is a different daemon, we have no way of making
279 * sure the command will reach it safely even with this state flag. This is why
280 * we consider that up to the state validation during command processing, the
281 * command is safe. After that, we can not guarantee the correctness of the
282 * client request vis-a-vis the consumer.
284 static enum consumerd_state ust_consumerd_state
;
285 static enum consumerd_state kernel_consumerd_state
;
288 * Socket timeout for receiving and sending in seconds.
290 static int app_socket_timeout
;
292 /* Set in main() with the current page size. */
295 /* Application health monitoring */
296 struct health_app
*health_sessiond
;
298 /* Agent TCP port for registration. Used by the agent thread. */
299 unsigned int agent_tcp_port
= DEFAULT_AGENT_TCP_PORT
;
301 /* Am I root or not. */
302 int is_root
; /* Set to 1 if the daemon is running as root */
304 const char * const config_section_name
= "sessiond";
306 /* Load session thread information to operate. */
307 struct load_session_thread_data
*load_info
;
309 /* Global hash tables */
310 struct lttng_ht
*agent_apps_ht_by_sock
= NULL
;
313 * Whether sessiond is ready for commands/health check requests.
314 * NR_LTTNG_SESSIOND_READY must match the number of calls to
315 * sessiond_notify_ready().
317 #define NR_LTTNG_SESSIOND_READY 3
318 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
320 /* Notify parents that we are ready for cmd and health check */
322 void sessiond_notify_ready(void)
324 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
326 * Notify parent pid that we are ready to accept command
327 * for client side. This ppid is the one from the
328 * external process that spawned us.
330 if (opt_sig_parent
) {
335 * Notify the parent of the fork() process that we are
338 if (opt_daemon
|| opt_background
) {
339 kill(child_ppid
, SIGUSR1
);
345 void setup_consumerd_path(void)
347 const char *bin
, *libdir
;
350 * Allow INSTALL_BIN_PATH to be used as a target path for the
351 * native architecture size consumer if CONFIG_CONSUMER*_PATH
352 * has not been defined.
354 #if (CAA_BITS_PER_LONG == 32)
355 if (!consumerd32_bin
[0]) {
356 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
358 if (!consumerd32_libdir
[0]) {
359 consumerd32_libdir
= INSTALL_LIB_PATH
;
361 #elif (CAA_BITS_PER_LONG == 64)
362 if (!consumerd64_bin
[0]) {
363 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
365 if (!consumerd64_libdir
[0]) {
366 consumerd64_libdir
= INSTALL_LIB_PATH
;
369 #error "Unknown bitness"
373 * runtime env. var. overrides the build default.
375 bin
= lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
377 consumerd32_bin
= bin
;
379 bin
= lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
381 consumerd64_bin
= bin
;
383 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
385 consumerd32_libdir
= libdir
;
387 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
389 consumerd64_libdir
= libdir
;
394 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
401 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
407 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
419 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
421 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
423 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
427 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
429 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event
*events
,
432 return __sessiond_set_thread_pollset(events
, size
,
433 ht_cleanup_quit_pipe
);
437 int __sessiond_check_thread_quit_pipe(int fd
, uint32_t events
, int a_pipe
)
439 if (fd
== a_pipe
&& (events
& LPOLLIN
)) {
446 * Check if the thread quit pipe was triggered.
448 * Return 1 if it was triggered else 0;
450 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
452 return __sessiond_check_thread_quit_pipe(fd
, events
,
453 thread_quit_pipe
[0]);
457 * Check if the ht_cleanup thread quit pipe was triggered.
459 * Return 1 if it was triggered else 0;
461 int sessiond_check_ht_cleanup_quit(int fd
, uint32_t events
)
463 return __sessiond_check_thread_quit_pipe(fd
, events
,
464 ht_cleanup_quit_pipe
[0]);
468 * Init thread quit pipe.
470 * Return -1 on error or 0 if all pipes are created.
472 static int __init_thread_quit_pipe(int *a_pipe
)
478 PERROR("thread quit pipe");
482 for (i
= 0; i
< 2; i
++) {
483 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
494 static int init_thread_quit_pipe(void)
496 return __init_thread_quit_pipe(thread_quit_pipe
);
499 static int init_ht_cleanup_quit_pipe(void)
501 return __init_thread_quit_pipe(ht_cleanup_quit_pipe
);
505 * Stop all threads by closing the thread quit pipe.
507 static void stop_threads(void)
511 /* Stopping all threads */
512 DBG("Terminating all threads");
513 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
515 ERR("write error on thread quit pipe");
518 /* Dispatch thread */
519 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
520 futex_nto1_wake(&ust_cmd_queue
.futex
);
524 * Close every consumer sockets.
526 static void close_consumer_sockets(void)
530 if (kconsumer_data
.err_sock
>= 0) {
531 ret
= close(kconsumer_data
.err_sock
);
533 PERROR("kernel consumer err_sock close");
536 if (ustconsumer32_data
.err_sock
>= 0) {
537 ret
= close(ustconsumer32_data
.err_sock
);
539 PERROR("UST consumerd32 err_sock close");
542 if (ustconsumer64_data
.err_sock
>= 0) {
543 ret
= close(ustconsumer64_data
.err_sock
);
545 PERROR("UST consumerd64 err_sock close");
548 if (kconsumer_data
.cmd_sock
>= 0) {
549 ret
= close(kconsumer_data
.cmd_sock
);
551 PERROR("kernel consumer cmd_sock close");
554 if (ustconsumer32_data
.cmd_sock
>= 0) {
555 ret
= close(ustconsumer32_data
.cmd_sock
);
557 PERROR("UST consumerd32 cmd_sock close");
560 if (ustconsumer64_data
.cmd_sock
>= 0) {
561 ret
= close(ustconsumer64_data
.cmd_sock
);
563 PERROR("UST consumerd64 cmd_sock close");
569 * Generate the full lock file path using the rundir.
571 * Return the snprintf() return value thus a negative value is an error.
573 static int generate_lock_file_path(char *path
, size_t len
)
580 /* Build lockfile path from rundir. */
581 ret
= snprintf(path
, len
, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE
, rundir
);
583 PERROR("snprintf lockfile path");
590 * Wait on consumer process termination.
592 * Need to be called with the consumer data lock held or from a context
593 * ensuring no concurrent access to data (e.g: cleanup).
595 static void wait_consumer(struct consumer_data
*consumer_data
)
600 if (consumer_data
->pid
<= 0) {
604 DBG("Waiting for complete teardown of consumerd (PID: %d)",
606 ret
= waitpid(consumer_data
->pid
, &status
, 0);
608 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
610 if (!WIFEXITED(status
)) {
611 ERR("consumerd termination with error: %d",
614 consumer_data
->pid
= 0;
618 * Cleanup the session daemon's data structures.
620 static void sessiond_cleanup(void)
623 struct ltt_session
*sess
, *stmp
;
626 DBG("Cleanup sessiond");
629 * Close the thread quit pipe. It has already done its job,
630 * since we are now called.
632 utils_close_pipe(thread_quit_pipe
);
635 * If opt_pidfile is undefined, the default file will be wiped when
636 * removing the rundir.
639 ret
= remove(opt_pidfile
);
641 PERROR("remove pidfile %s", opt_pidfile
);
645 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
648 snprintf(path
, PATH_MAX
,
650 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
651 DBG("Removing %s", path
);
654 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
655 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
);
656 DBG("Removing %s", path
);
660 snprintf(path
, PATH_MAX
,
661 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
663 DBG("Removing %s", path
);
666 snprintf(path
, PATH_MAX
,
667 DEFAULT_KCONSUMERD_PATH
,
669 DBG("Removing directory %s", path
);
672 /* ust consumerd 32 */
673 snprintf(path
, PATH_MAX
,
674 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
676 DBG("Removing %s", path
);
679 snprintf(path
, PATH_MAX
,
680 DEFAULT_USTCONSUMERD32_PATH
,
682 DBG("Removing directory %s", path
);
685 /* ust consumerd 64 */
686 snprintf(path
, PATH_MAX
,
687 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
689 DBG("Removing %s", path
);
692 snprintf(path
, PATH_MAX
,
693 DEFAULT_USTCONSUMERD64_PATH
,
695 DBG("Removing directory %s", path
);
698 DBG("Cleaning up all sessions");
700 /* Destroy session list mutex */
701 if (session_list_ptr
!= NULL
) {
702 pthread_mutex_destroy(&session_list_ptr
->lock
);
704 /* Cleanup ALL session */
705 cds_list_for_each_entry_safe(sess
, stmp
,
706 &session_list_ptr
->head
, list
) {
707 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
711 wait_consumer(&kconsumer_data
);
712 wait_consumer(&ustconsumer64_data
);
713 wait_consumer(&ustconsumer32_data
);
715 DBG("Cleaning up all agent apps");
716 agent_app_ht_clean();
718 DBG("Closing all UST sockets");
719 ust_app_clean_list();
720 buffer_reg_destroy_registries();
722 if (is_root
&& !opt_no_kernel
) {
723 DBG2("Closing kernel fd");
724 if (kernel_tracer_fd
>= 0) {
725 ret
= close(kernel_tracer_fd
);
730 DBG("Unloading kernel modules");
731 modprobe_remove_lttng_all();
735 close_consumer_sockets();
738 load_session_destroy_data(load_info
);
743 * Cleanup lock file by deleting it and finaly closing it which will
744 * release the file system lock.
746 if (lockfile_fd
>= 0) {
747 char lockfile_path
[PATH_MAX
];
749 ret
= generate_lock_file_path(lockfile_path
,
750 sizeof(lockfile_path
));
752 ret
= remove(lockfile_path
);
754 PERROR("remove lock file");
756 ret
= close(lockfile_fd
);
758 PERROR("close lock file");
764 * We do NOT rmdir rundir because there are other processes
765 * using it, for instance lttng-relayd, which can start in
766 * parallel with this teardown.
773 * Cleanup the daemon's option data structures.
775 static void sessiond_cleanup_options(void)
777 DBG("Cleaning up options");
780 * If the override option is set, the pointer points to a *non* const
781 * thus freeing it even though the variable type is set to const.
783 if (tracing_group_name_override
) {
784 free((void *) tracing_group_name
);
786 if (consumerd32_bin_override
) {
787 free((void *) consumerd32_bin
);
789 if (consumerd64_bin_override
) {
790 free((void *) consumerd64_bin
);
792 if (consumerd32_libdir_override
) {
793 free((void *) consumerd32_libdir
);
795 if (consumerd64_libdir_override
) {
796 free((void *) consumerd64_libdir
);
800 free(opt_load_session_path
);
801 free(kmod_probes_list
);
802 free(kmod_extra_probes_list
);
805 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
806 "Matthew, BEET driven development works!%c[%dm",
807 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
812 * Send data on a unix socket using the liblttsessiondcomm API.
814 * Return lttcomm error code.
816 static int send_unix_sock(int sock
, void *buf
, size_t len
)
818 /* Check valid length */
823 return lttcomm_send_unix_sock(sock
, buf
, len
);
827 * Free memory of a command context structure.
829 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
831 DBG("Clean command context structure");
833 if ((*cmd_ctx
)->llm
) {
834 free((*cmd_ctx
)->llm
);
836 if ((*cmd_ctx
)->lsm
) {
837 free((*cmd_ctx
)->lsm
);
845 * Notify UST applications using the shm mmap futex.
847 static int notify_ust_apps(int active
)
851 DBG("Notifying applications of session daemon state: %d", active
);
853 /* See shm.c for this call implying mmap, shm and futex calls */
854 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
855 if (wait_shm_mmap
== NULL
) {
859 /* Wake waiting process */
860 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
862 /* Apps notified successfully */
870 * Setup the outgoing data buffer for the response (llm) by allocating the
871 * right amount of memory and copying the original information from the lsm
874 * Return total size of the buffer pointed by buf.
876 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
882 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
883 if (cmd_ctx
->llm
== NULL
) {
889 /* Copy common data */
890 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
891 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
893 cmd_ctx
->llm
->data_size
= size
;
894 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
903 * Update the kernel poll set of all channel fd available over all tracing
904 * session. Add the wakeup pipe at the end of the set.
906 static int update_kernel_poll(struct lttng_poll_event
*events
)
909 struct ltt_session
*session
;
910 struct ltt_kernel_channel
*channel
;
912 DBG("Updating kernel poll set");
915 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
916 session_lock(session
);
917 if (session
->kernel_session
== NULL
) {
918 session_unlock(session
);
922 cds_list_for_each_entry(channel
,
923 &session
->kernel_session
->channel_list
.head
, list
) {
924 /* Add channel fd to the kernel poll set */
925 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
927 session_unlock(session
);
930 DBG("Channel fd %d added to kernel set", channel
->fd
);
932 session_unlock(session
);
934 session_unlock_list();
939 session_unlock_list();
944 * Find the channel fd from 'fd' over all tracing session. When found, check
945 * for new channel stream and send those stream fds to the kernel consumer.
947 * Useful for CPU hotplug feature.
949 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
952 struct ltt_session
*session
;
953 struct ltt_kernel_session
*ksess
;
954 struct ltt_kernel_channel
*channel
;
956 DBG("Updating kernel streams for channel fd %d", fd
);
959 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
960 session_lock(session
);
961 if (session
->kernel_session
== NULL
) {
962 session_unlock(session
);
965 ksess
= session
->kernel_session
;
967 cds_list_for_each_entry(channel
,
968 &ksess
->channel_list
.head
, list
) {
969 struct lttng_ht_iter iter
;
970 struct consumer_socket
*socket
;
972 if (channel
->fd
!= fd
) {
975 DBG("Channel found, updating kernel streams");
976 ret
= kernel_open_channel_stream(channel
);
980 /* Update the stream global counter */
981 ksess
->stream_count_global
+= ret
;
984 * Have we already sent fds to the consumer? If yes, it
985 * means that tracing is started so it is safe to send
986 * our updated stream fds.
988 if (ksess
->consumer_fds_sent
!= 1
989 || ksess
->consumer
== NULL
) {
995 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
996 &iter
.iter
, socket
, node
.node
) {
997 pthread_mutex_lock(socket
->lock
);
998 ret
= kernel_consumer_send_channel_stream(socket
,
1000 session
->output_traces
? 1 : 0);
1001 pthread_mutex_unlock(socket
->lock
);
1009 session_unlock(session
);
1011 session_unlock_list();
1015 session_unlock(session
);
1016 session_unlock_list();
1021 * For each tracing session, update newly registered apps. The session list
1022 * lock MUST be acquired before calling this.
1024 static void update_ust_app(int app_sock
)
1026 struct ltt_session
*sess
, *stmp
;
1028 /* Consumer is in an ERROR state. Stop any application update. */
1029 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
1030 /* Stop the update process since the consumer is dead. */
1034 /* For all tracing session(s) */
1035 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
1036 struct ust_app
*app
;
1039 if (!sess
->ust_session
) {
1040 goto unlock_session
;
1044 assert(app_sock
>= 0);
1045 app
= ust_app_find_by_sock(app_sock
);
1048 * Application can be unregistered before so
1049 * this is possible hence simply stopping the
1052 DBG3("UST app update failed to find app sock %d",
1056 ust_app_global_update(sess
->ust_session
, app
);
1060 session_unlock(sess
);
1065 * This thread manage event coming from the kernel.
1067 * Features supported in this thread:
1070 static void *thread_manage_kernel(void *data
)
1072 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1073 uint32_t revents
, nb_fd
;
1075 struct lttng_poll_event events
;
1077 DBG("[thread] Thread manage kernel started");
1079 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1082 * This first step of the while is to clean this structure which could free
1083 * non NULL pointers so initialize it before the loop.
1085 lttng_poll_init(&events
);
1087 if (testpoint(sessiond_thread_manage_kernel
)) {
1088 goto error_testpoint
;
1091 health_code_update();
1093 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1094 goto error_testpoint
;
1098 health_code_update();
1100 if (update_poll_flag
== 1) {
1101 /* Clean events object. We are about to populate it again. */
1102 lttng_poll_clean(&events
);
1104 ret
= sessiond_set_thread_pollset(&events
, 2);
1106 goto error_poll_create
;
1109 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1114 /* This will add the available kernel channel if any. */
1115 ret
= update_kernel_poll(&events
);
1119 update_poll_flag
= 0;
1122 DBG("Thread kernel polling");
1124 /* Poll infinite value of time */
1126 health_poll_entry();
1127 ret
= lttng_poll_wait(&events
, -1);
1128 DBG("Thread kernel return from poll on %d fds",
1129 LTTNG_POLL_GETNB(&events
));
1133 * Restart interrupted system call.
1135 if (errno
== EINTR
) {
1139 } else if (ret
== 0) {
1140 /* Should not happen since timeout is infinite */
1141 ERR("Return value of poll is 0 with an infinite timeout.\n"
1142 "This should not have happened! Continuing...");
1148 for (i
= 0; i
< nb_fd
; i
++) {
1149 /* Fetch once the poll data */
1150 revents
= LTTNG_POLL_GETEV(&events
, i
);
1151 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1153 health_code_update();
1156 /* No activity for this FD (poll implementation). */
1160 /* Thread quit pipe has been closed. Killing thread. */
1161 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1167 /* Check for data on kernel pipe */
1168 if (revents
& LPOLLIN
) {
1169 if (pollfd
== kernel_poll_pipe
[0]) {
1170 (void) lttng_read(kernel_poll_pipe
[0],
1173 * Ret value is useless here, if this pipe gets any actions an
1174 * update is required anyway.
1176 update_poll_flag
= 1;
1180 * New CPU detected by the kernel. Adding kernel stream to
1181 * kernel session and updating the kernel consumer
1183 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1189 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1190 update_poll_flag
= 1;
1193 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1201 lttng_poll_clean(&events
);
1204 utils_close_pipe(kernel_poll_pipe
);
1205 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1208 ERR("Health error occurred in %s", __func__
);
1209 WARN("Kernel thread died unexpectedly. "
1210 "Kernel tracing can continue but CPU hotplug is disabled.");
1212 health_unregister(health_sessiond
);
1213 DBG("Kernel thread dying");
1218 * Signal pthread condition of the consumer data that the thread.
1220 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1222 pthread_mutex_lock(&data
->cond_mutex
);
1225 * The state is set before signaling. It can be any value, it's the waiter
1226 * job to correctly interpret this condition variable associated to the
1227 * consumer pthread_cond.
1229 * A value of 0 means that the corresponding thread of the consumer data
1230 * was not started. 1 indicates that the thread has started and is ready
1231 * for action. A negative value means that there was an error during the
1234 data
->consumer_thread_is_ready
= state
;
1235 (void) pthread_cond_signal(&data
->cond
);
1237 pthread_mutex_unlock(&data
->cond_mutex
);
1241 * This thread manage the consumer error sent back to the session daemon.
1243 static void *thread_manage_consumer(void *data
)
1245 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1246 uint32_t revents
, nb_fd
;
1247 enum lttcomm_return_code code
;
1248 struct lttng_poll_event events
;
1249 struct consumer_data
*consumer_data
= data
;
1251 DBG("[thread] Manage consumer started");
1253 rcu_register_thread();
1254 rcu_thread_online();
1256 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1258 health_code_update();
1261 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1262 * metadata_sock. Nothing more will be added to this poll set.
1264 ret
= sessiond_set_thread_pollset(&events
, 3);
1270 * The error socket here is already in a listening state which was done
1271 * just before spawning this thread to avoid a race between the consumer
1272 * daemon exec trying to connect and the listen() call.
1274 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1279 health_code_update();
1281 /* Infinite blocking call, waiting for transmission */
1283 health_poll_entry();
1285 if (testpoint(sessiond_thread_manage_consumer
)) {
1289 ret
= lttng_poll_wait(&events
, -1);
1293 * Restart interrupted system call.
1295 if (errno
== EINTR
) {
1303 for (i
= 0; i
< nb_fd
; i
++) {
1304 /* Fetch once the poll data */
1305 revents
= LTTNG_POLL_GETEV(&events
, i
);
1306 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1308 health_code_update();
1311 /* No activity for this FD (poll implementation). */
1315 /* Thread quit pipe has been closed. Killing thread. */
1316 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1322 /* Event on the registration socket */
1323 if (pollfd
== consumer_data
->err_sock
) {
1324 if (revents
& LPOLLIN
) {
1326 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1327 ERR("consumer err socket poll error");
1330 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1336 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1342 * Set the CLOEXEC flag. Return code is useless because either way, the
1345 (void) utils_set_fd_cloexec(sock
);
1347 health_code_update();
1349 DBG2("Receiving code from consumer err_sock");
1351 /* Getting status code from kconsumerd */
1352 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1353 sizeof(enum lttcomm_return_code
));
1358 health_code_update();
1359 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1360 /* Connect both socket, command and metadata. */
1361 consumer_data
->cmd_sock
=
1362 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1363 consumer_data
->metadata_fd
=
1364 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1365 if (consumer_data
->cmd_sock
< 0
1366 || consumer_data
->metadata_fd
< 0) {
1367 PERROR("consumer connect cmd socket");
1368 /* On error, signal condition and quit. */
1369 signal_consumer_condition(consumer_data
, -1);
1372 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1373 /* Create metadata socket lock. */
1374 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1375 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1376 PERROR("zmalloc pthread mutex");
1380 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1382 signal_consumer_condition(consumer_data
, 1);
1383 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1384 DBG("Consumer metadata socket ready (fd: %d)",
1385 consumer_data
->metadata_fd
);
1387 ERR("consumer error when waiting for SOCK_READY : %s",
1388 lttcomm_get_readable_code(-code
));
1392 /* Remove the consumerd error sock since we've established a connexion */
1393 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1398 /* Add new accepted error socket. */
1399 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1404 /* Add metadata socket that is successfully connected. */
1405 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1406 LPOLLIN
| LPOLLRDHUP
);
1411 health_code_update();
1413 /* Infinite blocking call, waiting for transmission */
1416 health_code_update();
1418 /* Exit the thread because the thread quit pipe has been triggered. */
1420 /* Not a health error. */
1425 health_poll_entry();
1426 ret
= lttng_poll_wait(&events
, -1);
1430 * Restart interrupted system call.
1432 if (errno
== EINTR
) {
1440 for (i
= 0; i
< nb_fd
; i
++) {
1441 /* Fetch once the poll data */
1442 revents
= LTTNG_POLL_GETEV(&events
, i
);
1443 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1445 health_code_update();
1448 /* No activity for this FD (poll implementation). */
1453 * Thread quit pipe has been triggered, flag that we should stop
1454 * but continue the current loop to handle potential data from
1457 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1459 if (pollfd
== sock
) {
1460 /* Event on the consumerd socket */
1461 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1462 && !(revents
& LPOLLIN
)) {
1463 ERR("consumer err socket second poll error");
1466 health_code_update();
1467 /* Wait for any kconsumerd error */
1468 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1469 sizeof(enum lttcomm_return_code
));
1471 ERR("consumer closed the command socket");
1475 ERR("consumer return code : %s",
1476 lttcomm_get_readable_code(-code
));
1479 } else if (pollfd
== consumer_data
->metadata_fd
) {
1480 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1481 && !(revents
& LPOLLIN
)) {
1482 ERR("consumer err metadata socket second poll error");
1485 /* UST metadata requests */
1486 ret
= ust_consumer_metadata_request(
1487 &consumer_data
->metadata_sock
);
1489 ERR("Handling metadata request");
1493 /* No need for an else branch all FDs are tested prior. */
1495 health_code_update();
1501 * We lock here because we are about to close the sockets and some other
1502 * thread might be using them so get exclusive access which will abort all
1503 * other consumer command by other threads.
1505 pthread_mutex_lock(&consumer_data
->lock
);
1507 /* Immediately set the consumerd state to stopped */
1508 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1509 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1510 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1511 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1512 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1514 /* Code flow error... */
1518 if (consumer_data
->err_sock
>= 0) {
1519 ret
= close(consumer_data
->err_sock
);
1523 consumer_data
->err_sock
= -1;
1525 if (consumer_data
->cmd_sock
>= 0) {
1526 ret
= close(consumer_data
->cmd_sock
);
1530 consumer_data
->cmd_sock
= -1;
1532 if (consumer_data
->metadata_sock
.fd_ptr
&&
1533 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1534 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1546 unlink(consumer_data
->err_unix_sock_path
);
1547 unlink(consumer_data
->cmd_unix_sock_path
);
1548 pthread_mutex_unlock(&consumer_data
->lock
);
1550 /* Cleanup metadata socket mutex. */
1551 if (consumer_data
->metadata_sock
.lock
) {
1552 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1553 free(consumer_data
->metadata_sock
.lock
);
1555 lttng_poll_clean(&events
);
1559 ERR("Health error occurred in %s", __func__
);
1561 health_unregister(health_sessiond
);
1562 DBG("consumer thread cleanup completed");
1564 rcu_thread_offline();
1565 rcu_unregister_thread();
1571 * This thread manage application communication.
1573 static void *thread_manage_apps(void *data
)
1575 int i
, ret
, pollfd
, err
= -1;
1577 uint32_t revents
, nb_fd
;
1578 struct lttng_poll_event events
;
1580 DBG("[thread] Manage application started");
1582 rcu_register_thread();
1583 rcu_thread_online();
1585 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1587 if (testpoint(sessiond_thread_manage_apps
)) {
1588 goto error_testpoint
;
1591 health_code_update();
1593 ret
= sessiond_set_thread_pollset(&events
, 2);
1595 goto error_poll_create
;
1598 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1603 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1607 health_code_update();
1610 DBG("Apps thread polling");
1612 /* Inifinite blocking call, waiting for transmission */
1614 health_poll_entry();
1615 ret
= lttng_poll_wait(&events
, -1);
1616 DBG("Apps thread return from poll on %d fds",
1617 LTTNG_POLL_GETNB(&events
));
1621 * Restart interrupted system call.
1623 if (errno
== EINTR
) {
1631 for (i
= 0; i
< nb_fd
; i
++) {
1632 /* Fetch once the poll data */
1633 revents
= LTTNG_POLL_GETEV(&events
, i
);
1634 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1636 health_code_update();
1639 /* No activity for this FD (poll implementation). */
1643 /* Thread quit pipe has been closed. Killing thread. */
1644 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1650 /* Inspect the apps cmd pipe */
1651 if (pollfd
== apps_cmd_pipe
[0]) {
1652 if (revents
& LPOLLIN
) {
1656 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1657 if (size_ret
< sizeof(sock
)) {
1658 PERROR("read apps cmd pipe");
1662 health_code_update();
1665 * Since this is a command socket (write then read),
1666 * we only monitor the error events of the socket.
1668 ret
= lttng_poll_add(&events
, sock
,
1669 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1674 DBG("Apps with sock %d added to poll set", sock
);
1675 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1676 ERR("Apps command pipe error");
1679 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1684 * At this point, we know that a registered application made
1685 * the event at poll_wait.
1687 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1688 /* Removing from the poll set */
1689 ret
= lttng_poll_del(&events
, pollfd
);
1694 /* Socket closed on remote end. */
1695 ust_app_unregister(pollfd
);
1697 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1702 health_code_update();
1708 lttng_poll_clean(&events
);
1711 utils_close_pipe(apps_cmd_pipe
);
1712 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1715 * We don't clean the UST app hash table here since already registered
1716 * applications can still be controlled so let them be until the session
1717 * daemon dies or the applications stop.
1722 ERR("Health error occurred in %s", __func__
);
1724 health_unregister(health_sessiond
);
1725 DBG("Application communication apps thread cleanup complete");
1726 rcu_thread_offline();
1727 rcu_unregister_thread();
1732 * Send a socket to a thread This is called from the dispatch UST registration
1733 * thread once all sockets are set for the application.
1735 * The sock value can be invalid, we don't really care, the thread will handle
1736 * it and make the necessary cleanup if so.
1738 * On success, return 0 else a negative value being the errno message of the
1741 static int send_socket_to_thread(int fd
, int sock
)
1746 * It's possible that the FD is set as invalid with -1 concurrently just
1747 * before calling this function being a shutdown state of the thread.
1754 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1755 if (ret
< sizeof(sock
)) {
1756 PERROR("write apps pipe %d", fd
);
1763 /* All good. Don't send back the write positive ret value. */
1770 * Sanitize the wait queue of the dispatch registration thread meaning removing
1771 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1772 * notify socket is never received.
1774 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1776 int ret
, nb_fd
= 0, i
;
1777 unsigned int fd_added
= 0;
1778 struct lttng_poll_event events
;
1779 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1783 lttng_poll_init(&events
);
1785 /* Just skip everything for an empty queue. */
1786 if (!wait_queue
->count
) {
1790 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1795 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1796 &wait_queue
->head
, head
) {
1797 assert(wait_node
->app
);
1798 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1799 LPOLLHUP
| LPOLLERR
);
1812 * Poll but don't block so we can quickly identify the faulty events and
1813 * clean them afterwards from the wait queue.
1815 ret
= lttng_poll_wait(&events
, 0);
1821 for (i
= 0; i
< nb_fd
; i
++) {
1822 /* Get faulty FD. */
1823 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1824 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1827 /* No activity for this FD (poll implementation). */
1831 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1832 &wait_queue
->head
, head
) {
1833 if (pollfd
== wait_node
->app
->sock
&&
1834 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1835 cds_list_del(&wait_node
->head
);
1836 wait_queue
->count
--;
1837 ust_app_destroy(wait_node
->app
);
1840 * Silence warning of use-after-free in
1841 * cds_list_for_each_entry_safe which uses
1842 * __typeof__(*wait_node).
1847 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1854 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1858 lttng_poll_clean(&events
);
1862 lttng_poll_clean(&events
);
1864 ERR("Unable to sanitize wait queue");
1869 * Dispatch request from the registration threads to the application
1870 * communication thread.
1872 static void *thread_dispatch_ust_registration(void *data
)
1875 struct cds_wfcq_node
*node
;
1876 struct ust_command
*ust_cmd
= NULL
;
1877 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1878 struct ust_reg_wait_queue wait_queue
= {
1882 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1884 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1885 goto error_testpoint
;
1888 health_code_update();
1890 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1892 DBG("[thread] Dispatch UST command started");
1894 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1895 health_code_update();
1897 /* Atomically prepare the queue futex */
1898 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1901 struct ust_app
*app
= NULL
;
1905 * Make sure we don't have node(s) that have hung up before receiving
1906 * the notify socket. This is to clean the list in order to avoid
1907 * memory leaks from notify socket that are never seen.
1909 sanitize_wait_queue(&wait_queue
);
1911 health_code_update();
1912 /* Dequeue command for registration */
1913 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1915 DBG("Woken up but nothing in the UST command queue");
1916 /* Continue thread execution */
1920 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1922 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1923 " gid:%d sock:%d name:%s (version %d.%d)",
1924 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1925 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1926 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1927 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1929 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1930 wait_node
= zmalloc(sizeof(*wait_node
));
1932 PERROR("zmalloc wait_node dispatch");
1933 ret
= close(ust_cmd
->sock
);
1935 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1937 lttng_fd_put(LTTNG_FD_APPS
, 1);
1941 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1943 /* Create application object if socket is CMD. */
1944 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1946 if (!wait_node
->app
) {
1947 ret
= close(ust_cmd
->sock
);
1949 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1951 lttng_fd_put(LTTNG_FD_APPS
, 1);
1957 * Add application to the wait queue so we can set the notify
1958 * socket before putting this object in the global ht.
1960 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1965 * We have to continue here since we don't have the notify
1966 * socket and the application MUST be added to the hash table
1967 * only at that moment.
1972 * Look for the application in the local wait queue and set the
1973 * notify socket if found.
1975 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1976 &wait_queue
.head
, head
) {
1977 health_code_update();
1978 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1979 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1980 cds_list_del(&wait_node
->head
);
1982 app
= wait_node
->app
;
1984 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1990 * With no application at this stage the received socket is
1991 * basically useless so close it before we free the cmd data
1992 * structure for good.
1995 ret
= close(ust_cmd
->sock
);
1997 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1999 lttng_fd_put(LTTNG_FD_APPS
, 1);
2006 * @session_lock_list
2008 * Lock the global session list so from the register up to the
2009 * registration done message, no thread can see the application
2010 * and change its state.
2012 session_lock_list();
2016 * Add application to the global hash table. This needs to be
2017 * done before the update to the UST registry can locate the
2022 /* Set app version. This call will print an error if needed. */
2023 (void) ust_app_version(app
);
2025 /* Send notify socket through the notify pipe. */
2026 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
2030 session_unlock_list();
2032 * No notify thread, stop the UST tracing. However, this is
2033 * not an internal error of the this thread thus setting
2034 * the health error code to a normal exit.
2041 * Update newly registered application with the tracing
2042 * registry info already enabled information.
2044 update_ust_app(app
->sock
);
2047 * Don't care about return value. Let the manage apps threads
2048 * handle app unregistration upon socket close.
2050 (void) ust_app_register_done(app
);
2053 * Even if the application socket has been closed, send the app
2054 * to the thread and unregistration will take place at that
2057 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2060 session_unlock_list();
2062 * No apps. thread, stop the UST tracing. However, this is
2063 * not an internal error of the this thread thus setting
2064 * the health error code to a normal exit.
2071 session_unlock_list();
2073 } while (node
!= NULL
);
2075 health_poll_entry();
2076 /* Futex wait on queue. Blocking call on futex() */
2077 futex_nto1_wait(&ust_cmd_queue
.futex
);
2080 /* Normal exit, no error */
2084 /* Clean up wait queue. */
2085 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2086 &wait_queue
.head
, head
) {
2087 cds_list_del(&wait_node
->head
);
2092 /* Empty command queue. */
2094 /* Dequeue command for registration */
2095 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2099 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2100 ret
= close(ust_cmd
->sock
);
2102 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2104 lttng_fd_put(LTTNG_FD_APPS
, 1);
2109 DBG("Dispatch thread dying");
2112 ERR("Health error occurred in %s", __func__
);
2114 health_unregister(health_sessiond
);
2119 * This thread manage application registration.
2121 static void *thread_registration_apps(void *data
)
2123 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2124 uint32_t revents
, nb_fd
;
2125 struct lttng_poll_event events
;
2127 * Get allocated in this thread, enqueued to a global queue, dequeued and
2128 * freed in the manage apps thread.
2130 struct ust_command
*ust_cmd
= NULL
;
2132 DBG("[thread] Manage application registration started");
2134 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2136 if (testpoint(sessiond_thread_registration_apps
)) {
2137 goto error_testpoint
;
2140 ret
= lttcomm_listen_unix_sock(apps_sock
);
2146 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2147 * more will be added to this poll set.
2149 ret
= sessiond_set_thread_pollset(&events
, 2);
2151 goto error_create_poll
;
2154 /* Add the application registration socket */
2155 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2157 goto error_poll_add
;
2160 /* Notify all applications to register */
2161 ret
= notify_ust_apps(1);
2163 ERR("Failed to notify applications or create the wait shared memory.\n"
2164 "Execution continues but there might be problem for already\n"
2165 "running applications that wishes to register.");
2169 DBG("Accepting application registration");
2171 /* Inifinite blocking call, waiting for transmission */
2173 health_poll_entry();
2174 ret
= lttng_poll_wait(&events
, -1);
2178 * Restart interrupted system call.
2180 if (errno
== EINTR
) {
2188 for (i
= 0; i
< nb_fd
; i
++) {
2189 health_code_update();
2191 /* Fetch once the poll data */
2192 revents
= LTTNG_POLL_GETEV(&events
, i
);
2193 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2196 /* No activity for this FD (poll implementation). */
2200 /* Thread quit pipe has been closed. Killing thread. */
2201 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2207 /* Event on the registration socket */
2208 if (pollfd
== apps_sock
) {
2209 if (revents
& LPOLLIN
) {
2210 sock
= lttcomm_accept_unix_sock(apps_sock
);
2216 * Set socket timeout for both receiving and ending.
2217 * app_socket_timeout is in seconds, whereas
2218 * lttcomm_setsockopt_rcv_timeout and
2219 * lttcomm_setsockopt_snd_timeout expect msec as
2222 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2223 app_socket_timeout
* 1000);
2224 (void) lttcomm_setsockopt_snd_timeout(sock
,
2225 app_socket_timeout
* 1000);
2228 * Set the CLOEXEC flag. Return code is useless because
2229 * either way, the show must go on.
2231 (void) utils_set_fd_cloexec(sock
);
2233 /* Create UST registration command for enqueuing */
2234 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2235 if (ust_cmd
== NULL
) {
2236 PERROR("ust command zmalloc");
2245 * Using message-based transmissions to ensure we don't
2246 * have to deal with partially received messages.
2248 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2250 ERR("Exhausted file descriptors allowed for applications.");
2260 health_code_update();
2261 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2264 /* Close socket of the application. */
2269 lttng_fd_put(LTTNG_FD_APPS
, 1);
2273 health_code_update();
2275 ust_cmd
->sock
= sock
;
2278 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2279 " gid:%d sock:%d name:%s (version %d.%d)",
2280 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2281 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2282 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2283 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2286 * Lock free enqueue the registration request. The red pill
2287 * has been taken! This apps will be part of the *system*.
2289 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2292 * Wake the registration queue futex. Implicit memory
2293 * barrier with the exchange in cds_wfcq_enqueue.
2295 futex_nto1_wake(&ust_cmd_queue
.futex
);
2296 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2297 ERR("Register apps socket poll error");
2300 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2309 /* Notify that the registration thread is gone */
2312 if (apps_sock
>= 0) {
2313 ret
= close(apps_sock
);
2323 lttng_fd_put(LTTNG_FD_APPS
, 1);
2325 unlink(apps_unix_sock_path
);
2328 lttng_poll_clean(&events
);
2332 DBG("UST Registration thread cleanup complete");
2335 ERR("Health error occurred in %s", __func__
);
2337 health_unregister(health_sessiond
);
2343 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2344 * exec or it will fails.
2346 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2349 struct timespec timeout
;
2351 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2352 consumer_data
->consumer_thread_is_ready
= 0;
2354 /* Setup pthread condition */
2355 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2358 PERROR("pthread_condattr_init consumer data");
2363 * Set the monotonic clock in order to make sure we DO NOT jump in time
2364 * between the clock_gettime() call and the timedwait call. See bug #324
2365 * for a more details and how we noticed it.
2367 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2370 PERROR("pthread_condattr_setclock consumer data");
2374 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2377 PERROR("pthread_cond_init consumer data");
2381 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2385 PERROR("pthread_create consumer");
2390 /* We are about to wait on a pthread condition */
2391 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2393 /* Get time for sem_timedwait absolute timeout */
2394 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2396 * Set the timeout for the condition timed wait even if the clock gettime
2397 * call fails since we might loop on that call and we want to avoid to
2398 * increment the timeout too many times.
2400 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2403 * The following loop COULD be skipped in some conditions so this is why we
2404 * set ret to 0 in order to make sure at least one round of the loop is
2410 * Loop until the condition is reached or when a timeout is reached. Note
2411 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2412 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2413 * possible. This loop does not take any chances and works with both of
2416 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2417 if (clock_ret
< 0) {
2418 PERROR("clock_gettime spawn consumer");
2419 /* Infinite wait for the consumerd thread to be ready */
2420 ret
= pthread_cond_wait(&consumer_data
->cond
,
2421 &consumer_data
->cond_mutex
);
2423 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2424 &consumer_data
->cond_mutex
, &timeout
);
2428 /* Release the pthread condition */
2429 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2433 if (ret
== ETIMEDOUT
) {
2437 * Call has timed out so we kill the kconsumerd_thread and return
2440 ERR("Condition timed out. The consumer thread was never ready."
2442 pth_ret
= pthread_cancel(consumer_data
->thread
);
2444 PERROR("pthread_cancel consumer thread");
2447 PERROR("pthread_cond_wait failed consumer thread");
2449 /* Caller is expecting a negative value on failure. */
2454 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2455 if (consumer_data
->pid
== 0) {
2456 ERR("Consumerd did not start");
2457 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2460 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2469 * Join consumer thread
2471 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2475 /* Consumer pid must be a real one. */
2476 if (consumer_data
->pid
> 0) {
2478 ret
= kill(consumer_data
->pid
, SIGTERM
);
2480 PERROR("Error killing consumer daemon");
2483 return pthread_join(consumer_data
->thread
, &status
);
2490 * Fork and exec a consumer daemon (consumerd).
2492 * Return pid if successful else -1.
2494 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2498 const char *consumer_to_use
;
2499 const char *verbosity
;
2502 DBG("Spawning consumerd");
2509 if (opt_verbose_consumer
) {
2510 verbosity
= "--verbose";
2511 } else if (lttng_opt_quiet
) {
2512 verbosity
= "--quiet";
2517 switch (consumer_data
->type
) {
2518 case LTTNG_CONSUMER_KERNEL
:
2520 * Find out which consumerd to execute. We will first try the
2521 * 64-bit path, then the sessiond's installation directory, and
2522 * fallback on the 32-bit one,
2524 DBG3("Looking for a kernel consumer at these locations:");
2525 DBG3(" 1) %s", consumerd64_bin
);
2526 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2527 DBG3(" 3) %s", consumerd32_bin
);
2528 if (stat(consumerd64_bin
, &st
) == 0) {
2529 DBG3("Found location #1");
2530 consumer_to_use
= consumerd64_bin
;
2531 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2532 DBG3("Found location #2");
2533 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2534 } else if (stat(consumerd32_bin
, &st
) == 0) {
2535 DBG3("Found location #3");
2536 consumer_to_use
= consumerd32_bin
;
2538 DBG("Could not find any valid consumerd executable");
2542 DBG("Using kernel consumer at: %s", consumer_to_use
);
2543 ret
= execl(consumer_to_use
,
2544 "lttng-consumerd", verbosity
, "-k",
2545 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2546 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2547 "--group", tracing_group_name
,
2550 case LTTNG_CONSUMER64_UST
:
2552 char *tmpnew
= NULL
;
2554 if (consumerd64_libdir
[0] != '\0') {
2558 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2562 tmplen
= strlen("LD_LIBRARY_PATH=")
2563 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2564 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2569 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2570 strcat(tmpnew
, consumerd64_libdir
);
2571 if (tmp
[0] != '\0') {
2572 strcat(tmpnew
, ":");
2573 strcat(tmpnew
, tmp
);
2575 ret
= putenv(tmpnew
);
2582 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2583 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2584 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2585 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2586 "--group", tracing_group_name
,
2588 if (consumerd64_libdir
[0] != '\0') {
2593 case LTTNG_CONSUMER32_UST
:
2595 char *tmpnew
= NULL
;
2597 if (consumerd32_libdir
[0] != '\0') {
2601 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2605 tmplen
= strlen("LD_LIBRARY_PATH=")
2606 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2607 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2612 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2613 strcat(tmpnew
, consumerd32_libdir
);
2614 if (tmp
[0] != '\0') {
2615 strcat(tmpnew
, ":");
2616 strcat(tmpnew
, tmp
);
2618 ret
= putenv(tmpnew
);
2625 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2626 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2627 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2628 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2629 "--group", tracing_group_name
,
2631 if (consumerd32_libdir
[0] != '\0') {
2637 PERROR("unknown consumer type");
2641 PERROR("Consumer execl()");
2643 /* Reaching this point, we got a failure on our execl(). */
2645 } else if (pid
> 0) {
2648 PERROR("start consumer fork");
2656 * Spawn the consumerd daemon and session daemon thread.
2658 static int start_consumerd(struct consumer_data
*consumer_data
)
2663 * Set the listen() state on the socket since there is a possible race
2664 * between the exec() of the consumer daemon and this call if place in the
2665 * consumer thread. See bug #366 for more details.
2667 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2672 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2673 if (consumer_data
->pid
!= 0) {
2674 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2678 ret
= spawn_consumerd(consumer_data
);
2680 ERR("Spawning consumerd failed");
2681 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2685 /* Setting up the consumer_data pid */
2686 consumer_data
->pid
= ret
;
2687 DBG2("Consumer pid %d", consumer_data
->pid
);
2688 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2690 DBG2("Spawning consumer control thread");
2691 ret
= spawn_consumer_thread(consumer_data
);
2693 ERR("Fatal error spawning consumer control thread");
2701 /* Cleanup already created sockets on error. */
2702 if (consumer_data
->err_sock
>= 0) {
2705 err
= close(consumer_data
->err_sock
);
2707 PERROR("close consumer data error socket");
2714 * Setup necessary data for kernel tracer action.
2716 static int init_kernel_tracer(void)
2720 /* Modprobe lttng kernel modules */
2721 ret
= modprobe_lttng_control();
2726 /* Open debugfs lttng */
2727 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2728 if (kernel_tracer_fd
< 0) {
2729 DBG("Failed to open %s", module_proc_lttng
);
2734 /* Validate kernel version */
2735 ret
= kernel_validate_version(kernel_tracer_fd
);
2740 ret
= modprobe_lttng_data();
2745 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2749 modprobe_remove_lttng_control();
2750 ret
= close(kernel_tracer_fd
);
2754 kernel_tracer_fd
= -1;
2755 return LTTNG_ERR_KERN_VERSION
;
2758 ret
= close(kernel_tracer_fd
);
2764 modprobe_remove_lttng_control();
2767 WARN("No kernel tracer available");
2768 kernel_tracer_fd
= -1;
2770 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2772 return LTTNG_ERR_KERN_NA
;
2778 * Copy consumer output from the tracing session to the domain session. The
2779 * function also applies the right modification on a per domain basis for the
2780 * trace files destination directory.
2782 * Should *NOT* be called with RCU read-side lock held.
2784 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2787 const char *dir_name
;
2788 struct consumer_output
*consumer
;
2791 assert(session
->consumer
);
2794 case LTTNG_DOMAIN_KERNEL
:
2795 DBG3("Copying tracing session consumer output in kernel session");
2797 * XXX: We should audit the session creation and what this function
2798 * does "extra" in order to avoid a destroy since this function is used
2799 * in the domain session creation (kernel and ust) only. Same for UST
2802 if (session
->kernel_session
->consumer
) {
2803 consumer_output_put(session
->kernel_session
->consumer
);
2805 session
->kernel_session
->consumer
=
2806 consumer_copy_output(session
->consumer
);
2807 /* Ease our life a bit for the next part */
2808 consumer
= session
->kernel_session
->consumer
;
2809 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2811 case LTTNG_DOMAIN_JUL
:
2812 case LTTNG_DOMAIN_LOG4J
:
2813 case LTTNG_DOMAIN_PYTHON
:
2814 case LTTNG_DOMAIN_UST
:
2815 DBG3("Copying tracing session consumer output in UST session");
2816 if (session
->ust_session
->consumer
) {
2817 consumer_output_put(session
->ust_session
->consumer
);
2819 session
->ust_session
->consumer
=
2820 consumer_copy_output(session
->consumer
);
2821 /* Ease our life a bit for the next part */
2822 consumer
= session
->ust_session
->consumer
;
2823 dir_name
= DEFAULT_UST_TRACE_DIR
;
2826 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2830 /* Append correct directory to subdir */
2831 strncat(consumer
->subdir
, dir_name
,
2832 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2833 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2842 * Create an UST session and add it to the session ust list.
2844 * Should *NOT* be called with RCU read-side lock held.
2846 static int create_ust_session(struct ltt_session
*session
,
2847 struct lttng_domain
*domain
)
2850 struct ltt_ust_session
*lus
= NULL
;
2854 assert(session
->consumer
);
2856 switch (domain
->type
) {
2857 case LTTNG_DOMAIN_JUL
:
2858 case LTTNG_DOMAIN_LOG4J
:
2859 case LTTNG_DOMAIN_PYTHON
:
2860 case LTTNG_DOMAIN_UST
:
2863 ERR("Unknown UST domain on create session %d", domain
->type
);
2864 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2868 DBG("Creating UST session");
2870 lus
= trace_ust_create_session(session
->id
);
2872 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2876 lus
->uid
= session
->uid
;
2877 lus
->gid
= session
->gid
;
2878 lus
->output_traces
= session
->output_traces
;
2879 lus
->snapshot_mode
= session
->snapshot_mode
;
2880 lus
->live_timer_interval
= session
->live_timer
;
2881 session
->ust_session
= lus
;
2882 if (session
->shm_path
[0]) {
2883 strncpy(lus
->root_shm_path
, session
->shm_path
,
2884 sizeof(lus
->root_shm_path
));
2885 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2886 strncpy(lus
->shm_path
, session
->shm_path
,
2887 sizeof(lus
->shm_path
));
2888 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2889 strncat(lus
->shm_path
, "/ust",
2890 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2892 /* Copy session output to the newly created UST session */
2893 ret
= copy_session_consumer(domain
->type
, session
);
2894 if (ret
!= LTTNG_OK
) {
2902 session
->ust_session
= NULL
;
2907 * Create a kernel tracer session then create the default channel.
2909 static int create_kernel_session(struct ltt_session
*session
)
2913 DBG("Creating kernel session");
2915 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2917 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2921 /* Code flow safety */
2922 assert(session
->kernel_session
);
2924 /* Copy session output to the newly created Kernel session */
2925 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2926 if (ret
!= LTTNG_OK
) {
2930 /* Create directory(ies) on local filesystem. */
2931 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2932 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2933 ret
= run_as_mkdir_recursive(
2934 session
->kernel_session
->consumer
->dst
.trace_path
,
2935 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2937 if (errno
!= EEXIST
) {
2938 ERR("Trace directory creation error");
2944 session
->kernel_session
->uid
= session
->uid
;
2945 session
->kernel_session
->gid
= session
->gid
;
2946 session
->kernel_session
->output_traces
= session
->output_traces
;
2947 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2952 trace_kernel_destroy_session(session
->kernel_session
);
2953 session
->kernel_session
= NULL
;
2958 * Count number of session permitted by uid/gid.
2960 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2963 struct ltt_session
*session
;
2965 DBG("Counting number of available session for UID %d GID %d",
2967 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2969 * Only list the sessions the user can control.
2971 if (!session_access_ok(session
, uid
, gid
)) {
2980 * Process the command requested by the lttng client within the command
2981 * context structure. This function make sure that the return structure (llm)
2982 * is set and ready for transmission before returning.
2984 * Return any error encountered or 0 for success.
2986 * "sock" is only used for special-case var. len data.
2988 * Should *NOT* be called with RCU read-side lock held.
2990 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2994 int need_tracing_session
= 1;
2997 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2999 assert(!rcu_read_ongoing());
3003 switch (cmd_ctx
->lsm
->cmd_type
) {
3004 case LTTNG_CREATE_SESSION
:
3005 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3006 case LTTNG_CREATE_SESSION_LIVE
:
3007 case LTTNG_DESTROY_SESSION
:
3008 case LTTNG_LIST_SESSIONS
:
3009 case LTTNG_LIST_DOMAINS
:
3010 case LTTNG_START_TRACE
:
3011 case LTTNG_STOP_TRACE
:
3012 case LTTNG_DATA_PENDING
:
3013 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3014 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3015 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3016 case LTTNG_SNAPSHOT_RECORD
:
3017 case LTTNG_SAVE_SESSION
:
3018 case LTTNG_SET_SESSION_SHM_PATH
:
3025 if (opt_no_kernel
&& need_domain
3026 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3028 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3030 ret
= LTTNG_ERR_KERN_NA
;
3035 /* Deny register consumer if we already have a spawned consumer. */
3036 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3037 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3038 if (kconsumer_data
.pid
> 0) {
3039 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3040 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3043 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3047 * Check for command that don't needs to allocate a returned payload. We do
3048 * this here so we don't have to make the call for no payload at each
3051 switch(cmd_ctx
->lsm
->cmd_type
) {
3052 case LTTNG_LIST_SESSIONS
:
3053 case LTTNG_LIST_TRACEPOINTS
:
3054 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3055 case LTTNG_LIST_DOMAINS
:
3056 case LTTNG_LIST_CHANNELS
:
3057 case LTTNG_LIST_EVENTS
:
3058 case LTTNG_LIST_SYSCALLS
:
3059 case LTTNG_LIST_TRACKER_PIDS
:
3062 /* Setup lttng message with no payload */
3063 ret
= setup_lttng_msg(cmd_ctx
, 0);
3065 /* This label does not try to unlock the session */
3066 goto init_setup_error
;
3070 /* Commands that DO NOT need a session. */
3071 switch (cmd_ctx
->lsm
->cmd_type
) {
3072 case LTTNG_CREATE_SESSION
:
3073 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3074 case LTTNG_CREATE_SESSION_LIVE
:
3075 case LTTNG_CALIBRATE
:
3076 case LTTNG_LIST_SESSIONS
:
3077 case LTTNG_LIST_TRACEPOINTS
:
3078 case LTTNG_LIST_SYSCALLS
:
3079 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3080 case LTTNG_SAVE_SESSION
:
3081 need_tracing_session
= 0;
3084 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3086 * We keep the session list lock across _all_ commands
3087 * for now, because the per-session lock does not
3088 * handle teardown properly.
3090 session_lock_list();
3091 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3092 if (cmd_ctx
->session
== NULL
) {
3093 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3096 /* Acquire lock for the session */
3097 session_lock(cmd_ctx
->session
);
3103 * Commands that need a valid session but should NOT create one if none
3104 * exists. Instead of creating one and destroying it when the command is
3105 * handled, process that right before so we save some round trip in useless
3108 switch (cmd_ctx
->lsm
->cmd_type
) {
3109 case LTTNG_DISABLE_CHANNEL
:
3110 case LTTNG_DISABLE_EVENT
:
3111 switch (cmd_ctx
->lsm
->domain
.type
) {
3112 case LTTNG_DOMAIN_KERNEL
:
3113 if (!cmd_ctx
->session
->kernel_session
) {
3114 ret
= LTTNG_ERR_NO_CHANNEL
;
3118 case LTTNG_DOMAIN_JUL
:
3119 case LTTNG_DOMAIN_LOG4J
:
3120 case LTTNG_DOMAIN_PYTHON
:
3121 case LTTNG_DOMAIN_UST
:
3122 if (!cmd_ctx
->session
->ust_session
) {
3123 ret
= LTTNG_ERR_NO_CHANNEL
;
3128 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3140 * Check domain type for specific "pre-action".
3142 switch (cmd_ctx
->lsm
->domain
.type
) {
3143 case LTTNG_DOMAIN_KERNEL
:
3145 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3149 /* Kernel tracer check */
3150 if (kernel_tracer_fd
== -1) {
3151 /* Basically, load kernel tracer modules */
3152 ret
= init_kernel_tracer();
3158 /* Consumer is in an ERROR state. Report back to client */
3159 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3160 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3164 /* Need a session for kernel command */
3165 if (need_tracing_session
) {
3166 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3167 ret
= create_kernel_session(cmd_ctx
->session
);
3169 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3174 /* Start the kernel consumer daemon */
3175 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3176 if (kconsumer_data
.pid
== 0 &&
3177 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3178 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3179 ret
= start_consumerd(&kconsumer_data
);
3181 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3184 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3186 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3190 * The consumer was just spawned so we need to add the socket to
3191 * the consumer output of the session if exist.
3193 ret
= consumer_create_socket(&kconsumer_data
,
3194 cmd_ctx
->session
->kernel_session
->consumer
);
3201 case LTTNG_DOMAIN_JUL
:
3202 case LTTNG_DOMAIN_LOG4J
:
3203 case LTTNG_DOMAIN_PYTHON
:
3204 case LTTNG_DOMAIN_UST
:
3206 if (!ust_app_supported()) {
3207 ret
= LTTNG_ERR_NO_UST
;
3210 /* Consumer is in an ERROR state. Report back to client */
3211 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3212 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3216 if (need_tracing_session
) {
3217 /* Create UST session if none exist. */
3218 if (cmd_ctx
->session
->ust_session
== NULL
) {
3219 ret
= create_ust_session(cmd_ctx
->session
,
3220 &cmd_ctx
->lsm
->domain
);
3221 if (ret
!= LTTNG_OK
) {
3226 /* Start the UST consumer daemons */
3228 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3229 if (consumerd64_bin
[0] != '\0' &&
3230 ustconsumer64_data
.pid
== 0 &&
3231 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3232 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3233 ret
= start_consumerd(&ustconsumer64_data
);
3235 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3236 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3240 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3241 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3243 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3247 * Setup socket for consumer 64 bit. No need for atomic access
3248 * since it was set above and can ONLY be set in this thread.
3250 ret
= consumer_create_socket(&ustconsumer64_data
,
3251 cmd_ctx
->session
->ust_session
->consumer
);
3257 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3258 if (consumerd32_bin
[0] != '\0' &&
3259 ustconsumer32_data
.pid
== 0 &&
3260 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3261 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3262 ret
= start_consumerd(&ustconsumer32_data
);
3264 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3265 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3269 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3270 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3272 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3276 * Setup socket for consumer 64 bit. No need for atomic access
3277 * since it was set above and can ONLY be set in this thread.
3279 ret
= consumer_create_socket(&ustconsumer32_data
,
3280 cmd_ctx
->session
->ust_session
->consumer
);
3292 /* Validate consumer daemon state when start/stop trace command */
3293 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3294 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3295 switch (cmd_ctx
->lsm
->domain
.type
) {
3296 case LTTNG_DOMAIN_JUL
:
3297 case LTTNG_DOMAIN_LOG4J
:
3298 case LTTNG_DOMAIN_PYTHON
:
3299 case LTTNG_DOMAIN_UST
:
3300 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3301 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3305 case LTTNG_DOMAIN_KERNEL
:
3306 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3307 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3315 * Check that the UID or GID match that of the tracing session.
3316 * The root user can interact with all sessions.
3318 if (need_tracing_session
) {
3319 if (!session_access_ok(cmd_ctx
->session
,
3320 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3321 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3322 ret
= LTTNG_ERR_EPERM
;
3328 * Send relayd information to consumer as soon as we have a domain and a
3331 if (cmd_ctx
->session
&& need_domain
) {
3333 * Setup relayd if not done yet. If the relayd information was already
3334 * sent to the consumer, this call will gracefully return.
3336 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3337 if (ret
!= LTTNG_OK
) {
3342 /* Process by command type */
3343 switch (cmd_ctx
->lsm
->cmd_type
) {
3344 case LTTNG_ADD_CONTEXT
:
3346 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3347 cmd_ctx
->lsm
->u
.context
.channel_name
,
3348 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3351 case LTTNG_DISABLE_CHANNEL
:
3353 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3354 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3357 case LTTNG_DISABLE_EVENT
:
3361 * FIXME: handle filter; for now we just receive the filter's
3362 * bytecode along with the filter expression which are sent by
3363 * liblttng-ctl and discard them.
3365 * This fixes an issue where the client may block while sending
3366 * the filter payload and encounter an error because the session
3367 * daemon closes the socket without ever handling this data.
3369 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3370 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3373 char data
[LTTNG_FILTER_MAX_LEN
];
3375 DBG("Discarding disable event command payload of size %zu", count
);
3377 ret
= lttcomm_recv_unix_sock(sock
, data
,
3378 count
> sizeof(data
) ? sizeof(data
) : count
);
3383 count
-= (size_t) ret
;
3386 /* FIXME: passing packed structure to non-packed pointer */
3387 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3388 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3389 &cmd_ctx
->lsm
->u
.disable
.event
);
3392 case LTTNG_ENABLE_CHANNEL
:
3394 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3395 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3398 case LTTNG_TRACK_PID
:
3400 ret
= cmd_track_pid(cmd_ctx
->session
,
3401 cmd_ctx
->lsm
->domain
.type
,
3402 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3405 case LTTNG_UNTRACK_PID
:
3407 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3408 cmd_ctx
->lsm
->domain
.type
,
3409 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3412 case LTTNG_ENABLE_EVENT
:
3414 struct lttng_event_exclusion
*exclusion
= NULL
;
3415 struct lttng_filter_bytecode
*bytecode
= NULL
;
3416 char *filter_expression
= NULL
;
3418 /* Handle exclusion events and receive it from the client. */
3419 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3420 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3422 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3423 (count
* LTTNG_SYMBOL_NAME_LEN
));
3425 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3429 DBG("Receiving var len exclusion event list from client ...");
3430 exclusion
->count
= count
;
3431 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3432 count
* LTTNG_SYMBOL_NAME_LEN
);
3434 DBG("Nothing recv() from client var len data... continuing");
3437 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3442 /* Get filter expression from client. */
3443 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3444 size_t expression_len
=
3445 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3447 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3448 ret
= LTTNG_ERR_FILTER_INVAL
;
3453 filter_expression
= zmalloc(expression_len
);
3454 if (!filter_expression
) {
3456 ret
= LTTNG_ERR_FILTER_NOMEM
;
3460 /* Receive var. len. data */
3461 DBG("Receiving var len filter's expression from client ...");
3462 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3465 DBG("Nothing recv() from client car len data... continuing");
3467 free(filter_expression
);
3469 ret
= LTTNG_ERR_FILTER_INVAL
;
3474 /* Handle filter and get bytecode from client. */
3475 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3476 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3478 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3479 ret
= LTTNG_ERR_FILTER_INVAL
;
3480 free(filter_expression
);
3485 bytecode
= zmalloc(bytecode_len
);
3487 free(filter_expression
);
3489 ret
= LTTNG_ERR_FILTER_NOMEM
;
3493 /* Receive var. len. data */
3494 DBG("Receiving var len filter's bytecode from client ...");
3495 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3497 DBG("Nothing recv() from client car len data... continuing");
3499 free(filter_expression
);
3502 ret
= LTTNG_ERR_FILTER_INVAL
;
3506 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3507 free(filter_expression
);
3510 ret
= LTTNG_ERR_FILTER_INVAL
;
3515 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3516 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3517 &cmd_ctx
->lsm
->u
.enable
.event
,
3518 filter_expression
, bytecode
, exclusion
,
3519 kernel_poll_pipe
[1]);
3522 case LTTNG_LIST_TRACEPOINTS
:
3524 struct lttng_event
*events
;
3527 session_lock_list();
3528 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3529 session_unlock_list();
3530 if (nb_events
< 0) {
3531 /* Return value is a negative lttng_error_code. */
3537 * Setup lttng message with payload size set to the event list size in
3538 * bytes and then copy list into the llm payload.
3540 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3546 /* Copy event list into message payload */
3547 memcpy(cmd_ctx
->llm
->payload
, events
,
3548 sizeof(struct lttng_event
) * nb_events
);
3555 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3557 struct lttng_event_field
*fields
;
3560 session_lock_list();
3561 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3563 session_unlock_list();
3564 if (nb_fields
< 0) {
3565 /* Return value is a negative lttng_error_code. */
3571 * Setup lttng message with payload size set to the event list size in
3572 * bytes and then copy list into the llm payload.
3574 ret
= setup_lttng_msg(cmd_ctx
,
3575 sizeof(struct lttng_event_field
) * nb_fields
);
3581 /* Copy event list into message payload */
3582 memcpy(cmd_ctx
->llm
->payload
, fields
,
3583 sizeof(struct lttng_event_field
) * nb_fields
);
3590 case LTTNG_LIST_SYSCALLS
:
3592 struct lttng_event
*events
;
3595 nb_events
= cmd_list_syscalls(&events
);
3596 if (nb_events
< 0) {
3597 /* Return value is a negative lttng_error_code. */
3603 * Setup lttng message with payload size set to the event list size in
3604 * bytes and then copy list into the llm payload.
3606 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3612 /* Copy event list into message payload */
3613 memcpy(cmd_ctx
->llm
->payload
, events
,
3614 sizeof(struct lttng_event
) * nb_events
);
3621 case LTTNG_LIST_TRACKER_PIDS
:
3623 int32_t *pids
= NULL
;
3626 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3627 cmd_ctx
->lsm
->domain
.type
, &pids
);
3629 /* Return value is a negative lttng_error_code. */
3635 * Setup lttng message with payload size set to the event list size in
3636 * bytes and then copy list into the llm payload.
3638 ret
= setup_lttng_msg(cmd_ctx
, sizeof(int32_t) * nr_pids
);
3644 /* Copy event list into message payload */
3645 memcpy(cmd_ctx
->llm
->payload
, pids
,
3646 sizeof(int) * nr_pids
);
3653 case LTTNG_SET_CONSUMER_URI
:
3656 struct lttng_uri
*uris
;
3658 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3659 len
= nb_uri
* sizeof(struct lttng_uri
);
3662 ret
= LTTNG_ERR_INVALID
;
3666 uris
= zmalloc(len
);
3668 ret
= LTTNG_ERR_FATAL
;
3672 /* Receive variable len data */
3673 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3674 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3676 DBG("No URIs received from client... continuing");
3678 ret
= LTTNG_ERR_SESSION_FAIL
;
3683 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3685 if (ret
!= LTTNG_OK
) {
3692 case LTTNG_START_TRACE
:
3694 ret
= cmd_start_trace(cmd_ctx
->session
);
3697 case LTTNG_STOP_TRACE
:
3699 ret
= cmd_stop_trace(cmd_ctx
->session
);
3702 case LTTNG_CREATE_SESSION
:
3705 struct lttng_uri
*uris
= NULL
;
3707 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3708 len
= nb_uri
* sizeof(struct lttng_uri
);
3711 uris
= zmalloc(len
);
3713 ret
= LTTNG_ERR_FATAL
;
3717 /* Receive variable len data */
3718 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3719 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3721 DBG("No URIs received from client... continuing");
3723 ret
= LTTNG_ERR_SESSION_FAIL
;
3728 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3729 DBG("Creating session with ONE network URI is a bad call");
3730 ret
= LTTNG_ERR_SESSION_FAIL
;
3736 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3737 &cmd_ctx
->creds
, 0);
3743 case LTTNG_DESTROY_SESSION
:
3745 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3747 /* Set session to NULL so we do not unlock it after free. */
3748 cmd_ctx
->session
= NULL
;
3751 case LTTNG_LIST_DOMAINS
:
3754 struct lttng_domain
*domains
= NULL
;
3756 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3758 /* Return value is a negative lttng_error_code. */
3763 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3769 /* Copy event list into message payload */
3770 memcpy(cmd_ctx
->llm
->payload
, domains
,
3771 nb_dom
* sizeof(struct lttng_domain
));
3778 case LTTNG_LIST_CHANNELS
:
3781 struct lttng_channel
*channels
= NULL
;
3783 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3784 cmd_ctx
->session
, &channels
);
3786 /* Return value is a negative lttng_error_code. */
3791 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3797 /* Copy event list into message payload */
3798 memcpy(cmd_ctx
->llm
->payload
, channels
,
3799 nb_chan
* sizeof(struct lttng_channel
));
3806 case LTTNG_LIST_EVENTS
:
3809 struct lttng_event
*events
= NULL
;
3811 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3812 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3814 /* Return value is a negative lttng_error_code. */
3819 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3825 /* Copy event list into message payload */
3826 memcpy(cmd_ctx
->llm
->payload
, events
,
3827 nb_event
* sizeof(struct lttng_event
));
3834 case LTTNG_LIST_SESSIONS
:
3836 unsigned int nr_sessions
;
3838 session_lock_list();
3839 nr_sessions
= lttng_sessions_count(
3840 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3841 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3843 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3845 session_unlock_list();
3849 /* Filled the session array */
3850 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3851 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3852 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3854 session_unlock_list();
3859 case LTTNG_CALIBRATE
:
3861 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3862 &cmd_ctx
->lsm
->u
.calibrate
);
3865 case LTTNG_REGISTER_CONSUMER
:
3867 struct consumer_data
*cdata
;
3869 switch (cmd_ctx
->lsm
->domain
.type
) {
3870 case LTTNG_DOMAIN_KERNEL
:
3871 cdata
= &kconsumer_data
;
3874 ret
= LTTNG_ERR_UND
;
3878 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3879 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3882 case LTTNG_DATA_PENDING
:
3886 /* 1 byte to return whether or not data is pending */
3887 ret
= setup_lttng_msg(cmd_ctx
, 1);
3892 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
3896 * This function may returns 0 or 1 to indicate whether or not
3897 * there is data pending. In case of error, it should return an
3898 * LTTNG_ERR code. However, some code paths may still return
3899 * a nondescript error code, which we handle by returning an
3902 if (pending_ret
== 0 || pending_ret
== 1) {
3904 } else if (pending_ret
< 0) {
3905 ret
= LTTNG_ERR_UNK
;
3912 *cmd_ctx
->llm
->payload
= (uint8_t) pending_ret
;
3915 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3917 struct lttcomm_lttng_output_id reply
;
3919 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3920 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3921 if (ret
!= LTTNG_OK
) {
3925 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3930 /* Copy output list into message payload */
3931 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3935 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3937 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3938 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3941 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3944 struct lttng_snapshot_output
*outputs
= NULL
;
3946 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3947 if (nb_output
< 0) {
3952 ret
= setup_lttng_msg(cmd_ctx
,
3953 nb_output
* sizeof(struct lttng_snapshot_output
));
3960 /* Copy output list into message payload */
3961 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3962 nb_output
* sizeof(struct lttng_snapshot_output
));
3969 case LTTNG_SNAPSHOT_RECORD
:
3971 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3972 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3973 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3976 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3979 struct lttng_uri
*uris
= NULL
;
3981 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3982 len
= nb_uri
* sizeof(struct lttng_uri
);
3985 uris
= zmalloc(len
);
3987 ret
= LTTNG_ERR_FATAL
;
3991 /* Receive variable len data */
3992 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3993 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3995 DBG("No URIs received from client... continuing");
3997 ret
= LTTNG_ERR_SESSION_FAIL
;
4002 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4003 DBG("Creating session with ONE network URI is a bad call");
4004 ret
= LTTNG_ERR_SESSION_FAIL
;
4010 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
4011 nb_uri
, &cmd_ctx
->creds
);
4015 case LTTNG_CREATE_SESSION_LIVE
:
4018 struct lttng_uri
*uris
= NULL
;
4020 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4021 len
= nb_uri
* sizeof(struct lttng_uri
);
4024 uris
= zmalloc(len
);
4026 ret
= LTTNG_ERR_FATAL
;
4030 /* Receive variable len data */
4031 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4032 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4034 DBG("No URIs received from client... continuing");
4036 ret
= LTTNG_ERR_SESSION_FAIL
;
4041 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4042 DBG("Creating session with ONE network URI is a bad call");
4043 ret
= LTTNG_ERR_SESSION_FAIL
;
4049 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4050 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4054 case LTTNG_SAVE_SESSION
:
4056 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4060 case LTTNG_SET_SESSION_SHM_PATH
:
4062 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4063 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4067 ret
= LTTNG_ERR_UND
;
4072 if (cmd_ctx
->llm
== NULL
) {
4073 DBG("Missing llm structure. Allocating one.");
4074 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
4078 /* Set return code */
4079 cmd_ctx
->llm
->ret_code
= ret
;
4081 if (cmd_ctx
->session
) {
4082 session_unlock(cmd_ctx
->session
);
4084 if (need_tracing_session
) {
4085 session_unlock_list();
4088 assert(!rcu_read_ongoing());
4093 * Thread managing health check socket.
4095 static void *thread_manage_health(void *data
)
4097 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4098 uint32_t revents
, nb_fd
;
4099 struct lttng_poll_event events
;
4100 struct health_comm_msg msg
;
4101 struct health_comm_reply reply
;
4103 DBG("[thread] Manage health check started");
4105 rcu_register_thread();
4107 /* We might hit an error path before this is created. */
4108 lttng_poll_init(&events
);
4110 /* Create unix socket */
4111 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4113 ERR("Unable to create health check Unix socket");
4119 /* lttng health client socket path permissions */
4120 ret
= chown(health_unix_sock_path
, 0,
4121 utils_get_group_id(tracing_group_name
));
4123 ERR("Unable to set group on %s", health_unix_sock_path
);
4129 ret
= chmod(health_unix_sock_path
,
4130 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4132 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4140 * Set the CLOEXEC flag. Return code is useless because either way, the
4143 (void) utils_set_fd_cloexec(sock
);
4145 ret
= lttcomm_listen_unix_sock(sock
);
4151 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4152 * more will be added to this poll set.
4154 ret
= sessiond_set_thread_pollset(&events
, 2);
4159 /* Add the application registration socket */
4160 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4165 sessiond_notify_ready();
4168 DBG("Health check ready");
4170 /* Inifinite blocking call, waiting for transmission */
4172 ret
= lttng_poll_wait(&events
, -1);
4175 * Restart interrupted system call.
4177 if (errno
== EINTR
) {
4185 for (i
= 0; i
< nb_fd
; i
++) {
4186 /* Fetch once the poll data */
4187 revents
= LTTNG_POLL_GETEV(&events
, i
);
4188 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4191 /* No activity for this FD (poll implementation). */
4195 /* Thread quit pipe has been closed. Killing thread. */
4196 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4202 /* Event on the registration socket */
4203 if (pollfd
== sock
) {
4204 if (revents
& LPOLLIN
) {
4206 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4207 ERR("Health socket poll error");
4210 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4216 new_sock
= lttcomm_accept_unix_sock(sock
);
4222 * Set the CLOEXEC flag. Return code is useless because either way, the
4225 (void) utils_set_fd_cloexec(new_sock
);
4227 DBG("Receiving data from client for health...");
4228 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4230 DBG("Nothing recv() from client... continuing");
4231 ret
= close(new_sock
);
4239 rcu_thread_online();
4241 memset(&reply
, 0, sizeof(reply
));
4242 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4244 * health_check_state returns 0 if health is
4247 if (!health_check_state(health_sessiond
, i
)) {
4248 reply
.ret_code
|= 1ULL << i
;
4252 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4254 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4256 ERR("Failed to send health data back to client");
4259 /* End of transmission */
4260 ret
= close(new_sock
);
4270 ERR("Health error occurred in %s", __func__
);
4272 DBG("Health check thread dying");
4273 unlink(health_unix_sock_path
);
4281 lttng_poll_clean(&events
);
4283 rcu_unregister_thread();
4288 * This thread manage all clients request using the unix client socket for
4291 static void *thread_manage_clients(void *data
)
4293 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4295 uint32_t revents
, nb_fd
;
4296 struct command_ctx
*cmd_ctx
= NULL
;
4297 struct lttng_poll_event events
;
4299 DBG("[thread] Manage client started");
4301 rcu_register_thread();
4303 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4305 health_code_update();
4307 ret
= lttcomm_listen_unix_sock(client_sock
);
4313 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4314 * more will be added to this poll set.
4316 ret
= sessiond_set_thread_pollset(&events
, 2);
4318 goto error_create_poll
;
4321 /* Add the application registration socket */
4322 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4327 sessiond_notify_ready();
4328 ret
= sem_post(&load_info
->message_thread_ready
);
4330 PERROR("sem_post message_thread_ready");
4334 /* This testpoint is after we signal readiness to the parent. */
4335 if (testpoint(sessiond_thread_manage_clients
)) {
4339 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4343 health_code_update();
4346 DBG("Accepting client command ...");
4348 /* Inifinite blocking call, waiting for transmission */
4350 health_poll_entry();
4351 ret
= lttng_poll_wait(&events
, -1);
4355 * Restart interrupted system call.
4357 if (errno
== EINTR
) {
4365 for (i
= 0; i
< nb_fd
; i
++) {
4366 /* Fetch once the poll data */
4367 revents
= LTTNG_POLL_GETEV(&events
, i
);
4368 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4370 health_code_update();
4373 /* No activity for this FD (poll implementation). */
4377 /* Thread quit pipe has been closed. Killing thread. */
4378 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4384 /* Event on the registration socket */
4385 if (pollfd
== client_sock
) {
4386 if (revents
& LPOLLIN
) {
4388 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4389 ERR("Client socket poll error");
4392 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4398 DBG("Wait for client response");
4400 health_code_update();
4402 sock
= lttcomm_accept_unix_sock(client_sock
);
4408 * Set the CLOEXEC flag. Return code is useless because either way, the
4411 (void) utils_set_fd_cloexec(sock
);
4413 /* Set socket option for credentials retrieval */
4414 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4419 /* Allocate context command to process the client request */
4420 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4421 if (cmd_ctx
== NULL
) {
4422 PERROR("zmalloc cmd_ctx");
4426 /* Allocate data buffer for reception */
4427 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4428 if (cmd_ctx
->lsm
== NULL
) {
4429 PERROR("zmalloc cmd_ctx->lsm");
4433 cmd_ctx
->llm
= NULL
;
4434 cmd_ctx
->session
= NULL
;
4436 health_code_update();
4439 * Data is received from the lttng client. The struct
4440 * lttcomm_session_msg (lsm) contains the command and data request of
4443 DBG("Receiving data from client ...");
4444 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4445 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4447 DBG("Nothing recv() from client... continuing");
4453 clean_command_ctx(&cmd_ctx
);
4457 health_code_update();
4459 // TODO: Validate cmd_ctx including sanity check for
4460 // security purpose.
4462 rcu_thread_online();
4464 * This function dispatch the work to the kernel or userspace tracer
4465 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4466 * informations for the client. The command context struct contains
4467 * everything this function may needs.
4469 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4470 rcu_thread_offline();
4478 * TODO: Inform client somehow of the fatal error. At
4479 * this point, ret < 0 means that a zmalloc failed
4480 * (ENOMEM). Error detected but still accept
4481 * command, unless a socket error has been
4484 clean_command_ctx(&cmd_ctx
);
4488 health_code_update();
4490 DBG("Sending response (size: %d, retcode: %s (%d))",
4491 cmd_ctx
->lttng_msg_size
,
4492 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4493 cmd_ctx
->llm
->ret_code
);
4494 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4496 ERR("Failed to send data back to client");
4499 /* End of transmission */
4506 clean_command_ctx(&cmd_ctx
);
4508 health_code_update();
4520 lttng_poll_clean(&events
);
4521 clean_command_ctx(&cmd_ctx
);
4525 unlink(client_unix_sock_path
);
4526 if (client_sock
>= 0) {
4527 ret
= close(client_sock
);
4535 ERR("Health error occurred in %s", __func__
);
4538 health_unregister(health_sessiond
);
4540 DBG("Client thread dying");
4542 rcu_unregister_thread();
4545 * Since we are creating the consumer threads, we own them, so we need
4546 * to join them before our thread exits.
4548 ret
= join_consumer_thread(&kconsumer_data
);
4551 PERROR("join_consumer");
4554 ret
= join_consumer_thread(&ustconsumer32_data
);
4557 PERROR("join_consumer ust32");
4560 ret
= join_consumer_thread(&ustconsumer64_data
);
4563 PERROR("join_consumer ust64");
4570 * usage function on stderr
4572 static void usage(void)
4574 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4575 fprintf(stderr
, " -h, --help Display this usage.\n");
4576 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4577 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4578 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4579 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4580 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4581 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4582 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4583 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4584 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4585 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4586 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4587 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4588 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4589 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4590 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4591 fprintf(stderr
, " -V, --version Show version number.\n");
4592 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4593 fprintf(stderr
, " -q, --quiet No output at all.\n");
4594 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4595 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4596 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4597 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4598 fprintf(stderr
, " --agent-tcp-port Agent registration TCP port\n");
4599 fprintf(stderr
, " -f --config PATH Load daemon configuration file\n");
4600 fprintf(stderr
, " -l --load PATH Load session configuration\n");
4601 fprintf(stderr
, " --kmod-probes Specify kernel module probes to load\n");
4602 fprintf(stderr
, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4605 static int string_match(const char *str1
, const char *str2
)
4607 return (str1
&& str2
) && !strcmp(str1
, str2
);
4611 * Take an option from the getopt output and set it in the right variable to be
4614 * Return 0 on success else a negative value.
4616 static int set_option(int opt
, const char *arg
, const char *optname
)
4620 if (arg
&& arg
[0] == '\0') {
4622 * This only happens if the value is read from daemon config
4623 * file. This means the option requires an argument and the
4624 * configuration file contains a line such as:
4631 if (string_match(optname
, "client-sock") || opt
== 'c') {
4632 if (lttng_is_setuid_setgid()) {
4633 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4634 "-c, --client-sock");
4636 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4638 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4639 if (lttng_is_setuid_setgid()) {
4640 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4643 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4645 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4647 } else if (string_match(optname
, "background") || opt
== 'b') {
4649 } else if (string_match(optname
, "group") || opt
== 'g') {
4650 if (lttng_is_setuid_setgid()) {
4651 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4655 * If the override option is set, the pointer points to a
4656 * *non* const thus freeing it even though the variable type is
4659 if (tracing_group_name_override
) {
4660 free((void *) tracing_group_name
);
4662 tracing_group_name
= strdup(arg
);
4663 if (!tracing_group_name
) {
4667 tracing_group_name_override
= 1;
4669 } else if (string_match(optname
, "help") || opt
== 'h') {
4672 } else if (string_match(optname
, "version") || opt
== 'V') {
4673 fprintf(stdout
, "%s\n", VERSION
);
4675 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4677 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4678 if (lttng_is_setuid_setgid()) {
4679 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4680 "--kconsumerd-err-sock");
4682 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4684 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4685 if (lttng_is_setuid_setgid()) {
4686 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4687 "--kconsumerd-cmd-sock");
4689 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4691 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4692 if (lttng_is_setuid_setgid()) {
4693 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4694 "--ustconsumerd64-err-sock");
4696 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4698 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4699 if (lttng_is_setuid_setgid()) {
4700 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4701 "--ustconsumerd64-cmd-sock");
4703 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4705 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4706 if (lttng_is_setuid_setgid()) {
4707 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4708 "--ustconsumerd32-err-sock");
4710 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4712 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4713 if (lttng_is_setuid_setgid()) {
4714 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4715 "--ustconsumerd32-cmd-sock");
4717 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4719 } else if (string_match(optname
, "no-kernel")) {
4721 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4722 lttng_opt_quiet
= 1;
4723 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4724 /* Verbose level can increase using multiple -v */
4726 /* Value obtained from config file */
4727 lttng_opt_verbose
= config_parse_value(arg
);
4729 /* -v used on command line */
4730 lttng_opt_verbose
++;
4732 /* Clamp value to [0, 3] */
4733 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4734 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4735 } else if (string_match(optname
, "verbose-consumer")) {
4737 opt_verbose_consumer
= config_parse_value(arg
);
4739 opt_verbose_consumer
+= 1;
4741 } else if (string_match(optname
, "consumerd32-path")) {
4742 if (lttng_is_setuid_setgid()) {
4743 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4744 "--consumerd32-path");
4746 if (consumerd32_bin_override
) {
4747 free((void *) consumerd32_bin
);
4749 consumerd32_bin
= strdup(arg
);
4750 if (!consumerd32_bin
) {
4754 consumerd32_bin_override
= 1;
4756 } else if (string_match(optname
, "consumerd32-libdir")) {
4757 if (lttng_is_setuid_setgid()) {
4758 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4759 "--consumerd32-libdir");
4761 if (consumerd32_libdir_override
) {
4762 free((void *) consumerd32_libdir
);
4764 consumerd32_libdir
= strdup(arg
);
4765 if (!consumerd32_libdir
) {
4769 consumerd32_libdir_override
= 1;
4771 } else if (string_match(optname
, "consumerd64-path")) {
4772 if (lttng_is_setuid_setgid()) {
4773 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4774 "--consumerd64-path");
4776 if (consumerd64_bin_override
) {
4777 free((void *) consumerd64_bin
);
4779 consumerd64_bin
= strdup(arg
);
4780 if (!consumerd64_bin
) {
4784 consumerd64_bin_override
= 1;
4786 } else if (string_match(optname
, "consumerd64-libdir")) {
4787 if (lttng_is_setuid_setgid()) {
4788 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4789 "--consumerd64-libdir");
4791 if (consumerd64_libdir_override
) {
4792 free((void *) consumerd64_libdir
);
4794 consumerd64_libdir
= strdup(arg
);
4795 if (!consumerd64_libdir
) {
4799 consumerd64_libdir_override
= 1;
4801 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4802 if (lttng_is_setuid_setgid()) {
4803 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4807 opt_pidfile
= strdup(arg
);
4813 } else if (string_match(optname
, "agent-tcp-port")) {
4814 if (lttng_is_setuid_setgid()) {
4815 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4816 "--agent-tcp-port");
4825 v
= strtoul(arg
, NULL
, 0);
4826 if (errno
!= 0 || !isdigit(arg
[0])) {
4827 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4830 if (v
== 0 || v
>= 65535) {
4831 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4834 agent_tcp_port
= (uint32_t) v
;
4835 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4837 } else if (string_match(optname
, "load") || opt
== 'l') {
4838 if (lttng_is_setuid_setgid()) {
4839 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4842 free(opt_load_session_path
);
4843 opt_load_session_path
= strdup(arg
);
4844 if (!opt_load_session_path
) {
4849 } else if (string_match(optname
, "kmod-probes")) {
4850 if (lttng_is_setuid_setgid()) {
4851 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4854 free(kmod_probes_list
);
4855 kmod_probes_list
= strdup(arg
);
4856 if (!kmod_probes_list
) {
4861 } else if (string_match(optname
, "extra-kmod-probes")) {
4862 if (lttng_is_setuid_setgid()) {
4863 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4864 "--extra-kmod-probes");
4866 free(kmod_extra_probes_list
);
4867 kmod_extra_probes_list
= strdup(arg
);
4868 if (!kmod_extra_probes_list
) {
4873 } else if (string_match(optname
, "config") || opt
== 'f') {
4874 /* This is handled in set_options() thus silent skip. */
4877 /* Unknown option or other error.
4878 * Error is printed by getopt, just return */
4883 if (ret
== -EINVAL
) {
4884 const char *opt_name
= "unknown";
4887 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4889 if (opt
== long_options
[i
].val
) {
4890 opt_name
= long_options
[i
].name
;
4895 WARN("Invalid argument provided for option \"%s\", using default value.",
4903 * config_entry_handler_cb used to handle options read from a config file.
4904 * See config_entry_handler_cb comment in common/config/config.h for the
4905 * return value conventions.
4907 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4911 if (!entry
|| !entry
->name
|| !entry
->value
) {
4916 /* Check if the option is to be ignored */
4917 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4918 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4923 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4926 /* Ignore if not fully matched. */
4927 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4932 * If the option takes no argument on the command line, we have to
4933 * check if the value is "true". We support non-zero numeric values,
4936 if (!long_options
[i
].has_arg
) {
4937 ret
= config_parse_value(entry
->value
);
4940 WARN("Invalid configuration value \"%s\" for option %s",
4941 entry
->value
, entry
->name
);
4943 /* False, skip boolean config option. */
4948 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4952 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4959 * daemon configuration loading and argument parsing
4961 static int set_options(int argc
, char **argv
)
4963 int ret
= 0, c
= 0, option_index
= 0;
4964 int orig_optopt
= optopt
, orig_optind
= optind
;
4966 const char *config_path
= NULL
;
4968 optstring
= utils_generate_optstring(long_options
,
4969 sizeof(long_options
) / sizeof(struct option
));
4975 /* Check for the --config option */
4976 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4977 &option_index
)) != -1) {
4981 } else if (c
!= 'f') {
4982 /* if not equal to --config option. */
4986 if (lttng_is_setuid_setgid()) {
4987 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4990 config_path
= utils_expand_path(optarg
);
4992 ERR("Failed to resolve path: %s", optarg
);
4997 ret
= config_get_section_entries(config_path
, config_section_name
,
4998 config_entry_handler
, NULL
);
5001 ERR("Invalid configuration option at line %i", ret
);
5007 /* Reset getopt's global state */
5008 optopt
= orig_optopt
;
5009 optind
= orig_optind
;
5013 * getopt_long() will not set option_index if it encounters a
5016 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5023 * Pass NULL as the long option name if popt left the index
5026 ret
= set_option(c
, optarg
,
5027 option_index
< 0 ? NULL
:
5028 long_options
[option_index
].name
);
5040 * Creates the two needed socket by the daemon.
5041 * apps_sock - The communication socket for all UST apps.
5042 * client_sock - The communication of the cli tool (lttng).
5044 static int init_daemon_socket(void)
5049 old_umask
= umask(0);
5051 /* Create client tool unix socket */
5052 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
5053 if (client_sock
< 0) {
5054 ERR("Create unix sock failed: %s", client_unix_sock_path
);
5059 /* Set the cloexec flag */
5060 ret
= utils_set_fd_cloexec(client_sock
);
5062 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5063 "Continuing but note that the consumer daemon will have a "
5064 "reference to this socket on exec()", client_sock
);
5067 /* File permission MUST be 660 */
5068 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5070 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5075 /* Create the application unix socket */
5076 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5077 if (apps_sock
< 0) {
5078 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5083 /* Set the cloexec flag */
5084 ret
= utils_set_fd_cloexec(apps_sock
);
5086 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5087 "Continuing but note that the consumer daemon will have a "
5088 "reference to this socket on exec()", apps_sock
);
5091 /* File permission MUST be 666 */
5092 ret
= chmod(apps_unix_sock_path
,
5093 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5095 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5100 DBG3("Session daemon client socket %d and application socket %d created",
5101 client_sock
, apps_sock
);
5109 * Check if the global socket is available, and if a daemon is answering at the
5110 * other side. If yes, error is returned.
5112 static int check_existing_daemon(void)
5114 /* Is there anybody out there ? */
5115 if (lttng_session_daemon_alive()) {
5123 * Set the tracing group gid onto the client socket.
5125 * Race window between mkdir and chown is OK because we are going from more
5126 * permissive (root.root) to less permissive (root.tracing).
5128 static int set_permissions(char *rundir
)
5133 gid
= utils_get_group_id(tracing_group_name
);
5135 /* Set lttng run dir */
5136 ret
= chown(rundir
, 0, gid
);
5138 ERR("Unable to set group on %s", rundir
);
5143 * Ensure all applications and tracing group can search the run
5144 * dir. Allow everyone to read the directory, since it does not
5145 * buy us anything to hide its content.
5147 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5149 ERR("Unable to set permissions on %s", rundir
);
5153 /* lttng client socket path */
5154 ret
= chown(client_unix_sock_path
, 0, gid
);
5156 ERR("Unable to set group on %s", client_unix_sock_path
);
5160 /* kconsumer error socket path */
5161 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5163 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5167 /* 64-bit ustconsumer error socket path */
5168 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5170 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5174 /* 32-bit ustconsumer compat32 error socket path */
5175 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5177 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5181 DBG("All permissions are set");
5187 * Create the lttng run directory needed for all global sockets and pipe.
5189 static int create_lttng_rundir(const char *rundir
)
5193 DBG3("Creating LTTng run directory: %s", rundir
);
5195 ret
= mkdir(rundir
, S_IRWXU
);
5197 if (errno
!= EEXIST
) {
5198 ERR("Unable to create %s", rundir
);
5210 * Setup sockets and directory needed by the kconsumerd communication with the
5213 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5217 char path
[PATH_MAX
];
5219 switch (consumer_data
->type
) {
5220 case LTTNG_CONSUMER_KERNEL
:
5221 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5223 case LTTNG_CONSUMER64_UST
:
5224 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5226 case LTTNG_CONSUMER32_UST
:
5227 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5230 ERR("Consumer type unknown");
5235 DBG2("Creating consumer directory: %s", path
);
5237 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5239 if (errno
!= EEXIST
) {
5241 ERR("Failed to create %s", path
);
5247 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5249 ERR("Unable to set group on %s", path
);
5255 /* Create the kconsumerd error unix socket */
5256 consumer_data
->err_sock
=
5257 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5258 if (consumer_data
->err_sock
< 0) {
5259 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5265 * Set the CLOEXEC flag. Return code is useless because either way, the
5268 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5270 PERROR("utils_set_fd_cloexec");
5271 /* continue anyway */
5274 /* File permission MUST be 660 */
5275 ret
= chmod(consumer_data
->err_unix_sock_path
,
5276 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5278 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5288 * Signal handler for the daemon
5290 * Simply stop all worker threads, leaving main() return gracefully after
5291 * joining all threads and calling cleanup().
5293 static void sighandler(int sig
)
5297 DBG("SIGPIPE caught");
5300 DBG("SIGINT caught");
5304 DBG("SIGTERM caught");
5308 CMM_STORE_SHARED(recv_child_signal
, 1);
5316 * Setup signal handler for :
5317 * SIGINT, SIGTERM, SIGPIPE
5319 static int set_signal_handler(void)
5322 struct sigaction sa
;
5325 if ((ret
= sigemptyset(&sigset
)) < 0) {
5326 PERROR("sigemptyset");
5330 sa
.sa_handler
= sighandler
;
5331 sa
.sa_mask
= sigset
;
5333 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5334 PERROR("sigaction");
5338 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5339 PERROR("sigaction");
5343 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5344 PERROR("sigaction");
5348 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5349 PERROR("sigaction");
5353 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5359 * Set open files limit to unlimited. This daemon can open a large number of
5360 * file descriptors in order to consumer multiple kernel traces.
5362 static void set_ulimit(void)
5367 /* The kernel does not allowed an infinite limit for open files */
5368 lim
.rlim_cur
= 65535;
5369 lim
.rlim_max
= 65535;
5371 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5373 PERROR("failed to set open files limit");
5378 * Write pidfile using the rundir and opt_pidfile.
5380 static int write_pidfile(void)
5383 char pidfile_path
[PATH_MAX
];
5388 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
5390 /* Build pidfile path from rundir and opt_pidfile. */
5391 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5392 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5394 PERROR("snprintf pidfile path");
5400 * Create pid file in rundir.
5402 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5408 * Create lockfile using the rundir and return its fd.
5410 static int create_lockfile(void)
5413 char lockfile_path
[PATH_MAX
];
5415 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5420 ret
= utils_create_lock_file(lockfile_path
);
5426 * Write agent TCP port using the rundir.
5428 static int write_agent_port(void)
5431 char path
[PATH_MAX
];
5435 ret
= snprintf(path
, sizeof(path
), "%s/"
5436 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5438 PERROR("snprintf agent port path");
5443 * Create TCP agent port file in rundir.
5445 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5454 int main(int argc
, char **argv
)
5456 int ret
= 0, retval
= 0;
5458 const char *home_path
, *env_app_timeout
;
5460 init_kernel_workarounds();
5462 rcu_register_thread();
5464 if (set_signal_handler()) {
5466 goto exit_set_signal_handler
;
5469 setup_consumerd_path();
5471 page_size
= sysconf(_SC_PAGESIZE
);
5472 if (page_size
< 0) {
5473 PERROR("sysconf _SC_PAGESIZE");
5474 page_size
= LONG_MAX
;
5475 WARN("Fallback page size to %ld", page_size
);
5479 * Parse arguments and load the daemon configuration file.
5481 * We have an exit_options exit path to free memory reserved by
5482 * set_options. This is needed because the rest of sessiond_cleanup()
5483 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5484 * depends on set_options.
5487 if (set_options(argc
, argv
)) {
5493 if (opt_daemon
|| opt_background
) {
5496 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5504 * We are in the child. Make sure all other file descriptors are
5505 * closed, in case we are called with more opened file
5506 * descriptors than the standard ones.
5508 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5514 * Starting from here, we can create threads. This needs to be after
5515 * lttng_daemonize due to RCU.
5519 * Initialize the health check subsystem. This call should set the
5520 * appropriate time values.
5522 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5523 if (!health_sessiond
) {
5524 PERROR("health_app_create error");
5526 goto exit_health_sessiond_cleanup
;
5529 if (init_ht_cleanup_quit_pipe()) {
5531 goto exit_ht_cleanup_quit_pipe
;
5534 /* Setup the thread ht_cleanup communication pipe. */
5535 if (utils_create_pipe_cloexec(ht_cleanup_pipe
)) {
5537 goto exit_ht_cleanup_pipe
;
5540 /* Set up max poll set size */
5541 if (lttng_poll_set_max_size()) {
5543 goto exit_set_max_size
;
5546 /* Create thread to clean up RCU hash tables */
5547 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5548 thread_ht_cleanup
, (void *) NULL
);
5551 PERROR("pthread_create ht_cleanup");
5553 goto exit_ht_cleanup
;
5556 /* Create thread quit pipe */
5557 if (init_thread_quit_pipe()) {
5559 goto exit_init_data
;
5562 /* Check if daemon is UID = 0 */
5563 is_root
= !getuid();
5566 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5569 goto exit_init_data
;
5572 /* Create global run dir with root access */
5573 if (create_lttng_rundir(rundir
)) {
5575 goto exit_init_data
;
5578 if (strlen(apps_unix_sock_path
) == 0) {
5579 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5580 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5583 goto exit_init_data
;
5587 if (strlen(client_unix_sock_path
) == 0) {
5588 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5589 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5592 goto exit_init_data
;
5596 /* Set global SHM for ust */
5597 if (strlen(wait_shm_path
) == 0) {
5598 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5599 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5602 goto exit_init_data
;
5606 if (strlen(health_unix_sock_path
) == 0) {
5607 ret
= snprintf(health_unix_sock_path
,
5608 sizeof(health_unix_sock_path
),
5609 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5612 goto exit_init_data
;
5616 /* Setup kernel consumerd path */
5617 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5618 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5621 goto exit_init_data
;
5623 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5624 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5627 goto exit_init_data
;
5630 DBG2("Kernel consumer err path: %s",
5631 kconsumer_data
.err_unix_sock_path
);
5632 DBG2("Kernel consumer cmd path: %s",
5633 kconsumer_data
.cmd_unix_sock_path
);
5635 home_path
= utils_get_home_dir();
5636 if (home_path
== NULL
) {
5637 /* TODO: Add --socket PATH option */
5638 ERR("Can't get HOME directory for sockets creation.");
5640 goto exit_init_data
;
5644 * Create rundir from home path. This will create something like
5647 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5650 goto exit_init_data
;
5653 if (create_lttng_rundir(rundir
)) {
5655 goto exit_init_data
;
5658 if (strlen(apps_unix_sock_path
) == 0) {
5659 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5660 DEFAULT_HOME_APPS_UNIX_SOCK
,
5664 goto exit_init_data
;
5668 /* Set the cli tool unix socket path */
5669 if (strlen(client_unix_sock_path
) == 0) {
5670 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5671 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5675 goto exit_init_data
;
5679 /* Set global SHM for ust */
5680 if (strlen(wait_shm_path
) == 0) {
5681 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5682 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5686 goto exit_init_data
;
5690 /* Set health check Unix path */
5691 if (strlen(health_unix_sock_path
) == 0) {
5692 ret
= snprintf(health_unix_sock_path
,
5693 sizeof(health_unix_sock_path
),
5694 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5698 goto exit_init_data
;
5703 lockfile_fd
= create_lockfile();
5704 if (lockfile_fd
< 0) {
5706 goto exit_init_data
;
5709 /* Set consumer initial state */
5710 kernel_consumerd_state
= CONSUMER_STOPPED
;
5711 ust_consumerd_state
= CONSUMER_STOPPED
;
5713 DBG("Client socket path %s", client_unix_sock_path
);
5714 DBG("Application socket path %s", apps_unix_sock_path
);
5715 DBG("Application wait path %s", wait_shm_path
);
5716 DBG("LTTng run directory path: %s", rundir
);
5718 /* 32 bits consumerd path setup */
5719 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5720 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5722 PERROR("snprintf 32-bit consumer error socket path");
5724 goto exit_init_data
;
5726 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5727 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5729 PERROR("snprintf 32-bit consumer command socket path");
5731 goto exit_init_data
;
5734 DBG2("UST consumer 32 bits err path: %s",
5735 ustconsumer32_data
.err_unix_sock_path
);
5736 DBG2("UST consumer 32 bits cmd path: %s",
5737 ustconsumer32_data
.cmd_unix_sock_path
);
5739 /* 64 bits consumerd path setup */
5740 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5741 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5743 PERROR("snprintf 64-bit consumer error socket path");
5745 goto exit_init_data
;
5747 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5748 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5750 PERROR("snprintf 64-bit consumer command socket path");
5752 goto exit_init_data
;
5755 DBG2("UST consumer 64 bits err path: %s",
5756 ustconsumer64_data
.err_unix_sock_path
);
5757 DBG2("UST consumer 64 bits cmd path: %s",
5758 ustconsumer64_data
.cmd_unix_sock_path
);
5761 * See if daemon already exist.
5763 if (check_existing_daemon()) {
5764 ERR("Already running daemon.\n");
5766 * We do not goto exit because we must not cleanup()
5767 * because a daemon is already running.
5770 goto exit_init_data
;
5774 * Init UST app hash table. Alloc hash table before this point since
5775 * cleanup() can get called after that point.
5777 if (ust_app_ht_alloc()) {
5778 ERR("Failed to allocate UST app hash table");
5780 goto exit_init_data
;
5784 * Initialize agent app hash table. We allocate the hash table here
5785 * since cleanup() can get called after this point.
5787 if (agent_app_ht_alloc()) {
5788 ERR("Failed to allocate Agent app hash table");
5790 goto exit_init_data
;
5794 * These actions must be executed as root. We do that *after* setting up
5795 * the sockets path because we MUST make the check for another daemon using
5796 * those paths *before* trying to set the kernel consumer sockets and init
5800 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5802 goto exit_init_data
;
5805 /* Setup kernel tracer */
5806 if (!opt_no_kernel
) {
5807 init_kernel_tracer();
5808 if (kernel_tracer_fd
>= 0) {
5809 ret
= syscall_init_table();
5811 ERR("Unable to populate syscall table. "
5812 "Syscall tracing won't work "
5813 "for this session daemon.");
5818 /* Set ulimit for open files */
5821 /* init lttng_fd tracking must be done after set_ulimit. */
5824 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5826 goto exit_init_data
;
5829 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5831 goto exit_init_data
;
5834 /* Setup the needed unix socket */
5835 if (init_daemon_socket()) {
5837 goto exit_init_data
;
5840 /* Set credentials to socket */
5841 if (is_root
&& set_permissions(rundir
)) {
5843 goto exit_init_data
;
5846 /* Get parent pid if -S, --sig-parent is specified. */
5847 if (opt_sig_parent
) {
5851 /* Setup the kernel pipe for waking up the kernel thread */
5852 if (is_root
&& !opt_no_kernel
) {
5853 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5855 goto exit_init_data
;
5859 /* Setup the thread apps communication pipe. */
5860 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5862 goto exit_init_data
;
5865 /* Setup the thread apps notify communication pipe. */
5866 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5868 goto exit_init_data
;
5871 /* Initialize global buffer per UID and PID registry. */
5872 buffer_reg_init_uid_registry();
5873 buffer_reg_init_pid_registry();
5875 /* Init UST command queue. */
5876 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5879 * Get session list pointer. This pointer MUST NOT be free'd. This list
5880 * is statically declared in session.c
5882 session_list_ptr
= session_get_list();
5886 /* Check for the application socket timeout env variable. */
5887 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5888 if (env_app_timeout
) {
5889 app_socket_timeout
= atoi(env_app_timeout
);
5891 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5894 ret
= write_pidfile();
5896 ERR("Error in write_pidfile");
5898 goto exit_init_data
;
5900 ret
= write_agent_port();
5902 ERR("Error in write_agent_port");
5904 goto exit_init_data
;
5907 /* Initialize communication library */
5909 /* Initialize TCP timeout values */
5910 lttcomm_inet_init();
5912 if (load_session_init_data(&load_info
) < 0) {
5914 goto exit_init_data
;
5916 load_info
->path
= opt_load_session_path
;
5918 /* Create health-check thread */
5919 ret
= pthread_create(&health_thread
, NULL
,
5920 thread_manage_health
, (void *) NULL
);
5923 PERROR("pthread_create health");
5928 /* Create thread to manage the client socket */
5929 ret
= pthread_create(&client_thread
, NULL
,
5930 thread_manage_clients
, (void *) NULL
);
5933 PERROR("pthread_create clients");
5938 /* Create thread to dispatch registration */
5939 ret
= pthread_create(&dispatch_thread
, NULL
,
5940 thread_dispatch_ust_registration
, (void *) NULL
);
5943 PERROR("pthread_create dispatch");
5948 /* Create thread to manage application registration. */
5949 ret
= pthread_create(®_apps_thread
, NULL
,
5950 thread_registration_apps
, (void *) NULL
);
5953 PERROR("pthread_create registration");
5958 /* Create thread to manage application socket */
5959 ret
= pthread_create(&apps_thread
, NULL
,
5960 thread_manage_apps
, (void *) NULL
);
5963 PERROR("pthread_create apps");
5968 /* Create thread to manage application notify socket */
5969 ret
= pthread_create(&apps_notify_thread
, NULL
,
5970 ust_thread_manage_notify
, (void *) NULL
);
5973 PERROR("pthread_create notify");
5975 goto exit_apps_notify
;
5978 /* Create agent registration thread. */
5979 ret
= pthread_create(&agent_reg_thread
, NULL
,
5980 agent_thread_manage_registration
, (void *) NULL
);
5983 PERROR("pthread_create agent");
5985 goto exit_agent_reg
;
5988 /* Don't start this thread if kernel tracing is not requested nor root */
5989 if (is_root
&& !opt_no_kernel
) {
5990 /* Create kernel thread to manage kernel event */
5991 ret
= pthread_create(&kernel_thread
, NULL
,
5992 thread_manage_kernel
, (void *) NULL
);
5995 PERROR("pthread_create kernel");
6001 /* Create session loading thread. */
6002 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
6006 PERROR("pthread_create load_session_thread");
6008 goto exit_load_session
;
6012 * This is where we start awaiting program completion (e.g. through
6013 * signal that asks threads to teardown).
6016 ret
= pthread_join(load_session_thread
, &status
);
6019 PERROR("pthread_join load_session_thread");
6024 if (is_root
&& !opt_no_kernel
) {
6025 ret
= pthread_join(kernel_thread
, &status
);
6028 PERROR("pthread_join");
6034 ret
= pthread_join(agent_reg_thread
, &status
);
6037 PERROR("pthread_join agent");
6042 ret
= pthread_join(apps_notify_thread
, &status
);
6045 PERROR("pthread_join apps notify");
6050 ret
= pthread_join(apps_thread
, &status
);
6053 PERROR("pthread_join apps");
6058 ret
= pthread_join(reg_apps_thread
, &status
);
6061 PERROR("pthread_join");
6067 * Join dispatch thread after joining reg_apps_thread to ensure
6068 * we don't leak applications in the queue.
6070 ret
= pthread_join(dispatch_thread
, &status
);
6073 PERROR("pthread_join");
6078 ret
= pthread_join(client_thread
, &status
);
6081 PERROR("pthread_join");
6086 ret
= pthread_join(health_thread
, &status
);
6089 PERROR("pthread_join health thread");
6096 * sessiond_cleanup() is called when no other thread is running, except
6097 * the ht_cleanup thread, which is needed to destroy the hash tables.
6099 rcu_thread_online();
6101 rcu_thread_offline();
6102 rcu_unregister_thread();
6104 ret
= notify_thread_pipe(ht_cleanup_quit_pipe
[1]);
6106 ERR("write error on ht_cleanup quit pipe");
6110 ret
= pthread_join(ht_cleanup_thread
, &status
);
6113 PERROR("pthread_join ht cleanup thread");
6119 utils_close_pipe(ht_cleanup_pipe
);
6120 exit_ht_cleanup_pipe
:
6123 * Close the ht_cleanup quit pipe.
6125 utils_close_pipe(ht_cleanup_quit_pipe
);
6126 exit_ht_cleanup_quit_pipe
:
6128 health_app_destroy(health_sessiond
);
6129 exit_health_sessiond_cleanup
:
6132 sessiond_cleanup_options();
6134 exit_set_signal_handler
:
6135 /* Ensure all prior call_rcu are done. */