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 (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
1169 (void) lttng_read(kernel_poll_pipe
[0],
1172 * Ret value is useless here, if this pipe gets any actions an
1173 * update is required anyway.
1175 update_poll_flag
= 1;
1179 * New CPU detected by the kernel. Adding kernel stream to
1180 * kernel session and updating the kernel consumer
1182 if (revents
& LPOLLIN
) {
1183 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1189 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1190 * and unregister kernel stream at this point.
1199 lttng_poll_clean(&events
);
1202 utils_close_pipe(kernel_poll_pipe
);
1203 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1206 ERR("Health error occurred in %s", __func__
);
1207 WARN("Kernel thread died unexpectedly. "
1208 "Kernel tracing can continue but CPU hotplug is disabled.");
1210 health_unregister(health_sessiond
);
1211 DBG("Kernel thread dying");
1216 * Signal pthread condition of the consumer data that the thread.
1218 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1220 pthread_mutex_lock(&data
->cond_mutex
);
1223 * The state is set before signaling. It can be any value, it's the waiter
1224 * job to correctly interpret this condition variable associated to the
1225 * consumer pthread_cond.
1227 * A value of 0 means that the corresponding thread of the consumer data
1228 * was not started. 1 indicates that the thread has started and is ready
1229 * for action. A negative value means that there was an error during the
1232 data
->consumer_thread_is_ready
= state
;
1233 (void) pthread_cond_signal(&data
->cond
);
1235 pthread_mutex_unlock(&data
->cond_mutex
);
1239 * This thread manage the consumer error sent back to the session daemon.
1241 static void *thread_manage_consumer(void *data
)
1243 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1244 uint32_t revents
, nb_fd
;
1245 enum lttcomm_return_code code
;
1246 struct lttng_poll_event events
;
1247 struct consumer_data
*consumer_data
= data
;
1249 DBG("[thread] Manage consumer started");
1251 rcu_register_thread();
1252 rcu_thread_online();
1254 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1256 health_code_update();
1259 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1260 * metadata_sock. Nothing more will be added to this poll set.
1262 ret
= sessiond_set_thread_pollset(&events
, 3);
1268 * The error socket here is already in a listening state which was done
1269 * just before spawning this thread to avoid a race between the consumer
1270 * daemon exec trying to connect and the listen() call.
1272 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1277 health_code_update();
1279 /* Infinite blocking call, waiting for transmission */
1281 health_poll_entry();
1283 if (testpoint(sessiond_thread_manage_consumer
)) {
1287 ret
= lttng_poll_wait(&events
, -1);
1291 * Restart interrupted system call.
1293 if (errno
== EINTR
) {
1301 for (i
= 0; i
< nb_fd
; i
++) {
1302 /* Fetch once the poll data */
1303 revents
= LTTNG_POLL_GETEV(&events
, i
);
1304 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1306 health_code_update();
1309 /* No activity for this FD (poll implementation). */
1313 /* Thread quit pipe has been closed. Killing thread. */
1314 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1320 /* Event on the registration socket */
1321 if (pollfd
== consumer_data
->err_sock
) {
1322 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1323 ERR("consumer err socket poll error");
1329 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1335 * Set the CLOEXEC flag. Return code is useless because either way, the
1338 (void) utils_set_fd_cloexec(sock
);
1340 health_code_update();
1342 DBG2("Receiving code from consumer err_sock");
1344 /* Getting status code from kconsumerd */
1345 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1346 sizeof(enum lttcomm_return_code
));
1351 health_code_update();
1352 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1353 /* Connect both socket, command and metadata. */
1354 consumer_data
->cmd_sock
=
1355 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1356 consumer_data
->metadata_fd
=
1357 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1358 if (consumer_data
->cmd_sock
< 0
1359 || consumer_data
->metadata_fd
< 0) {
1360 PERROR("consumer connect cmd socket");
1361 /* On error, signal condition and quit. */
1362 signal_consumer_condition(consumer_data
, -1);
1365 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1366 /* Create metadata socket lock. */
1367 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1368 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1369 PERROR("zmalloc pthread mutex");
1373 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1375 signal_consumer_condition(consumer_data
, 1);
1376 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1377 DBG("Consumer metadata socket ready (fd: %d)",
1378 consumer_data
->metadata_fd
);
1380 ERR("consumer error when waiting for SOCK_READY : %s",
1381 lttcomm_get_readable_code(-code
));
1385 /* Remove the consumerd error sock since we've established a connexion */
1386 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1391 /* Add new accepted error socket. */
1392 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1397 /* Add metadata socket that is successfully connected. */
1398 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1399 LPOLLIN
| LPOLLRDHUP
);
1404 health_code_update();
1406 /* Infinite blocking call, waiting for transmission */
1409 health_code_update();
1411 /* Exit the thread because the thread quit pipe has been triggered. */
1413 /* Not a health error. */
1418 health_poll_entry();
1419 ret
= lttng_poll_wait(&events
, -1);
1423 * Restart interrupted system call.
1425 if (errno
== EINTR
) {
1433 for (i
= 0; i
< nb_fd
; i
++) {
1434 /* Fetch once the poll data */
1435 revents
= LTTNG_POLL_GETEV(&events
, i
);
1436 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1438 health_code_update();
1441 /* No activity for this FD (poll implementation). */
1446 * Thread quit pipe has been triggered, flag that we should stop
1447 * but continue the current loop to handle potential data from
1450 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1452 if (pollfd
== sock
) {
1453 /* Event on the consumerd socket */
1454 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1455 ERR("consumer err socket second poll error");
1458 health_code_update();
1459 /* Wait for any kconsumerd error */
1460 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1461 sizeof(enum lttcomm_return_code
));
1463 ERR("consumer closed the command socket");
1467 ERR("consumer return code : %s",
1468 lttcomm_get_readable_code(-code
));
1471 } else if (pollfd
== consumer_data
->metadata_fd
) {
1472 /* UST metadata requests */
1473 ret
= ust_consumer_metadata_request(
1474 &consumer_data
->metadata_sock
);
1476 ERR("Handling metadata request");
1480 /* No need for an else branch all FDs are tested prior. */
1482 health_code_update();
1488 * We lock here because we are about to close the sockets and some other
1489 * thread might be using them so get exclusive access which will abort all
1490 * other consumer command by other threads.
1492 pthread_mutex_lock(&consumer_data
->lock
);
1494 /* Immediately set the consumerd state to stopped */
1495 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1496 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1497 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1498 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1499 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1501 /* Code flow error... */
1505 if (consumer_data
->err_sock
>= 0) {
1506 ret
= close(consumer_data
->err_sock
);
1510 consumer_data
->err_sock
= -1;
1512 if (consumer_data
->cmd_sock
>= 0) {
1513 ret
= close(consumer_data
->cmd_sock
);
1517 consumer_data
->cmd_sock
= -1;
1519 if (consumer_data
->metadata_sock
.fd_ptr
&&
1520 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1521 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1533 unlink(consumer_data
->err_unix_sock_path
);
1534 unlink(consumer_data
->cmd_unix_sock_path
);
1535 pthread_mutex_unlock(&consumer_data
->lock
);
1537 /* Cleanup metadata socket mutex. */
1538 if (consumer_data
->metadata_sock
.lock
) {
1539 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1540 free(consumer_data
->metadata_sock
.lock
);
1542 lttng_poll_clean(&events
);
1546 ERR("Health error occurred in %s", __func__
);
1548 health_unregister(health_sessiond
);
1549 DBG("consumer thread cleanup completed");
1551 rcu_thread_offline();
1552 rcu_unregister_thread();
1558 * This thread manage application communication.
1560 static void *thread_manage_apps(void *data
)
1562 int i
, ret
, pollfd
, err
= -1;
1564 uint32_t revents
, nb_fd
;
1565 struct lttng_poll_event events
;
1567 DBG("[thread] Manage application started");
1569 rcu_register_thread();
1570 rcu_thread_online();
1572 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1574 if (testpoint(sessiond_thread_manage_apps
)) {
1575 goto error_testpoint
;
1578 health_code_update();
1580 ret
= sessiond_set_thread_pollset(&events
, 2);
1582 goto error_poll_create
;
1585 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1590 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1594 health_code_update();
1597 DBG("Apps thread polling");
1599 /* Inifinite blocking call, waiting for transmission */
1601 health_poll_entry();
1602 ret
= lttng_poll_wait(&events
, -1);
1603 DBG("Apps thread return from poll on %d fds",
1604 LTTNG_POLL_GETNB(&events
));
1608 * Restart interrupted system call.
1610 if (errno
== EINTR
) {
1618 for (i
= 0; i
< nb_fd
; i
++) {
1619 /* Fetch once the poll data */
1620 revents
= LTTNG_POLL_GETEV(&events
, i
);
1621 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1623 health_code_update();
1626 /* No activity for this FD (poll implementation). */
1630 /* Thread quit pipe has been closed. Killing thread. */
1631 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1637 /* Inspect the apps cmd pipe */
1638 if (pollfd
== apps_cmd_pipe
[0]) {
1639 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1640 ERR("Apps command pipe error");
1642 } else if (revents
& LPOLLIN
) {
1646 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1647 if (size_ret
< sizeof(sock
)) {
1648 PERROR("read apps cmd pipe");
1652 health_code_update();
1655 * We only monitor the error events of the socket. This
1656 * thread does not handle any incoming data from UST
1659 ret
= lttng_poll_add(&events
, sock
,
1660 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1665 DBG("Apps with sock %d added to poll set", sock
);
1669 * At this point, we know that a registered application made
1670 * the event at poll_wait.
1672 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1673 /* Removing from the poll set */
1674 ret
= lttng_poll_del(&events
, pollfd
);
1679 /* Socket closed on remote end. */
1680 ust_app_unregister(pollfd
);
1684 health_code_update();
1690 lttng_poll_clean(&events
);
1693 utils_close_pipe(apps_cmd_pipe
);
1694 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1697 * We don't clean the UST app hash table here since already registered
1698 * applications can still be controlled so let them be until the session
1699 * daemon dies or the applications stop.
1704 ERR("Health error occurred in %s", __func__
);
1706 health_unregister(health_sessiond
);
1707 DBG("Application communication apps thread cleanup complete");
1708 rcu_thread_offline();
1709 rcu_unregister_thread();
1714 * Send a socket to a thread This is called from the dispatch UST registration
1715 * thread once all sockets are set for the application.
1717 * The sock value can be invalid, we don't really care, the thread will handle
1718 * it and make the necessary cleanup if so.
1720 * On success, return 0 else a negative value being the errno message of the
1723 static int send_socket_to_thread(int fd
, int sock
)
1728 * It's possible that the FD is set as invalid with -1 concurrently just
1729 * before calling this function being a shutdown state of the thread.
1736 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1737 if (ret
< sizeof(sock
)) {
1738 PERROR("write apps pipe %d", fd
);
1745 /* All good. Don't send back the write positive ret value. */
1752 * Sanitize the wait queue of the dispatch registration thread meaning removing
1753 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1754 * notify socket is never received.
1756 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1758 int ret
, nb_fd
= 0, i
;
1759 unsigned int fd_added
= 0;
1760 struct lttng_poll_event events
;
1761 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1765 lttng_poll_init(&events
);
1767 /* Just skip everything for an empty queue. */
1768 if (!wait_queue
->count
) {
1772 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1777 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1778 &wait_queue
->head
, head
) {
1779 assert(wait_node
->app
);
1780 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1781 LPOLLHUP
| LPOLLERR
);
1794 * Poll but don't block so we can quickly identify the faulty events and
1795 * clean them afterwards from the wait queue.
1797 ret
= lttng_poll_wait(&events
, 0);
1803 for (i
= 0; i
< nb_fd
; i
++) {
1804 /* Get faulty FD. */
1805 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1806 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1809 /* No activity for this FD (poll implementation). */
1813 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1814 &wait_queue
->head
, head
) {
1815 if (pollfd
== wait_node
->app
->sock
&&
1816 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1817 cds_list_del(&wait_node
->head
);
1818 wait_queue
->count
--;
1819 ust_app_destroy(wait_node
->app
);
1822 * Silence warning of use-after-free in
1823 * cds_list_for_each_entry_safe which uses
1824 * __typeof__(*wait_node).
1833 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1837 lttng_poll_clean(&events
);
1841 lttng_poll_clean(&events
);
1843 ERR("Unable to sanitize wait queue");
1848 * Dispatch request from the registration threads to the application
1849 * communication thread.
1851 static void *thread_dispatch_ust_registration(void *data
)
1854 struct cds_wfcq_node
*node
;
1855 struct ust_command
*ust_cmd
= NULL
;
1856 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1857 struct ust_reg_wait_queue wait_queue
= {
1861 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1863 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1864 goto error_testpoint
;
1867 health_code_update();
1869 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1871 DBG("[thread] Dispatch UST command started");
1873 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1874 health_code_update();
1876 /* Atomically prepare the queue futex */
1877 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1880 struct ust_app
*app
= NULL
;
1884 * Make sure we don't have node(s) that have hung up before receiving
1885 * the notify socket. This is to clean the list in order to avoid
1886 * memory leaks from notify socket that are never seen.
1888 sanitize_wait_queue(&wait_queue
);
1890 health_code_update();
1891 /* Dequeue command for registration */
1892 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1894 DBG("Woken up but nothing in the UST command queue");
1895 /* Continue thread execution */
1899 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1901 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1902 " gid:%d sock:%d name:%s (version %d.%d)",
1903 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1904 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1905 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1906 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1908 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1909 wait_node
= zmalloc(sizeof(*wait_node
));
1911 PERROR("zmalloc wait_node dispatch");
1912 ret
= close(ust_cmd
->sock
);
1914 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1916 lttng_fd_put(LTTNG_FD_APPS
, 1);
1920 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1922 /* Create application object if socket is CMD. */
1923 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1925 if (!wait_node
->app
) {
1926 ret
= close(ust_cmd
->sock
);
1928 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1930 lttng_fd_put(LTTNG_FD_APPS
, 1);
1936 * Add application to the wait queue so we can set the notify
1937 * socket before putting this object in the global ht.
1939 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1944 * We have to continue here since we don't have the notify
1945 * socket and the application MUST be added to the hash table
1946 * only at that moment.
1951 * Look for the application in the local wait queue and set the
1952 * notify socket if found.
1954 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1955 &wait_queue
.head
, head
) {
1956 health_code_update();
1957 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1958 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1959 cds_list_del(&wait_node
->head
);
1961 app
= wait_node
->app
;
1963 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1969 * With no application at this stage the received socket is
1970 * basically useless so close it before we free the cmd data
1971 * structure for good.
1974 ret
= close(ust_cmd
->sock
);
1976 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1978 lttng_fd_put(LTTNG_FD_APPS
, 1);
1985 * @session_lock_list
1987 * Lock the global session list so from the register up to the
1988 * registration done message, no thread can see the application
1989 * and change its state.
1991 session_lock_list();
1995 * Add application to the global hash table. This needs to be
1996 * done before the update to the UST registry can locate the
2001 /* Set app version. This call will print an error if needed. */
2002 (void) ust_app_version(app
);
2004 /* Send notify socket through the notify pipe. */
2005 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
2009 session_unlock_list();
2011 * No notify thread, stop the UST tracing. However, this is
2012 * not an internal error of the this thread thus setting
2013 * the health error code to a normal exit.
2020 * Update newly registered application with the tracing
2021 * registry info already enabled information.
2023 update_ust_app(app
->sock
);
2026 * Don't care about return value. Let the manage apps threads
2027 * handle app unregistration upon socket close.
2029 (void) ust_app_register_done(app
);
2032 * Even if the application socket has been closed, send the app
2033 * to the thread and unregistration will take place at that
2036 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2039 session_unlock_list();
2041 * No apps. thread, stop the UST tracing. However, this is
2042 * not an internal error of the this thread thus setting
2043 * the health error code to a normal exit.
2050 session_unlock_list();
2052 } while (node
!= NULL
);
2054 health_poll_entry();
2055 /* Futex wait on queue. Blocking call on futex() */
2056 futex_nto1_wait(&ust_cmd_queue
.futex
);
2059 /* Normal exit, no error */
2063 /* Clean up wait queue. */
2064 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2065 &wait_queue
.head
, head
) {
2066 cds_list_del(&wait_node
->head
);
2071 /* Empty command queue. */
2073 /* Dequeue command for registration */
2074 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2078 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2079 ret
= close(ust_cmd
->sock
);
2081 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2083 lttng_fd_put(LTTNG_FD_APPS
, 1);
2088 DBG("Dispatch thread dying");
2091 ERR("Health error occurred in %s", __func__
);
2093 health_unregister(health_sessiond
);
2098 * This thread manage application registration.
2100 static void *thread_registration_apps(void *data
)
2102 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2103 uint32_t revents
, nb_fd
;
2104 struct lttng_poll_event events
;
2106 * Get allocated in this thread, enqueued to a global queue, dequeued and
2107 * freed in the manage apps thread.
2109 struct ust_command
*ust_cmd
= NULL
;
2111 DBG("[thread] Manage application registration started");
2113 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2115 if (testpoint(sessiond_thread_registration_apps
)) {
2116 goto error_testpoint
;
2119 ret
= lttcomm_listen_unix_sock(apps_sock
);
2125 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2126 * more will be added to this poll set.
2128 ret
= sessiond_set_thread_pollset(&events
, 2);
2130 goto error_create_poll
;
2133 /* Add the application registration socket */
2134 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2136 goto error_poll_add
;
2139 /* Notify all applications to register */
2140 ret
= notify_ust_apps(1);
2142 ERR("Failed to notify applications or create the wait shared memory.\n"
2143 "Execution continues but there might be problem for already\n"
2144 "running applications that wishes to register.");
2148 DBG("Accepting application registration");
2150 /* Inifinite blocking call, waiting for transmission */
2152 health_poll_entry();
2153 ret
= lttng_poll_wait(&events
, -1);
2157 * Restart interrupted system call.
2159 if (errno
== EINTR
) {
2167 for (i
= 0; i
< nb_fd
; i
++) {
2168 health_code_update();
2170 /* Fetch once the poll data */
2171 revents
= LTTNG_POLL_GETEV(&events
, i
);
2172 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2175 /* No activity for this FD (poll implementation). */
2179 /* Thread quit pipe has been closed. Killing thread. */
2180 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2186 /* Event on the registration socket */
2187 if (pollfd
== apps_sock
) {
2188 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2189 ERR("Register apps socket poll error");
2191 } else if (revents
& LPOLLIN
) {
2192 sock
= lttcomm_accept_unix_sock(apps_sock
);
2198 * Set socket timeout for both receiving and ending.
2199 * app_socket_timeout is in seconds, whereas
2200 * lttcomm_setsockopt_rcv_timeout and
2201 * lttcomm_setsockopt_snd_timeout expect msec as
2204 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2205 app_socket_timeout
* 1000);
2206 (void) lttcomm_setsockopt_snd_timeout(sock
,
2207 app_socket_timeout
* 1000);
2210 * Set the CLOEXEC flag. Return code is useless because
2211 * either way, the show must go on.
2213 (void) utils_set_fd_cloexec(sock
);
2215 /* Create UST registration command for enqueuing */
2216 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2217 if (ust_cmd
== NULL
) {
2218 PERROR("ust command zmalloc");
2227 * Using message-based transmissions to ensure we don't
2228 * have to deal with partially received messages.
2230 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2232 ERR("Exhausted file descriptors allowed for applications.");
2242 health_code_update();
2243 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2246 /* Close socket of the application. */
2251 lttng_fd_put(LTTNG_FD_APPS
, 1);
2255 health_code_update();
2257 ust_cmd
->sock
= sock
;
2260 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2261 " gid:%d sock:%d name:%s (version %d.%d)",
2262 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2263 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2264 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2265 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2268 * Lock free enqueue the registration request. The red pill
2269 * has been taken! This apps will be part of the *system*.
2271 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2274 * Wake the registration queue futex. Implicit memory
2275 * barrier with the exchange in cds_wfcq_enqueue.
2277 futex_nto1_wake(&ust_cmd_queue
.futex
);
2285 /* Notify that the registration thread is gone */
2288 if (apps_sock
>= 0) {
2289 ret
= close(apps_sock
);
2299 lttng_fd_put(LTTNG_FD_APPS
, 1);
2301 unlink(apps_unix_sock_path
);
2304 lttng_poll_clean(&events
);
2308 DBG("UST Registration thread cleanup complete");
2311 ERR("Health error occurred in %s", __func__
);
2313 health_unregister(health_sessiond
);
2319 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2320 * exec or it will fails.
2322 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2325 struct timespec timeout
;
2327 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2328 consumer_data
->consumer_thread_is_ready
= 0;
2330 /* Setup pthread condition */
2331 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2334 PERROR("pthread_condattr_init consumer data");
2339 * Set the monotonic clock in order to make sure we DO NOT jump in time
2340 * between the clock_gettime() call and the timedwait call. See bug #324
2341 * for a more details and how we noticed it.
2343 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2346 PERROR("pthread_condattr_setclock consumer data");
2350 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2353 PERROR("pthread_cond_init consumer data");
2357 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2361 PERROR("pthread_create consumer");
2366 /* We are about to wait on a pthread condition */
2367 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2369 /* Get time for sem_timedwait absolute timeout */
2370 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2372 * Set the timeout for the condition timed wait even if the clock gettime
2373 * call fails since we might loop on that call and we want to avoid to
2374 * increment the timeout too many times.
2376 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2379 * The following loop COULD be skipped in some conditions so this is why we
2380 * set ret to 0 in order to make sure at least one round of the loop is
2386 * Loop until the condition is reached or when a timeout is reached. Note
2387 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2388 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2389 * possible. This loop does not take any chances and works with both of
2392 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2393 if (clock_ret
< 0) {
2394 PERROR("clock_gettime spawn consumer");
2395 /* Infinite wait for the consumerd thread to be ready */
2396 ret
= pthread_cond_wait(&consumer_data
->cond
,
2397 &consumer_data
->cond_mutex
);
2399 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2400 &consumer_data
->cond_mutex
, &timeout
);
2404 /* Release the pthread condition */
2405 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2409 if (ret
== ETIMEDOUT
) {
2413 * Call has timed out so we kill the kconsumerd_thread and return
2416 ERR("Condition timed out. The consumer thread was never ready."
2418 pth_ret
= pthread_cancel(consumer_data
->thread
);
2420 PERROR("pthread_cancel consumer thread");
2423 PERROR("pthread_cond_wait failed consumer thread");
2425 /* Caller is expecting a negative value on failure. */
2430 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2431 if (consumer_data
->pid
== 0) {
2432 ERR("Consumerd did not start");
2433 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2436 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2445 * Join consumer thread
2447 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2451 /* Consumer pid must be a real one. */
2452 if (consumer_data
->pid
> 0) {
2454 ret
= kill(consumer_data
->pid
, SIGTERM
);
2456 PERROR("Error killing consumer daemon");
2459 return pthread_join(consumer_data
->thread
, &status
);
2466 * Fork and exec a consumer daemon (consumerd).
2468 * Return pid if successful else -1.
2470 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2474 const char *consumer_to_use
;
2475 const char *verbosity
;
2478 DBG("Spawning consumerd");
2485 if (opt_verbose_consumer
) {
2486 verbosity
= "--verbose";
2487 } else if (lttng_opt_quiet
) {
2488 verbosity
= "--quiet";
2493 switch (consumer_data
->type
) {
2494 case LTTNG_CONSUMER_KERNEL
:
2496 * Find out which consumerd to execute. We will first try the
2497 * 64-bit path, then the sessiond's installation directory, and
2498 * fallback on the 32-bit one,
2500 DBG3("Looking for a kernel consumer at these locations:");
2501 DBG3(" 1) %s", consumerd64_bin
);
2502 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2503 DBG3(" 3) %s", consumerd32_bin
);
2504 if (stat(consumerd64_bin
, &st
) == 0) {
2505 DBG3("Found location #1");
2506 consumer_to_use
= consumerd64_bin
;
2507 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2508 DBG3("Found location #2");
2509 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2510 } else if (stat(consumerd32_bin
, &st
) == 0) {
2511 DBG3("Found location #3");
2512 consumer_to_use
= consumerd32_bin
;
2514 DBG("Could not find any valid consumerd executable");
2518 DBG("Using kernel consumer at: %s", consumer_to_use
);
2519 ret
= execl(consumer_to_use
,
2520 "lttng-consumerd", verbosity
, "-k",
2521 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2522 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2523 "--group", tracing_group_name
,
2526 case LTTNG_CONSUMER64_UST
:
2528 char *tmpnew
= NULL
;
2530 if (consumerd64_libdir
[0] != '\0') {
2534 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2538 tmplen
= strlen("LD_LIBRARY_PATH=")
2539 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2540 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2545 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2546 strcat(tmpnew
, consumerd64_libdir
);
2547 if (tmp
[0] != '\0') {
2548 strcat(tmpnew
, ":");
2549 strcat(tmpnew
, tmp
);
2551 ret
= putenv(tmpnew
);
2558 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2559 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2560 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2561 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2562 "--group", tracing_group_name
,
2564 if (consumerd64_libdir
[0] != '\0') {
2569 case LTTNG_CONSUMER32_UST
:
2571 char *tmpnew
= NULL
;
2573 if (consumerd32_libdir
[0] != '\0') {
2577 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2581 tmplen
= strlen("LD_LIBRARY_PATH=")
2582 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2583 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2588 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2589 strcat(tmpnew
, consumerd32_libdir
);
2590 if (tmp
[0] != '\0') {
2591 strcat(tmpnew
, ":");
2592 strcat(tmpnew
, tmp
);
2594 ret
= putenv(tmpnew
);
2601 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2602 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2603 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2604 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2605 "--group", tracing_group_name
,
2607 if (consumerd32_libdir
[0] != '\0') {
2613 PERROR("unknown consumer type");
2617 PERROR("Consumer execl()");
2619 /* Reaching this point, we got a failure on our execl(). */
2621 } else if (pid
> 0) {
2624 PERROR("start consumer fork");
2632 * Spawn the consumerd daemon and session daemon thread.
2634 static int start_consumerd(struct consumer_data
*consumer_data
)
2639 * Set the listen() state on the socket since there is a possible race
2640 * between the exec() of the consumer daemon and this call if place in the
2641 * consumer thread. See bug #366 for more details.
2643 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2648 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2649 if (consumer_data
->pid
!= 0) {
2650 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2654 ret
= spawn_consumerd(consumer_data
);
2656 ERR("Spawning consumerd failed");
2657 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2661 /* Setting up the consumer_data pid */
2662 consumer_data
->pid
= ret
;
2663 DBG2("Consumer pid %d", consumer_data
->pid
);
2664 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2666 DBG2("Spawning consumer control thread");
2667 ret
= spawn_consumer_thread(consumer_data
);
2669 ERR("Fatal error spawning consumer control thread");
2677 /* Cleanup already created sockets on error. */
2678 if (consumer_data
->err_sock
>= 0) {
2681 err
= close(consumer_data
->err_sock
);
2683 PERROR("close consumer data error socket");
2690 * Setup necessary data for kernel tracer action.
2692 static int init_kernel_tracer(void)
2696 /* Modprobe lttng kernel modules */
2697 ret
= modprobe_lttng_control();
2702 /* Open debugfs lttng */
2703 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2704 if (kernel_tracer_fd
< 0) {
2705 DBG("Failed to open %s", module_proc_lttng
);
2710 /* Validate kernel version */
2711 ret
= kernel_validate_version(kernel_tracer_fd
);
2716 ret
= modprobe_lttng_data();
2721 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2725 modprobe_remove_lttng_control();
2726 ret
= close(kernel_tracer_fd
);
2730 kernel_tracer_fd
= -1;
2731 return LTTNG_ERR_KERN_VERSION
;
2734 ret
= close(kernel_tracer_fd
);
2740 modprobe_remove_lttng_control();
2743 WARN("No kernel tracer available");
2744 kernel_tracer_fd
= -1;
2746 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2748 return LTTNG_ERR_KERN_NA
;
2754 * Copy consumer output from the tracing session to the domain session. The
2755 * function also applies the right modification on a per domain basis for the
2756 * trace files destination directory.
2758 * Should *NOT* be called with RCU read-side lock held.
2760 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2763 const char *dir_name
;
2764 struct consumer_output
*consumer
;
2767 assert(session
->consumer
);
2770 case LTTNG_DOMAIN_KERNEL
:
2771 DBG3("Copying tracing session consumer output in kernel session");
2773 * XXX: We should audit the session creation and what this function
2774 * does "extra" in order to avoid a destroy since this function is used
2775 * in the domain session creation (kernel and ust) only. Same for UST
2778 if (session
->kernel_session
->consumer
) {
2779 consumer_output_put(session
->kernel_session
->consumer
);
2781 session
->kernel_session
->consumer
=
2782 consumer_copy_output(session
->consumer
);
2783 /* Ease our life a bit for the next part */
2784 consumer
= session
->kernel_session
->consumer
;
2785 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2787 case LTTNG_DOMAIN_JUL
:
2788 case LTTNG_DOMAIN_LOG4J
:
2789 case LTTNG_DOMAIN_PYTHON
:
2790 case LTTNG_DOMAIN_UST
:
2791 DBG3("Copying tracing session consumer output in UST session");
2792 if (session
->ust_session
->consumer
) {
2793 consumer_output_put(session
->ust_session
->consumer
);
2795 session
->ust_session
->consumer
=
2796 consumer_copy_output(session
->consumer
);
2797 /* Ease our life a bit for the next part */
2798 consumer
= session
->ust_session
->consumer
;
2799 dir_name
= DEFAULT_UST_TRACE_DIR
;
2802 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2806 /* Append correct directory to subdir */
2807 strncat(consumer
->subdir
, dir_name
,
2808 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2809 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2818 * Create an UST session and add it to the session ust list.
2820 * Should *NOT* be called with RCU read-side lock held.
2822 static int create_ust_session(struct ltt_session
*session
,
2823 struct lttng_domain
*domain
)
2826 struct ltt_ust_session
*lus
= NULL
;
2830 assert(session
->consumer
);
2832 switch (domain
->type
) {
2833 case LTTNG_DOMAIN_JUL
:
2834 case LTTNG_DOMAIN_LOG4J
:
2835 case LTTNG_DOMAIN_PYTHON
:
2836 case LTTNG_DOMAIN_UST
:
2839 ERR("Unknown UST domain on create session %d", domain
->type
);
2840 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2844 DBG("Creating UST session");
2846 lus
= trace_ust_create_session(session
->id
);
2848 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2852 lus
->uid
= session
->uid
;
2853 lus
->gid
= session
->gid
;
2854 lus
->output_traces
= session
->output_traces
;
2855 lus
->snapshot_mode
= session
->snapshot_mode
;
2856 lus
->live_timer_interval
= session
->live_timer
;
2857 session
->ust_session
= lus
;
2858 if (session
->shm_path
[0]) {
2859 strncpy(lus
->root_shm_path
, session
->shm_path
,
2860 sizeof(lus
->root_shm_path
));
2861 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2862 strncpy(lus
->shm_path
, session
->shm_path
,
2863 sizeof(lus
->shm_path
));
2864 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2865 strncat(lus
->shm_path
, "/ust",
2866 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2868 /* Copy session output to the newly created UST session */
2869 ret
= copy_session_consumer(domain
->type
, session
);
2870 if (ret
!= LTTNG_OK
) {
2878 session
->ust_session
= NULL
;
2883 * Create a kernel tracer session then create the default channel.
2885 static int create_kernel_session(struct ltt_session
*session
)
2889 DBG("Creating kernel session");
2891 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2893 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2897 /* Code flow safety */
2898 assert(session
->kernel_session
);
2900 /* Copy session output to the newly created Kernel session */
2901 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2902 if (ret
!= LTTNG_OK
) {
2906 /* Create directory(ies) on local filesystem. */
2907 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2908 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2909 ret
= run_as_mkdir_recursive(
2910 session
->kernel_session
->consumer
->dst
.trace_path
,
2911 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2913 if (errno
!= EEXIST
) {
2914 ERR("Trace directory creation error");
2920 session
->kernel_session
->uid
= session
->uid
;
2921 session
->kernel_session
->gid
= session
->gid
;
2922 session
->kernel_session
->output_traces
= session
->output_traces
;
2923 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2928 trace_kernel_destroy_session(session
->kernel_session
);
2929 session
->kernel_session
= NULL
;
2934 * Count number of session permitted by uid/gid.
2936 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2939 struct ltt_session
*session
;
2941 DBG("Counting number of available session for UID %d GID %d",
2943 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2945 * Only list the sessions the user can control.
2947 if (!session_access_ok(session
, uid
, gid
)) {
2956 * Process the command requested by the lttng client within the command
2957 * context structure. This function make sure that the return structure (llm)
2958 * is set and ready for transmission before returning.
2960 * Return any error encountered or 0 for success.
2962 * "sock" is only used for special-case var. len data.
2964 * Should *NOT* be called with RCU read-side lock held.
2966 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2970 int need_tracing_session
= 1;
2973 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2977 switch (cmd_ctx
->lsm
->cmd_type
) {
2978 case LTTNG_CREATE_SESSION
:
2979 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2980 case LTTNG_CREATE_SESSION_LIVE
:
2981 case LTTNG_DESTROY_SESSION
:
2982 case LTTNG_LIST_SESSIONS
:
2983 case LTTNG_LIST_DOMAINS
:
2984 case LTTNG_START_TRACE
:
2985 case LTTNG_STOP_TRACE
:
2986 case LTTNG_DATA_PENDING
:
2987 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2988 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2989 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2990 case LTTNG_SNAPSHOT_RECORD
:
2991 case LTTNG_SAVE_SESSION
:
2992 case LTTNG_SET_SESSION_SHM_PATH
:
2999 if (opt_no_kernel
&& need_domain
3000 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3002 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3004 ret
= LTTNG_ERR_KERN_NA
;
3009 /* Deny register consumer if we already have a spawned consumer. */
3010 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3011 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3012 if (kconsumer_data
.pid
> 0) {
3013 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3014 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3017 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3021 * Check for command that don't needs to allocate a returned payload. We do
3022 * this here so we don't have to make the call for no payload at each
3025 switch(cmd_ctx
->lsm
->cmd_type
) {
3026 case LTTNG_LIST_SESSIONS
:
3027 case LTTNG_LIST_TRACEPOINTS
:
3028 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3029 case LTTNG_LIST_DOMAINS
:
3030 case LTTNG_LIST_CHANNELS
:
3031 case LTTNG_LIST_EVENTS
:
3032 case LTTNG_LIST_SYSCALLS
:
3033 case LTTNG_LIST_TRACKER_PIDS
:
3036 /* Setup lttng message with no payload */
3037 ret
= setup_lttng_msg(cmd_ctx
, 0);
3039 /* This label does not try to unlock the session */
3040 goto init_setup_error
;
3044 /* Commands that DO NOT need a session. */
3045 switch (cmd_ctx
->lsm
->cmd_type
) {
3046 case LTTNG_CREATE_SESSION
:
3047 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3048 case LTTNG_CREATE_SESSION_LIVE
:
3049 case LTTNG_CALIBRATE
:
3050 case LTTNG_LIST_SESSIONS
:
3051 case LTTNG_LIST_TRACEPOINTS
:
3052 case LTTNG_LIST_SYSCALLS
:
3053 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3054 case LTTNG_SAVE_SESSION
:
3055 need_tracing_session
= 0;
3058 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3060 * We keep the session list lock across _all_ commands
3061 * for now, because the per-session lock does not
3062 * handle teardown properly.
3064 session_lock_list();
3065 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3066 if (cmd_ctx
->session
== NULL
) {
3067 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3070 /* Acquire lock for the session */
3071 session_lock(cmd_ctx
->session
);
3077 * Commands that need a valid session but should NOT create one if none
3078 * exists. Instead of creating one and destroying it when the command is
3079 * handled, process that right before so we save some round trip in useless
3082 switch (cmd_ctx
->lsm
->cmd_type
) {
3083 case LTTNG_DISABLE_CHANNEL
:
3084 case LTTNG_DISABLE_EVENT
:
3085 switch (cmd_ctx
->lsm
->domain
.type
) {
3086 case LTTNG_DOMAIN_KERNEL
:
3087 if (!cmd_ctx
->session
->kernel_session
) {
3088 ret
= LTTNG_ERR_NO_CHANNEL
;
3092 case LTTNG_DOMAIN_JUL
:
3093 case LTTNG_DOMAIN_LOG4J
:
3094 case LTTNG_DOMAIN_PYTHON
:
3095 case LTTNG_DOMAIN_UST
:
3096 if (!cmd_ctx
->session
->ust_session
) {
3097 ret
= LTTNG_ERR_NO_CHANNEL
;
3102 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3114 * Check domain type for specific "pre-action".
3116 switch (cmd_ctx
->lsm
->domain
.type
) {
3117 case LTTNG_DOMAIN_KERNEL
:
3119 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3123 /* Kernel tracer check */
3124 if (kernel_tracer_fd
== -1) {
3125 /* Basically, load kernel tracer modules */
3126 ret
= init_kernel_tracer();
3132 /* Consumer is in an ERROR state. Report back to client */
3133 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3134 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3138 /* Need a session for kernel command */
3139 if (need_tracing_session
) {
3140 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3141 ret
= create_kernel_session(cmd_ctx
->session
);
3143 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3148 /* Start the kernel consumer daemon */
3149 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3150 if (kconsumer_data
.pid
== 0 &&
3151 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3152 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3153 ret
= start_consumerd(&kconsumer_data
);
3155 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3158 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3160 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3164 * The consumer was just spawned so we need to add the socket to
3165 * the consumer output of the session if exist.
3167 ret
= consumer_create_socket(&kconsumer_data
,
3168 cmd_ctx
->session
->kernel_session
->consumer
);
3175 case LTTNG_DOMAIN_JUL
:
3176 case LTTNG_DOMAIN_LOG4J
:
3177 case LTTNG_DOMAIN_PYTHON
:
3178 case LTTNG_DOMAIN_UST
:
3180 if (!ust_app_supported()) {
3181 ret
= LTTNG_ERR_NO_UST
;
3184 /* Consumer is in an ERROR state. Report back to client */
3185 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3186 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3190 if (need_tracing_session
) {
3191 /* Create UST session if none exist. */
3192 if (cmd_ctx
->session
->ust_session
== NULL
) {
3193 ret
= create_ust_session(cmd_ctx
->session
,
3194 &cmd_ctx
->lsm
->domain
);
3195 if (ret
!= LTTNG_OK
) {
3200 /* Start the UST consumer daemons */
3202 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3203 if (consumerd64_bin
[0] != '\0' &&
3204 ustconsumer64_data
.pid
== 0 &&
3205 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3206 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3207 ret
= start_consumerd(&ustconsumer64_data
);
3209 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3210 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3214 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3215 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3217 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3221 * Setup socket for consumer 64 bit. No need for atomic access
3222 * since it was set above and can ONLY be set in this thread.
3224 ret
= consumer_create_socket(&ustconsumer64_data
,
3225 cmd_ctx
->session
->ust_session
->consumer
);
3231 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3232 if (consumerd32_bin
[0] != '\0' &&
3233 ustconsumer32_data
.pid
== 0 &&
3234 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3235 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3236 ret
= start_consumerd(&ustconsumer32_data
);
3238 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3239 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3243 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3244 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3246 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3250 * Setup socket for consumer 64 bit. No need for atomic access
3251 * since it was set above and can ONLY be set in this thread.
3253 ret
= consumer_create_socket(&ustconsumer32_data
,
3254 cmd_ctx
->session
->ust_session
->consumer
);
3266 /* Validate consumer daemon state when start/stop trace command */
3267 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3268 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3269 switch (cmd_ctx
->lsm
->domain
.type
) {
3270 case LTTNG_DOMAIN_JUL
:
3271 case LTTNG_DOMAIN_LOG4J
:
3272 case LTTNG_DOMAIN_PYTHON
:
3273 case LTTNG_DOMAIN_UST
:
3274 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3275 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3279 case LTTNG_DOMAIN_KERNEL
:
3280 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3281 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3289 * Check that the UID or GID match that of the tracing session.
3290 * The root user can interact with all sessions.
3292 if (need_tracing_session
) {
3293 if (!session_access_ok(cmd_ctx
->session
,
3294 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3295 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3296 ret
= LTTNG_ERR_EPERM
;
3302 * Send relayd information to consumer as soon as we have a domain and a
3305 if (cmd_ctx
->session
&& need_domain
) {
3307 * Setup relayd if not done yet. If the relayd information was already
3308 * sent to the consumer, this call will gracefully return.
3310 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3311 if (ret
!= LTTNG_OK
) {
3316 /* Process by command type */
3317 switch (cmd_ctx
->lsm
->cmd_type
) {
3318 case LTTNG_ADD_CONTEXT
:
3320 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3321 cmd_ctx
->lsm
->u
.context
.channel_name
,
3322 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3325 case LTTNG_DISABLE_CHANNEL
:
3327 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3328 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3331 case LTTNG_DISABLE_EVENT
:
3335 * FIXME: handle filter; for now we just receive the filter's
3336 * bytecode along with the filter expression which are sent by
3337 * liblttng-ctl and discard them.
3339 * This fixes an issue where the client may block while sending
3340 * the filter payload and encounter an error because the session
3341 * daemon closes the socket without ever handling this data.
3343 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3344 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3347 char data
[LTTNG_FILTER_MAX_LEN
];
3349 DBG("Discarding disable event command payload of size %zu", count
);
3351 ret
= lttcomm_recv_unix_sock(sock
, data
,
3352 count
> sizeof(data
) ? sizeof(data
) : count
);
3357 count
-= (size_t) ret
;
3360 /* FIXME: passing packed structure to non-packed pointer */
3361 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3362 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3363 &cmd_ctx
->lsm
->u
.disable
.event
);
3366 case LTTNG_ENABLE_CHANNEL
:
3368 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3369 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3372 case LTTNG_TRACK_PID
:
3374 ret
= cmd_track_pid(cmd_ctx
->session
,
3375 cmd_ctx
->lsm
->domain
.type
,
3376 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3379 case LTTNG_UNTRACK_PID
:
3381 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3382 cmd_ctx
->lsm
->domain
.type
,
3383 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3386 case LTTNG_ENABLE_EVENT
:
3388 struct lttng_event_exclusion
*exclusion
= NULL
;
3389 struct lttng_filter_bytecode
*bytecode
= NULL
;
3390 char *filter_expression
= NULL
;
3392 /* Handle exclusion events and receive it from the client. */
3393 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3394 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3396 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3397 (count
* LTTNG_SYMBOL_NAME_LEN
));
3399 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3403 DBG("Receiving var len exclusion event list from client ...");
3404 exclusion
->count
= count
;
3405 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3406 count
* LTTNG_SYMBOL_NAME_LEN
);
3408 DBG("Nothing recv() from client var len data... continuing");
3411 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3416 /* Get filter expression from client. */
3417 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3418 size_t expression_len
=
3419 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3421 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3422 ret
= LTTNG_ERR_FILTER_INVAL
;
3427 filter_expression
= zmalloc(expression_len
);
3428 if (!filter_expression
) {
3430 ret
= LTTNG_ERR_FILTER_NOMEM
;
3434 /* Receive var. len. data */
3435 DBG("Receiving var len filter's expression from client ...");
3436 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3439 DBG("Nothing recv() from client car len data... continuing");
3441 free(filter_expression
);
3443 ret
= LTTNG_ERR_FILTER_INVAL
;
3448 /* Handle filter and get bytecode from client. */
3449 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3450 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3452 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3453 ret
= LTTNG_ERR_FILTER_INVAL
;
3454 free(filter_expression
);
3459 bytecode
= zmalloc(bytecode_len
);
3461 free(filter_expression
);
3463 ret
= LTTNG_ERR_FILTER_NOMEM
;
3467 /* Receive var. len. data */
3468 DBG("Receiving var len filter's bytecode from client ...");
3469 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3471 DBG("Nothing recv() from client car len data... continuing");
3473 free(filter_expression
);
3476 ret
= LTTNG_ERR_FILTER_INVAL
;
3480 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3481 free(filter_expression
);
3484 ret
= LTTNG_ERR_FILTER_INVAL
;
3489 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3490 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3491 &cmd_ctx
->lsm
->u
.enable
.event
,
3492 filter_expression
, bytecode
, exclusion
,
3493 kernel_poll_pipe
[1]);
3496 case LTTNG_LIST_TRACEPOINTS
:
3498 struct lttng_event
*events
;
3501 session_lock_list();
3502 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3503 session_unlock_list();
3504 if (nb_events
< 0) {
3505 /* Return value is a negative lttng_error_code. */
3511 * Setup lttng message with payload size set to the event list size in
3512 * bytes and then copy list into the llm payload.
3514 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3520 /* Copy event list into message payload */
3521 memcpy(cmd_ctx
->llm
->payload
, events
,
3522 sizeof(struct lttng_event
) * nb_events
);
3529 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3531 struct lttng_event_field
*fields
;
3534 session_lock_list();
3535 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3537 session_unlock_list();
3538 if (nb_fields
< 0) {
3539 /* Return value is a negative lttng_error_code. */
3545 * Setup lttng message with payload size set to the event list size in
3546 * bytes and then copy list into the llm payload.
3548 ret
= setup_lttng_msg(cmd_ctx
,
3549 sizeof(struct lttng_event_field
) * nb_fields
);
3555 /* Copy event list into message payload */
3556 memcpy(cmd_ctx
->llm
->payload
, fields
,
3557 sizeof(struct lttng_event_field
) * nb_fields
);
3564 case LTTNG_LIST_SYSCALLS
:
3566 struct lttng_event
*events
;
3569 nb_events
= cmd_list_syscalls(&events
);
3570 if (nb_events
< 0) {
3571 /* Return value is a negative lttng_error_code. */
3577 * Setup lttng message with payload size set to the event list size in
3578 * bytes and then copy list into the llm payload.
3580 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3586 /* Copy event list into message payload */
3587 memcpy(cmd_ctx
->llm
->payload
, events
,
3588 sizeof(struct lttng_event
) * nb_events
);
3595 case LTTNG_LIST_TRACKER_PIDS
:
3597 int32_t *pids
= NULL
;
3600 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3601 cmd_ctx
->lsm
->domain
.type
, &pids
);
3603 /* Return value is a negative lttng_error_code. */
3609 * Setup lttng message with payload size set to the event list size in
3610 * bytes and then copy list into the llm payload.
3612 ret
= setup_lttng_msg(cmd_ctx
, sizeof(int32_t) * nr_pids
);
3618 /* Copy event list into message payload */
3619 memcpy(cmd_ctx
->llm
->payload
, pids
,
3620 sizeof(int) * nr_pids
);
3627 case LTTNG_SET_CONSUMER_URI
:
3630 struct lttng_uri
*uris
;
3632 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3633 len
= nb_uri
* sizeof(struct lttng_uri
);
3636 ret
= LTTNG_ERR_INVALID
;
3640 uris
= zmalloc(len
);
3642 ret
= LTTNG_ERR_FATAL
;
3646 /* Receive variable len data */
3647 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3648 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3650 DBG("No URIs received from client... continuing");
3652 ret
= LTTNG_ERR_SESSION_FAIL
;
3657 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3659 if (ret
!= LTTNG_OK
) {
3666 case LTTNG_START_TRACE
:
3668 ret
= cmd_start_trace(cmd_ctx
->session
);
3671 case LTTNG_STOP_TRACE
:
3673 ret
= cmd_stop_trace(cmd_ctx
->session
);
3676 case LTTNG_CREATE_SESSION
:
3679 struct lttng_uri
*uris
= NULL
;
3681 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3682 len
= nb_uri
* sizeof(struct lttng_uri
);
3685 uris
= zmalloc(len
);
3687 ret
= LTTNG_ERR_FATAL
;
3691 /* Receive variable len data */
3692 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3693 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3695 DBG("No URIs received from client... continuing");
3697 ret
= LTTNG_ERR_SESSION_FAIL
;
3702 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3703 DBG("Creating session with ONE network URI is a bad call");
3704 ret
= LTTNG_ERR_SESSION_FAIL
;
3710 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3711 &cmd_ctx
->creds
, 0);
3717 case LTTNG_DESTROY_SESSION
:
3719 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3721 /* Set session to NULL so we do not unlock it after free. */
3722 cmd_ctx
->session
= NULL
;
3725 case LTTNG_LIST_DOMAINS
:
3728 struct lttng_domain
*domains
= NULL
;
3730 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3732 /* Return value is a negative lttng_error_code. */
3737 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3743 /* Copy event list into message payload */
3744 memcpy(cmd_ctx
->llm
->payload
, domains
,
3745 nb_dom
* sizeof(struct lttng_domain
));
3752 case LTTNG_LIST_CHANNELS
:
3755 struct lttng_channel
*channels
= NULL
;
3757 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3758 cmd_ctx
->session
, &channels
);
3760 /* Return value is a negative lttng_error_code. */
3765 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3771 /* Copy event list into message payload */
3772 memcpy(cmd_ctx
->llm
->payload
, channels
,
3773 nb_chan
* sizeof(struct lttng_channel
));
3780 case LTTNG_LIST_EVENTS
:
3783 struct lttng_event
*events
= NULL
;
3785 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3786 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3788 /* Return value is a negative lttng_error_code. */
3793 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3799 /* Copy event list into message payload */
3800 memcpy(cmd_ctx
->llm
->payload
, events
,
3801 nb_event
* sizeof(struct lttng_event
));
3808 case LTTNG_LIST_SESSIONS
:
3810 unsigned int nr_sessions
;
3812 session_lock_list();
3813 nr_sessions
= lttng_sessions_count(
3814 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3815 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3817 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3819 session_unlock_list();
3823 /* Filled the session array */
3824 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3825 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3826 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3828 session_unlock_list();
3833 case LTTNG_CALIBRATE
:
3835 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3836 &cmd_ctx
->lsm
->u
.calibrate
);
3839 case LTTNG_REGISTER_CONSUMER
:
3841 struct consumer_data
*cdata
;
3843 switch (cmd_ctx
->lsm
->domain
.type
) {
3844 case LTTNG_DOMAIN_KERNEL
:
3845 cdata
= &kconsumer_data
;
3848 ret
= LTTNG_ERR_UND
;
3852 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3853 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3856 case LTTNG_DATA_PENDING
:
3860 /* 1 byte to return whether or not data is pending */
3861 ret
= setup_lttng_msg(cmd_ctx
, 1);
3866 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
3870 * This function may returns 0 or 1 to indicate whether or not
3871 * there is data pending. In case of error, it should return an
3872 * LTTNG_ERR code. However, some code paths may still return
3873 * a nondescript error code, which we handle by returning an
3876 if (pending_ret
== 0 || pending_ret
== 1) {
3878 } else if (pending_ret
< 0) {
3879 ret
= LTTNG_ERR_UNK
;
3886 *cmd_ctx
->llm
->payload
= (uint8_t) pending_ret
;
3889 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3891 struct lttcomm_lttng_output_id reply
;
3893 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3894 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3895 if (ret
!= LTTNG_OK
) {
3899 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3904 /* Copy output list into message payload */
3905 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3909 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3911 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3912 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3915 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3918 struct lttng_snapshot_output
*outputs
= NULL
;
3920 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3921 if (nb_output
< 0) {
3926 ret
= setup_lttng_msg(cmd_ctx
,
3927 nb_output
* sizeof(struct lttng_snapshot_output
));
3934 /* Copy output list into message payload */
3935 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3936 nb_output
* sizeof(struct lttng_snapshot_output
));
3943 case LTTNG_SNAPSHOT_RECORD
:
3945 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3946 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3947 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3950 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3953 struct lttng_uri
*uris
= NULL
;
3955 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3956 len
= nb_uri
* sizeof(struct lttng_uri
);
3959 uris
= zmalloc(len
);
3961 ret
= LTTNG_ERR_FATAL
;
3965 /* Receive variable len data */
3966 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3967 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3969 DBG("No URIs received from client... continuing");
3971 ret
= LTTNG_ERR_SESSION_FAIL
;
3976 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3977 DBG("Creating session with ONE network URI is a bad call");
3978 ret
= LTTNG_ERR_SESSION_FAIL
;
3984 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3985 nb_uri
, &cmd_ctx
->creds
);
3989 case LTTNG_CREATE_SESSION_LIVE
:
3992 struct lttng_uri
*uris
= NULL
;
3994 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3995 len
= nb_uri
* sizeof(struct lttng_uri
);
3998 uris
= zmalloc(len
);
4000 ret
= LTTNG_ERR_FATAL
;
4004 /* Receive variable len data */
4005 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4006 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4008 DBG("No URIs received from client... continuing");
4010 ret
= LTTNG_ERR_SESSION_FAIL
;
4015 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4016 DBG("Creating session with ONE network URI is a bad call");
4017 ret
= LTTNG_ERR_SESSION_FAIL
;
4023 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4024 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4028 case LTTNG_SAVE_SESSION
:
4030 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4034 case LTTNG_SET_SESSION_SHM_PATH
:
4036 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4037 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4041 ret
= LTTNG_ERR_UND
;
4046 if (cmd_ctx
->llm
== NULL
) {
4047 DBG("Missing llm structure. Allocating one.");
4048 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
4052 /* Set return code */
4053 cmd_ctx
->llm
->ret_code
= ret
;
4055 if (cmd_ctx
->session
) {
4056 session_unlock(cmd_ctx
->session
);
4058 if (need_tracing_session
) {
4059 session_unlock_list();
4066 * Thread managing health check socket.
4068 static void *thread_manage_health(void *data
)
4070 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4071 uint32_t revents
, nb_fd
;
4072 struct lttng_poll_event events
;
4073 struct health_comm_msg msg
;
4074 struct health_comm_reply reply
;
4076 DBG("[thread] Manage health check started");
4078 rcu_register_thread();
4080 /* We might hit an error path before this is created. */
4081 lttng_poll_init(&events
);
4083 /* Create unix socket */
4084 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4086 ERR("Unable to create health check Unix socket");
4092 /* lttng health client socket path permissions */
4093 ret
= chown(health_unix_sock_path
, 0,
4094 utils_get_group_id(tracing_group_name
));
4096 ERR("Unable to set group on %s", health_unix_sock_path
);
4102 ret
= chmod(health_unix_sock_path
,
4103 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4105 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4113 * Set the CLOEXEC flag. Return code is useless because either way, the
4116 (void) utils_set_fd_cloexec(sock
);
4118 ret
= lttcomm_listen_unix_sock(sock
);
4124 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4125 * more will be added to this poll set.
4127 ret
= sessiond_set_thread_pollset(&events
, 2);
4132 /* Add the application registration socket */
4133 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4138 sessiond_notify_ready();
4141 DBG("Health check ready");
4143 /* Inifinite blocking call, waiting for transmission */
4145 ret
= lttng_poll_wait(&events
, -1);
4148 * Restart interrupted system call.
4150 if (errno
== EINTR
) {
4158 for (i
= 0; i
< nb_fd
; i
++) {
4159 /* Fetch once the poll data */
4160 revents
= LTTNG_POLL_GETEV(&events
, i
);
4161 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4164 /* No activity for this FD (poll implementation). */
4168 /* Thread quit pipe has been closed. Killing thread. */
4169 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4175 /* Event on the registration socket */
4176 if (pollfd
== sock
) {
4177 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4178 ERR("Health socket poll error");
4184 new_sock
= lttcomm_accept_unix_sock(sock
);
4190 * Set the CLOEXEC flag. Return code is useless because either way, the
4193 (void) utils_set_fd_cloexec(new_sock
);
4195 DBG("Receiving data from client for health...");
4196 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4198 DBG("Nothing recv() from client... continuing");
4199 ret
= close(new_sock
);
4207 rcu_thread_online();
4209 memset(&reply
, 0, sizeof(reply
));
4210 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4212 * health_check_state returns 0 if health is
4215 if (!health_check_state(health_sessiond
, i
)) {
4216 reply
.ret_code
|= 1ULL << i
;
4220 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4222 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4224 ERR("Failed to send health data back to client");
4227 /* End of transmission */
4228 ret
= close(new_sock
);
4238 ERR("Health error occurred in %s", __func__
);
4240 DBG("Health check thread dying");
4241 unlink(health_unix_sock_path
);
4249 lttng_poll_clean(&events
);
4251 rcu_unregister_thread();
4256 * This thread manage all clients request using the unix client socket for
4259 static void *thread_manage_clients(void *data
)
4261 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4263 uint32_t revents
, nb_fd
;
4264 struct command_ctx
*cmd_ctx
= NULL
;
4265 struct lttng_poll_event events
;
4267 DBG("[thread] Manage client started");
4269 rcu_register_thread();
4271 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4273 health_code_update();
4275 ret
= lttcomm_listen_unix_sock(client_sock
);
4281 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4282 * more will be added to this poll set.
4284 ret
= sessiond_set_thread_pollset(&events
, 2);
4286 goto error_create_poll
;
4289 /* Add the application registration socket */
4290 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4295 sessiond_notify_ready();
4296 ret
= sem_post(&load_info
->message_thread_ready
);
4298 PERROR("sem_post message_thread_ready");
4302 /* This testpoint is after we signal readiness to the parent. */
4303 if (testpoint(sessiond_thread_manage_clients
)) {
4307 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4311 health_code_update();
4314 DBG("Accepting client command ...");
4316 /* Inifinite blocking call, waiting for transmission */
4318 health_poll_entry();
4319 ret
= lttng_poll_wait(&events
, -1);
4323 * Restart interrupted system call.
4325 if (errno
== EINTR
) {
4333 for (i
= 0; i
< nb_fd
; i
++) {
4334 /* Fetch once the poll data */
4335 revents
= LTTNG_POLL_GETEV(&events
, i
);
4336 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4338 health_code_update();
4341 /* No activity for this FD (poll implementation). */
4345 /* Thread quit pipe has been closed. Killing thread. */
4346 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4352 /* Event on the registration socket */
4353 if (pollfd
== client_sock
) {
4354 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4355 ERR("Client socket poll error");
4361 DBG("Wait for client response");
4363 health_code_update();
4365 sock
= lttcomm_accept_unix_sock(client_sock
);
4371 * Set the CLOEXEC flag. Return code is useless because either way, the
4374 (void) utils_set_fd_cloexec(sock
);
4376 /* Set socket option for credentials retrieval */
4377 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4382 /* Allocate context command to process the client request */
4383 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4384 if (cmd_ctx
== NULL
) {
4385 PERROR("zmalloc cmd_ctx");
4389 /* Allocate data buffer for reception */
4390 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4391 if (cmd_ctx
->lsm
== NULL
) {
4392 PERROR("zmalloc cmd_ctx->lsm");
4396 cmd_ctx
->llm
= NULL
;
4397 cmd_ctx
->session
= NULL
;
4399 health_code_update();
4402 * Data is received from the lttng client. The struct
4403 * lttcomm_session_msg (lsm) contains the command and data request of
4406 DBG("Receiving data from client ...");
4407 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4408 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4410 DBG("Nothing recv() from client... continuing");
4416 clean_command_ctx(&cmd_ctx
);
4420 health_code_update();
4422 // TODO: Validate cmd_ctx including sanity check for
4423 // security purpose.
4425 rcu_thread_online();
4427 * This function dispatch the work to the kernel or userspace tracer
4428 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4429 * informations for the client. The command context struct contains
4430 * everything this function may needs.
4432 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4433 rcu_thread_offline();
4441 * TODO: Inform client somehow of the fatal error. At
4442 * this point, ret < 0 means that a zmalloc failed
4443 * (ENOMEM). Error detected but still accept
4444 * command, unless a socket error has been
4447 clean_command_ctx(&cmd_ctx
);
4451 health_code_update();
4453 DBG("Sending response (size: %d, retcode: %s (%d))",
4454 cmd_ctx
->lttng_msg_size
,
4455 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4456 cmd_ctx
->llm
->ret_code
);
4457 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4459 ERR("Failed to send data back to client");
4462 /* End of transmission */
4469 clean_command_ctx(&cmd_ctx
);
4471 health_code_update();
4483 lttng_poll_clean(&events
);
4484 clean_command_ctx(&cmd_ctx
);
4488 unlink(client_unix_sock_path
);
4489 if (client_sock
>= 0) {
4490 ret
= close(client_sock
);
4498 ERR("Health error occurred in %s", __func__
);
4501 health_unregister(health_sessiond
);
4503 DBG("Client thread dying");
4505 rcu_unregister_thread();
4508 * Since we are creating the consumer threads, we own them, so we need
4509 * to join them before our thread exits.
4511 ret
= join_consumer_thread(&kconsumer_data
);
4514 PERROR("join_consumer");
4517 ret
= join_consumer_thread(&ustconsumer32_data
);
4520 PERROR("join_consumer ust32");
4523 ret
= join_consumer_thread(&ustconsumer64_data
);
4526 PERROR("join_consumer ust64");
4533 * usage function on stderr
4535 static void usage(void)
4537 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4538 fprintf(stderr
, " -h, --help Display this usage.\n");
4539 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4540 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4541 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4542 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4543 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4544 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4545 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4546 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4547 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4548 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4549 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4550 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4551 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4552 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4553 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4554 fprintf(stderr
, " -V, --version Show version number.\n");
4555 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4556 fprintf(stderr
, " -q, --quiet No output at all.\n");
4557 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4558 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4559 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4560 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4561 fprintf(stderr
, " --agent-tcp-port Agent registration TCP port\n");
4562 fprintf(stderr
, " -f --config PATH Load daemon configuration file\n");
4563 fprintf(stderr
, " -l --load PATH Load session configuration\n");
4564 fprintf(stderr
, " --kmod-probes Specify kernel module probes to load\n");
4565 fprintf(stderr
, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4568 static int string_match(const char *str1
, const char *str2
)
4570 return (str1
&& str2
) && !strcmp(str1
, str2
);
4574 * Take an option from the getopt output and set it in the right variable to be
4577 * Return 0 on success else a negative value.
4579 static int set_option(int opt
, const char *arg
, const char *optname
)
4583 if (arg
&& arg
[0] == '\0') {
4585 * This only happens if the value is read from daemon config
4586 * file. This means the option requires an argument and the
4587 * configuration file contains a line such as:
4594 if (string_match(optname
, "client-sock") || opt
== 'c') {
4595 if (lttng_is_setuid_setgid()) {
4596 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4597 "-c, --client-sock");
4599 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4601 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4602 if (lttng_is_setuid_setgid()) {
4603 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4606 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4608 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4610 } else if (string_match(optname
, "background") || opt
== 'b') {
4612 } else if (string_match(optname
, "group") || opt
== 'g') {
4613 if (lttng_is_setuid_setgid()) {
4614 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4618 * If the override option is set, the pointer points to a
4619 * *non* const thus freeing it even though the variable type is
4622 if (tracing_group_name_override
) {
4623 free((void *) tracing_group_name
);
4625 tracing_group_name
= strdup(arg
);
4626 if (!tracing_group_name
) {
4630 tracing_group_name_override
= 1;
4632 } else if (string_match(optname
, "help") || opt
== 'h') {
4635 } else if (string_match(optname
, "version") || opt
== 'V') {
4636 fprintf(stdout
, "%s\n", VERSION
);
4638 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4640 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4641 if (lttng_is_setuid_setgid()) {
4642 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4643 "--kconsumerd-err-sock");
4645 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4647 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4648 if (lttng_is_setuid_setgid()) {
4649 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4650 "--kconsumerd-cmd-sock");
4652 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4654 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4655 if (lttng_is_setuid_setgid()) {
4656 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4657 "--ustconsumerd64-err-sock");
4659 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4661 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4662 if (lttng_is_setuid_setgid()) {
4663 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4664 "--ustconsumerd64-cmd-sock");
4666 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4668 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4669 if (lttng_is_setuid_setgid()) {
4670 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4671 "--ustconsumerd32-err-sock");
4673 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4675 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4676 if (lttng_is_setuid_setgid()) {
4677 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4678 "--ustconsumerd32-cmd-sock");
4680 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4682 } else if (string_match(optname
, "no-kernel")) {
4684 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4685 lttng_opt_quiet
= 1;
4686 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4687 /* Verbose level can increase using multiple -v */
4689 /* Value obtained from config file */
4690 lttng_opt_verbose
= config_parse_value(arg
);
4692 /* -v used on command line */
4693 lttng_opt_verbose
++;
4695 /* Clamp value to [0, 3] */
4696 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4697 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4698 } else if (string_match(optname
, "verbose-consumer")) {
4700 opt_verbose_consumer
= config_parse_value(arg
);
4702 opt_verbose_consumer
+= 1;
4704 } else if (string_match(optname
, "consumerd32-path")) {
4705 if (lttng_is_setuid_setgid()) {
4706 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4707 "--consumerd32-path");
4709 if (consumerd32_bin_override
) {
4710 free((void *) consumerd32_bin
);
4712 consumerd32_bin
= strdup(arg
);
4713 if (!consumerd32_bin
) {
4717 consumerd32_bin_override
= 1;
4719 } else if (string_match(optname
, "consumerd32-libdir")) {
4720 if (lttng_is_setuid_setgid()) {
4721 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4722 "--consumerd32-libdir");
4724 if (consumerd32_libdir_override
) {
4725 free((void *) consumerd32_libdir
);
4727 consumerd32_libdir
= strdup(arg
);
4728 if (!consumerd32_libdir
) {
4732 consumerd32_libdir_override
= 1;
4734 } else if (string_match(optname
, "consumerd64-path")) {
4735 if (lttng_is_setuid_setgid()) {
4736 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4737 "--consumerd64-path");
4739 if (consumerd64_bin_override
) {
4740 free((void *) consumerd64_bin
);
4742 consumerd64_bin
= strdup(arg
);
4743 if (!consumerd64_bin
) {
4747 consumerd64_bin_override
= 1;
4749 } else if (string_match(optname
, "consumerd64-libdir")) {
4750 if (lttng_is_setuid_setgid()) {
4751 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4752 "--consumerd64-libdir");
4754 if (consumerd64_libdir_override
) {
4755 free((void *) consumerd64_libdir
);
4757 consumerd64_libdir
= strdup(arg
);
4758 if (!consumerd64_libdir
) {
4762 consumerd64_libdir_override
= 1;
4764 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4765 if (lttng_is_setuid_setgid()) {
4766 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4770 opt_pidfile
= strdup(arg
);
4776 } else if (string_match(optname
, "agent-tcp-port")) {
4777 if (lttng_is_setuid_setgid()) {
4778 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4779 "--agent-tcp-port");
4788 v
= strtoul(arg
, NULL
, 0);
4789 if (errno
!= 0 || !isdigit(arg
[0])) {
4790 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4793 if (v
== 0 || v
>= 65535) {
4794 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4797 agent_tcp_port
= (uint32_t) v
;
4798 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4800 } else if (string_match(optname
, "load") || opt
== 'l') {
4801 if (lttng_is_setuid_setgid()) {
4802 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4805 free(opt_load_session_path
);
4806 opt_load_session_path
= strdup(arg
);
4807 if (!opt_load_session_path
) {
4812 } else if (string_match(optname
, "kmod-probes")) {
4813 if (lttng_is_setuid_setgid()) {
4814 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4817 free(kmod_probes_list
);
4818 kmod_probes_list
= strdup(arg
);
4819 if (!kmod_probes_list
) {
4824 } else if (string_match(optname
, "extra-kmod-probes")) {
4825 if (lttng_is_setuid_setgid()) {
4826 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4827 "--extra-kmod-probes");
4829 free(kmod_extra_probes_list
);
4830 kmod_extra_probes_list
= strdup(arg
);
4831 if (!kmod_extra_probes_list
) {
4836 } else if (string_match(optname
, "config") || opt
== 'f') {
4837 /* This is handled in set_options() thus silent skip. */
4840 /* Unknown option or other error.
4841 * Error is printed by getopt, just return */
4846 if (ret
== -EINVAL
) {
4847 const char *opt_name
= "unknown";
4850 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4852 if (opt
== long_options
[i
].val
) {
4853 opt_name
= long_options
[i
].name
;
4858 WARN("Invalid argument provided for option \"%s\", using default value.",
4866 * config_entry_handler_cb used to handle options read from a config file.
4867 * See config_entry_handler_cb comment in common/config/config.h for the
4868 * return value conventions.
4870 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4874 if (!entry
|| !entry
->name
|| !entry
->value
) {
4879 /* Check if the option is to be ignored */
4880 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4881 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4886 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4889 /* Ignore if not fully matched. */
4890 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4895 * If the option takes no argument on the command line, we have to
4896 * check if the value is "true". We support non-zero numeric values,
4899 if (!long_options
[i
].has_arg
) {
4900 ret
= config_parse_value(entry
->value
);
4903 WARN("Invalid configuration value \"%s\" for option %s",
4904 entry
->value
, entry
->name
);
4906 /* False, skip boolean config option. */
4911 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4915 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4922 * daemon configuration loading and argument parsing
4924 static int set_options(int argc
, char **argv
)
4926 int ret
= 0, c
= 0, option_index
= 0;
4927 int orig_optopt
= optopt
, orig_optind
= optind
;
4929 const char *config_path
= NULL
;
4931 optstring
= utils_generate_optstring(long_options
,
4932 sizeof(long_options
) / sizeof(struct option
));
4938 /* Check for the --config option */
4939 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4940 &option_index
)) != -1) {
4944 } else if (c
!= 'f') {
4945 /* if not equal to --config option. */
4949 if (lttng_is_setuid_setgid()) {
4950 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4953 config_path
= utils_expand_path(optarg
);
4955 ERR("Failed to resolve path: %s", optarg
);
4960 ret
= config_get_section_entries(config_path
, config_section_name
,
4961 config_entry_handler
, NULL
);
4964 ERR("Invalid configuration option at line %i", ret
);
4970 /* Reset getopt's global state */
4971 optopt
= orig_optopt
;
4972 optind
= orig_optind
;
4976 * getopt_long() will not set option_index if it encounters a
4979 c
= getopt_long(argc
, argv
, optstring
, long_options
,
4986 * Pass NULL as the long option name if popt left the index
4989 ret
= set_option(c
, optarg
,
4990 option_index
< 0 ? NULL
:
4991 long_options
[option_index
].name
);
5003 * Creates the two needed socket by the daemon.
5004 * apps_sock - The communication socket for all UST apps.
5005 * client_sock - The communication of the cli tool (lttng).
5007 static int init_daemon_socket(void)
5012 old_umask
= umask(0);
5014 /* Create client tool unix socket */
5015 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
5016 if (client_sock
< 0) {
5017 ERR("Create unix sock failed: %s", client_unix_sock_path
);
5022 /* Set the cloexec flag */
5023 ret
= utils_set_fd_cloexec(client_sock
);
5025 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5026 "Continuing but note that the consumer daemon will have a "
5027 "reference to this socket on exec()", client_sock
);
5030 /* File permission MUST be 660 */
5031 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5033 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5038 /* Create the application unix socket */
5039 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5040 if (apps_sock
< 0) {
5041 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5046 /* Set the cloexec flag */
5047 ret
= utils_set_fd_cloexec(apps_sock
);
5049 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5050 "Continuing but note that the consumer daemon will have a "
5051 "reference to this socket on exec()", apps_sock
);
5054 /* File permission MUST be 666 */
5055 ret
= chmod(apps_unix_sock_path
,
5056 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5058 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5063 DBG3("Session daemon client socket %d and application socket %d created",
5064 client_sock
, apps_sock
);
5072 * Check if the global socket is available, and if a daemon is answering at the
5073 * other side. If yes, error is returned.
5075 static int check_existing_daemon(void)
5077 /* Is there anybody out there ? */
5078 if (lttng_session_daemon_alive()) {
5086 * Set the tracing group gid onto the client socket.
5088 * Race window between mkdir and chown is OK because we are going from more
5089 * permissive (root.root) to less permissive (root.tracing).
5091 static int set_permissions(char *rundir
)
5096 gid
= utils_get_group_id(tracing_group_name
);
5098 /* Set lttng run dir */
5099 ret
= chown(rundir
, 0, gid
);
5101 ERR("Unable to set group on %s", rundir
);
5106 * Ensure all applications and tracing group can search the run
5107 * dir. Allow everyone to read the directory, since it does not
5108 * buy us anything to hide its content.
5110 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5112 ERR("Unable to set permissions on %s", rundir
);
5116 /* lttng client socket path */
5117 ret
= chown(client_unix_sock_path
, 0, gid
);
5119 ERR("Unable to set group on %s", client_unix_sock_path
);
5123 /* kconsumer error socket path */
5124 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5126 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5130 /* 64-bit ustconsumer error socket path */
5131 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5133 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5137 /* 32-bit ustconsumer compat32 error socket path */
5138 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5140 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5144 DBG("All permissions are set");
5150 * Create the lttng run directory needed for all global sockets and pipe.
5152 static int create_lttng_rundir(const char *rundir
)
5156 DBG3("Creating LTTng run directory: %s", rundir
);
5158 ret
= mkdir(rundir
, S_IRWXU
);
5160 if (errno
!= EEXIST
) {
5161 ERR("Unable to create %s", rundir
);
5173 * Setup sockets and directory needed by the kconsumerd communication with the
5176 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5180 char path
[PATH_MAX
];
5182 switch (consumer_data
->type
) {
5183 case LTTNG_CONSUMER_KERNEL
:
5184 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5186 case LTTNG_CONSUMER64_UST
:
5187 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5189 case LTTNG_CONSUMER32_UST
:
5190 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5193 ERR("Consumer type unknown");
5198 DBG2("Creating consumer directory: %s", path
);
5200 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5202 if (errno
!= EEXIST
) {
5204 ERR("Failed to create %s", path
);
5210 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5212 ERR("Unable to set group on %s", path
);
5218 /* Create the kconsumerd error unix socket */
5219 consumer_data
->err_sock
=
5220 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5221 if (consumer_data
->err_sock
< 0) {
5222 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5228 * Set the CLOEXEC flag. Return code is useless because either way, the
5231 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5233 PERROR("utils_set_fd_cloexec");
5234 /* continue anyway */
5237 /* File permission MUST be 660 */
5238 ret
= chmod(consumer_data
->err_unix_sock_path
,
5239 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5241 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5251 * Signal handler for the daemon
5253 * Simply stop all worker threads, leaving main() return gracefully after
5254 * joining all threads and calling cleanup().
5256 static void sighandler(int sig
)
5260 DBG("SIGPIPE caught");
5263 DBG("SIGINT caught");
5267 DBG("SIGTERM caught");
5271 CMM_STORE_SHARED(recv_child_signal
, 1);
5279 * Setup signal handler for :
5280 * SIGINT, SIGTERM, SIGPIPE
5282 static int set_signal_handler(void)
5285 struct sigaction sa
;
5288 if ((ret
= sigemptyset(&sigset
)) < 0) {
5289 PERROR("sigemptyset");
5293 sa
.sa_handler
= sighandler
;
5294 sa
.sa_mask
= sigset
;
5296 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5297 PERROR("sigaction");
5301 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5302 PERROR("sigaction");
5306 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5307 PERROR("sigaction");
5311 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5312 PERROR("sigaction");
5316 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5322 * Set open files limit to unlimited. This daemon can open a large number of
5323 * file descriptors in order to consumer multiple kernel traces.
5325 static void set_ulimit(void)
5330 /* The kernel does not allowed an infinite limit for open files */
5331 lim
.rlim_cur
= 65535;
5332 lim
.rlim_max
= 65535;
5334 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5336 PERROR("failed to set open files limit");
5341 * Write pidfile using the rundir and opt_pidfile.
5343 static int write_pidfile(void)
5346 char pidfile_path
[PATH_MAX
];
5351 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
5353 /* Build pidfile path from rundir and opt_pidfile. */
5354 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5355 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5357 PERROR("snprintf pidfile path");
5363 * Create pid file in rundir.
5365 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5371 * Create lockfile using the rundir and return its fd.
5373 static int create_lockfile(void)
5376 char lockfile_path
[PATH_MAX
];
5378 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5383 ret
= utils_create_lock_file(lockfile_path
);
5389 * Write agent TCP port using the rundir.
5391 static int write_agent_port(void)
5394 char path
[PATH_MAX
];
5398 ret
= snprintf(path
, sizeof(path
), "%s/"
5399 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5401 PERROR("snprintf agent port path");
5406 * Create TCP agent port file in rundir.
5408 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5417 int main(int argc
, char **argv
)
5419 int ret
= 0, retval
= 0;
5421 const char *home_path
, *env_app_timeout
;
5423 init_kernel_workarounds();
5425 rcu_register_thread();
5427 if (set_signal_handler()) {
5429 goto exit_set_signal_handler
;
5432 setup_consumerd_path();
5434 page_size
= sysconf(_SC_PAGESIZE
);
5435 if (page_size
< 0) {
5436 PERROR("sysconf _SC_PAGESIZE");
5437 page_size
= LONG_MAX
;
5438 WARN("Fallback page size to %ld", page_size
);
5442 * Parse arguments and load the daemon configuration file.
5444 * We have an exit_options exit path to free memory reserved by
5445 * set_options. This is needed because the rest of sessiond_cleanup()
5446 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5447 * depends on set_options.
5450 if (set_options(argc
, argv
)) {
5456 if (opt_daemon
|| opt_background
) {
5459 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5467 * We are in the child. Make sure all other file descriptors are
5468 * closed, in case we are called with more opened file
5469 * descriptors than the standard ones.
5471 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5477 * Starting from here, we can create threads. This needs to be after
5478 * lttng_daemonize due to RCU.
5482 * Initialize the health check subsystem. This call should set the
5483 * appropriate time values.
5485 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5486 if (!health_sessiond
) {
5487 PERROR("health_app_create error");
5489 goto exit_health_sessiond_cleanup
;
5492 if (init_ht_cleanup_quit_pipe()) {
5494 goto exit_ht_cleanup_quit_pipe
;
5497 /* Setup the thread ht_cleanup communication pipe. */
5498 if (utils_create_pipe_cloexec(ht_cleanup_pipe
)) {
5500 goto exit_ht_cleanup_pipe
;
5503 /* Set up max poll set size */
5504 if (lttng_poll_set_max_size()) {
5506 goto exit_set_max_size
;
5509 /* Create thread to clean up RCU hash tables */
5510 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5511 thread_ht_cleanup
, (void *) NULL
);
5514 PERROR("pthread_create ht_cleanup");
5516 goto exit_ht_cleanup
;
5519 /* Create thread quit pipe */
5520 if (init_thread_quit_pipe()) {
5522 goto exit_init_data
;
5525 /* Check if daemon is UID = 0 */
5526 is_root
= !getuid();
5529 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5532 goto exit_init_data
;
5535 /* Create global run dir with root access */
5536 if (create_lttng_rundir(rundir
)) {
5538 goto exit_init_data
;
5541 if (strlen(apps_unix_sock_path
) == 0) {
5542 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5543 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5546 goto exit_init_data
;
5550 if (strlen(client_unix_sock_path
) == 0) {
5551 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5552 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5555 goto exit_init_data
;
5559 /* Set global SHM for ust */
5560 if (strlen(wait_shm_path
) == 0) {
5561 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5562 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5565 goto exit_init_data
;
5569 if (strlen(health_unix_sock_path
) == 0) {
5570 ret
= snprintf(health_unix_sock_path
,
5571 sizeof(health_unix_sock_path
),
5572 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5575 goto exit_init_data
;
5579 /* Setup kernel consumerd path */
5580 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5581 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5584 goto exit_init_data
;
5586 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5587 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5590 goto exit_init_data
;
5593 DBG2("Kernel consumer err path: %s",
5594 kconsumer_data
.err_unix_sock_path
);
5595 DBG2("Kernel consumer cmd path: %s",
5596 kconsumer_data
.cmd_unix_sock_path
);
5598 home_path
= utils_get_home_dir();
5599 if (home_path
== NULL
) {
5600 /* TODO: Add --socket PATH option */
5601 ERR("Can't get HOME directory for sockets creation.");
5603 goto exit_init_data
;
5607 * Create rundir from home path. This will create something like
5610 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5613 goto exit_init_data
;
5616 if (create_lttng_rundir(rundir
)) {
5618 goto exit_init_data
;
5621 if (strlen(apps_unix_sock_path
) == 0) {
5622 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5623 DEFAULT_HOME_APPS_UNIX_SOCK
,
5627 goto exit_init_data
;
5631 /* Set the cli tool unix socket path */
5632 if (strlen(client_unix_sock_path
) == 0) {
5633 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5634 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5638 goto exit_init_data
;
5642 /* Set global SHM for ust */
5643 if (strlen(wait_shm_path
) == 0) {
5644 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5645 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5649 goto exit_init_data
;
5653 /* Set health check Unix path */
5654 if (strlen(health_unix_sock_path
) == 0) {
5655 ret
= snprintf(health_unix_sock_path
,
5656 sizeof(health_unix_sock_path
),
5657 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5661 goto exit_init_data
;
5666 lockfile_fd
= create_lockfile();
5667 if (lockfile_fd
< 0) {
5669 goto exit_init_data
;
5672 /* Set consumer initial state */
5673 kernel_consumerd_state
= CONSUMER_STOPPED
;
5674 ust_consumerd_state
= CONSUMER_STOPPED
;
5676 DBG("Client socket path %s", client_unix_sock_path
);
5677 DBG("Application socket path %s", apps_unix_sock_path
);
5678 DBG("Application wait path %s", wait_shm_path
);
5679 DBG("LTTng run directory path: %s", rundir
);
5681 /* 32 bits consumerd path setup */
5682 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5683 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5685 PERROR("snprintf 32-bit consumer error socket path");
5687 goto exit_init_data
;
5689 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5690 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5692 PERROR("snprintf 32-bit consumer command socket path");
5694 goto exit_init_data
;
5697 DBG2("UST consumer 32 bits err path: %s",
5698 ustconsumer32_data
.err_unix_sock_path
);
5699 DBG2("UST consumer 32 bits cmd path: %s",
5700 ustconsumer32_data
.cmd_unix_sock_path
);
5702 /* 64 bits consumerd path setup */
5703 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5704 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5706 PERROR("snprintf 64-bit consumer error socket path");
5708 goto exit_init_data
;
5710 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5711 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5713 PERROR("snprintf 64-bit consumer command socket path");
5715 goto exit_init_data
;
5718 DBG2("UST consumer 64 bits err path: %s",
5719 ustconsumer64_data
.err_unix_sock_path
);
5720 DBG2("UST consumer 64 bits cmd path: %s",
5721 ustconsumer64_data
.cmd_unix_sock_path
);
5724 * See if daemon already exist.
5726 if (check_existing_daemon()) {
5727 ERR("Already running daemon.\n");
5729 * We do not goto exit because we must not cleanup()
5730 * because a daemon is already running.
5733 goto exit_init_data
;
5737 * Init UST app hash table. Alloc hash table before this point since
5738 * cleanup() can get called after that point.
5740 if (ust_app_ht_alloc()) {
5741 ERR("Failed to allocate UST app hash table");
5743 goto exit_init_data
;
5747 * Initialize agent app hash table. We allocate the hash table here
5748 * since cleanup() can get called after this point.
5750 if (agent_app_ht_alloc()) {
5751 ERR("Failed to allocate Agent app hash table");
5753 goto exit_init_data
;
5757 * These actions must be executed as root. We do that *after* setting up
5758 * the sockets path because we MUST make the check for another daemon using
5759 * those paths *before* trying to set the kernel consumer sockets and init
5763 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5765 goto exit_init_data
;
5768 /* Setup kernel tracer */
5769 if (!opt_no_kernel
) {
5770 init_kernel_tracer();
5771 if (kernel_tracer_fd
>= 0) {
5772 ret
= syscall_init_table();
5774 ERR("Unable to populate syscall table. "
5775 "Syscall tracing won't work "
5776 "for this session daemon.");
5781 /* Set ulimit for open files */
5784 /* init lttng_fd tracking must be done after set_ulimit. */
5787 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5789 goto exit_init_data
;
5792 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5794 goto exit_init_data
;
5797 /* Setup the needed unix socket */
5798 if (init_daemon_socket()) {
5800 goto exit_init_data
;
5803 /* Set credentials to socket */
5804 if (is_root
&& set_permissions(rundir
)) {
5806 goto exit_init_data
;
5809 /* Get parent pid if -S, --sig-parent is specified. */
5810 if (opt_sig_parent
) {
5814 /* Setup the kernel pipe for waking up the kernel thread */
5815 if (is_root
&& !opt_no_kernel
) {
5816 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5818 goto exit_init_data
;
5822 /* Setup the thread apps communication pipe. */
5823 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5825 goto exit_init_data
;
5828 /* Setup the thread apps notify communication pipe. */
5829 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5831 goto exit_init_data
;
5834 /* Initialize global buffer per UID and PID registry. */
5835 buffer_reg_init_uid_registry();
5836 buffer_reg_init_pid_registry();
5838 /* Init UST command queue. */
5839 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5842 * Get session list pointer. This pointer MUST NOT be free'd. This list
5843 * is statically declared in session.c
5845 session_list_ptr
= session_get_list();
5849 /* Check for the application socket timeout env variable. */
5850 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5851 if (env_app_timeout
) {
5852 app_socket_timeout
= atoi(env_app_timeout
);
5854 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5857 ret
= write_pidfile();
5859 ERR("Error in write_pidfile");
5861 goto exit_init_data
;
5863 ret
= write_agent_port();
5865 ERR("Error in write_agent_port");
5867 goto exit_init_data
;
5870 /* Initialize communication library */
5872 /* Initialize TCP timeout values */
5873 lttcomm_inet_init();
5875 if (load_session_init_data(&load_info
) < 0) {
5877 goto exit_init_data
;
5879 load_info
->path
= opt_load_session_path
;
5881 /* Create health-check thread */
5882 ret
= pthread_create(&health_thread
, NULL
,
5883 thread_manage_health
, (void *) NULL
);
5886 PERROR("pthread_create health");
5891 /* Create thread to manage the client socket */
5892 ret
= pthread_create(&client_thread
, NULL
,
5893 thread_manage_clients
, (void *) NULL
);
5896 PERROR("pthread_create clients");
5901 /* Create thread to dispatch registration */
5902 ret
= pthread_create(&dispatch_thread
, NULL
,
5903 thread_dispatch_ust_registration
, (void *) NULL
);
5906 PERROR("pthread_create dispatch");
5911 /* Create thread to manage application registration. */
5912 ret
= pthread_create(®_apps_thread
, NULL
,
5913 thread_registration_apps
, (void *) NULL
);
5916 PERROR("pthread_create registration");
5921 /* Create thread to manage application socket */
5922 ret
= pthread_create(&apps_thread
, NULL
,
5923 thread_manage_apps
, (void *) NULL
);
5926 PERROR("pthread_create apps");
5931 /* Create thread to manage application notify socket */
5932 ret
= pthread_create(&apps_notify_thread
, NULL
,
5933 ust_thread_manage_notify
, (void *) NULL
);
5936 PERROR("pthread_create notify");
5938 goto exit_apps_notify
;
5941 /* Create agent registration thread. */
5942 ret
= pthread_create(&agent_reg_thread
, NULL
,
5943 agent_thread_manage_registration
, (void *) NULL
);
5946 PERROR("pthread_create agent");
5948 goto exit_agent_reg
;
5951 /* Don't start this thread if kernel tracing is not requested nor root */
5952 if (is_root
&& !opt_no_kernel
) {
5953 /* Create kernel thread to manage kernel event */
5954 ret
= pthread_create(&kernel_thread
, NULL
,
5955 thread_manage_kernel
, (void *) NULL
);
5958 PERROR("pthread_create kernel");
5964 /* Create session loading thread. */
5965 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
5969 PERROR("pthread_create load_session_thread");
5971 goto exit_load_session
;
5975 * This is where we start awaiting program completion (e.g. through
5976 * signal that asks threads to teardown).
5979 ret
= pthread_join(load_session_thread
, &status
);
5982 PERROR("pthread_join load_session_thread");
5987 if (is_root
&& !opt_no_kernel
) {
5988 ret
= pthread_join(kernel_thread
, &status
);
5991 PERROR("pthread_join");
5997 ret
= pthread_join(agent_reg_thread
, &status
);
6000 PERROR("pthread_join agent");
6005 ret
= pthread_join(apps_notify_thread
, &status
);
6008 PERROR("pthread_join apps notify");
6013 ret
= pthread_join(apps_thread
, &status
);
6016 PERROR("pthread_join apps");
6021 ret
= pthread_join(reg_apps_thread
, &status
);
6024 PERROR("pthread_join");
6030 * Join dispatch thread after joining reg_apps_thread to ensure
6031 * we don't leak applications in the queue.
6033 ret
= pthread_join(dispatch_thread
, &status
);
6036 PERROR("pthread_join");
6041 ret
= pthread_join(client_thread
, &status
);
6044 PERROR("pthread_join");
6049 ret
= pthread_join(health_thread
, &status
);
6052 PERROR("pthread_join health thread");
6059 * sessiond_cleanup() is called when no other thread is running, except
6060 * the ht_cleanup thread, which is needed to destroy the hash tables.
6062 rcu_thread_online();
6064 rcu_thread_offline();
6065 rcu_unregister_thread();
6067 ret
= notify_thread_pipe(ht_cleanup_quit_pipe
[1]);
6069 ERR("write error on ht_cleanup quit pipe");
6073 ret
= pthread_join(ht_cleanup_thread
, &status
);
6076 PERROR("pthread_join ht cleanup thread");
6082 utils_close_pipe(ht_cleanup_pipe
);
6083 exit_ht_cleanup_pipe
:
6086 * Close the ht_cleanup quit pipe.
6088 utils_close_pipe(ht_cleanup_quit_pipe
);
6089 exit_ht_cleanup_quit_pipe
:
6091 health_app_destroy(health_sessiond
);
6092 exit_health_sessiond_cleanup
:
6095 sessiond_cleanup_options();
6097 exit_set_signal_handler
:
6098 /* Ensure all prior call_rcu are done. */