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
);
716 DBG("Cleaning up all agent apps");
717 agent_app_ht_clean();
719 DBG("Closing all UST sockets");
720 ust_app_clean_list();
721 buffer_reg_destroy_registries();
723 if (is_root
&& !opt_no_kernel
) {
724 DBG2("Closing kernel fd");
725 if (kernel_tracer_fd
>= 0) {
726 ret
= close(kernel_tracer_fd
);
731 DBG("Unloading kernel modules");
732 modprobe_remove_lttng_all();
736 close_consumer_sockets();
739 load_session_destroy_data(load_info
);
744 * Cleanup lock file by deleting it and finaly closing it which will
745 * release the file system lock.
747 if (lockfile_fd
>= 0) {
748 char lockfile_path
[PATH_MAX
];
750 ret
= generate_lock_file_path(lockfile_path
,
751 sizeof(lockfile_path
));
753 ret
= remove(lockfile_path
);
755 PERROR("remove lock file");
757 ret
= close(lockfile_fd
);
759 PERROR("close lock file");
765 * We do NOT rmdir rundir because there are other processes
766 * using it, for instance lttng-relayd, which can start in
767 * parallel with this teardown.
774 * Cleanup the daemon's option data structures.
776 static void sessiond_cleanup_options(void)
778 DBG("Cleaning up options");
781 * If the override option is set, the pointer points to a *non* const
782 * thus freeing it even though the variable type is set to const.
784 if (tracing_group_name_override
) {
785 free((void *) tracing_group_name
);
787 if (consumerd32_bin_override
) {
788 free((void *) consumerd32_bin
);
790 if (consumerd64_bin_override
) {
791 free((void *) consumerd64_bin
);
793 if (consumerd32_libdir_override
) {
794 free((void *) consumerd32_libdir
);
796 if (consumerd64_libdir_override
) {
797 free((void *) consumerd64_libdir
);
801 free(opt_load_session_path
);
802 free(kmod_probes_list
);
803 free(kmod_extra_probes_list
);
806 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
807 "Matthew, BEET driven development works!%c[%dm",
808 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
813 * Send data on a unix socket using the liblttsessiondcomm API.
815 * Return lttcomm error code.
817 static int send_unix_sock(int sock
, void *buf
, size_t len
)
819 /* Check valid length */
824 return lttcomm_send_unix_sock(sock
, buf
, len
);
828 * Free memory of a command context structure.
830 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
832 DBG("Clean command context structure");
834 if ((*cmd_ctx
)->llm
) {
835 free((*cmd_ctx
)->llm
);
837 if ((*cmd_ctx
)->lsm
) {
838 free((*cmd_ctx
)->lsm
);
846 * Notify UST applications using the shm mmap futex.
848 static int notify_ust_apps(int active
)
852 DBG("Notifying applications of session daemon state: %d", active
);
854 /* See shm.c for this call implying mmap, shm and futex calls */
855 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
856 if (wait_shm_mmap
== NULL
) {
860 /* Wake waiting process */
861 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
863 /* Apps notified successfully */
871 * Setup the outgoing data buffer for the response (llm) by allocating the
872 * right amount of memory and copying the original information from the lsm
875 * Return total size of the buffer pointed by buf.
877 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
883 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
884 if (cmd_ctx
->llm
== NULL
) {
890 /* Copy common data */
891 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
892 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
894 cmd_ctx
->llm
->data_size
= size
;
895 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
904 * Update the kernel poll set of all channel fd available over all tracing
905 * session. Add the wakeup pipe at the end of the set.
907 static int update_kernel_poll(struct lttng_poll_event
*events
)
910 struct ltt_session
*session
;
911 struct ltt_kernel_channel
*channel
;
913 DBG("Updating kernel poll set");
916 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
917 session_lock(session
);
918 if (session
->kernel_session
== NULL
) {
919 session_unlock(session
);
923 cds_list_for_each_entry(channel
,
924 &session
->kernel_session
->channel_list
.head
, list
) {
925 /* Add channel fd to the kernel poll set */
926 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
928 session_unlock(session
);
931 DBG("Channel fd %d added to kernel set", channel
->fd
);
933 session_unlock(session
);
935 session_unlock_list();
940 session_unlock_list();
945 * Find the channel fd from 'fd' over all tracing session. When found, check
946 * for new channel stream and send those stream fds to the kernel consumer.
948 * Useful for CPU hotplug feature.
950 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
953 struct ltt_session
*session
;
954 struct ltt_kernel_session
*ksess
;
955 struct ltt_kernel_channel
*channel
;
957 DBG("Updating kernel streams for channel fd %d", fd
);
960 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
961 session_lock(session
);
962 if (session
->kernel_session
== NULL
) {
963 session_unlock(session
);
966 ksess
= session
->kernel_session
;
968 cds_list_for_each_entry(channel
,
969 &ksess
->channel_list
.head
, list
) {
970 struct lttng_ht_iter iter
;
971 struct consumer_socket
*socket
;
973 if (channel
->fd
!= fd
) {
976 DBG("Channel found, updating kernel streams");
977 ret
= kernel_open_channel_stream(channel
);
981 /* Update the stream global counter */
982 ksess
->stream_count_global
+= ret
;
985 * Have we already sent fds to the consumer? If yes, it
986 * means that tracing is started so it is safe to send
987 * our updated stream fds.
989 if (ksess
->consumer_fds_sent
!= 1
990 || ksess
->consumer
== NULL
) {
996 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
997 &iter
.iter
, socket
, node
.node
) {
998 pthread_mutex_lock(socket
->lock
);
999 ret
= kernel_consumer_send_channel_stream(socket
,
1001 session
->output_traces
? 1 : 0);
1002 pthread_mutex_unlock(socket
->lock
);
1010 session_unlock(session
);
1012 session_unlock_list();
1016 session_unlock(session
);
1017 session_unlock_list();
1022 * For each tracing session, update newly registered apps. The session list
1023 * lock MUST be acquired before calling this.
1025 static void update_ust_app(int app_sock
)
1027 struct ltt_session
*sess
, *stmp
;
1029 /* Consumer is in an ERROR state. Stop any application update. */
1030 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
1031 /* Stop the update process since the consumer is dead. */
1035 /* For all tracing session(s) */
1036 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
1037 struct ust_app
*app
;
1040 if (!sess
->ust_session
) {
1041 goto unlock_session
;
1045 assert(app_sock
>= 0);
1046 app
= ust_app_find_by_sock(app_sock
);
1049 * Application can be unregistered before so
1050 * this is possible hence simply stopping the
1053 DBG3("UST app update failed to find app sock %d",
1057 ust_app_global_update(sess
->ust_session
, app
);
1061 session_unlock(sess
);
1066 * This thread manage event coming from the kernel.
1068 * Features supported in this thread:
1071 static void *thread_manage_kernel(void *data
)
1073 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1074 uint32_t revents
, nb_fd
;
1076 struct lttng_poll_event events
;
1078 DBG("[thread] Thread manage kernel started");
1080 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1083 * This first step of the while is to clean this structure which could free
1084 * non NULL pointers so initialize it before the loop.
1086 lttng_poll_init(&events
);
1088 if (testpoint(sessiond_thread_manage_kernel
)) {
1089 goto error_testpoint
;
1092 health_code_update();
1094 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1095 goto error_testpoint
;
1099 health_code_update();
1101 if (update_poll_flag
== 1) {
1102 /* Clean events object. We are about to populate it again. */
1103 lttng_poll_clean(&events
);
1105 ret
= sessiond_set_thread_pollset(&events
, 2);
1107 goto error_poll_create
;
1110 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1115 /* This will add the available kernel channel if any. */
1116 ret
= update_kernel_poll(&events
);
1120 update_poll_flag
= 0;
1123 DBG("Thread kernel polling");
1125 /* Poll infinite value of time */
1127 health_poll_entry();
1128 ret
= lttng_poll_wait(&events
, -1);
1129 DBG("Thread kernel return from poll on %d fds",
1130 LTTNG_POLL_GETNB(&events
));
1134 * Restart interrupted system call.
1136 if (errno
== EINTR
) {
1140 } else if (ret
== 0) {
1141 /* Should not happen since timeout is infinite */
1142 ERR("Return value of poll is 0 with an infinite timeout.\n"
1143 "This should not have happened! Continuing...");
1149 for (i
= 0; i
< nb_fd
; i
++) {
1150 /* Fetch once the poll data */
1151 revents
= LTTNG_POLL_GETEV(&events
, i
);
1152 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1154 health_code_update();
1157 /* No activity for this FD (poll implementation). */
1161 /* Thread quit pipe has been closed. Killing thread. */
1162 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1168 /* Check for data on kernel pipe */
1169 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
1170 (void) lttng_read(kernel_poll_pipe
[0],
1173 * Ret value is useless here, if this pipe gets any actions an
1174 * update is required anyway.
1176 update_poll_flag
= 1;
1180 * New CPU detected by the kernel. Adding kernel stream to
1181 * kernel session and updating the kernel consumer
1183 if (revents
& LPOLLIN
) {
1184 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1190 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1191 * and unregister kernel stream at this point.
1200 lttng_poll_clean(&events
);
1203 utils_close_pipe(kernel_poll_pipe
);
1204 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1207 ERR("Health error occurred in %s", __func__
);
1208 WARN("Kernel thread died unexpectedly. "
1209 "Kernel tracing can continue but CPU hotplug is disabled.");
1211 health_unregister(health_sessiond
);
1212 DBG("Kernel thread dying");
1217 * Signal pthread condition of the consumer data that the thread.
1219 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1221 pthread_mutex_lock(&data
->cond_mutex
);
1224 * The state is set before signaling. It can be any value, it's the waiter
1225 * job to correctly interpret this condition variable associated to the
1226 * consumer pthread_cond.
1228 * A value of 0 means that the corresponding thread of the consumer data
1229 * was not started. 1 indicates that the thread has started and is ready
1230 * for action. A negative value means that there was an error during the
1233 data
->consumer_thread_is_ready
= state
;
1234 (void) pthread_cond_signal(&data
->cond
);
1236 pthread_mutex_unlock(&data
->cond_mutex
);
1240 * This thread manage the consumer error sent back to the session daemon.
1242 static void *thread_manage_consumer(void *data
)
1244 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1245 uint32_t revents
, nb_fd
;
1246 enum lttcomm_return_code code
;
1247 struct lttng_poll_event events
;
1248 struct consumer_data
*consumer_data
= data
;
1250 DBG("[thread] Manage consumer started");
1252 rcu_register_thread();
1253 rcu_thread_online();
1255 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1257 health_code_update();
1260 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1261 * metadata_sock. Nothing more will be added to this poll set.
1263 ret
= sessiond_set_thread_pollset(&events
, 3);
1269 * The error socket here is already in a listening state which was done
1270 * just before spawning this thread to avoid a race between the consumer
1271 * daemon exec trying to connect and the listen() call.
1273 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1278 health_code_update();
1280 /* Infinite blocking call, waiting for transmission */
1282 health_poll_entry();
1284 if (testpoint(sessiond_thread_manage_consumer
)) {
1288 ret
= lttng_poll_wait(&events
, -1);
1292 * Restart interrupted system call.
1294 if (errno
== EINTR
) {
1302 for (i
= 0; i
< nb_fd
; i
++) {
1303 /* Fetch once the poll data */
1304 revents
= LTTNG_POLL_GETEV(&events
, i
);
1305 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1307 health_code_update();
1310 /* No activity for this FD (poll implementation). */
1314 /* Thread quit pipe has been closed. Killing thread. */
1315 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1321 /* Event on the registration socket */
1322 if (pollfd
== consumer_data
->err_sock
) {
1323 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1324 ERR("consumer err socket poll error");
1330 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1336 * Set the CLOEXEC flag. Return code is useless because either way, the
1339 (void) utils_set_fd_cloexec(sock
);
1341 health_code_update();
1343 DBG2("Receiving code from consumer err_sock");
1345 /* Getting status code from kconsumerd */
1346 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1347 sizeof(enum lttcomm_return_code
));
1352 health_code_update();
1353 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1354 /* Connect both socket, command and metadata. */
1355 consumer_data
->cmd_sock
=
1356 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1357 consumer_data
->metadata_fd
=
1358 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1359 if (consumer_data
->cmd_sock
< 0
1360 || consumer_data
->metadata_fd
< 0) {
1361 PERROR("consumer connect cmd socket");
1362 /* On error, signal condition and quit. */
1363 signal_consumer_condition(consumer_data
, -1);
1366 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1367 /* Create metadata socket lock. */
1368 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1369 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1370 PERROR("zmalloc pthread mutex");
1374 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1376 signal_consumer_condition(consumer_data
, 1);
1377 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1378 DBG("Consumer metadata socket ready (fd: %d)",
1379 consumer_data
->metadata_fd
);
1381 ERR("consumer error when waiting for SOCK_READY : %s",
1382 lttcomm_get_readable_code(-code
));
1386 /* Remove the consumerd error sock since we've established a connexion */
1387 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1392 /* Add new accepted error socket. */
1393 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1398 /* Add metadata socket that is successfully connected. */
1399 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1400 LPOLLIN
| LPOLLRDHUP
);
1405 health_code_update();
1407 /* Infinite blocking call, waiting for transmission */
1410 health_code_update();
1412 /* Exit the thread because the thread quit pipe has been triggered. */
1414 /* Not a health error. */
1419 health_poll_entry();
1420 ret
= lttng_poll_wait(&events
, -1);
1424 * Restart interrupted system call.
1426 if (errno
== EINTR
) {
1434 for (i
= 0; i
< nb_fd
; i
++) {
1435 /* Fetch once the poll data */
1436 revents
= LTTNG_POLL_GETEV(&events
, i
);
1437 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1439 health_code_update();
1442 /* No activity for this FD (poll implementation). */
1447 * Thread quit pipe has been triggered, flag that we should stop
1448 * but continue the current loop to handle potential data from
1451 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1453 if (pollfd
== sock
) {
1454 /* Event on the consumerd socket */
1455 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1456 ERR("consumer err socket second poll error");
1459 health_code_update();
1460 /* Wait for any kconsumerd error */
1461 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1462 sizeof(enum lttcomm_return_code
));
1464 ERR("consumer closed the command socket");
1468 ERR("consumer return code : %s",
1469 lttcomm_get_readable_code(-code
));
1472 } else if (pollfd
== consumer_data
->metadata_fd
) {
1473 /* UST metadata requests */
1474 ret
= ust_consumer_metadata_request(
1475 &consumer_data
->metadata_sock
);
1477 ERR("Handling metadata request");
1481 /* No need for an else branch all FDs are tested prior. */
1483 health_code_update();
1489 * We lock here because we are about to close the sockets and some other
1490 * thread might be using them so get exclusive access which will abort all
1491 * other consumer command by other threads.
1493 pthread_mutex_lock(&consumer_data
->lock
);
1495 /* Immediately set the consumerd state to stopped */
1496 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1497 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1498 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1499 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1500 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1502 /* Code flow error... */
1506 if (consumer_data
->err_sock
>= 0) {
1507 ret
= close(consumer_data
->err_sock
);
1511 consumer_data
->err_sock
= -1;
1513 if (consumer_data
->cmd_sock
>= 0) {
1514 ret
= close(consumer_data
->cmd_sock
);
1518 consumer_data
->cmd_sock
= -1;
1520 if (consumer_data
->metadata_sock
.fd_ptr
&&
1521 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1522 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1534 unlink(consumer_data
->err_unix_sock_path
);
1535 unlink(consumer_data
->cmd_unix_sock_path
);
1536 pthread_mutex_unlock(&consumer_data
->lock
);
1538 /* Cleanup metadata socket mutex. */
1539 if (consumer_data
->metadata_sock
.lock
) {
1540 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1541 free(consumer_data
->metadata_sock
.lock
);
1543 lttng_poll_clean(&events
);
1547 ERR("Health error occurred in %s", __func__
);
1549 health_unregister(health_sessiond
);
1550 DBG("consumer thread cleanup completed");
1552 rcu_thread_offline();
1553 rcu_unregister_thread();
1559 * This thread manage application communication.
1561 static void *thread_manage_apps(void *data
)
1563 int i
, ret
, pollfd
, err
= -1;
1565 uint32_t revents
, nb_fd
;
1566 struct lttng_poll_event events
;
1568 DBG("[thread] Manage application started");
1570 rcu_register_thread();
1571 rcu_thread_online();
1573 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1575 if (testpoint(sessiond_thread_manage_apps
)) {
1576 goto error_testpoint
;
1579 health_code_update();
1581 ret
= sessiond_set_thread_pollset(&events
, 2);
1583 goto error_poll_create
;
1586 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1591 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1595 health_code_update();
1598 DBG("Apps thread polling");
1600 /* Inifinite blocking call, waiting for transmission */
1602 health_poll_entry();
1603 ret
= lttng_poll_wait(&events
, -1);
1604 DBG("Apps thread return from poll on %d fds",
1605 LTTNG_POLL_GETNB(&events
));
1609 * Restart interrupted system call.
1611 if (errno
== EINTR
) {
1619 for (i
= 0; i
< nb_fd
; i
++) {
1620 /* Fetch once the poll data */
1621 revents
= LTTNG_POLL_GETEV(&events
, i
);
1622 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1624 health_code_update();
1627 /* No activity for this FD (poll implementation). */
1631 /* Thread quit pipe has been closed. Killing thread. */
1632 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1638 /* Inspect the apps cmd pipe */
1639 if (pollfd
== apps_cmd_pipe
[0]) {
1640 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1641 ERR("Apps command pipe error");
1643 } else if (revents
& LPOLLIN
) {
1647 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1648 if (size_ret
< sizeof(sock
)) {
1649 PERROR("read apps cmd pipe");
1653 health_code_update();
1656 * We only monitor the error events of the socket. This
1657 * thread does not handle any incoming data from UST
1660 ret
= lttng_poll_add(&events
, sock
,
1661 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1666 DBG("Apps with sock %d added to poll set", sock
);
1670 * At this point, we know that a registered application made
1671 * the event at poll_wait.
1673 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1674 /* Removing from the poll set */
1675 ret
= lttng_poll_del(&events
, pollfd
);
1680 /* Socket closed on remote end. */
1681 ust_app_unregister(pollfd
);
1685 health_code_update();
1691 lttng_poll_clean(&events
);
1694 utils_close_pipe(apps_cmd_pipe
);
1695 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1698 * We don't clean the UST app hash table here since already registered
1699 * applications can still be controlled so let them be until the session
1700 * daemon dies or the applications stop.
1705 ERR("Health error occurred in %s", __func__
);
1707 health_unregister(health_sessiond
);
1708 DBG("Application communication apps thread cleanup complete");
1709 rcu_thread_offline();
1710 rcu_unregister_thread();
1715 * Send a socket to a thread This is called from the dispatch UST registration
1716 * thread once all sockets are set for the application.
1718 * The sock value can be invalid, we don't really care, the thread will handle
1719 * it and make the necessary cleanup if so.
1721 * On success, return 0 else a negative value being the errno message of the
1724 static int send_socket_to_thread(int fd
, int sock
)
1729 * It's possible that the FD is set as invalid with -1 concurrently just
1730 * before calling this function being a shutdown state of the thread.
1737 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1738 if (ret
< sizeof(sock
)) {
1739 PERROR("write apps pipe %d", fd
);
1746 /* All good. Don't send back the write positive ret value. */
1753 * Sanitize the wait queue of the dispatch registration thread meaning removing
1754 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1755 * notify socket is never received.
1757 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1759 int ret
, nb_fd
= 0, i
;
1760 unsigned int fd_added
= 0;
1761 struct lttng_poll_event events
;
1762 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1766 lttng_poll_init(&events
);
1768 /* Just skip everything for an empty queue. */
1769 if (!wait_queue
->count
) {
1773 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1778 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1779 &wait_queue
->head
, head
) {
1780 assert(wait_node
->app
);
1781 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1782 LPOLLHUP
| LPOLLERR
);
1795 * Poll but don't block so we can quickly identify the faulty events and
1796 * clean them afterwards from the wait queue.
1798 ret
= lttng_poll_wait(&events
, 0);
1804 for (i
= 0; i
< nb_fd
; i
++) {
1805 /* Get faulty FD. */
1806 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1807 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1810 /* No activity for this FD (poll implementation). */
1814 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1815 &wait_queue
->head
, head
) {
1816 if (pollfd
== wait_node
->app
->sock
&&
1817 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1818 cds_list_del(&wait_node
->head
);
1819 wait_queue
->count
--;
1820 ust_app_destroy(wait_node
->app
);
1823 * Silence warning of use-after-free in
1824 * cds_list_for_each_entry_safe which uses
1825 * __typeof__(*wait_node).
1834 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1838 lttng_poll_clean(&events
);
1842 lttng_poll_clean(&events
);
1844 ERR("Unable to sanitize wait queue");
1849 * Dispatch request from the registration threads to the application
1850 * communication thread.
1852 static void *thread_dispatch_ust_registration(void *data
)
1855 struct cds_wfcq_node
*node
;
1856 struct ust_command
*ust_cmd
= NULL
;
1857 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1858 struct ust_reg_wait_queue wait_queue
= {
1862 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1864 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1865 goto error_testpoint
;
1868 health_code_update();
1870 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1872 DBG("[thread] Dispatch UST command started");
1874 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1875 health_code_update();
1877 /* Atomically prepare the queue futex */
1878 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1881 struct ust_app
*app
= NULL
;
1885 * Make sure we don't have node(s) that have hung up before receiving
1886 * the notify socket. This is to clean the list in order to avoid
1887 * memory leaks from notify socket that are never seen.
1889 sanitize_wait_queue(&wait_queue
);
1891 health_code_update();
1892 /* Dequeue command for registration */
1893 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1895 DBG("Woken up but nothing in the UST command queue");
1896 /* Continue thread execution */
1900 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1902 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1903 " gid:%d sock:%d name:%s (version %d.%d)",
1904 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1905 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1906 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1907 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1909 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1910 wait_node
= zmalloc(sizeof(*wait_node
));
1912 PERROR("zmalloc wait_node dispatch");
1913 ret
= close(ust_cmd
->sock
);
1915 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1917 lttng_fd_put(LTTNG_FD_APPS
, 1);
1921 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1923 /* Create application object if socket is CMD. */
1924 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1926 if (!wait_node
->app
) {
1927 ret
= close(ust_cmd
->sock
);
1929 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1931 lttng_fd_put(LTTNG_FD_APPS
, 1);
1937 * Add application to the wait queue so we can set the notify
1938 * socket before putting this object in the global ht.
1940 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1945 * We have to continue here since we don't have the notify
1946 * socket and the application MUST be added to the hash table
1947 * only at that moment.
1952 * Look for the application in the local wait queue and set the
1953 * notify socket if found.
1955 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1956 &wait_queue
.head
, head
) {
1957 health_code_update();
1958 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1959 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1960 cds_list_del(&wait_node
->head
);
1962 app
= wait_node
->app
;
1964 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1970 * With no application at this stage the received socket is
1971 * basically useless so close it before we free the cmd data
1972 * structure for good.
1975 ret
= close(ust_cmd
->sock
);
1977 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1979 lttng_fd_put(LTTNG_FD_APPS
, 1);
1986 * @session_lock_list
1988 * Lock the global session list so from the register up to the
1989 * registration done message, no thread can see the application
1990 * and change its state.
1992 session_lock_list();
1996 * Add application to the global hash table. This needs to be
1997 * done before the update to the UST registry can locate the
2002 /* Set app version. This call will print an error if needed. */
2003 (void) ust_app_version(app
);
2005 /* Send notify socket through the notify pipe. */
2006 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
2010 session_unlock_list();
2012 * No notify thread, stop the UST tracing. However, this is
2013 * not an internal error of the this thread thus setting
2014 * the health error code to a normal exit.
2021 * Update newly registered application with the tracing
2022 * registry info already enabled information.
2024 update_ust_app(app
->sock
);
2027 * Don't care about return value. Let the manage apps threads
2028 * handle app unregistration upon socket close.
2030 (void) ust_app_register_done(app
);
2033 * Even if the application socket has been closed, send the app
2034 * to the thread and unregistration will take place at that
2037 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2040 session_unlock_list();
2042 * No apps. thread, stop the UST tracing. However, this is
2043 * not an internal error of the this thread thus setting
2044 * the health error code to a normal exit.
2051 session_unlock_list();
2053 } while (node
!= NULL
);
2055 health_poll_entry();
2056 /* Futex wait on queue. Blocking call on futex() */
2057 futex_nto1_wait(&ust_cmd_queue
.futex
);
2060 /* Normal exit, no error */
2064 /* Clean up wait queue. */
2065 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2066 &wait_queue
.head
, head
) {
2067 cds_list_del(&wait_node
->head
);
2072 /* Empty command queue. */
2074 /* Dequeue command for registration */
2075 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2079 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2080 ret
= close(ust_cmd
->sock
);
2082 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2084 lttng_fd_put(LTTNG_FD_APPS
, 1);
2089 DBG("Dispatch thread dying");
2092 ERR("Health error occurred in %s", __func__
);
2094 health_unregister(health_sessiond
);
2099 * This thread manage application registration.
2101 static void *thread_registration_apps(void *data
)
2103 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2104 uint32_t revents
, nb_fd
;
2105 struct lttng_poll_event events
;
2107 * Get allocated in this thread, enqueued to a global queue, dequeued and
2108 * freed in the manage apps thread.
2110 struct ust_command
*ust_cmd
= NULL
;
2112 DBG("[thread] Manage application registration started");
2114 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2116 if (testpoint(sessiond_thread_registration_apps
)) {
2117 goto error_testpoint
;
2120 ret
= lttcomm_listen_unix_sock(apps_sock
);
2126 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2127 * more will be added to this poll set.
2129 ret
= sessiond_set_thread_pollset(&events
, 2);
2131 goto error_create_poll
;
2134 /* Add the application registration socket */
2135 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2137 goto error_poll_add
;
2140 /* Notify all applications to register */
2141 ret
= notify_ust_apps(1);
2143 ERR("Failed to notify applications or create the wait shared memory.\n"
2144 "Execution continues but there might be problem for already\n"
2145 "running applications that wishes to register.");
2149 DBG("Accepting application registration");
2151 /* Inifinite blocking call, waiting for transmission */
2153 health_poll_entry();
2154 ret
= lttng_poll_wait(&events
, -1);
2158 * Restart interrupted system call.
2160 if (errno
== EINTR
) {
2168 for (i
= 0; i
< nb_fd
; i
++) {
2169 health_code_update();
2171 /* Fetch once the poll data */
2172 revents
= LTTNG_POLL_GETEV(&events
, i
);
2173 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2176 /* No activity for this FD (poll implementation). */
2180 /* Thread quit pipe has been closed. Killing thread. */
2181 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2187 /* Event on the registration socket */
2188 if (pollfd
== apps_sock
) {
2189 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2190 ERR("Register apps socket poll error");
2192 } else if (revents
& LPOLLIN
) {
2193 sock
= lttcomm_accept_unix_sock(apps_sock
);
2199 * Set socket timeout for both receiving and ending.
2200 * app_socket_timeout is in seconds, whereas
2201 * lttcomm_setsockopt_rcv_timeout and
2202 * lttcomm_setsockopt_snd_timeout expect msec as
2205 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2206 app_socket_timeout
* 1000);
2207 (void) lttcomm_setsockopt_snd_timeout(sock
,
2208 app_socket_timeout
* 1000);
2211 * Set the CLOEXEC flag. Return code is useless because
2212 * either way, the show must go on.
2214 (void) utils_set_fd_cloexec(sock
);
2216 /* Create UST registration command for enqueuing */
2217 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2218 if (ust_cmd
== NULL
) {
2219 PERROR("ust command zmalloc");
2228 * Using message-based transmissions to ensure we don't
2229 * have to deal with partially received messages.
2231 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2233 ERR("Exhausted file descriptors allowed for applications.");
2243 health_code_update();
2244 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2247 /* Close socket of the application. */
2252 lttng_fd_put(LTTNG_FD_APPS
, 1);
2256 health_code_update();
2258 ust_cmd
->sock
= sock
;
2261 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2262 " gid:%d sock:%d name:%s (version %d.%d)",
2263 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2264 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2265 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2266 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2269 * Lock free enqueue the registration request. The red pill
2270 * has been taken! This apps will be part of the *system*.
2272 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2275 * Wake the registration queue futex. Implicit memory
2276 * barrier with the exchange in cds_wfcq_enqueue.
2278 futex_nto1_wake(&ust_cmd_queue
.futex
);
2286 /* Notify that the registration thread is gone */
2289 if (apps_sock
>= 0) {
2290 ret
= close(apps_sock
);
2300 lttng_fd_put(LTTNG_FD_APPS
, 1);
2302 unlink(apps_unix_sock_path
);
2305 lttng_poll_clean(&events
);
2309 DBG("UST Registration thread cleanup complete");
2312 ERR("Health error occurred in %s", __func__
);
2314 health_unregister(health_sessiond
);
2320 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2321 * exec or it will fails.
2323 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2326 struct timespec timeout
;
2328 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2329 consumer_data
->consumer_thread_is_ready
= 0;
2331 /* Setup pthread condition */
2332 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2335 PERROR("pthread_condattr_init consumer data");
2340 * Set the monotonic clock in order to make sure we DO NOT jump in time
2341 * between the clock_gettime() call and the timedwait call. See bug #324
2342 * for a more details and how we noticed it.
2344 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2347 PERROR("pthread_condattr_setclock consumer data");
2351 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2354 PERROR("pthread_cond_init consumer data");
2358 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2362 PERROR("pthread_create consumer");
2367 /* We are about to wait on a pthread condition */
2368 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2370 /* Get time for sem_timedwait absolute timeout */
2371 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2373 * Set the timeout for the condition timed wait even if the clock gettime
2374 * call fails since we might loop on that call and we want to avoid to
2375 * increment the timeout too many times.
2377 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2380 * The following loop COULD be skipped in some conditions so this is why we
2381 * set ret to 0 in order to make sure at least one round of the loop is
2387 * Loop until the condition is reached or when a timeout is reached. Note
2388 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2389 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2390 * possible. This loop does not take any chances and works with both of
2393 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2394 if (clock_ret
< 0) {
2395 PERROR("clock_gettime spawn consumer");
2396 /* Infinite wait for the consumerd thread to be ready */
2397 ret
= pthread_cond_wait(&consumer_data
->cond
,
2398 &consumer_data
->cond_mutex
);
2400 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2401 &consumer_data
->cond_mutex
, &timeout
);
2405 /* Release the pthread condition */
2406 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2410 if (ret
== ETIMEDOUT
) {
2414 * Call has timed out so we kill the kconsumerd_thread and return
2417 ERR("Condition timed out. The consumer thread was never ready."
2419 pth_ret
= pthread_cancel(consumer_data
->thread
);
2421 PERROR("pthread_cancel consumer thread");
2424 PERROR("pthread_cond_wait failed consumer thread");
2426 /* Caller is expecting a negative value on failure. */
2431 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2432 if (consumer_data
->pid
== 0) {
2433 ERR("Consumerd did not start");
2434 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2437 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2446 * Join consumer thread
2448 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2452 /* Consumer pid must be a real one. */
2453 if (consumer_data
->pid
> 0) {
2455 ret
= kill(consumer_data
->pid
, SIGTERM
);
2457 PERROR("Error killing consumer daemon");
2460 return pthread_join(consumer_data
->thread
, &status
);
2467 * Fork and exec a consumer daemon (consumerd).
2469 * Return pid if successful else -1.
2471 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2475 const char *consumer_to_use
;
2476 const char *verbosity
;
2479 DBG("Spawning consumerd");
2486 if (opt_verbose_consumer
) {
2487 verbosity
= "--verbose";
2488 } else if (lttng_opt_quiet
) {
2489 verbosity
= "--quiet";
2494 switch (consumer_data
->type
) {
2495 case LTTNG_CONSUMER_KERNEL
:
2497 * Find out which consumerd to execute. We will first try the
2498 * 64-bit path, then the sessiond's installation directory, and
2499 * fallback on the 32-bit one,
2501 DBG3("Looking for a kernel consumer at these locations:");
2502 DBG3(" 1) %s", consumerd64_bin
);
2503 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2504 DBG3(" 3) %s", consumerd32_bin
);
2505 if (stat(consumerd64_bin
, &st
) == 0) {
2506 DBG3("Found location #1");
2507 consumer_to_use
= consumerd64_bin
;
2508 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2509 DBG3("Found location #2");
2510 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2511 } else if (stat(consumerd32_bin
, &st
) == 0) {
2512 DBG3("Found location #3");
2513 consumer_to_use
= consumerd32_bin
;
2515 DBG("Could not find any valid consumerd executable");
2519 DBG("Using kernel consumer at: %s", consumer_to_use
);
2520 ret
= execl(consumer_to_use
,
2521 "lttng-consumerd", verbosity
, "-k",
2522 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2523 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2524 "--group", tracing_group_name
,
2527 case LTTNG_CONSUMER64_UST
:
2529 char *tmpnew
= NULL
;
2531 if (consumerd64_libdir
[0] != '\0') {
2535 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2539 tmplen
= strlen("LD_LIBRARY_PATH=")
2540 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2541 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2546 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2547 strcat(tmpnew
, consumerd64_libdir
);
2548 if (tmp
[0] != '\0') {
2549 strcat(tmpnew
, ":");
2550 strcat(tmpnew
, tmp
);
2552 ret
= putenv(tmpnew
);
2559 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2560 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2561 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2562 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2563 "--group", tracing_group_name
,
2565 if (consumerd64_libdir
[0] != '\0') {
2570 case LTTNG_CONSUMER32_UST
:
2572 char *tmpnew
= NULL
;
2574 if (consumerd32_libdir
[0] != '\0') {
2578 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2582 tmplen
= strlen("LD_LIBRARY_PATH=")
2583 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2584 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2589 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2590 strcat(tmpnew
, consumerd32_libdir
);
2591 if (tmp
[0] != '\0') {
2592 strcat(tmpnew
, ":");
2593 strcat(tmpnew
, tmp
);
2595 ret
= putenv(tmpnew
);
2602 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2603 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2604 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2605 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2606 "--group", tracing_group_name
,
2608 if (consumerd32_libdir
[0] != '\0') {
2614 PERROR("unknown consumer type");
2618 PERROR("Consumer execl()");
2620 /* Reaching this point, we got a failure on our execl(). */
2622 } else if (pid
> 0) {
2625 PERROR("start consumer fork");
2633 * Spawn the consumerd daemon and session daemon thread.
2635 static int start_consumerd(struct consumer_data
*consumer_data
)
2640 * Set the listen() state on the socket since there is a possible race
2641 * between the exec() of the consumer daemon and this call if place in the
2642 * consumer thread. See bug #366 for more details.
2644 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2649 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2650 if (consumer_data
->pid
!= 0) {
2651 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2655 ret
= spawn_consumerd(consumer_data
);
2657 ERR("Spawning consumerd failed");
2658 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2662 /* Setting up the consumer_data pid */
2663 consumer_data
->pid
= ret
;
2664 DBG2("Consumer pid %d", consumer_data
->pid
);
2665 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2667 DBG2("Spawning consumer control thread");
2668 ret
= spawn_consumer_thread(consumer_data
);
2670 ERR("Fatal error spawning consumer control thread");
2678 /* Cleanup already created sockets on error. */
2679 if (consumer_data
->err_sock
>= 0) {
2682 err
= close(consumer_data
->err_sock
);
2684 PERROR("close consumer data error socket");
2691 * Setup necessary data for kernel tracer action.
2693 static int init_kernel_tracer(void)
2697 /* Modprobe lttng kernel modules */
2698 ret
= modprobe_lttng_control();
2703 /* Open debugfs lttng */
2704 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2705 if (kernel_tracer_fd
< 0) {
2706 DBG("Failed to open %s", module_proc_lttng
);
2711 /* Validate kernel version */
2712 ret
= kernel_validate_version(kernel_tracer_fd
);
2717 ret
= modprobe_lttng_data();
2722 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2726 modprobe_remove_lttng_control();
2727 ret
= close(kernel_tracer_fd
);
2731 kernel_tracer_fd
= -1;
2732 return LTTNG_ERR_KERN_VERSION
;
2735 ret
= close(kernel_tracer_fd
);
2741 modprobe_remove_lttng_control();
2744 WARN("No kernel tracer available");
2745 kernel_tracer_fd
= -1;
2747 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2749 return LTTNG_ERR_KERN_NA
;
2755 * Copy consumer output from the tracing session to the domain session. The
2756 * function also applies the right modification on a per domain basis for the
2757 * trace files destination directory.
2759 * Should *NOT* be called with RCU read-side lock held.
2761 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2764 const char *dir_name
;
2765 struct consumer_output
*consumer
;
2768 assert(session
->consumer
);
2771 case LTTNG_DOMAIN_KERNEL
:
2772 DBG3("Copying tracing session consumer output in kernel session");
2774 * XXX: We should audit the session creation and what this function
2775 * does "extra" in order to avoid a destroy since this function is used
2776 * in the domain session creation (kernel and ust) only. Same for UST
2779 if (session
->kernel_session
->consumer
) {
2780 consumer_output_put(session
->kernel_session
->consumer
);
2782 session
->kernel_session
->consumer
=
2783 consumer_copy_output(session
->consumer
);
2784 /* Ease our life a bit for the next part */
2785 consumer
= session
->kernel_session
->consumer
;
2786 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2788 case LTTNG_DOMAIN_JUL
:
2789 case LTTNG_DOMAIN_LOG4J
:
2790 case LTTNG_DOMAIN_PYTHON
:
2791 case LTTNG_DOMAIN_UST
:
2792 DBG3("Copying tracing session consumer output in UST session");
2793 if (session
->ust_session
->consumer
) {
2794 consumer_output_put(session
->ust_session
->consumer
);
2796 session
->ust_session
->consumer
=
2797 consumer_copy_output(session
->consumer
);
2798 /* Ease our life a bit for the next part */
2799 consumer
= session
->ust_session
->consumer
;
2800 dir_name
= DEFAULT_UST_TRACE_DIR
;
2803 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2807 /* Append correct directory to subdir */
2808 strncat(consumer
->subdir
, dir_name
,
2809 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2810 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2819 * Create an UST session and add it to the session ust list.
2821 * Should *NOT* be called with RCU read-side lock held.
2823 static int create_ust_session(struct ltt_session
*session
,
2824 struct lttng_domain
*domain
)
2827 struct ltt_ust_session
*lus
= NULL
;
2831 assert(session
->consumer
);
2833 switch (domain
->type
) {
2834 case LTTNG_DOMAIN_JUL
:
2835 case LTTNG_DOMAIN_LOG4J
:
2836 case LTTNG_DOMAIN_PYTHON
:
2837 case LTTNG_DOMAIN_UST
:
2840 ERR("Unknown UST domain on create session %d", domain
->type
);
2841 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2845 DBG("Creating UST session");
2847 lus
= trace_ust_create_session(session
->id
);
2849 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2853 lus
->uid
= session
->uid
;
2854 lus
->gid
= session
->gid
;
2855 lus
->output_traces
= session
->output_traces
;
2856 lus
->snapshot_mode
= session
->snapshot_mode
;
2857 lus
->live_timer_interval
= session
->live_timer
;
2858 session
->ust_session
= lus
;
2859 if (session
->shm_path
[0]) {
2860 strncpy(lus
->root_shm_path
, session
->shm_path
,
2861 sizeof(lus
->root_shm_path
));
2862 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2863 strncpy(lus
->shm_path
, session
->shm_path
,
2864 sizeof(lus
->shm_path
));
2865 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2866 strncat(lus
->shm_path
, "/ust",
2867 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2869 /* Copy session output to the newly created UST session */
2870 ret
= copy_session_consumer(domain
->type
, session
);
2871 if (ret
!= LTTNG_OK
) {
2879 session
->ust_session
= NULL
;
2884 * Create a kernel tracer session then create the default channel.
2886 static int create_kernel_session(struct ltt_session
*session
)
2890 DBG("Creating kernel session");
2892 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2894 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2898 /* Code flow safety */
2899 assert(session
->kernel_session
);
2901 /* Copy session output to the newly created Kernel session */
2902 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2903 if (ret
!= LTTNG_OK
) {
2907 /* Create directory(ies) on local filesystem. */
2908 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2909 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2910 ret
= run_as_mkdir_recursive(
2911 session
->kernel_session
->consumer
->dst
.trace_path
,
2912 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2914 if (errno
!= EEXIST
) {
2915 ERR("Trace directory creation error");
2921 session
->kernel_session
->uid
= session
->uid
;
2922 session
->kernel_session
->gid
= session
->gid
;
2923 session
->kernel_session
->output_traces
= session
->output_traces
;
2924 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2929 trace_kernel_destroy_session(session
->kernel_session
);
2930 session
->kernel_session
= NULL
;
2935 * Count number of session permitted by uid/gid.
2937 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2940 struct ltt_session
*session
;
2942 DBG("Counting number of available session for UID %d GID %d",
2944 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2946 * Only list the sessions the user can control.
2948 if (!session_access_ok(session
, uid
, gid
)) {
2957 * Process the command requested by the lttng client within the command
2958 * context structure. This function make sure that the return structure (llm)
2959 * is set and ready for transmission before returning.
2961 * Return any error encountered or 0 for success.
2963 * "sock" is only used for special-case var. len data.
2965 * Should *NOT* be called with RCU read-side lock held.
2967 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2971 int need_tracing_session
= 1;
2974 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2978 switch (cmd_ctx
->lsm
->cmd_type
) {
2979 case LTTNG_CREATE_SESSION
:
2980 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2981 case LTTNG_CREATE_SESSION_LIVE
:
2982 case LTTNG_DESTROY_SESSION
:
2983 case LTTNG_LIST_SESSIONS
:
2984 case LTTNG_LIST_DOMAINS
:
2985 case LTTNG_START_TRACE
:
2986 case LTTNG_STOP_TRACE
:
2987 case LTTNG_DATA_PENDING
:
2988 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2989 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2990 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2991 case LTTNG_SNAPSHOT_RECORD
:
2992 case LTTNG_SAVE_SESSION
:
2993 case LTTNG_SET_SESSION_SHM_PATH
:
3000 if (opt_no_kernel
&& need_domain
3001 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3003 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3005 ret
= LTTNG_ERR_KERN_NA
;
3010 /* Deny register consumer if we already have a spawned consumer. */
3011 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3012 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3013 if (kconsumer_data
.pid
> 0) {
3014 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3015 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3018 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3022 * Check for command that don't needs to allocate a returned payload. We do
3023 * this here so we don't have to make the call for no payload at each
3026 switch(cmd_ctx
->lsm
->cmd_type
) {
3027 case LTTNG_LIST_SESSIONS
:
3028 case LTTNG_LIST_TRACEPOINTS
:
3029 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3030 case LTTNG_LIST_DOMAINS
:
3031 case LTTNG_LIST_CHANNELS
:
3032 case LTTNG_LIST_EVENTS
:
3033 case LTTNG_LIST_SYSCALLS
:
3034 case LTTNG_LIST_TRACKER_PIDS
:
3037 /* Setup lttng message with no payload */
3038 ret
= setup_lttng_msg(cmd_ctx
, 0);
3040 /* This label does not try to unlock the session */
3041 goto init_setup_error
;
3045 /* Commands that DO NOT need a session. */
3046 switch (cmd_ctx
->lsm
->cmd_type
) {
3047 case LTTNG_CREATE_SESSION
:
3048 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3049 case LTTNG_CREATE_SESSION_LIVE
:
3050 case LTTNG_CALIBRATE
:
3051 case LTTNG_LIST_SESSIONS
:
3052 case LTTNG_LIST_TRACEPOINTS
:
3053 case LTTNG_LIST_SYSCALLS
:
3054 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3055 case LTTNG_SAVE_SESSION
:
3056 need_tracing_session
= 0;
3059 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3061 * We keep the session list lock across _all_ commands
3062 * for now, because the per-session lock does not
3063 * handle teardown properly.
3065 session_lock_list();
3066 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3067 if (cmd_ctx
->session
== NULL
) {
3068 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3071 /* Acquire lock for the session */
3072 session_lock(cmd_ctx
->session
);
3078 * Commands that need a valid session but should NOT create one if none
3079 * exists. Instead of creating one and destroying it when the command is
3080 * handled, process that right before so we save some round trip in useless
3083 switch (cmd_ctx
->lsm
->cmd_type
) {
3084 case LTTNG_DISABLE_CHANNEL
:
3085 case LTTNG_DISABLE_EVENT
:
3086 switch (cmd_ctx
->lsm
->domain
.type
) {
3087 case LTTNG_DOMAIN_KERNEL
:
3088 if (!cmd_ctx
->session
->kernel_session
) {
3089 ret
= LTTNG_ERR_NO_CHANNEL
;
3093 case LTTNG_DOMAIN_JUL
:
3094 case LTTNG_DOMAIN_LOG4J
:
3095 case LTTNG_DOMAIN_PYTHON
:
3096 case LTTNG_DOMAIN_UST
:
3097 if (!cmd_ctx
->session
->ust_session
) {
3098 ret
= LTTNG_ERR_NO_CHANNEL
;
3103 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3115 * Check domain type for specific "pre-action".
3117 switch (cmd_ctx
->lsm
->domain
.type
) {
3118 case LTTNG_DOMAIN_KERNEL
:
3120 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3124 /* Kernel tracer check */
3125 if (kernel_tracer_fd
== -1) {
3126 /* Basically, load kernel tracer modules */
3127 ret
= init_kernel_tracer();
3133 /* Consumer is in an ERROR state. Report back to client */
3134 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3135 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3139 /* Need a session for kernel command */
3140 if (need_tracing_session
) {
3141 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3142 ret
= create_kernel_session(cmd_ctx
->session
);
3144 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3149 /* Start the kernel consumer daemon */
3150 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3151 if (kconsumer_data
.pid
== 0 &&
3152 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3153 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3154 ret
= start_consumerd(&kconsumer_data
);
3156 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3159 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3161 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3165 * The consumer was just spawned so we need to add the socket to
3166 * the consumer output of the session if exist.
3168 ret
= consumer_create_socket(&kconsumer_data
,
3169 cmd_ctx
->session
->kernel_session
->consumer
);
3176 case LTTNG_DOMAIN_JUL
:
3177 case LTTNG_DOMAIN_LOG4J
:
3178 case LTTNG_DOMAIN_PYTHON
:
3179 case LTTNG_DOMAIN_UST
:
3181 if (!ust_app_supported()) {
3182 ret
= LTTNG_ERR_NO_UST
;
3185 /* Consumer is in an ERROR state. Report back to client */
3186 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3187 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3191 if (need_tracing_session
) {
3192 /* Create UST session if none exist. */
3193 if (cmd_ctx
->session
->ust_session
== NULL
) {
3194 ret
= create_ust_session(cmd_ctx
->session
,
3195 &cmd_ctx
->lsm
->domain
);
3196 if (ret
!= LTTNG_OK
) {
3201 /* Start the UST consumer daemons */
3203 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3204 if (consumerd64_bin
[0] != '\0' &&
3205 ustconsumer64_data
.pid
== 0 &&
3206 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3207 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3208 ret
= start_consumerd(&ustconsumer64_data
);
3210 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3211 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3215 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3216 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3218 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3222 * Setup socket for consumer 64 bit. No need for atomic access
3223 * since it was set above and can ONLY be set in this thread.
3225 ret
= consumer_create_socket(&ustconsumer64_data
,
3226 cmd_ctx
->session
->ust_session
->consumer
);
3232 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3233 if (consumerd32_bin
[0] != '\0' &&
3234 ustconsumer32_data
.pid
== 0 &&
3235 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3236 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3237 ret
= start_consumerd(&ustconsumer32_data
);
3239 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3240 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3244 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3245 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3247 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3251 * Setup socket for consumer 64 bit. No need for atomic access
3252 * since it was set above and can ONLY be set in this thread.
3254 ret
= consumer_create_socket(&ustconsumer32_data
,
3255 cmd_ctx
->session
->ust_session
->consumer
);
3267 /* Validate consumer daemon state when start/stop trace command */
3268 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3269 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3270 switch (cmd_ctx
->lsm
->domain
.type
) {
3271 case LTTNG_DOMAIN_JUL
:
3272 case LTTNG_DOMAIN_LOG4J
:
3273 case LTTNG_DOMAIN_PYTHON
:
3274 case LTTNG_DOMAIN_UST
:
3275 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3276 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3280 case LTTNG_DOMAIN_KERNEL
:
3281 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3282 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3290 * Check that the UID or GID match that of the tracing session.
3291 * The root user can interact with all sessions.
3293 if (need_tracing_session
) {
3294 if (!session_access_ok(cmd_ctx
->session
,
3295 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3296 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3297 ret
= LTTNG_ERR_EPERM
;
3303 * Send relayd information to consumer as soon as we have a domain and a
3306 if (cmd_ctx
->session
&& need_domain
) {
3308 * Setup relayd if not done yet. If the relayd information was already
3309 * sent to the consumer, this call will gracefully return.
3311 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3312 if (ret
!= LTTNG_OK
) {
3317 /* Process by command type */
3318 switch (cmd_ctx
->lsm
->cmd_type
) {
3319 case LTTNG_ADD_CONTEXT
:
3321 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3322 cmd_ctx
->lsm
->u
.context
.channel_name
,
3323 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3326 case LTTNG_DISABLE_CHANNEL
:
3328 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3329 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3332 case LTTNG_DISABLE_EVENT
:
3336 * FIXME: handle filter; for now we just receive the filter's
3337 * bytecode along with the filter expression which are sent by
3338 * liblttng-ctl and discard them.
3340 * This fixes an issue where the client may block while sending
3341 * the filter payload and encounter an error because the session
3342 * daemon closes the socket without ever handling this data.
3344 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3345 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3348 char data
[LTTNG_FILTER_MAX_LEN
];
3350 DBG("Discarding disable event command payload of size %zu", count
);
3352 ret
= lttcomm_recv_unix_sock(sock
, data
,
3353 count
> sizeof(data
) ? sizeof(data
) : count
);
3358 count
-= (size_t) ret
;
3361 /* FIXME: passing packed structure to non-packed pointer */
3362 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3363 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3364 &cmd_ctx
->lsm
->u
.disable
.event
);
3367 case LTTNG_ENABLE_CHANNEL
:
3369 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3370 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3373 case LTTNG_TRACK_PID
:
3375 ret
= cmd_track_pid(cmd_ctx
->session
,
3376 cmd_ctx
->lsm
->domain
.type
,
3377 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3380 case LTTNG_UNTRACK_PID
:
3382 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3383 cmd_ctx
->lsm
->domain
.type
,
3384 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3387 case LTTNG_ENABLE_EVENT
:
3389 struct lttng_event_exclusion
*exclusion
= NULL
;
3390 struct lttng_filter_bytecode
*bytecode
= NULL
;
3391 char *filter_expression
= NULL
;
3393 /* Handle exclusion events and receive it from the client. */
3394 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3395 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3397 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3398 (count
* LTTNG_SYMBOL_NAME_LEN
));
3400 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3404 DBG("Receiving var len exclusion event list from client ...");
3405 exclusion
->count
= count
;
3406 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3407 count
* LTTNG_SYMBOL_NAME_LEN
);
3409 DBG("Nothing recv() from client var len data... continuing");
3412 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3417 /* Get filter expression from client. */
3418 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3419 size_t expression_len
=
3420 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3422 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3423 ret
= LTTNG_ERR_FILTER_INVAL
;
3428 filter_expression
= zmalloc(expression_len
);
3429 if (!filter_expression
) {
3431 ret
= LTTNG_ERR_FILTER_NOMEM
;
3435 /* Receive var. len. data */
3436 DBG("Receiving var len filter's expression from client ...");
3437 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3440 DBG("Nothing recv() from client car len data... continuing");
3442 free(filter_expression
);
3444 ret
= LTTNG_ERR_FILTER_INVAL
;
3449 /* Handle filter and get bytecode from client. */
3450 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3451 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3453 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3454 ret
= LTTNG_ERR_FILTER_INVAL
;
3455 free(filter_expression
);
3460 bytecode
= zmalloc(bytecode_len
);
3462 free(filter_expression
);
3464 ret
= LTTNG_ERR_FILTER_NOMEM
;
3468 /* Receive var. len. data */
3469 DBG("Receiving var len filter's bytecode from client ...");
3470 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3472 DBG("Nothing recv() from client car len data... continuing");
3474 free(filter_expression
);
3477 ret
= LTTNG_ERR_FILTER_INVAL
;
3481 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3482 free(filter_expression
);
3485 ret
= LTTNG_ERR_FILTER_INVAL
;
3490 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3491 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3492 &cmd_ctx
->lsm
->u
.enable
.event
,
3493 filter_expression
, bytecode
, exclusion
,
3494 kernel_poll_pipe
[1]);
3497 case LTTNG_LIST_TRACEPOINTS
:
3499 struct lttng_event
*events
;
3502 session_lock_list();
3503 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3504 session_unlock_list();
3505 if (nb_events
< 0) {
3506 /* Return value is a negative lttng_error_code. */
3512 * Setup lttng message with payload size set to the event list size in
3513 * bytes and then copy list into the llm payload.
3515 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3521 /* Copy event list into message payload */
3522 memcpy(cmd_ctx
->llm
->payload
, events
,
3523 sizeof(struct lttng_event
) * nb_events
);
3530 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3532 struct lttng_event_field
*fields
;
3535 session_lock_list();
3536 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3538 session_unlock_list();
3539 if (nb_fields
< 0) {
3540 /* Return value is a negative lttng_error_code. */
3546 * Setup lttng message with payload size set to the event list size in
3547 * bytes and then copy list into the llm payload.
3549 ret
= setup_lttng_msg(cmd_ctx
,
3550 sizeof(struct lttng_event_field
) * nb_fields
);
3556 /* Copy event list into message payload */
3557 memcpy(cmd_ctx
->llm
->payload
, fields
,
3558 sizeof(struct lttng_event_field
) * nb_fields
);
3565 case LTTNG_LIST_SYSCALLS
:
3567 struct lttng_event
*events
;
3570 nb_events
= cmd_list_syscalls(&events
);
3571 if (nb_events
< 0) {
3572 /* Return value is a negative lttng_error_code. */
3578 * Setup lttng message with payload size set to the event list size in
3579 * bytes and then copy list into the llm payload.
3581 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3587 /* Copy event list into message payload */
3588 memcpy(cmd_ctx
->llm
->payload
, events
,
3589 sizeof(struct lttng_event
) * nb_events
);
3596 case LTTNG_LIST_TRACKER_PIDS
:
3598 int32_t *pids
= NULL
;
3601 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3602 cmd_ctx
->lsm
->domain
.type
, &pids
);
3604 /* Return value is a negative lttng_error_code. */
3610 * Setup lttng message with payload size set to the event list size in
3611 * bytes and then copy list into the llm payload.
3613 ret
= setup_lttng_msg(cmd_ctx
, sizeof(int32_t) * nr_pids
);
3619 /* Copy event list into message payload */
3620 memcpy(cmd_ctx
->llm
->payload
, pids
,
3621 sizeof(int) * nr_pids
);
3628 case LTTNG_SET_CONSUMER_URI
:
3631 struct lttng_uri
*uris
;
3633 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3634 len
= nb_uri
* sizeof(struct lttng_uri
);
3637 ret
= LTTNG_ERR_INVALID
;
3641 uris
= zmalloc(len
);
3643 ret
= LTTNG_ERR_FATAL
;
3647 /* Receive variable len data */
3648 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3649 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3651 DBG("No URIs received from client... continuing");
3653 ret
= LTTNG_ERR_SESSION_FAIL
;
3658 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3660 if (ret
!= LTTNG_OK
) {
3667 case LTTNG_START_TRACE
:
3669 ret
= cmd_start_trace(cmd_ctx
->session
);
3672 case LTTNG_STOP_TRACE
:
3674 ret
= cmd_stop_trace(cmd_ctx
->session
);
3677 case LTTNG_CREATE_SESSION
:
3680 struct lttng_uri
*uris
= NULL
;
3682 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3683 len
= nb_uri
* sizeof(struct lttng_uri
);
3686 uris
= zmalloc(len
);
3688 ret
= LTTNG_ERR_FATAL
;
3692 /* Receive variable len data */
3693 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3694 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3696 DBG("No URIs received from client... continuing");
3698 ret
= LTTNG_ERR_SESSION_FAIL
;
3703 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3704 DBG("Creating session with ONE network URI is a bad call");
3705 ret
= LTTNG_ERR_SESSION_FAIL
;
3711 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3712 &cmd_ctx
->creds
, 0);
3718 case LTTNG_DESTROY_SESSION
:
3720 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3722 /* Set session to NULL so we do not unlock it after free. */
3723 cmd_ctx
->session
= NULL
;
3726 case LTTNG_LIST_DOMAINS
:
3729 struct lttng_domain
*domains
= NULL
;
3731 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3733 /* Return value is a negative lttng_error_code. */
3738 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3744 /* Copy event list into message payload */
3745 memcpy(cmd_ctx
->llm
->payload
, domains
,
3746 nb_dom
* sizeof(struct lttng_domain
));
3753 case LTTNG_LIST_CHANNELS
:
3756 struct lttng_channel
*channels
= NULL
;
3758 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3759 cmd_ctx
->session
, &channels
);
3761 /* Return value is a negative lttng_error_code. */
3766 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3772 /* Copy event list into message payload */
3773 memcpy(cmd_ctx
->llm
->payload
, channels
,
3774 nb_chan
* sizeof(struct lttng_channel
));
3781 case LTTNG_LIST_EVENTS
:
3784 struct lttng_event
*events
= NULL
;
3786 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3787 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3789 /* Return value is a negative lttng_error_code. */
3794 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3800 /* Copy event list into message payload */
3801 memcpy(cmd_ctx
->llm
->payload
, events
,
3802 nb_event
* sizeof(struct lttng_event
));
3809 case LTTNG_LIST_SESSIONS
:
3811 unsigned int nr_sessions
;
3813 session_lock_list();
3814 nr_sessions
= lttng_sessions_count(
3815 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3816 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3818 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3820 session_unlock_list();
3824 /* Filled the session array */
3825 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3826 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3827 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3829 session_unlock_list();
3834 case LTTNG_CALIBRATE
:
3836 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3837 &cmd_ctx
->lsm
->u
.calibrate
);
3840 case LTTNG_REGISTER_CONSUMER
:
3842 struct consumer_data
*cdata
;
3844 switch (cmd_ctx
->lsm
->domain
.type
) {
3845 case LTTNG_DOMAIN_KERNEL
:
3846 cdata
= &kconsumer_data
;
3849 ret
= LTTNG_ERR_UND
;
3853 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3854 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3857 case LTTNG_DATA_PENDING
:
3861 /* 1 byte to return whether or not data is pending */
3862 ret
= setup_lttng_msg(cmd_ctx
, 1);
3867 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
3871 * This function may returns 0 or 1 to indicate whether or not
3872 * there is data pending. In case of error, it should return an
3873 * LTTNG_ERR code. However, some code paths may still return
3874 * a nondescript error code, which we handle by returning an
3877 if (pending_ret
== 0 || pending_ret
== 1) {
3879 } else if (pending_ret
< 0) {
3880 ret
= LTTNG_ERR_UNK
;
3887 *cmd_ctx
->llm
->payload
= (uint8_t) pending_ret
;
3890 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3892 struct lttcomm_lttng_output_id reply
;
3894 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3895 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3896 if (ret
!= LTTNG_OK
) {
3900 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3905 /* Copy output list into message payload */
3906 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3910 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3912 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3913 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3916 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3919 struct lttng_snapshot_output
*outputs
= NULL
;
3921 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3922 if (nb_output
< 0) {
3927 ret
= setup_lttng_msg(cmd_ctx
,
3928 nb_output
* sizeof(struct lttng_snapshot_output
));
3935 /* Copy output list into message payload */
3936 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3937 nb_output
* sizeof(struct lttng_snapshot_output
));
3944 case LTTNG_SNAPSHOT_RECORD
:
3946 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3947 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3948 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3951 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3954 struct lttng_uri
*uris
= NULL
;
3956 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3957 len
= nb_uri
* sizeof(struct lttng_uri
);
3960 uris
= zmalloc(len
);
3962 ret
= LTTNG_ERR_FATAL
;
3966 /* Receive variable len data */
3967 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3968 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3970 DBG("No URIs received from client... continuing");
3972 ret
= LTTNG_ERR_SESSION_FAIL
;
3977 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3978 DBG("Creating session with ONE network URI is a bad call");
3979 ret
= LTTNG_ERR_SESSION_FAIL
;
3985 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3986 nb_uri
, &cmd_ctx
->creds
);
3990 case LTTNG_CREATE_SESSION_LIVE
:
3993 struct lttng_uri
*uris
= NULL
;
3995 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3996 len
= nb_uri
* sizeof(struct lttng_uri
);
3999 uris
= zmalloc(len
);
4001 ret
= LTTNG_ERR_FATAL
;
4005 /* Receive variable len data */
4006 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4007 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4009 DBG("No URIs received from client... continuing");
4011 ret
= LTTNG_ERR_SESSION_FAIL
;
4016 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4017 DBG("Creating session with ONE network URI is a bad call");
4018 ret
= LTTNG_ERR_SESSION_FAIL
;
4024 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4025 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4029 case LTTNG_SAVE_SESSION
:
4031 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4035 case LTTNG_SET_SESSION_SHM_PATH
:
4037 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4038 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4042 ret
= LTTNG_ERR_UND
;
4047 if (cmd_ctx
->llm
== NULL
) {
4048 DBG("Missing llm structure. Allocating one.");
4049 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
4053 /* Set return code */
4054 cmd_ctx
->llm
->ret_code
= ret
;
4056 if (cmd_ctx
->session
) {
4057 session_unlock(cmd_ctx
->session
);
4059 if (need_tracing_session
) {
4060 session_unlock_list();
4067 * Thread managing health check socket.
4069 static void *thread_manage_health(void *data
)
4071 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4072 uint32_t revents
, nb_fd
;
4073 struct lttng_poll_event events
;
4074 struct health_comm_msg msg
;
4075 struct health_comm_reply reply
;
4077 DBG("[thread] Manage health check started");
4079 rcu_register_thread();
4081 /* We might hit an error path before this is created. */
4082 lttng_poll_init(&events
);
4084 /* Create unix socket */
4085 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4087 ERR("Unable to create health check Unix socket");
4093 /* lttng health client socket path permissions */
4094 ret
= chown(health_unix_sock_path
, 0,
4095 utils_get_group_id(tracing_group_name
));
4097 ERR("Unable to set group on %s", health_unix_sock_path
);
4103 ret
= chmod(health_unix_sock_path
,
4104 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4106 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4114 * Set the CLOEXEC flag. Return code is useless because either way, the
4117 (void) utils_set_fd_cloexec(sock
);
4119 ret
= lttcomm_listen_unix_sock(sock
);
4125 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4126 * more will be added to this poll set.
4128 ret
= sessiond_set_thread_pollset(&events
, 2);
4133 /* Add the application registration socket */
4134 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4139 sessiond_notify_ready();
4142 DBG("Health check ready");
4144 /* Inifinite blocking call, waiting for transmission */
4146 ret
= lttng_poll_wait(&events
, -1);
4149 * Restart interrupted system call.
4151 if (errno
== EINTR
) {
4159 for (i
= 0; i
< nb_fd
; i
++) {
4160 /* Fetch once the poll data */
4161 revents
= LTTNG_POLL_GETEV(&events
, i
);
4162 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4165 /* No activity for this FD (poll implementation). */
4169 /* Thread quit pipe has been closed. Killing thread. */
4170 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4176 /* Event on the registration socket */
4177 if (pollfd
== sock
) {
4178 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4179 ERR("Health socket poll error");
4185 new_sock
= lttcomm_accept_unix_sock(sock
);
4191 * Set the CLOEXEC flag. Return code is useless because either way, the
4194 (void) utils_set_fd_cloexec(new_sock
);
4196 DBG("Receiving data from client for health...");
4197 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4199 DBG("Nothing recv() from client... continuing");
4200 ret
= close(new_sock
);
4208 rcu_thread_online();
4210 memset(&reply
, 0, sizeof(reply
));
4211 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4213 * health_check_state returns 0 if health is
4216 if (!health_check_state(health_sessiond
, i
)) {
4217 reply
.ret_code
|= 1ULL << i
;
4221 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4223 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4225 ERR("Failed to send health data back to client");
4228 /* End of transmission */
4229 ret
= close(new_sock
);
4239 ERR("Health error occurred in %s", __func__
);
4241 DBG("Health check thread dying");
4242 unlink(health_unix_sock_path
);
4250 lttng_poll_clean(&events
);
4252 rcu_unregister_thread();
4257 * This thread manage all clients request using the unix client socket for
4260 static void *thread_manage_clients(void *data
)
4262 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4264 uint32_t revents
, nb_fd
;
4265 struct command_ctx
*cmd_ctx
= NULL
;
4266 struct lttng_poll_event events
;
4268 DBG("[thread] Manage client started");
4270 rcu_register_thread();
4272 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4274 health_code_update();
4276 ret
= lttcomm_listen_unix_sock(client_sock
);
4282 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4283 * more will be added to this poll set.
4285 ret
= sessiond_set_thread_pollset(&events
, 2);
4287 goto error_create_poll
;
4290 /* Add the application registration socket */
4291 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4296 sessiond_notify_ready();
4297 ret
= sem_post(&load_info
->message_thread_ready
);
4299 PERROR("sem_post message_thread_ready");
4303 /* This testpoint is after we signal readiness to the parent. */
4304 if (testpoint(sessiond_thread_manage_clients
)) {
4308 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4312 health_code_update();
4315 DBG("Accepting client command ...");
4317 /* Inifinite blocking call, waiting for transmission */
4319 health_poll_entry();
4320 ret
= lttng_poll_wait(&events
, -1);
4324 * Restart interrupted system call.
4326 if (errno
== EINTR
) {
4334 for (i
= 0; i
< nb_fd
; i
++) {
4335 /* Fetch once the poll data */
4336 revents
= LTTNG_POLL_GETEV(&events
, i
);
4337 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4339 health_code_update();
4342 /* No activity for this FD (poll implementation). */
4346 /* Thread quit pipe has been closed. Killing thread. */
4347 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4353 /* Event on the registration socket */
4354 if (pollfd
== client_sock
) {
4355 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4356 ERR("Client socket poll error");
4362 DBG("Wait for client response");
4364 health_code_update();
4366 sock
= lttcomm_accept_unix_sock(client_sock
);
4372 * Set the CLOEXEC flag. Return code is useless because either way, the
4375 (void) utils_set_fd_cloexec(sock
);
4377 /* Set socket option for credentials retrieval */
4378 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4383 /* Allocate context command to process the client request */
4384 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4385 if (cmd_ctx
== NULL
) {
4386 PERROR("zmalloc cmd_ctx");
4390 /* Allocate data buffer for reception */
4391 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4392 if (cmd_ctx
->lsm
== NULL
) {
4393 PERROR("zmalloc cmd_ctx->lsm");
4397 cmd_ctx
->llm
= NULL
;
4398 cmd_ctx
->session
= NULL
;
4400 health_code_update();
4403 * Data is received from the lttng client. The struct
4404 * lttcomm_session_msg (lsm) contains the command and data request of
4407 DBG("Receiving data from client ...");
4408 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4409 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4411 DBG("Nothing recv() from client... continuing");
4417 clean_command_ctx(&cmd_ctx
);
4421 health_code_update();
4423 // TODO: Validate cmd_ctx including sanity check for
4424 // security purpose.
4426 rcu_thread_online();
4428 * This function dispatch the work to the kernel or userspace tracer
4429 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4430 * informations for the client. The command context struct contains
4431 * everything this function may needs.
4433 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4434 rcu_thread_offline();
4442 * TODO: Inform client somehow of the fatal error. At
4443 * this point, ret < 0 means that a zmalloc failed
4444 * (ENOMEM). Error detected but still accept
4445 * command, unless a socket error has been
4448 clean_command_ctx(&cmd_ctx
);
4452 health_code_update();
4454 DBG("Sending response (size: %d, retcode: %s (%d))",
4455 cmd_ctx
->lttng_msg_size
,
4456 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4457 cmd_ctx
->llm
->ret_code
);
4458 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4460 ERR("Failed to send data back to client");
4463 /* End of transmission */
4470 clean_command_ctx(&cmd_ctx
);
4472 health_code_update();
4484 lttng_poll_clean(&events
);
4485 clean_command_ctx(&cmd_ctx
);
4489 unlink(client_unix_sock_path
);
4490 if (client_sock
>= 0) {
4491 ret
= close(client_sock
);
4499 ERR("Health error occurred in %s", __func__
);
4502 health_unregister(health_sessiond
);
4504 DBG("Client thread dying");
4506 rcu_unregister_thread();
4509 * Since we are creating the consumer threads, we own them, so we need
4510 * to join them before our thread exits.
4512 ret
= join_consumer_thread(&kconsumer_data
);
4515 PERROR("join_consumer");
4518 ret
= join_consumer_thread(&ustconsumer32_data
);
4521 PERROR("join_consumer ust32");
4524 ret
= join_consumer_thread(&ustconsumer64_data
);
4527 PERROR("join_consumer ust64");
4534 * usage function on stderr
4536 static void usage(void)
4538 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4539 fprintf(stderr
, " -h, --help Display this usage.\n");
4540 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4541 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4542 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4543 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4544 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4545 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4546 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4547 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4548 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4549 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4550 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4551 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4552 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4553 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4554 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4555 fprintf(stderr
, " -V, --version Show version number.\n");
4556 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4557 fprintf(stderr
, " -q, --quiet No output at all.\n");
4558 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4559 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4560 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4561 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4562 fprintf(stderr
, " --agent-tcp-port Agent registration TCP port\n");
4563 fprintf(stderr
, " -f --config PATH Load daemon configuration file\n");
4564 fprintf(stderr
, " -l --load PATH Load session configuration\n");
4565 fprintf(stderr
, " --kmod-probes Specify kernel module probes to load\n");
4566 fprintf(stderr
, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4569 static int string_match(const char *str1
, const char *str2
)
4571 return (str1
&& str2
) && !strcmp(str1
, str2
);
4575 * Take an option from the getopt output and set it in the right variable to be
4578 * Return 0 on success else a negative value.
4580 static int set_option(int opt
, const char *arg
, const char *optname
)
4584 if (arg
&& arg
[0] == '\0') {
4586 * This only happens if the value is read from daemon config
4587 * file. This means the option requires an argument and the
4588 * configuration file contains a line such as:
4595 if (string_match(optname
, "client-sock") || opt
== 'c') {
4596 if (lttng_is_setuid_setgid()) {
4597 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4598 "-c, --client-sock");
4600 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4602 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4603 if (lttng_is_setuid_setgid()) {
4604 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4607 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4609 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4611 } else if (string_match(optname
, "background") || opt
== 'b') {
4613 } else if (string_match(optname
, "group") || opt
== 'g') {
4614 if (lttng_is_setuid_setgid()) {
4615 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4619 * If the override option is set, the pointer points to a
4620 * *non* const thus freeing it even though the variable type is
4623 if (tracing_group_name_override
) {
4624 free((void *) tracing_group_name
);
4626 tracing_group_name
= strdup(arg
);
4627 if (!tracing_group_name
) {
4631 tracing_group_name_override
= 1;
4633 } else if (string_match(optname
, "help") || opt
== 'h') {
4636 } else if (string_match(optname
, "version") || opt
== 'V') {
4637 fprintf(stdout
, "%s\n", VERSION
);
4639 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4641 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4642 if (lttng_is_setuid_setgid()) {
4643 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4644 "--kconsumerd-err-sock");
4646 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4648 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4649 if (lttng_is_setuid_setgid()) {
4650 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4651 "--kconsumerd-cmd-sock");
4653 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4655 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4656 if (lttng_is_setuid_setgid()) {
4657 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4658 "--ustconsumerd64-err-sock");
4660 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4662 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4663 if (lttng_is_setuid_setgid()) {
4664 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4665 "--ustconsumerd64-cmd-sock");
4667 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4669 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4670 if (lttng_is_setuid_setgid()) {
4671 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4672 "--ustconsumerd32-err-sock");
4674 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4676 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4677 if (lttng_is_setuid_setgid()) {
4678 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4679 "--ustconsumerd32-cmd-sock");
4681 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4683 } else if (string_match(optname
, "no-kernel")) {
4685 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4686 lttng_opt_quiet
= 1;
4687 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4688 /* Verbose level can increase using multiple -v */
4690 /* Value obtained from config file */
4691 lttng_opt_verbose
= config_parse_value(arg
);
4693 /* -v used on command line */
4694 lttng_opt_verbose
++;
4696 /* Clamp value to [0, 3] */
4697 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4698 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4699 } else if (string_match(optname
, "verbose-consumer")) {
4701 opt_verbose_consumer
= config_parse_value(arg
);
4703 opt_verbose_consumer
+= 1;
4705 } else if (string_match(optname
, "consumerd32-path")) {
4706 if (lttng_is_setuid_setgid()) {
4707 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4708 "--consumerd32-path");
4710 if (consumerd32_bin_override
) {
4711 free((void *) consumerd32_bin
);
4713 consumerd32_bin
= strdup(arg
);
4714 if (!consumerd32_bin
) {
4718 consumerd32_bin_override
= 1;
4720 } else if (string_match(optname
, "consumerd32-libdir")) {
4721 if (lttng_is_setuid_setgid()) {
4722 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4723 "--consumerd32-libdir");
4725 if (consumerd32_libdir_override
) {
4726 free((void *) consumerd32_libdir
);
4728 consumerd32_libdir
= strdup(arg
);
4729 if (!consumerd32_libdir
) {
4733 consumerd32_libdir_override
= 1;
4735 } else if (string_match(optname
, "consumerd64-path")) {
4736 if (lttng_is_setuid_setgid()) {
4737 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4738 "--consumerd64-path");
4740 if (consumerd64_bin_override
) {
4741 free((void *) consumerd64_bin
);
4743 consumerd64_bin
= strdup(arg
);
4744 if (!consumerd64_bin
) {
4748 consumerd64_bin_override
= 1;
4750 } else if (string_match(optname
, "consumerd64-libdir")) {
4751 if (lttng_is_setuid_setgid()) {
4752 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4753 "--consumerd64-libdir");
4755 if (consumerd64_libdir_override
) {
4756 free((void *) consumerd64_libdir
);
4758 consumerd64_libdir
= strdup(arg
);
4759 if (!consumerd64_libdir
) {
4763 consumerd64_libdir_override
= 1;
4765 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4766 if (lttng_is_setuid_setgid()) {
4767 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4771 opt_pidfile
= strdup(arg
);
4777 } else if (string_match(optname
, "agent-tcp-port")) {
4778 if (lttng_is_setuid_setgid()) {
4779 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4780 "--agent-tcp-port");
4789 v
= strtoul(arg
, NULL
, 0);
4790 if (errno
!= 0 || !isdigit(arg
[0])) {
4791 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4794 if (v
== 0 || v
>= 65535) {
4795 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4798 agent_tcp_port
= (uint32_t) v
;
4799 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4801 } else if (string_match(optname
, "load") || opt
== 'l') {
4802 if (lttng_is_setuid_setgid()) {
4803 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4806 free(opt_load_session_path
);
4807 opt_load_session_path
= strdup(arg
);
4808 if (!opt_load_session_path
) {
4813 } else if (string_match(optname
, "kmod-probes")) {
4814 if (lttng_is_setuid_setgid()) {
4815 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4818 free(kmod_probes_list
);
4819 kmod_probes_list
= strdup(arg
);
4820 if (!kmod_probes_list
) {
4825 } else if (string_match(optname
, "extra-kmod-probes")) {
4826 if (lttng_is_setuid_setgid()) {
4827 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4828 "--extra-kmod-probes");
4830 free(kmod_extra_probes_list
);
4831 kmod_extra_probes_list
= strdup(arg
);
4832 if (!kmod_extra_probes_list
) {
4837 } else if (string_match(optname
, "config") || opt
== 'f') {
4838 /* This is handled in set_options() thus silent skip. */
4841 /* Unknown option or other error.
4842 * Error is printed by getopt, just return */
4847 if (ret
== -EINVAL
) {
4848 const char *opt_name
= "unknown";
4851 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4853 if (opt
== long_options
[i
].val
) {
4854 opt_name
= long_options
[i
].name
;
4859 WARN("Invalid argument provided for option \"%s\", using default value.",
4867 * config_entry_handler_cb used to handle options read from a config file.
4868 * See config_entry_handler_cb comment in common/config/config.h for the
4869 * return value conventions.
4871 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4875 if (!entry
|| !entry
->name
|| !entry
->value
) {
4880 /* Check if the option is to be ignored */
4881 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4882 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4887 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4890 /* Ignore if not fully matched. */
4891 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4896 * If the option takes no argument on the command line, we have to
4897 * check if the value is "true". We support non-zero numeric values,
4900 if (!long_options
[i
].has_arg
) {
4901 ret
= config_parse_value(entry
->value
);
4904 WARN("Invalid configuration value \"%s\" for option %s",
4905 entry
->value
, entry
->name
);
4907 /* False, skip boolean config option. */
4912 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4916 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4923 * daemon configuration loading and argument parsing
4925 static int set_options(int argc
, char **argv
)
4927 int ret
= 0, c
= 0, option_index
= 0;
4928 int orig_optopt
= optopt
, orig_optind
= optind
;
4930 const char *config_path
= NULL
;
4932 optstring
= utils_generate_optstring(long_options
,
4933 sizeof(long_options
) / sizeof(struct option
));
4939 /* Check for the --config option */
4940 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4941 &option_index
)) != -1) {
4945 } else if (c
!= 'f') {
4946 /* if not equal to --config option. */
4950 if (lttng_is_setuid_setgid()) {
4951 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4954 config_path
= utils_expand_path(optarg
);
4956 ERR("Failed to resolve path: %s", optarg
);
4961 ret
= config_get_section_entries(config_path
, config_section_name
,
4962 config_entry_handler
, NULL
);
4965 ERR("Invalid configuration option at line %i", ret
);
4971 /* Reset getopt's global state */
4972 optopt
= orig_optopt
;
4973 optind
= orig_optind
;
4977 * getopt_long() will not set option_index if it encounters a
4980 c
= getopt_long(argc
, argv
, optstring
, long_options
,
4987 * Pass NULL as the long option name if popt left the index
4990 ret
= set_option(c
, optarg
,
4991 option_index
< 0 ? NULL
:
4992 long_options
[option_index
].name
);
5004 * Creates the two needed socket by the daemon.
5005 * apps_sock - The communication socket for all UST apps.
5006 * client_sock - The communication of the cli tool (lttng).
5008 static int init_daemon_socket(void)
5013 old_umask
= umask(0);
5015 /* Create client tool unix socket */
5016 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
5017 if (client_sock
< 0) {
5018 ERR("Create unix sock failed: %s", client_unix_sock_path
);
5023 /* Set the cloexec flag */
5024 ret
= utils_set_fd_cloexec(client_sock
);
5026 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5027 "Continuing but note that the consumer daemon will have a "
5028 "reference to this socket on exec()", client_sock
);
5031 /* File permission MUST be 660 */
5032 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5034 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5039 /* Create the application unix socket */
5040 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5041 if (apps_sock
< 0) {
5042 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5047 /* Set the cloexec flag */
5048 ret
= utils_set_fd_cloexec(apps_sock
);
5050 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5051 "Continuing but note that the consumer daemon will have a "
5052 "reference to this socket on exec()", apps_sock
);
5055 /* File permission MUST be 666 */
5056 ret
= chmod(apps_unix_sock_path
,
5057 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5059 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5064 DBG3("Session daemon client socket %d and application socket %d created",
5065 client_sock
, apps_sock
);
5073 * Check if the global socket is available, and if a daemon is answering at the
5074 * other side. If yes, error is returned.
5076 static int check_existing_daemon(void)
5078 /* Is there anybody out there ? */
5079 if (lttng_session_daemon_alive()) {
5087 * Set the tracing group gid onto the client socket.
5089 * Race window between mkdir and chown is OK because we are going from more
5090 * permissive (root.root) to less permissive (root.tracing).
5092 static int set_permissions(char *rundir
)
5097 gid
= utils_get_group_id(tracing_group_name
);
5099 /* Set lttng run dir */
5100 ret
= chown(rundir
, 0, gid
);
5102 ERR("Unable to set group on %s", rundir
);
5107 * Ensure all applications and tracing group can search the run
5108 * dir. Allow everyone to read the directory, since it does not
5109 * buy us anything to hide its content.
5111 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5113 ERR("Unable to set permissions on %s", rundir
);
5117 /* lttng client socket path */
5118 ret
= chown(client_unix_sock_path
, 0, gid
);
5120 ERR("Unable to set group on %s", client_unix_sock_path
);
5124 /* kconsumer error socket path */
5125 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5127 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5131 /* 64-bit ustconsumer error socket path */
5132 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5134 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5138 /* 32-bit ustconsumer compat32 error socket path */
5139 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5141 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5145 DBG("All permissions are set");
5151 * Create the lttng run directory needed for all global sockets and pipe.
5153 static int create_lttng_rundir(const char *rundir
)
5157 DBG3("Creating LTTng run directory: %s", rundir
);
5159 ret
= mkdir(rundir
, S_IRWXU
);
5161 if (errno
!= EEXIST
) {
5162 ERR("Unable to create %s", rundir
);
5174 * Setup sockets and directory needed by the kconsumerd communication with the
5177 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5181 char path
[PATH_MAX
];
5183 switch (consumer_data
->type
) {
5184 case LTTNG_CONSUMER_KERNEL
:
5185 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5187 case LTTNG_CONSUMER64_UST
:
5188 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5190 case LTTNG_CONSUMER32_UST
:
5191 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5194 ERR("Consumer type unknown");
5199 DBG2("Creating consumer directory: %s", path
);
5201 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5203 if (errno
!= EEXIST
) {
5205 ERR("Failed to create %s", path
);
5211 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5213 ERR("Unable to set group on %s", path
);
5219 /* Create the kconsumerd error unix socket */
5220 consumer_data
->err_sock
=
5221 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5222 if (consumer_data
->err_sock
< 0) {
5223 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5229 * Set the CLOEXEC flag. Return code is useless because either way, the
5232 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5234 PERROR("utils_set_fd_cloexec");
5235 /* continue anyway */
5238 /* File permission MUST be 660 */
5239 ret
= chmod(consumer_data
->err_unix_sock_path
,
5240 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5242 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5252 * Signal handler for the daemon
5254 * Simply stop all worker threads, leaving main() return gracefully after
5255 * joining all threads and calling cleanup().
5257 static void sighandler(int sig
)
5261 DBG("SIGPIPE caught");
5264 DBG("SIGINT caught");
5268 DBG("SIGTERM caught");
5272 CMM_STORE_SHARED(recv_child_signal
, 1);
5280 * Setup signal handler for :
5281 * SIGINT, SIGTERM, SIGPIPE
5283 static int set_signal_handler(void)
5286 struct sigaction sa
;
5289 if ((ret
= sigemptyset(&sigset
)) < 0) {
5290 PERROR("sigemptyset");
5294 sa
.sa_handler
= sighandler
;
5295 sa
.sa_mask
= sigset
;
5297 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5298 PERROR("sigaction");
5302 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5303 PERROR("sigaction");
5307 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5308 PERROR("sigaction");
5312 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5313 PERROR("sigaction");
5317 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5323 * Set open files limit to unlimited. This daemon can open a large number of
5324 * file descriptors in order to consumer multiple kernel traces.
5326 static void set_ulimit(void)
5331 /* The kernel does not allowed an infinite limit for open files */
5332 lim
.rlim_cur
= 65535;
5333 lim
.rlim_max
= 65535;
5335 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5337 PERROR("failed to set open files limit");
5342 * Write pidfile using the rundir and opt_pidfile.
5344 static int write_pidfile(void)
5347 char pidfile_path
[PATH_MAX
];
5352 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
5354 /* Build pidfile path from rundir and opt_pidfile. */
5355 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5356 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5358 PERROR("snprintf pidfile path");
5364 * Create pid file in rundir.
5366 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5372 * Create lockfile using the rundir and return its fd.
5374 static int create_lockfile(void)
5377 char lockfile_path
[PATH_MAX
];
5379 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5384 ret
= utils_create_lock_file(lockfile_path
);
5390 * Write agent TCP port using the rundir.
5392 static int write_agent_port(void)
5395 char path
[PATH_MAX
];
5399 ret
= snprintf(path
, sizeof(path
), "%s/"
5400 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5402 PERROR("snprintf agent port path");
5407 * Create TCP agent port file in rundir.
5409 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5418 int main(int argc
, char **argv
)
5420 int ret
= 0, retval
= 0;
5422 const char *home_path
, *env_app_timeout
;
5424 init_kernel_workarounds();
5426 rcu_register_thread();
5428 if (set_signal_handler()) {
5430 goto exit_set_signal_handler
;
5433 setup_consumerd_path();
5435 page_size
= sysconf(_SC_PAGESIZE
);
5436 if (page_size
< 0) {
5437 PERROR("sysconf _SC_PAGESIZE");
5438 page_size
= LONG_MAX
;
5439 WARN("Fallback page size to %ld", page_size
);
5443 * Parse arguments and load the daemon configuration file.
5445 * We have an exit_options exit path to free memory reserved by
5446 * set_options. This is needed because the rest of sessiond_cleanup()
5447 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5448 * depends on set_options.
5451 if (set_options(argc
, argv
)) {
5457 if (opt_daemon
|| opt_background
) {
5460 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5468 * We are in the child. Make sure all other file descriptors are
5469 * closed, in case we are called with more opened file
5470 * descriptors than the standard ones.
5472 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5478 * Starting from here, we can create threads. This needs to be after
5479 * lttng_daemonize due to RCU.
5483 * Initialize the health check subsystem. This call should set the
5484 * appropriate time values.
5486 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5487 if (!health_sessiond
) {
5488 PERROR("health_app_create error");
5490 goto exit_health_sessiond_cleanup
;
5493 if (init_ht_cleanup_quit_pipe()) {
5495 goto exit_ht_cleanup_quit_pipe
;
5498 /* Setup the thread ht_cleanup communication pipe. */
5499 if (utils_create_pipe_cloexec(ht_cleanup_pipe
)) {
5501 goto exit_ht_cleanup_pipe
;
5504 /* Set up max poll set size */
5505 if (lttng_poll_set_max_size()) {
5507 goto exit_set_max_size
;
5510 /* Create thread to clean up RCU hash tables */
5511 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5512 thread_ht_cleanup
, (void *) NULL
);
5515 PERROR("pthread_create ht_cleanup");
5517 goto exit_ht_cleanup
;
5520 /* Create thread quit pipe */
5521 if (init_thread_quit_pipe()) {
5523 goto exit_init_data
;
5526 /* Check if daemon is UID = 0 */
5527 is_root
= !getuid();
5530 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5533 goto exit_init_data
;
5536 /* Create global run dir with root access */
5537 if (create_lttng_rundir(rundir
)) {
5539 goto exit_init_data
;
5542 if (strlen(apps_unix_sock_path
) == 0) {
5543 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5544 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5547 goto exit_init_data
;
5551 if (strlen(client_unix_sock_path
) == 0) {
5552 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5553 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5556 goto exit_init_data
;
5560 /* Set global SHM for ust */
5561 if (strlen(wait_shm_path
) == 0) {
5562 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5563 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5566 goto exit_init_data
;
5570 if (strlen(health_unix_sock_path
) == 0) {
5571 ret
= snprintf(health_unix_sock_path
,
5572 sizeof(health_unix_sock_path
),
5573 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5576 goto exit_init_data
;
5580 /* Setup kernel consumerd path */
5581 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5582 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5585 goto exit_init_data
;
5587 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5588 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5591 goto exit_init_data
;
5594 DBG2("Kernel consumer err path: %s",
5595 kconsumer_data
.err_unix_sock_path
);
5596 DBG2("Kernel consumer cmd path: %s",
5597 kconsumer_data
.cmd_unix_sock_path
);
5599 home_path
= utils_get_home_dir();
5600 if (home_path
== NULL
) {
5601 /* TODO: Add --socket PATH option */
5602 ERR("Can't get HOME directory for sockets creation.");
5604 goto exit_init_data
;
5608 * Create rundir from home path. This will create something like
5611 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5614 goto exit_init_data
;
5617 if (create_lttng_rundir(rundir
)) {
5619 goto exit_init_data
;
5622 if (strlen(apps_unix_sock_path
) == 0) {
5623 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5624 DEFAULT_HOME_APPS_UNIX_SOCK
,
5628 goto exit_init_data
;
5632 /* Set the cli tool unix socket path */
5633 if (strlen(client_unix_sock_path
) == 0) {
5634 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5635 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5639 goto exit_init_data
;
5643 /* Set global SHM for ust */
5644 if (strlen(wait_shm_path
) == 0) {
5645 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5646 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5650 goto exit_init_data
;
5654 /* Set health check Unix path */
5655 if (strlen(health_unix_sock_path
) == 0) {
5656 ret
= snprintf(health_unix_sock_path
,
5657 sizeof(health_unix_sock_path
),
5658 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5662 goto exit_init_data
;
5667 lockfile_fd
= create_lockfile();
5668 if (lockfile_fd
< 0) {
5670 goto exit_init_data
;
5673 /* Set consumer initial state */
5674 kernel_consumerd_state
= CONSUMER_STOPPED
;
5675 ust_consumerd_state
= CONSUMER_STOPPED
;
5677 DBG("Client socket path %s", client_unix_sock_path
);
5678 DBG("Application socket path %s", apps_unix_sock_path
);
5679 DBG("Application wait path %s", wait_shm_path
);
5680 DBG("LTTng run directory path: %s", rundir
);
5682 /* 32 bits consumerd path setup */
5683 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5684 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5686 PERROR("snprintf 32-bit consumer error socket path");
5688 goto exit_init_data
;
5690 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5691 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5693 PERROR("snprintf 32-bit consumer command socket path");
5695 goto exit_init_data
;
5698 DBG2("UST consumer 32 bits err path: %s",
5699 ustconsumer32_data
.err_unix_sock_path
);
5700 DBG2("UST consumer 32 bits cmd path: %s",
5701 ustconsumer32_data
.cmd_unix_sock_path
);
5703 /* 64 bits consumerd path setup */
5704 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5705 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5707 PERROR("snprintf 64-bit consumer error socket path");
5709 goto exit_init_data
;
5711 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5712 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5714 PERROR("snprintf 64-bit consumer command socket path");
5716 goto exit_init_data
;
5719 DBG2("UST consumer 64 bits err path: %s",
5720 ustconsumer64_data
.err_unix_sock_path
);
5721 DBG2("UST consumer 64 bits cmd path: %s",
5722 ustconsumer64_data
.cmd_unix_sock_path
);
5725 * See if daemon already exist.
5727 if (check_existing_daemon()) {
5728 ERR("Already running daemon.\n");
5730 * We do not goto exit because we must not cleanup()
5731 * because a daemon is already running.
5734 goto exit_init_data
;
5738 * Init UST app hash table. Alloc hash table before this point since
5739 * cleanup() can get called after that point.
5741 if (ust_app_ht_alloc()) {
5742 ERR("Failed to allocate UST app hash table");
5744 goto exit_init_data
;
5748 * Initialize agent app hash table. We allocate the hash table here
5749 * since cleanup() can get called after this point.
5751 if (agent_app_ht_alloc()) {
5752 ERR("Failed to allocate Agent app hash table");
5754 goto exit_init_data
;
5758 * These actions must be executed as root. We do that *after* setting up
5759 * the sockets path because we MUST make the check for another daemon using
5760 * those paths *before* trying to set the kernel consumer sockets and init
5764 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5766 goto exit_init_data
;
5769 /* Setup kernel tracer */
5770 if (!opt_no_kernel
) {
5771 init_kernel_tracer();
5772 if (kernel_tracer_fd
>= 0) {
5773 ret
= syscall_init_table();
5775 ERR("Unable to populate syscall table. "
5776 "Syscall tracing won't work "
5777 "for this session daemon.");
5782 /* Set ulimit for open files */
5785 /* init lttng_fd tracking must be done after set_ulimit. */
5788 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5790 goto exit_init_data
;
5793 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5795 goto exit_init_data
;
5798 /* Setup the needed unix socket */
5799 if (init_daemon_socket()) {
5801 goto exit_init_data
;
5804 /* Set credentials to socket */
5805 if (is_root
&& set_permissions(rundir
)) {
5807 goto exit_init_data
;
5810 /* Get parent pid if -S, --sig-parent is specified. */
5811 if (opt_sig_parent
) {
5815 /* Setup the kernel pipe for waking up the kernel thread */
5816 if (is_root
&& !opt_no_kernel
) {
5817 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5819 goto exit_init_data
;
5823 /* Setup the thread apps communication pipe. */
5824 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5826 goto exit_init_data
;
5829 /* Setup the thread apps notify communication pipe. */
5830 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5832 goto exit_init_data
;
5835 /* Initialize global buffer per UID and PID registry. */
5836 buffer_reg_init_uid_registry();
5837 buffer_reg_init_pid_registry();
5839 /* Init UST command queue. */
5840 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5843 * Get session list pointer. This pointer MUST NOT be free'd. This list
5844 * is statically declared in session.c
5846 session_list_ptr
= session_get_list();
5850 /* Check for the application socket timeout env variable. */
5851 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5852 if (env_app_timeout
) {
5853 app_socket_timeout
= atoi(env_app_timeout
);
5855 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5858 ret
= write_pidfile();
5860 ERR("Error in write_pidfile");
5862 goto exit_init_data
;
5864 ret
= write_agent_port();
5866 ERR("Error in write_agent_port");
5868 goto exit_init_data
;
5871 /* Initialize communication library */
5873 /* Initialize TCP timeout values */
5874 lttcomm_inet_init();
5876 if (load_session_init_data(&load_info
) < 0) {
5878 goto exit_init_data
;
5880 load_info
->path
= opt_load_session_path
;
5882 /* Create health-check thread */
5883 ret
= pthread_create(&health_thread
, NULL
,
5884 thread_manage_health
, (void *) NULL
);
5887 PERROR("pthread_create health");
5892 /* Create thread to manage the client socket */
5893 ret
= pthread_create(&client_thread
, NULL
,
5894 thread_manage_clients
, (void *) NULL
);
5897 PERROR("pthread_create clients");
5902 /* Create thread to dispatch registration */
5903 ret
= pthread_create(&dispatch_thread
, NULL
,
5904 thread_dispatch_ust_registration
, (void *) NULL
);
5907 PERROR("pthread_create dispatch");
5912 /* Create thread to manage application registration. */
5913 ret
= pthread_create(®_apps_thread
, NULL
,
5914 thread_registration_apps
, (void *) NULL
);
5917 PERROR("pthread_create registration");
5922 /* Create thread to manage application socket */
5923 ret
= pthread_create(&apps_thread
, NULL
,
5924 thread_manage_apps
, (void *) NULL
);
5927 PERROR("pthread_create apps");
5932 /* Create thread to manage application notify socket */
5933 ret
= pthread_create(&apps_notify_thread
, NULL
,
5934 ust_thread_manage_notify
, (void *) NULL
);
5937 PERROR("pthread_create notify");
5939 goto exit_apps_notify
;
5942 /* Create agent registration thread. */
5943 ret
= pthread_create(&agent_reg_thread
, NULL
,
5944 agent_thread_manage_registration
, (void *) NULL
);
5947 PERROR("pthread_create agent");
5949 goto exit_agent_reg
;
5952 /* Don't start this thread if kernel tracing is not requested nor root */
5953 if (is_root
&& !opt_no_kernel
) {
5954 /* Create kernel thread to manage kernel event */
5955 ret
= pthread_create(&kernel_thread
, NULL
,
5956 thread_manage_kernel
, (void *) NULL
);
5959 PERROR("pthread_create kernel");
5965 /* Create session loading thread. */
5966 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
5970 PERROR("pthread_create load_session_thread");
5972 goto exit_load_session
;
5976 * This is where we start awaiting program completion (e.g. through
5977 * signal that asks threads to teardown).
5980 ret
= pthread_join(load_session_thread
, &status
);
5983 PERROR("pthread_join load_session_thread");
5988 if (is_root
&& !opt_no_kernel
) {
5989 ret
= pthread_join(kernel_thread
, &status
);
5992 PERROR("pthread_join");
5998 ret
= pthread_join(agent_reg_thread
, &status
);
6001 PERROR("pthread_join agent");
6006 ret
= pthread_join(apps_notify_thread
, &status
);
6009 PERROR("pthread_join apps notify");
6014 ret
= pthread_join(apps_thread
, &status
);
6017 PERROR("pthread_join apps");
6022 ret
= pthread_join(reg_apps_thread
, &status
);
6025 PERROR("pthread_join");
6031 * Join dispatch thread after joining reg_apps_thread to ensure
6032 * we don't leak applications in the queue.
6034 ret
= pthread_join(dispatch_thread
, &status
);
6037 PERROR("pthread_join");
6042 ret
= pthread_join(client_thread
, &status
);
6045 PERROR("pthread_join");
6050 ret
= pthread_join(health_thread
, &status
);
6053 PERROR("pthread_join health thread");
6060 * sessiond_cleanup() is called when no other thread is running, except
6061 * the ht_cleanup thread, which is needed to destroy the hash tables.
6063 rcu_thread_online();
6065 rcu_thread_offline();
6066 rcu_unregister_thread();
6068 ret
= notify_thread_pipe(ht_cleanup_quit_pipe
[1]);
6070 ERR("write error on ht_cleanup quit pipe");
6074 ret
= pthread_join(ht_cleanup_thread
, &status
);
6077 PERROR("pthread_join ht cleanup thread");
6083 utils_close_pipe(ht_cleanup_pipe
);
6084 exit_ht_cleanup_pipe
:
6087 * Close the ht_cleanup quit pipe.
6089 utils_close_pipe(ht_cleanup_quit_pipe
);
6090 exit_ht_cleanup_quit_pipe
:
6092 health_app_destroy(health_sessiond
);
6093 exit_health_sessiond_cleanup
:
6096 sessiond_cleanup_options();
6098 exit_set_signal_handler
:
6099 /* Ensure all prior call_rcu are done. */