2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
33 #include <sys/mount.h>
34 #include <sys/resource.h>
35 #include <sys/socket.h>
37 #include <sys/types.h>
39 #include <urcu/uatomic.h>
43 #include <common/common.h>
44 #include <common/compat/socket.h>
45 #include <common/compat/getenv.h>
46 #include <common/defaults.h>
47 #include <common/kernel-consumer/kernel-consumer.h>
48 #include <common/futex.h>
49 #include <common/relayd/relayd.h>
50 #include <common/utils.h>
51 #include <common/daemonize.h>
52 #include <common/config/config.h>
54 #include "lttng-sessiond.h"
55 #include "buffer-registry.h"
62 #include "kernel-consumer.h"
66 #include "ust-consumer.h"
69 #include "health-sessiond.h"
70 #include "testpoint.h"
71 #include "ust-thread.h"
72 #include "agent-thread.h"
74 #include "load-session-thread.h"
77 #define CONSUMERD_FILE "lttng-consumerd"
80 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
81 static int tracing_group_name_override
;
82 static char *opt_pidfile
;
83 static int opt_sig_parent
;
84 static int opt_verbose_consumer
;
85 static int opt_daemon
, opt_background
;
86 static int opt_no_kernel
;
87 static char *opt_load_session_path
;
88 static pid_t ppid
; /* Parent PID for --sig-parent option */
89 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
91 static int lockfile_fd
= -1;
93 /* Set to 1 when a SIGUSR1 signal is received. */
94 static int recv_child_signal
;
97 * Consumer daemon specific control data. Every value not initialized here is
98 * set to 0 by the static definition.
100 static struct consumer_data kconsumer_data
= {
101 .type
= LTTNG_CONSUMER_KERNEL
,
102 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
103 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
106 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 .lock
= PTHREAD_MUTEX_INITIALIZER
,
108 .cond
= PTHREAD_COND_INITIALIZER
,
109 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
111 static struct consumer_data ustconsumer64_data
= {
112 .type
= LTTNG_CONSUMER64_UST
,
113 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
114 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
117 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
118 .lock
= PTHREAD_MUTEX_INITIALIZER
,
119 .cond
= PTHREAD_COND_INITIALIZER
,
120 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
122 static struct consumer_data ustconsumer32_data
= {
123 .type
= LTTNG_CONSUMER32_UST
,
124 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
125 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
128 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
129 .lock
= PTHREAD_MUTEX_INITIALIZER
,
130 .cond
= PTHREAD_COND_INITIALIZER
,
131 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
134 /* Command line options */
135 static const struct option long_options
[] = {
136 { "client-sock", required_argument
, 0, 'c' },
137 { "apps-sock", required_argument
, 0, 'a' },
138 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
139 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
140 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
141 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
142 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
143 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
144 { "consumerd32-path", required_argument
, 0, '\0' },
145 { "consumerd32-libdir", required_argument
, 0, '\0' },
146 { "consumerd64-path", required_argument
, 0, '\0' },
147 { "consumerd64-libdir", required_argument
, 0, '\0' },
148 { "daemonize", no_argument
, 0, 'd' },
149 { "background", no_argument
, 0, 'b' },
150 { "sig-parent", no_argument
, 0, 'S' },
151 { "help", no_argument
, 0, 'h' },
152 { "group", required_argument
, 0, 'g' },
153 { "version", no_argument
, 0, 'V' },
154 { "quiet", no_argument
, 0, 'q' },
155 { "verbose", no_argument
, 0, 'v' },
156 { "verbose-consumer", no_argument
, 0, '\0' },
157 { "no-kernel", no_argument
, 0, '\0' },
158 { "pidfile", required_argument
, 0, 'p' },
159 { "agent-tcp-port", required_argument
, 0, '\0' },
160 { "config", required_argument
, 0, 'f' },
161 { "load", required_argument
, 0, 'l' },
162 { "kmod-probes", required_argument
, 0, '\0' },
163 { "extra-kmod-probes", required_argument
, 0, '\0' },
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options
[] = { "help", "version", "config" };
170 /* Shared between threads */
171 static int dispatch_thread_exit
;
173 /* Global application Unix socket path */
174 static char apps_unix_sock_path
[PATH_MAX
];
175 /* Global client Unix socket path */
176 static char client_unix_sock_path
[PATH_MAX
];
177 /* global wait shm path for UST */
178 static char wait_shm_path
[PATH_MAX
];
179 /* Global health check unix path */
180 static char health_unix_sock_path
[PATH_MAX
];
182 /* Sockets and FDs */
183 static int client_sock
= -1;
184 static int apps_sock
= -1;
185 int kernel_tracer_fd
= -1;
186 static int kernel_poll_pipe
[2] = { -1, -1 };
189 * Quit pipe for all threads. This permits a single cancellation point
190 * for all threads when receiving an event on the pipe.
192 static int thread_quit_pipe
[2] = { -1, -1 };
193 static int ht_cleanup_quit_pipe
[2] = { -1, -1 };
196 * This pipe is used to inform the thread managing application communication
197 * that a command is queued and ready to be processed.
199 static int apps_cmd_pipe
[2] = { -1, -1 };
201 int apps_cmd_notify_pipe
[2] = { -1, -1 };
203 /* Pthread, Mutexes and Semaphores */
204 static pthread_t apps_thread
;
205 static pthread_t apps_notify_thread
;
206 static pthread_t reg_apps_thread
;
207 static pthread_t client_thread
;
208 static pthread_t kernel_thread
;
209 static pthread_t dispatch_thread
;
210 static pthread_t health_thread
;
211 static pthread_t ht_cleanup_thread
;
212 static pthread_t agent_reg_thread
;
213 static pthread_t load_session_thread
;
216 * UST registration command queue. This queue is tied with a futex and uses a N
217 * wakers / 1 waiter implemented and detailed in futex.c/.h
219 * The thread_registration_apps and thread_dispatch_ust_registration uses this
220 * queue along with the wait/wake scheme. The thread_manage_apps receives down
221 * the line new application socket and monitors it for any I/O error or clean
222 * close that triggers an unregistration of the application.
224 static struct ust_cmd_queue ust_cmd_queue
;
227 * Pointer initialized before thread creation.
229 * This points to the tracing session list containing the session count and a
230 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
231 * MUST NOT be taken if you call a public function in session.c.
233 * The lock is nested inside the structure: session_list_ptr->lock. Please use
234 * session_lock_list and session_unlock_list for lock acquisition.
236 static struct ltt_session_list
*session_list_ptr
;
238 int ust_consumerd64_fd
= -1;
239 int ust_consumerd32_fd
= -1;
241 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
242 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
243 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
244 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
245 static int consumerd32_bin_override
;
246 static int consumerd64_bin_override
;
247 static int consumerd32_libdir_override
;
248 static int consumerd64_libdir_override
;
250 static const char *module_proc_lttng
= "/proc/lttng";
253 * Consumer daemon state which is changed when spawning it, killing it or in
254 * case of a fatal error.
256 enum consumerd_state
{
257 CONSUMER_STARTED
= 1,
258 CONSUMER_STOPPED
= 2,
263 * This consumer daemon state is used to validate if a client command will be
264 * able to reach the consumer. If not, the client is informed. For instance,
265 * doing a "lttng start" when the consumer state is set to ERROR will return an
266 * error to the client.
268 * The following example shows a possible race condition of this scheme:
270 * consumer thread error happens
272 * client cmd checks state -> still OK
273 * consumer thread exit, sets error
274 * client cmd try to talk to consumer
277 * However, since the consumer is a different daemon, we have no way of making
278 * sure the command will reach it safely even with this state flag. This is why
279 * we consider that up to the state validation during command processing, the
280 * command is safe. After that, we can not guarantee the correctness of the
281 * client request vis-a-vis the consumer.
283 static enum consumerd_state ust_consumerd_state
;
284 static enum consumerd_state kernel_consumerd_state
;
287 * Socket timeout for receiving and sending in seconds.
289 static int app_socket_timeout
;
291 /* Set in main() with the current page size. */
294 /* Application health monitoring */
295 struct health_app
*health_sessiond
;
297 /* Agent TCP port for registration. Used by the agent thread. */
298 unsigned int agent_tcp_port
= DEFAULT_AGENT_TCP_PORT
;
300 /* Am I root or not. */
301 int is_root
; /* Set to 1 if the daemon is running as root */
303 const char * const config_section_name
= "sessiond";
305 /* Load session thread information to operate. */
306 struct load_session_thread_data
*load_info
;
309 * Whether sessiond is ready for commands/health check requests.
310 * NR_LTTNG_SESSIOND_READY must match the number of calls to
311 * sessiond_notify_ready().
313 #define NR_LTTNG_SESSIOND_READY 3
314 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
316 /* Notify parents that we are ready for cmd and health check */
318 void sessiond_notify_ready(void)
320 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
322 * Notify parent pid that we are ready to accept command
323 * for client side. This ppid is the one from the
324 * external process that spawned us.
326 if (opt_sig_parent
) {
331 * Notify the parent of the fork() process that we are
334 if (opt_daemon
|| opt_background
) {
335 kill(child_ppid
, SIGUSR1
);
341 void setup_consumerd_path(void)
343 const char *bin
, *libdir
;
346 * Allow INSTALL_BIN_PATH to be used as a target path for the
347 * native architecture size consumer if CONFIG_CONSUMER*_PATH
348 * has not been defined.
350 #if (CAA_BITS_PER_LONG == 32)
351 if (!consumerd32_bin
[0]) {
352 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
354 if (!consumerd32_libdir
[0]) {
355 consumerd32_libdir
= INSTALL_LIB_PATH
;
357 #elif (CAA_BITS_PER_LONG == 64)
358 if (!consumerd64_bin
[0]) {
359 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
361 if (!consumerd64_libdir
[0]) {
362 consumerd64_libdir
= INSTALL_LIB_PATH
;
365 #error "Unknown bitness"
369 * runtime env. var. overrides the build default.
371 bin
= lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
373 consumerd32_bin
= bin
;
375 bin
= lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
377 consumerd64_bin
= bin
;
379 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
381 consumerd32_libdir
= libdir
;
383 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
385 consumerd64_libdir
= libdir
;
390 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
397 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
403 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
415 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
417 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
419 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
423 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
425 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event
*events
,
428 return __sessiond_set_thread_pollset(events
, size
,
429 ht_cleanup_quit_pipe
);
433 int __sessiond_check_thread_quit_pipe(int fd
, uint32_t events
, int a_pipe
)
435 if (fd
== a_pipe
&& (events
& LPOLLIN
)) {
442 * Check if the thread quit pipe was triggered.
444 * Return 1 if it was triggered else 0;
446 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
448 return __sessiond_check_thread_quit_pipe(fd
, events
,
449 thread_quit_pipe
[0]);
453 * Check if the ht_cleanup thread quit pipe was triggered.
455 * Return 1 if it was triggered else 0;
457 int sessiond_check_ht_cleanup_quit(int fd
, uint32_t events
)
459 return __sessiond_check_thread_quit_pipe(fd
, events
,
460 ht_cleanup_quit_pipe
[0]);
464 * Init thread quit pipe.
466 * Return -1 on error or 0 if all pipes are created.
468 static int __init_thread_quit_pipe(int *a_pipe
)
474 PERROR("thread quit pipe");
478 for (i
= 0; i
< 2; i
++) {
479 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
490 static int init_thread_quit_pipe(void)
492 return __init_thread_quit_pipe(thread_quit_pipe
);
495 static int init_ht_cleanup_quit_pipe(void)
497 return __init_thread_quit_pipe(ht_cleanup_quit_pipe
);
501 * Stop all threads by closing the thread quit pipe.
503 static void stop_threads(void)
507 /* Stopping all threads */
508 DBG("Terminating all threads");
509 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
511 ERR("write error on thread quit pipe");
514 /* Dispatch thread */
515 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
516 futex_nto1_wake(&ust_cmd_queue
.futex
);
520 * Close every consumer sockets.
522 static void close_consumer_sockets(void)
526 if (kconsumer_data
.err_sock
>= 0) {
527 ret
= close(kconsumer_data
.err_sock
);
529 PERROR("kernel consumer err_sock close");
532 if (ustconsumer32_data
.err_sock
>= 0) {
533 ret
= close(ustconsumer32_data
.err_sock
);
535 PERROR("UST consumerd32 err_sock close");
538 if (ustconsumer64_data
.err_sock
>= 0) {
539 ret
= close(ustconsumer64_data
.err_sock
);
541 PERROR("UST consumerd64 err_sock close");
544 if (kconsumer_data
.cmd_sock
>= 0) {
545 ret
= close(kconsumer_data
.cmd_sock
);
547 PERROR("kernel consumer cmd_sock close");
550 if (ustconsumer32_data
.cmd_sock
>= 0) {
551 ret
= close(ustconsumer32_data
.cmd_sock
);
553 PERROR("UST consumerd32 cmd_sock close");
556 if (ustconsumer64_data
.cmd_sock
>= 0) {
557 ret
= close(ustconsumer64_data
.cmd_sock
);
559 PERROR("UST consumerd64 cmd_sock close");
565 * Generate the full lock file path using the rundir.
567 * Return the snprintf() return value thus a negative value is an error.
569 static int generate_lock_file_path(char *path
, size_t len
)
576 /* Build lockfile path from rundir. */
577 ret
= snprintf(path
, len
, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE
, rundir
);
579 PERROR("snprintf lockfile path");
586 * Wait on consumer process termination.
588 * Need to be called with the consumer data lock held or from a context
589 * ensuring no concurrent access to data (e.g: cleanup).
591 static void wait_consumer(struct consumer_data
*consumer_data
)
596 if (consumer_data
->pid
<= 0) {
600 DBG("Waiting for complete teardown of consumerd (PID: %d)",
602 ret
= waitpid(consumer_data
->pid
, &status
, 0);
604 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
606 if (!WIFEXITED(status
)) {
607 ERR("consumerd termination with error: %d",
610 consumer_data
->pid
= 0;
614 * Cleanup the session daemon's data structures.
616 static void sessiond_cleanup(void)
619 struct ltt_session
*sess
, *stmp
;
622 DBG("Cleanup sessiond");
625 * Close the thread quit pipe. It has already done its job,
626 * since we are now called.
628 utils_close_pipe(thread_quit_pipe
);
631 * If opt_pidfile is undefined, the default file will be wiped when
632 * removing the rundir.
635 ret
= remove(opt_pidfile
);
637 PERROR("remove pidfile %s", opt_pidfile
);
641 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
644 snprintf(path
, PATH_MAX
,
646 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
647 DBG("Removing %s", path
);
650 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
651 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
);
652 DBG("Removing %s", path
);
656 snprintf(path
, PATH_MAX
,
657 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
659 DBG("Removing %s", path
);
662 snprintf(path
, PATH_MAX
,
663 DEFAULT_KCONSUMERD_PATH
,
665 DBG("Removing directory %s", path
);
668 /* ust consumerd 32 */
669 snprintf(path
, PATH_MAX
,
670 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
672 DBG("Removing %s", path
);
675 snprintf(path
, PATH_MAX
,
676 DEFAULT_USTCONSUMERD32_PATH
,
678 DBG("Removing directory %s", path
);
681 /* ust consumerd 64 */
682 snprintf(path
, PATH_MAX
,
683 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
685 DBG("Removing %s", path
);
688 snprintf(path
, PATH_MAX
,
689 DEFAULT_USTCONSUMERD64_PATH
,
691 DBG("Removing directory %s", path
);
694 DBG("Cleaning up all sessions");
696 /* Destroy session list mutex */
697 if (session_list_ptr
!= NULL
) {
698 pthread_mutex_destroy(&session_list_ptr
->lock
);
700 /* Cleanup ALL session */
701 cds_list_for_each_entry_safe(sess
, stmp
,
702 &session_list_ptr
->head
, list
) {
703 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
707 wait_consumer(&kconsumer_data
);
708 wait_consumer(&ustconsumer64_data
);
709 wait_consumer(&ustconsumer32_data
);
711 DBG("Cleaning up all agent apps");
712 agent_app_ht_clean();
714 DBG("Closing all UST sockets");
715 ust_app_clean_list();
716 buffer_reg_destroy_registries();
718 if (is_root
&& !opt_no_kernel
) {
719 DBG2("Closing kernel fd");
720 if (kernel_tracer_fd
>= 0) {
721 ret
= close(kernel_tracer_fd
);
726 DBG("Unloading kernel modules");
727 modprobe_remove_lttng_all();
731 close_consumer_sockets();
734 load_session_destroy_data(load_info
);
739 * Cleanup lock file by deleting it and finaly closing it which will
740 * release the file system lock.
742 if (lockfile_fd
>= 0) {
743 char lockfile_path
[PATH_MAX
];
745 ret
= generate_lock_file_path(lockfile_path
,
746 sizeof(lockfile_path
));
748 ret
= remove(lockfile_path
);
750 PERROR("remove lock file");
752 ret
= close(lockfile_fd
);
754 PERROR("close lock file");
760 * We do NOT rmdir rundir because there are other processes
761 * using it, for instance lttng-relayd, which can start in
762 * parallel with this teardown.
769 * Cleanup the daemon's option data structures.
771 static void sessiond_cleanup_options(void)
773 DBG("Cleaning up options");
776 * If the override option is set, the pointer points to a *non* const
777 * thus freeing it even though the variable type is set to const.
779 if (tracing_group_name_override
) {
780 free((void *) tracing_group_name
);
782 if (consumerd32_bin_override
) {
783 free((void *) consumerd32_bin
);
785 if (consumerd64_bin_override
) {
786 free((void *) consumerd64_bin
);
788 if (consumerd32_libdir_override
) {
789 free((void *) consumerd32_libdir
);
791 if (consumerd64_libdir_override
) {
792 free((void *) consumerd64_libdir
);
796 free(opt_load_session_path
);
797 free(kmod_probes_list
);
798 free(kmod_extra_probes_list
);
801 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
802 "Matthew, BEET driven development works!%c[%dm",
803 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
808 * Send data on a unix socket using the liblttsessiondcomm API.
810 * Return lttcomm error code.
812 static int send_unix_sock(int sock
, void *buf
, size_t len
)
814 /* Check valid length */
819 return lttcomm_send_unix_sock(sock
, buf
, len
);
823 * Free memory of a command context structure.
825 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
827 DBG("Clean command context structure");
829 if ((*cmd_ctx
)->llm
) {
830 free((*cmd_ctx
)->llm
);
832 if ((*cmd_ctx
)->lsm
) {
833 free((*cmd_ctx
)->lsm
);
841 * Notify UST applications using the shm mmap futex.
843 static int notify_ust_apps(int active
)
847 DBG("Notifying applications of session daemon state: %d", active
);
849 /* See shm.c for this call implying mmap, shm and futex calls */
850 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
851 if (wait_shm_mmap
== NULL
) {
855 /* Wake waiting process */
856 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
858 /* Apps notified successfully */
866 * Setup the outgoing data buffer for the response (llm) by allocating the
867 * right amount of memory and copying the original information from the lsm
870 * Return total size of the buffer pointed by buf.
872 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
878 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
879 if (cmd_ctx
->llm
== NULL
) {
885 /* Copy common data */
886 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
887 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
889 cmd_ctx
->llm
->data_size
= size
;
890 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
899 * Update the kernel poll set of all channel fd available over all tracing
900 * session. Add the wakeup pipe at the end of the set.
902 static int update_kernel_poll(struct lttng_poll_event
*events
)
905 struct ltt_session
*session
;
906 struct ltt_kernel_channel
*channel
;
908 DBG("Updating kernel poll set");
911 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
912 session_lock(session
);
913 if (session
->kernel_session
== NULL
) {
914 session_unlock(session
);
918 cds_list_for_each_entry(channel
,
919 &session
->kernel_session
->channel_list
.head
, list
) {
920 /* Add channel fd to the kernel poll set */
921 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
923 session_unlock(session
);
926 DBG("Channel fd %d added to kernel set", channel
->fd
);
928 session_unlock(session
);
930 session_unlock_list();
935 session_unlock_list();
940 * Find the channel fd from 'fd' over all tracing session. When found, check
941 * for new channel stream and send those stream fds to the kernel consumer.
943 * Useful for CPU hotplug feature.
945 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
948 struct ltt_session
*session
;
949 struct ltt_kernel_session
*ksess
;
950 struct ltt_kernel_channel
*channel
;
952 DBG("Updating kernel streams for channel fd %d", fd
);
955 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
956 session_lock(session
);
957 if (session
->kernel_session
== NULL
) {
958 session_unlock(session
);
961 ksess
= session
->kernel_session
;
963 cds_list_for_each_entry(channel
,
964 &ksess
->channel_list
.head
, list
) {
965 struct lttng_ht_iter iter
;
966 struct consumer_socket
*socket
;
968 if (channel
->fd
!= fd
) {
971 DBG("Channel found, updating kernel streams");
972 ret
= kernel_open_channel_stream(channel
);
976 /* Update the stream global counter */
977 ksess
->stream_count_global
+= ret
;
980 * Have we already sent fds to the consumer? If yes, it
981 * means that tracing is started so it is safe to send
982 * our updated stream fds.
984 if (ksess
->consumer_fds_sent
!= 1
985 || ksess
->consumer
== NULL
) {
991 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
992 &iter
.iter
, socket
, node
.node
) {
993 pthread_mutex_lock(socket
->lock
);
994 ret
= kernel_consumer_send_channel_stream(socket
,
996 session
->output_traces
? 1 : 0);
997 pthread_mutex_unlock(socket
->lock
);
1005 session_unlock(session
);
1007 session_unlock_list();
1011 session_unlock(session
);
1012 session_unlock_list();
1017 * For each tracing session, update newly registered apps. The session list
1018 * lock MUST be acquired before calling this.
1020 static void update_ust_app(int app_sock
)
1022 struct ltt_session
*sess
, *stmp
;
1024 /* Consumer is in an ERROR state. Stop any application update. */
1025 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
1026 /* Stop the update process since the consumer is dead. */
1030 /* For all tracing session(s) */
1031 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
1032 struct ust_app
*app
;
1035 if (!sess
->ust_session
) {
1036 goto unlock_session
;
1040 assert(app_sock
>= 0);
1041 app
= ust_app_find_by_sock(app_sock
);
1044 * Application can be unregistered before so
1045 * this is possible hence simply stopping the
1048 DBG3("UST app update failed to find app sock %d",
1052 ust_app_global_update(sess
->ust_session
, app
);
1056 session_unlock(sess
);
1061 * This thread manage event coming from the kernel.
1063 * Features supported in this thread:
1066 static void *thread_manage_kernel(void *data
)
1068 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1069 uint32_t revents
, nb_fd
;
1071 struct lttng_poll_event events
;
1073 DBG("[thread] Thread manage kernel started");
1075 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1078 * This first step of the while is to clean this structure which could free
1079 * non NULL pointers so initialize it before the loop.
1081 lttng_poll_init(&events
);
1083 if (testpoint(sessiond_thread_manage_kernel
)) {
1084 goto error_testpoint
;
1087 health_code_update();
1089 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1090 goto error_testpoint
;
1094 health_code_update();
1096 if (update_poll_flag
== 1) {
1097 /* Clean events object. We are about to populate it again. */
1098 lttng_poll_clean(&events
);
1100 ret
= sessiond_set_thread_pollset(&events
, 2);
1102 goto error_poll_create
;
1105 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1110 /* This will add the available kernel channel if any. */
1111 ret
= update_kernel_poll(&events
);
1115 update_poll_flag
= 0;
1118 DBG("Thread kernel polling");
1120 /* Poll infinite value of time */
1122 health_poll_entry();
1123 ret
= lttng_poll_wait(&events
, -1);
1124 DBG("Thread kernel return from poll on %d fds",
1125 LTTNG_POLL_GETNB(&events
));
1129 * Restart interrupted system call.
1131 if (errno
== EINTR
) {
1135 } else if (ret
== 0) {
1136 /* Should not happen since timeout is infinite */
1137 ERR("Return value of poll is 0 with an infinite timeout.\n"
1138 "This should not have happened! Continuing...");
1144 for (i
= 0; i
< nb_fd
; i
++) {
1145 /* Fetch once the poll data */
1146 revents
= LTTNG_POLL_GETEV(&events
, i
);
1147 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1149 health_code_update();
1152 /* No activity for this FD (poll implementation). */
1156 /* Thread quit pipe has been closed. Killing thread. */
1157 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1163 /* Check for data on kernel pipe */
1164 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
1165 (void) lttng_read(kernel_poll_pipe
[0],
1168 * Ret value is useless here, if this pipe gets any actions an
1169 * update is required anyway.
1171 update_poll_flag
= 1;
1175 * New CPU detected by the kernel. Adding kernel stream to
1176 * kernel session and updating the kernel consumer
1178 if (revents
& LPOLLIN
) {
1179 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1185 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1186 * and unregister kernel stream at this point.
1195 lttng_poll_clean(&events
);
1198 utils_close_pipe(kernel_poll_pipe
);
1199 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1202 ERR("Health error occurred in %s", __func__
);
1203 WARN("Kernel thread died unexpectedly. "
1204 "Kernel tracing can continue but CPU hotplug is disabled.");
1206 health_unregister(health_sessiond
);
1207 DBG("Kernel thread dying");
1212 * Signal pthread condition of the consumer data that the thread.
1214 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1216 pthread_mutex_lock(&data
->cond_mutex
);
1219 * The state is set before signaling. It can be any value, it's the waiter
1220 * job to correctly interpret this condition variable associated to the
1221 * consumer pthread_cond.
1223 * A value of 0 means that the corresponding thread of the consumer data
1224 * was not started. 1 indicates that the thread has started and is ready
1225 * for action. A negative value means that there was an error during the
1228 data
->consumer_thread_is_ready
= state
;
1229 (void) pthread_cond_signal(&data
->cond
);
1231 pthread_mutex_unlock(&data
->cond_mutex
);
1235 * This thread manage the consumer error sent back to the session daemon.
1237 static void *thread_manage_consumer(void *data
)
1239 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1240 uint32_t revents
, nb_fd
;
1241 enum lttcomm_return_code code
;
1242 struct lttng_poll_event events
;
1243 struct consumer_data
*consumer_data
= data
;
1245 DBG("[thread] Manage consumer started");
1247 rcu_register_thread();
1248 rcu_thread_online();
1250 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1252 health_code_update();
1255 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1256 * metadata_sock. Nothing more will be added to this poll set.
1258 ret
= sessiond_set_thread_pollset(&events
, 3);
1264 * The error socket here is already in a listening state which was done
1265 * just before spawning this thread to avoid a race between the consumer
1266 * daemon exec trying to connect and the listen() call.
1268 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1273 health_code_update();
1275 /* Infinite blocking call, waiting for transmission */
1277 health_poll_entry();
1279 if (testpoint(sessiond_thread_manage_consumer
)) {
1283 ret
= lttng_poll_wait(&events
, -1);
1287 * Restart interrupted system call.
1289 if (errno
== EINTR
) {
1297 for (i
= 0; i
< nb_fd
; i
++) {
1298 /* Fetch once the poll data */
1299 revents
= LTTNG_POLL_GETEV(&events
, i
);
1300 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1302 health_code_update();
1305 /* No activity for this FD (poll implementation). */
1309 /* Thread quit pipe has been closed. Killing thread. */
1310 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1316 /* Event on the registration socket */
1317 if (pollfd
== consumer_data
->err_sock
) {
1318 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1319 ERR("consumer err socket poll error");
1325 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1331 * Set the CLOEXEC flag. Return code is useless because either way, the
1334 (void) utils_set_fd_cloexec(sock
);
1336 health_code_update();
1338 DBG2("Receiving code from consumer err_sock");
1340 /* Getting status code from kconsumerd */
1341 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1342 sizeof(enum lttcomm_return_code
));
1347 health_code_update();
1348 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1349 /* Connect both socket, command and metadata. */
1350 consumer_data
->cmd_sock
=
1351 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1352 consumer_data
->metadata_fd
=
1353 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1354 if (consumer_data
->cmd_sock
< 0
1355 || consumer_data
->metadata_fd
< 0) {
1356 PERROR("consumer connect cmd socket");
1357 /* On error, signal condition and quit. */
1358 signal_consumer_condition(consumer_data
, -1);
1361 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1362 /* Create metadata socket lock. */
1363 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1364 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1365 PERROR("zmalloc pthread mutex");
1369 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1371 signal_consumer_condition(consumer_data
, 1);
1372 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1373 DBG("Consumer metadata socket ready (fd: %d)",
1374 consumer_data
->metadata_fd
);
1376 ERR("consumer error when waiting for SOCK_READY : %s",
1377 lttcomm_get_readable_code(-code
));
1381 /* Remove the consumerd error sock since we've established a connexion */
1382 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1387 /* Add new accepted error socket. */
1388 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1393 /* Add metadata socket that is successfully connected. */
1394 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1395 LPOLLIN
| LPOLLRDHUP
);
1400 health_code_update();
1402 /* Infinite blocking call, waiting for transmission */
1405 health_code_update();
1407 /* Exit the thread because the thread quit pipe has been triggered. */
1409 /* Not a health error. */
1414 health_poll_entry();
1415 ret
= lttng_poll_wait(&events
, -1);
1419 * Restart interrupted system call.
1421 if (errno
== EINTR
) {
1429 for (i
= 0; i
< nb_fd
; i
++) {
1430 /* Fetch once the poll data */
1431 revents
= LTTNG_POLL_GETEV(&events
, i
);
1432 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1434 health_code_update();
1437 /* No activity for this FD (poll implementation). */
1442 * Thread quit pipe has been triggered, flag that we should stop
1443 * but continue the current loop to handle potential data from
1446 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1448 if (pollfd
== sock
) {
1449 /* Event on the consumerd socket */
1450 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1451 ERR("consumer err socket second poll error");
1454 health_code_update();
1455 /* Wait for any kconsumerd error */
1456 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1457 sizeof(enum lttcomm_return_code
));
1459 ERR("consumer closed the command socket");
1463 ERR("consumer return code : %s",
1464 lttcomm_get_readable_code(-code
));
1467 } else if (pollfd
== consumer_data
->metadata_fd
) {
1468 /* UST metadata requests */
1469 ret
= ust_consumer_metadata_request(
1470 &consumer_data
->metadata_sock
);
1472 ERR("Handling metadata request");
1476 /* No need for an else branch all FDs are tested prior. */
1478 health_code_update();
1484 * We lock here because we are about to close the sockets and some other
1485 * thread might be using them so get exclusive access which will abort all
1486 * other consumer command by other threads.
1488 pthread_mutex_lock(&consumer_data
->lock
);
1490 /* Immediately set the consumerd state to stopped */
1491 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1492 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1493 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1494 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1495 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1497 /* Code flow error... */
1501 if (consumer_data
->err_sock
>= 0) {
1502 ret
= close(consumer_data
->err_sock
);
1506 consumer_data
->err_sock
= -1;
1508 if (consumer_data
->cmd_sock
>= 0) {
1509 ret
= close(consumer_data
->cmd_sock
);
1513 consumer_data
->cmd_sock
= -1;
1515 if (consumer_data
->metadata_sock
.fd_ptr
&&
1516 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1517 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1529 unlink(consumer_data
->err_unix_sock_path
);
1530 unlink(consumer_data
->cmd_unix_sock_path
);
1531 pthread_mutex_unlock(&consumer_data
->lock
);
1533 /* Cleanup metadata socket mutex. */
1534 if (consumer_data
->metadata_sock
.lock
) {
1535 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1536 free(consumer_data
->metadata_sock
.lock
);
1538 lttng_poll_clean(&events
);
1542 ERR("Health error occurred in %s", __func__
);
1544 health_unregister(health_sessiond
);
1545 DBG("consumer thread cleanup completed");
1547 rcu_thread_offline();
1548 rcu_unregister_thread();
1554 * This thread manage application communication.
1556 static void *thread_manage_apps(void *data
)
1558 int i
, ret
, pollfd
, err
= -1;
1560 uint32_t revents
, nb_fd
;
1561 struct lttng_poll_event events
;
1563 DBG("[thread] Manage application started");
1565 rcu_register_thread();
1566 rcu_thread_online();
1568 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1570 if (testpoint(sessiond_thread_manage_apps
)) {
1571 goto error_testpoint
;
1574 health_code_update();
1576 ret
= sessiond_set_thread_pollset(&events
, 2);
1578 goto error_poll_create
;
1581 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1586 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1590 health_code_update();
1593 DBG("Apps thread polling");
1595 /* Inifinite blocking call, waiting for transmission */
1597 health_poll_entry();
1598 ret
= lttng_poll_wait(&events
, -1);
1599 DBG("Apps thread return from poll on %d fds",
1600 LTTNG_POLL_GETNB(&events
));
1604 * Restart interrupted system call.
1606 if (errno
== EINTR
) {
1614 for (i
= 0; i
< nb_fd
; i
++) {
1615 /* Fetch once the poll data */
1616 revents
= LTTNG_POLL_GETEV(&events
, i
);
1617 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1619 health_code_update();
1622 /* No activity for this FD (poll implementation). */
1626 /* Thread quit pipe has been closed. Killing thread. */
1627 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1633 /* Inspect the apps cmd pipe */
1634 if (pollfd
== apps_cmd_pipe
[0]) {
1635 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1636 ERR("Apps command pipe error");
1638 } else if (revents
& LPOLLIN
) {
1642 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1643 if (size_ret
< sizeof(sock
)) {
1644 PERROR("read apps cmd pipe");
1648 health_code_update();
1651 * We only monitor the error events of the socket. This
1652 * thread does not handle any incoming data from UST
1655 ret
= lttng_poll_add(&events
, sock
,
1656 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1661 DBG("Apps with sock %d added to poll set", sock
);
1665 * At this point, we know that a registered application made
1666 * the event at poll_wait.
1668 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1669 /* Removing from the poll set */
1670 ret
= lttng_poll_del(&events
, pollfd
);
1675 /* Socket closed on remote end. */
1676 ust_app_unregister(pollfd
);
1680 health_code_update();
1686 lttng_poll_clean(&events
);
1689 utils_close_pipe(apps_cmd_pipe
);
1690 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1693 * We don't clean the UST app hash table here since already registered
1694 * applications can still be controlled so let them be until the session
1695 * daemon dies or the applications stop.
1700 ERR("Health error occurred in %s", __func__
);
1702 health_unregister(health_sessiond
);
1703 DBG("Application communication apps thread cleanup complete");
1704 rcu_thread_offline();
1705 rcu_unregister_thread();
1710 * Send a socket to a thread This is called from the dispatch UST registration
1711 * thread once all sockets are set for the application.
1713 * The sock value can be invalid, we don't really care, the thread will handle
1714 * it and make the necessary cleanup if so.
1716 * On success, return 0 else a negative value being the errno message of the
1719 static int send_socket_to_thread(int fd
, int sock
)
1724 * It's possible that the FD is set as invalid with -1 concurrently just
1725 * before calling this function being a shutdown state of the thread.
1732 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1733 if (ret
< sizeof(sock
)) {
1734 PERROR("write apps pipe %d", fd
);
1741 /* All good. Don't send back the write positive ret value. */
1748 * Sanitize the wait queue of the dispatch registration thread meaning removing
1749 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1750 * notify socket is never received.
1752 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1754 int ret
, nb_fd
= 0, i
;
1755 unsigned int fd_added
= 0;
1756 struct lttng_poll_event events
;
1757 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1761 lttng_poll_init(&events
);
1763 /* Just skip everything for an empty queue. */
1764 if (!wait_queue
->count
) {
1768 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1773 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1774 &wait_queue
->head
, head
) {
1775 assert(wait_node
->app
);
1776 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1777 LPOLLHUP
| LPOLLERR
);
1790 * Poll but don't block so we can quickly identify the faulty events and
1791 * clean them afterwards from the wait queue.
1793 ret
= lttng_poll_wait(&events
, 0);
1799 for (i
= 0; i
< nb_fd
; i
++) {
1800 /* Get faulty FD. */
1801 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1802 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1805 /* No activity for this FD (poll implementation). */
1809 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1810 &wait_queue
->head
, head
) {
1811 if (pollfd
== wait_node
->app
->sock
&&
1812 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1813 cds_list_del(&wait_node
->head
);
1814 wait_queue
->count
--;
1815 ust_app_destroy(wait_node
->app
);
1823 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1827 lttng_poll_clean(&events
);
1831 lttng_poll_clean(&events
);
1833 ERR("Unable to sanitize wait queue");
1838 * Dispatch request from the registration threads to the application
1839 * communication thread.
1841 static void *thread_dispatch_ust_registration(void *data
)
1844 struct cds_wfcq_node
*node
;
1845 struct ust_command
*ust_cmd
= NULL
;
1846 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1847 struct ust_reg_wait_queue wait_queue
= {
1851 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1853 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1854 goto error_testpoint
;
1857 health_code_update();
1859 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1861 DBG("[thread] Dispatch UST command started");
1863 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1864 health_code_update();
1866 /* Atomically prepare the queue futex */
1867 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1870 struct ust_app
*app
= NULL
;
1874 * Make sure we don't have node(s) that have hung up before receiving
1875 * the notify socket. This is to clean the list in order to avoid
1876 * memory leaks from notify socket that are never seen.
1878 sanitize_wait_queue(&wait_queue
);
1880 health_code_update();
1881 /* Dequeue command for registration */
1882 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1884 DBG("Woken up but nothing in the UST command queue");
1885 /* Continue thread execution */
1889 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1891 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1892 " gid:%d sock:%d name:%s (version %d.%d)",
1893 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1894 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1895 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1896 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1898 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1899 wait_node
= zmalloc(sizeof(*wait_node
));
1901 PERROR("zmalloc wait_node dispatch");
1902 ret
= close(ust_cmd
->sock
);
1904 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1906 lttng_fd_put(LTTNG_FD_APPS
, 1);
1910 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1912 /* Create application object if socket is CMD. */
1913 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1915 if (!wait_node
->app
) {
1916 ret
= close(ust_cmd
->sock
);
1918 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1920 lttng_fd_put(LTTNG_FD_APPS
, 1);
1926 * Add application to the wait queue so we can set the notify
1927 * socket before putting this object in the global ht.
1929 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1934 * We have to continue here since we don't have the notify
1935 * socket and the application MUST be added to the hash table
1936 * only at that moment.
1941 * Look for the application in the local wait queue and set the
1942 * notify socket if found.
1944 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1945 &wait_queue
.head
, head
) {
1946 health_code_update();
1947 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1948 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1949 cds_list_del(&wait_node
->head
);
1951 app
= wait_node
->app
;
1953 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1959 * With no application at this stage the received socket is
1960 * basically useless so close it before we free the cmd data
1961 * structure for good.
1964 ret
= close(ust_cmd
->sock
);
1966 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1968 lttng_fd_put(LTTNG_FD_APPS
, 1);
1975 * @session_lock_list
1977 * Lock the global session list so from the register up to the
1978 * registration done message, no thread can see the application
1979 * and change its state.
1981 session_lock_list();
1985 * Add application to the global hash table. This needs to be
1986 * done before the update to the UST registry can locate the
1991 /* Set app version. This call will print an error if needed. */
1992 (void) ust_app_version(app
);
1994 /* Send notify socket through the notify pipe. */
1995 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1999 session_unlock_list();
2001 * No notify thread, stop the UST tracing. However, this is
2002 * not an internal error of the this thread thus setting
2003 * the health error code to a normal exit.
2010 * Update newly registered application with the tracing
2011 * registry info already enabled information.
2013 update_ust_app(app
->sock
);
2016 * Don't care about return value. Let the manage apps threads
2017 * handle app unregistration upon socket close.
2019 (void) ust_app_register_done(app
);
2022 * Even if the application socket has been closed, send the app
2023 * to the thread and unregistration will take place at that
2026 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2029 session_unlock_list();
2031 * No apps. thread, stop the UST tracing. However, this is
2032 * not an internal error of the this thread thus setting
2033 * the health error code to a normal exit.
2040 session_unlock_list();
2042 } while (node
!= NULL
);
2044 health_poll_entry();
2045 /* Futex wait on queue. Blocking call on futex() */
2046 futex_nto1_wait(&ust_cmd_queue
.futex
);
2049 /* Normal exit, no error */
2053 /* Clean up wait queue. */
2054 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2055 &wait_queue
.head
, head
) {
2056 cds_list_del(&wait_node
->head
);
2061 /* Empty command queue. */
2063 /* Dequeue command for registration */
2064 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2068 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2069 ret
= close(ust_cmd
->sock
);
2071 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2073 lttng_fd_put(LTTNG_FD_APPS
, 1);
2078 DBG("Dispatch thread dying");
2081 ERR("Health error occurred in %s", __func__
);
2083 health_unregister(health_sessiond
);
2088 * This thread manage application registration.
2090 static void *thread_registration_apps(void *data
)
2092 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2093 uint32_t revents
, nb_fd
;
2094 struct lttng_poll_event events
;
2096 * Get allocated in this thread, enqueued to a global queue, dequeued and
2097 * freed in the manage apps thread.
2099 struct ust_command
*ust_cmd
= NULL
;
2101 DBG("[thread] Manage application registration started");
2103 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2105 if (testpoint(sessiond_thread_registration_apps
)) {
2106 goto error_testpoint
;
2109 ret
= lttcomm_listen_unix_sock(apps_sock
);
2115 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2116 * more will be added to this poll set.
2118 ret
= sessiond_set_thread_pollset(&events
, 2);
2120 goto error_create_poll
;
2123 /* Add the application registration socket */
2124 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2126 goto error_poll_add
;
2129 /* Notify all applications to register */
2130 ret
= notify_ust_apps(1);
2132 ERR("Failed to notify applications or create the wait shared memory.\n"
2133 "Execution continues but there might be problem for already\n"
2134 "running applications that wishes to register.");
2138 DBG("Accepting application registration");
2140 /* Inifinite blocking call, waiting for transmission */
2142 health_poll_entry();
2143 ret
= lttng_poll_wait(&events
, -1);
2147 * Restart interrupted system call.
2149 if (errno
== EINTR
) {
2157 for (i
= 0; i
< nb_fd
; i
++) {
2158 health_code_update();
2160 /* Fetch once the poll data */
2161 revents
= LTTNG_POLL_GETEV(&events
, i
);
2162 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2165 /* No activity for this FD (poll implementation). */
2169 /* Thread quit pipe has been closed. Killing thread. */
2170 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2176 /* Event on the registration socket */
2177 if (pollfd
== apps_sock
) {
2178 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2179 ERR("Register apps socket poll error");
2181 } else if (revents
& LPOLLIN
) {
2182 sock
= lttcomm_accept_unix_sock(apps_sock
);
2188 * Set socket timeout for both receiving and ending.
2189 * app_socket_timeout is in seconds, whereas
2190 * lttcomm_setsockopt_rcv_timeout and
2191 * lttcomm_setsockopt_snd_timeout expect msec as
2194 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2195 app_socket_timeout
* 1000);
2196 (void) lttcomm_setsockopt_snd_timeout(sock
,
2197 app_socket_timeout
* 1000);
2200 * Set the CLOEXEC flag. Return code is useless because
2201 * either way, the show must go on.
2203 (void) utils_set_fd_cloexec(sock
);
2205 /* Create UST registration command for enqueuing */
2206 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2207 if (ust_cmd
== NULL
) {
2208 PERROR("ust command zmalloc");
2217 * Using message-based transmissions to ensure we don't
2218 * have to deal with partially received messages.
2220 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2222 ERR("Exhausted file descriptors allowed for applications.");
2232 health_code_update();
2233 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2236 /* Close socket of the application. */
2241 lttng_fd_put(LTTNG_FD_APPS
, 1);
2245 health_code_update();
2247 ust_cmd
->sock
= sock
;
2250 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2251 " gid:%d sock:%d name:%s (version %d.%d)",
2252 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2253 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2254 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2255 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2258 * Lock free enqueue the registration request. The red pill
2259 * has been taken! This apps will be part of the *system*.
2261 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2264 * Wake the registration queue futex. Implicit memory
2265 * barrier with the exchange in cds_wfcq_enqueue.
2267 futex_nto1_wake(&ust_cmd_queue
.futex
);
2275 /* Notify that the registration thread is gone */
2278 if (apps_sock
>= 0) {
2279 ret
= close(apps_sock
);
2289 lttng_fd_put(LTTNG_FD_APPS
, 1);
2291 unlink(apps_unix_sock_path
);
2294 lttng_poll_clean(&events
);
2298 DBG("UST Registration thread cleanup complete");
2301 ERR("Health error occurred in %s", __func__
);
2303 health_unregister(health_sessiond
);
2309 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2310 * exec or it will fails.
2312 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2315 struct timespec timeout
;
2317 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2318 consumer_data
->consumer_thread_is_ready
= 0;
2320 /* Setup pthread condition */
2321 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2324 PERROR("pthread_condattr_init consumer data");
2329 * Set the monotonic clock in order to make sure we DO NOT jump in time
2330 * between the clock_gettime() call and the timedwait call. See bug #324
2331 * for a more details and how we noticed it.
2333 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2336 PERROR("pthread_condattr_setclock consumer data");
2340 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2343 PERROR("pthread_cond_init consumer data");
2347 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2351 PERROR("pthread_create consumer");
2356 /* We are about to wait on a pthread condition */
2357 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2359 /* Get time for sem_timedwait absolute timeout */
2360 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2362 * Set the timeout for the condition timed wait even if the clock gettime
2363 * call fails since we might loop on that call and we want to avoid to
2364 * increment the timeout too many times.
2366 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2369 * The following loop COULD be skipped in some conditions so this is why we
2370 * set ret to 0 in order to make sure at least one round of the loop is
2376 * Loop until the condition is reached or when a timeout is reached. Note
2377 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2378 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2379 * possible. This loop does not take any chances and works with both of
2382 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2383 if (clock_ret
< 0) {
2384 PERROR("clock_gettime spawn consumer");
2385 /* Infinite wait for the consumerd thread to be ready */
2386 ret
= pthread_cond_wait(&consumer_data
->cond
,
2387 &consumer_data
->cond_mutex
);
2389 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2390 &consumer_data
->cond_mutex
, &timeout
);
2394 /* Release the pthread condition */
2395 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2399 if (ret
== ETIMEDOUT
) {
2403 * Call has timed out so we kill the kconsumerd_thread and return
2406 ERR("Condition timed out. The consumer thread was never ready."
2408 pth_ret
= pthread_cancel(consumer_data
->thread
);
2410 PERROR("pthread_cancel consumer thread");
2413 PERROR("pthread_cond_wait failed consumer thread");
2415 /* Caller is expecting a negative value on failure. */
2420 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2421 if (consumer_data
->pid
== 0) {
2422 ERR("Consumerd did not start");
2423 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2426 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2435 * Join consumer thread
2437 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2441 /* Consumer pid must be a real one. */
2442 if (consumer_data
->pid
> 0) {
2444 ret
= kill(consumer_data
->pid
, SIGTERM
);
2446 PERROR("Error killing consumer daemon");
2449 return pthread_join(consumer_data
->thread
, &status
);
2456 * Fork and exec a consumer daemon (consumerd).
2458 * Return pid if successful else -1.
2460 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2464 const char *consumer_to_use
;
2465 const char *verbosity
;
2468 DBG("Spawning consumerd");
2475 if (opt_verbose_consumer
) {
2476 verbosity
= "--verbose";
2477 } else if (lttng_opt_quiet
) {
2478 verbosity
= "--quiet";
2483 switch (consumer_data
->type
) {
2484 case LTTNG_CONSUMER_KERNEL
:
2486 * Find out which consumerd to execute. We will first try the
2487 * 64-bit path, then the sessiond's installation directory, and
2488 * fallback on the 32-bit one,
2490 DBG3("Looking for a kernel consumer at these locations:");
2491 DBG3(" 1) %s", consumerd64_bin
);
2492 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2493 DBG3(" 3) %s", consumerd32_bin
);
2494 if (stat(consumerd64_bin
, &st
) == 0) {
2495 DBG3("Found location #1");
2496 consumer_to_use
= consumerd64_bin
;
2497 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2498 DBG3("Found location #2");
2499 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2500 } else if (stat(consumerd32_bin
, &st
) == 0) {
2501 DBG3("Found location #3");
2502 consumer_to_use
= consumerd32_bin
;
2504 DBG("Could not find any valid consumerd executable");
2508 DBG("Using kernel consumer at: %s", consumer_to_use
);
2509 ret
= execl(consumer_to_use
,
2510 "lttng-consumerd", verbosity
, "-k",
2511 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2512 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2513 "--group", tracing_group_name
,
2516 case LTTNG_CONSUMER64_UST
:
2518 char *tmpnew
= NULL
;
2520 if (consumerd64_libdir
[0] != '\0') {
2524 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2528 tmplen
= strlen("LD_LIBRARY_PATH=")
2529 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2530 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2535 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2536 strcat(tmpnew
, consumerd64_libdir
);
2537 if (tmp
[0] != '\0') {
2538 strcat(tmpnew
, ":");
2539 strcat(tmpnew
, tmp
);
2541 ret
= putenv(tmpnew
);
2548 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2549 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2550 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2551 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2552 "--group", tracing_group_name
,
2554 if (consumerd64_libdir
[0] != '\0') {
2559 case LTTNG_CONSUMER32_UST
:
2561 char *tmpnew
= NULL
;
2563 if (consumerd32_libdir
[0] != '\0') {
2567 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2571 tmplen
= strlen("LD_LIBRARY_PATH=")
2572 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2573 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2578 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2579 strcat(tmpnew
, consumerd32_libdir
);
2580 if (tmp
[0] != '\0') {
2581 strcat(tmpnew
, ":");
2582 strcat(tmpnew
, tmp
);
2584 ret
= putenv(tmpnew
);
2591 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2592 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2593 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2594 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2595 "--group", tracing_group_name
,
2597 if (consumerd32_libdir
[0] != '\0') {
2603 PERROR("unknown consumer type");
2607 PERROR("Consumer execl()");
2609 /* Reaching this point, we got a failure on our execl(). */
2611 } else if (pid
> 0) {
2614 PERROR("start consumer fork");
2622 * Spawn the consumerd daemon and session daemon thread.
2624 static int start_consumerd(struct consumer_data
*consumer_data
)
2629 * Set the listen() state on the socket since there is a possible race
2630 * between the exec() of the consumer daemon and this call if place in the
2631 * consumer thread. See bug #366 for more details.
2633 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2638 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2639 if (consumer_data
->pid
!= 0) {
2640 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2644 ret
= spawn_consumerd(consumer_data
);
2646 ERR("Spawning consumerd failed");
2647 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2651 /* Setting up the consumer_data pid */
2652 consumer_data
->pid
= ret
;
2653 DBG2("Consumer pid %d", consumer_data
->pid
);
2654 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2656 DBG2("Spawning consumer control thread");
2657 ret
= spawn_consumer_thread(consumer_data
);
2659 ERR("Fatal error spawning consumer control thread");
2667 /* Cleanup already created sockets on error. */
2668 if (consumer_data
->err_sock
>= 0) {
2671 err
= close(consumer_data
->err_sock
);
2673 PERROR("close consumer data error socket");
2680 * Setup necessary data for kernel tracer action.
2682 static int init_kernel_tracer(void)
2686 /* Modprobe lttng kernel modules */
2687 ret
= modprobe_lttng_control();
2692 /* Open debugfs lttng */
2693 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2694 if (kernel_tracer_fd
< 0) {
2695 DBG("Failed to open %s", module_proc_lttng
);
2700 /* Validate kernel version */
2701 ret
= kernel_validate_version(kernel_tracer_fd
);
2706 ret
= modprobe_lttng_data();
2711 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2715 modprobe_remove_lttng_control();
2716 ret
= close(kernel_tracer_fd
);
2720 kernel_tracer_fd
= -1;
2721 return LTTNG_ERR_KERN_VERSION
;
2724 ret
= close(kernel_tracer_fd
);
2730 modprobe_remove_lttng_control();
2733 WARN("No kernel tracer available");
2734 kernel_tracer_fd
= -1;
2736 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2738 return LTTNG_ERR_KERN_NA
;
2744 * Copy consumer output from the tracing session to the domain session. The
2745 * function also applies the right modification on a per domain basis for the
2746 * trace files destination directory.
2748 * Should *NOT* be called with RCU read-side lock held.
2750 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2753 const char *dir_name
;
2754 struct consumer_output
*consumer
;
2757 assert(session
->consumer
);
2760 case LTTNG_DOMAIN_KERNEL
:
2761 DBG3("Copying tracing session consumer output in kernel session");
2763 * XXX: We should audit the session creation and what this function
2764 * does "extra" in order to avoid a destroy since this function is used
2765 * in the domain session creation (kernel and ust) only. Same for UST
2768 if (session
->kernel_session
->consumer
) {
2769 consumer_output_put(session
->kernel_session
->consumer
);
2771 session
->kernel_session
->consumer
=
2772 consumer_copy_output(session
->consumer
);
2773 /* Ease our life a bit for the next part */
2774 consumer
= session
->kernel_session
->consumer
;
2775 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2777 case LTTNG_DOMAIN_JUL
:
2778 case LTTNG_DOMAIN_LOG4J
:
2779 case LTTNG_DOMAIN_PYTHON
:
2780 case LTTNG_DOMAIN_UST
:
2781 DBG3("Copying tracing session consumer output in UST session");
2782 if (session
->ust_session
->consumer
) {
2783 consumer_output_put(session
->ust_session
->consumer
);
2785 session
->ust_session
->consumer
=
2786 consumer_copy_output(session
->consumer
);
2787 /* Ease our life a bit for the next part */
2788 consumer
= session
->ust_session
->consumer
;
2789 dir_name
= DEFAULT_UST_TRACE_DIR
;
2792 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2796 /* Append correct directory to subdir */
2797 strncat(consumer
->subdir
, dir_name
,
2798 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2799 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2808 * Create an UST session and add it to the session ust list.
2810 * Should *NOT* be called with RCU read-side lock held.
2812 static int create_ust_session(struct ltt_session
*session
,
2813 struct lttng_domain
*domain
)
2816 struct ltt_ust_session
*lus
= NULL
;
2820 assert(session
->consumer
);
2822 switch (domain
->type
) {
2823 case LTTNG_DOMAIN_JUL
:
2824 case LTTNG_DOMAIN_LOG4J
:
2825 case LTTNG_DOMAIN_PYTHON
:
2826 case LTTNG_DOMAIN_UST
:
2829 ERR("Unknown UST domain on create session %d", domain
->type
);
2830 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2834 DBG("Creating UST session");
2836 lus
= trace_ust_create_session(session
->id
);
2838 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2842 lus
->uid
= session
->uid
;
2843 lus
->gid
= session
->gid
;
2844 lus
->output_traces
= session
->output_traces
;
2845 lus
->snapshot_mode
= session
->snapshot_mode
;
2846 lus
->live_timer_interval
= session
->live_timer
;
2847 session
->ust_session
= lus
;
2848 if (session
->shm_path
[0]) {
2849 strncpy(lus
->root_shm_path
, session
->shm_path
,
2850 sizeof(lus
->root_shm_path
));
2851 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2852 strncpy(lus
->shm_path
, session
->shm_path
,
2853 sizeof(lus
->shm_path
));
2854 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2855 strncat(lus
->shm_path
, "/ust",
2856 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2858 /* Copy session output to the newly created UST session */
2859 ret
= copy_session_consumer(domain
->type
, session
);
2860 if (ret
!= LTTNG_OK
) {
2868 session
->ust_session
= NULL
;
2873 * Create a kernel tracer session then create the default channel.
2875 static int create_kernel_session(struct ltt_session
*session
)
2879 DBG("Creating kernel session");
2881 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2883 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2887 /* Code flow safety */
2888 assert(session
->kernel_session
);
2890 /* Copy session output to the newly created Kernel session */
2891 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2892 if (ret
!= LTTNG_OK
) {
2896 /* Create directory(ies) on local filesystem. */
2897 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2898 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2899 ret
= run_as_mkdir_recursive(
2900 session
->kernel_session
->consumer
->dst
.trace_path
,
2901 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2903 if (errno
!= EEXIST
) {
2904 ERR("Trace directory creation error");
2910 session
->kernel_session
->uid
= session
->uid
;
2911 session
->kernel_session
->gid
= session
->gid
;
2912 session
->kernel_session
->output_traces
= session
->output_traces
;
2913 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2918 trace_kernel_destroy_session(session
->kernel_session
);
2919 session
->kernel_session
= NULL
;
2924 * Count number of session permitted by uid/gid.
2926 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2929 struct ltt_session
*session
;
2931 DBG("Counting number of available session for UID %d GID %d",
2933 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2935 * Only list the sessions the user can control.
2937 if (!session_access_ok(session
, uid
, gid
)) {
2946 * Process the command requested by the lttng client within the command
2947 * context structure. This function make sure that the return structure (llm)
2948 * is set and ready for transmission before returning.
2950 * Return any error encountered or 0 for success.
2952 * "sock" is only used for special-case var. len data.
2954 * Should *NOT* be called with RCU read-side lock held.
2956 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2960 int need_tracing_session
= 1;
2963 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2965 assert(!rcu_read_ongoing());
2969 switch (cmd_ctx
->lsm
->cmd_type
) {
2970 case LTTNG_CREATE_SESSION
:
2971 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2972 case LTTNG_CREATE_SESSION_LIVE
:
2973 case LTTNG_DESTROY_SESSION
:
2974 case LTTNG_LIST_SESSIONS
:
2975 case LTTNG_LIST_DOMAINS
:
2976 case LTTNG_START_TRACE
:
2977 case LTTNG_STOP_TRACE
:
2978 case LTTNG_DATA_PENDING
:
2979 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2980 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2981 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2982 case LTTNG_SNAPSHOT_RECORD
:
2983 case LTTNG_SAVE_SESSION
:
2984 case LTTNG_SET_SESSION_SHM_PATH
:
2991 if (opt_no_kernel
&& need_domain
2992 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2994 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2996 ret
= LTTNG_ERR_KERN_NA
;
3001 /* Deny register consumer if we already have a spawned consumer. */
3002 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3003 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3004 if (kconsumer_data
.pid
> 0) {
3005 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3006 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3009 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3013 * Check for command that don't needs to allocate a returned payload. We do
3014 * this here so we don't have to make the call for no payload at each
3017 switch(cmd_ctx
->lsm
->cmd_type
) {
3018 case LTTNG_LIST_SESSIONS
:
3019 case LTTNG_LIST_TRACEPOINTS
:
3020 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3021 case LTTNG_LIST_DOMAINS
:
3022 case LTTNG_LIST_CHANNELS
:
3023 case LTTNG_LIST_EVENTS
:
3024 case LTTNG_LIST_SYSCALLS
:
3025 case LTTNG_LIST_TRACKER_PIDS
:
3028 /* Setup lttng message with no payload */
3029 ret
= setup_lttng_msg(cmd_ctx
, 0);
3031 /* This label does not try to unlock the session */
3032 goto init_setup_error
;
3036 /* Commands that DO NOT need a session. */
3037 switch (cmd_ctx
->lsm
->cmd_type
) {
3038 case LTTNG_CREATE_SESSION
:
3039 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3040 case LTTNG_CREATE_SESSION_LIVE
:
3041 case LTTNG_CALIBRATE
:
3042 case LTTNG_LIST_SESSIONS
:
3043 case LTTNG_LIST_TRACEPOINTS
:
3044 case LTTNG_LIST_SYSCALLS
:
3045 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3046 case LTTNG_SAVE_SESSION
:
3047 need_tracing_session
= 0;
3050 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3052 * We keep the session list lock across _all_ commands
3053 * for now, because the per-session lock does not
3054 * handle teardown properly.
3056 session_lock_list();
3057 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3058 if (cmd_ctx
->session
== NULL
) {
3059 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3062 /* Acquire lock for the session */
3063 session_lock(cmd_ctx
->session
);
3069 * Commands that need a valid session but should NOT create one if none
3070 * exists. Instead of creating one and destroying it when the command is
3071 * handled, process that right before so we save some round trip in useless
3074 switch (cmd_ctx
->lsm
->cmd_type
) {
3075 case LTTNG_DISABLE_CHANNEL
:
3076 case LTTNG_DISABLE_EVENT
:
3077 switch (cmd_ctx
->lsm
->domain
.type
) {
3078 case LTTNG_DOMAIN_KERNEL
:
3079 if (!cmd_ctx
->session
->kernel_session
) {
3080 ret
= LTTNG_ERR_NO_CHANNEL
;
3084 case LTTNG_DOMAIN_JUL
:
3085 case LTTNG_DOMAIN_LOG4J
:
3086 case LTTNG_DOMAIN_PYTHON
:
3087 case LTTNG_DOMAIN_UST
:
3088 if (!cmd_ctx
->session
->ust_session
) {
3089 ret
= LTTNG_ERR_NO_CHANNEL
;
3094 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3106 * Check domain type for specific "pre-action".
3108 switch (cmd_ctx
->lsm
->domain
.type
) {
3109 case LTTNG_DOMAIN_KERNEL
:
3111 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3115 /* Kernel tracer check */
3116 if (kernel_tracer_fd
== -1) {
3117 /* Basically, load kernel tracer modules */
3118 ret
= init_kernel_tracer();
3124 /* Consumer is in an ERROR state. Report back to client */
3125 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3126 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3130 /* Need a session for kernel command */
3131 if (need_tracing_session
) {
3132 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3133 ret
= create_kernel_session(cmd_ctx
->session
);
3135 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3140 /* Start the kernel consumer daemon */
3141 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3142 if (kconsumer_data
.pid
== 0 &&
3143 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3144 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3145 ret
= start_consumerd(&kconsumer_data
);
3147 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3150 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3152 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3156 * The consumer was just spawned so we need to add the socket to
3157 * the consumer output of the session if exist.
3159 ret
= consumer_create_socket(&kconsumer_data
,
3160 cmd_ctx
->session
->kernel_session
->consumer
);
3167 case LTTNG_DOMAIN_JUL
:
3168 case LTTNG_DOMAIN_LOG4J
:
3169 case LTTNG_DOMAIN_PYTHON
:
3170 case LTTNG_DOMAIN_UST
:
3172 if (!ust_app_supported()) {
3173 ret
= LTTNG_ERR_NO_UST
;
3176 /* Consumer is in an ERROR state. Report back to client */
3177 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3178 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3182 if (need_tracing_session
) {
3183 /* Create UST session if none exist. */
3184 if (cmd_ctx
->session
->ust_session
== NULL
) {
3185 ret
= create_ust_session(cmd_ctx
->session
,
3186 &cmd_ctx
->lsm
->domain
);
3187 if (ret
!= LTTNG_OK
) {
3192 /* Start the UST consumer daemons */
3194 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3195 if (consumerd64_bin
[0] != '\0' &&
3196 ustconsumer64_data
.pid
== 0 &&
3197 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3198 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3199 ret
= start_consumerd(&ustconsumer64_data
);
3201 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3202 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3206 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3207 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3209 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3213 * Setup socket for consumer 64 bit. No need for atomic access
3214 * since it was set above and can ONLY be set in this thread.
3216 ret
= consumer_create_socket(&ustconsumer64_data
,
3217 cmd_ctx
->session
->ust_session
->consumer
);
3223 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3224 if (consumerd32_bin
[0] != '\0' &&
3225 ustconsumer32_data
.pid
== 0 &&
3226 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3227 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3228 ret
= start_consumerd(&ustconsumer32_data
);
3230 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3231 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3235 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3236 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3238 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3242 * Setup socket for consumer 64 bit. No need for atomic access
3243 * since it was set above and can ONLY be set in this thread.
3245 ret
= consumer_create_socket(&ustconsumer32_data
,
3246 cmd_ctx
->session
->ust_session
->consumer
);
3258 /* Validate consumer daemon state when start/stop trace command */
3259 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3260 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3261 switch (cmd_ctx
->lsm
->domain
.type
) {
3262 case LTTNG_DOMAIN_JUL
:
3263 case LTTNG_DOMAIN_LOG4J
:
3264 case LTTNG_DOMAIN_PYTHON
:
3265 case LTTNG_DOMAIN_UST
:
3266 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3267 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3271 case LTTNG_DOMAIN_KERNEL
:
3272 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3273 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3281 * Check that the UID or GID match that of the tracing session.
3282 * The root user can interact with all sessions.
3284 if (need_tracing_session
) {
3285 if (!session_access_ok(cmd_ctx
->session
,
3286 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3287 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3288 ret
= LTTNG_ERR_EPERM
;
3294 * Send relayd information to consumer as soon as we have a domain and a
3297 if (cmd_ctx
->session
&& need_domain
) {
3299 * Setup relayd if not done yet. If the relayd information was already
3300 * sent to the consumer, this call will gracefully return.
3302 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3303 if (ret
!= LTTNG_OK
) {
3308 /* Process by command type */
3309 switch (cmd_ctx
->lsm
->cmd_type
) {
3310 case LTTNG_ADD_CONTEXT
:
3312 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3313 cmd_ctx
->lsm
->u
.context
.channel_name
,
3314 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3317 case LTTNG_DISABLE_CHANNEL
:
3319 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3320 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3323 case LTTNG_DISABLE_EVENT
:
3327 * FIXME: handle filter; for now we just receive the filter's
3328 * bytecode along with the filter expression which are sent by
3329 * liblttng-ctl and discard them.
3331 * This fixes an issue where the client may block while sending
3332 * the filter payload and encounter an error because the session
3333 * daemon closes the socket without ever handling this data.
3335 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3336 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3339 char data
[LTTNG_FILTER_MAX_LEN
];
3341 DBG("Discarding disable event command payload of size %zu", count
);
3343 ret
= lttcomm_recv_unix_sock(sock
, data
,
3344 count
> sizeof(data
) ? sizeof(data
) : count
);
3349 count
-= (size_t) ret
;
3352 /* FIXME: passing packed structure to non-packed pointer */
3353 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3354 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3355 &cmd_ctx
->lsm
->u
.disable
.event
);
3358 case LTTNG_ENABLE_CHANNEL
:
3360 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3361 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3364 case LTTNG_TRACK_PID
:
3366 ret
= cmd_track_pid(cmd_ctx
->session
,
3367 cmd_ctx
->lsm
->domain
.type
,
3368 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3371 case LTTNG_UNTRACK_PID
:
3373 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3374 cmd_ctx
->lsm
->domain
.type
,
3375 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3378 case LTTNG_ENABLE_EVENT
:
3380 struct lttng_event_exclusion
*exclusion
= NULL
;
3381 struct lttng_filter_bytecode
*bytecode
= NULL
;
3382 char *filter_expression
= NULL
;
3384 /* Handle exclusion events and receive it from the client. */
3385 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3386 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3388 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3389 (count
* LTTNG_SYMBOL_NAME_LEN
));
3391 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3395 DBG("Receiving var len exclusion event list from client ...");
3396 exclusion
->count
= count
;
3397 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3398 count
* LTTNG_SYMBOL_NAME_LEN
);
3400 DBG("Nothing recv() from client var len data... continuing");
3403 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3408 /* Get filter expression from client. */
3409 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3410 size_t expression_len
=
3411 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3413 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3414 ret
= LTTNG_ERR_FILTER_INVAL
;
3419 filter_expression
= zmalloc(expression_len
);
3420 if (!filter_expression
) {
3422 ret
= LTTNG_ERR_FILTER_NOMEM
;
3426 /* Receive var. len. data */
3427 DBG("Receiving var len filter's expression from client ...");
3428 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3431 DBG("Nothing recv() from client car len data... continuing");
3433 free(filter_expression
);
3435 ret
= LTTNG_ERR_FILTER_INVAL
;
3440 /* Handle filter and get bytecode from client. */
3441 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3442 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3444 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3445 ret
= LTTNG_ERR_FILTER_INVAL
;
3446 free(filter_expression
);
3451 bytecode
= zmalloc(bytecode_len
);
3453 free(filter_expression
);
3455 ret
= LTTNG_ERR_FILTER_NOMEM
;
3459 /* Receive var. len. data */
3460 DBG("Receiving var len filter's bytecode from client ...");
3461 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3463 DBG("Nothing recv() from client car len data... continuing");
3465 free(filter_expression
);
3468 ret
= LTTNG_ERR_FILTER_INVAL
;
3472 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3473 free(filter_expression
);
3476 ret
= LTTNG_ERR_FILTER_INVAL
;
3481 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3482 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3483 &cmd_ctx
->lsm
->u
.enable
.event
,
3484 filter_expression
, bytecode
, exclusion
,
3485 kernel_poll_pipe
[1]);
3488 case LTTNG_LIST_TRACEPOINTS
:
3490 struct lttng_event
*events
;
3493 session_lock_list();
3494 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3495 session_unlock_list();
3496 if (nb_events
< 0) {
3497 /* Return value is a negative lttng_error_code. */
3503 * Setup lttng message with payload size set to the event list size in
3504 * bytes and then copy list into the llm payload.
3506 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3512 /* Copy event list into message payload */
3513 memcpy(cmd_ctx
->llm
->payload
, events
,
3514 sizeof(struct lttng_event
) * nb_events
);
3521 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3523 struct lttng_event_field
*fields
;
3526 session_lock_list();
3527 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3529 session_unlock_list();
3530 if (nb_fields
< 0) {
3531 /* Return value is a negative lttng_error_code. */
3537 * Setup lttng message with payload size set to the event list size in
3538 * bytes and then copy list into the llm payload.
3540 ret
= setup_lttng_msg(cmd_ctx
,
3541 sizeof(struct lttng_event_field
) * nb_fields
);
3547 /* Copy event list into message payload */
3548 memcpy(cmd_ctx
->llm
->payload
, fields
,
3549 sizeof(struct lttng_event_field
) * nb_fields
);
3556 case LTTNG_LIST_SYSCALLS
:
3558 struct lttng_event
*events
;
3561 nb_events
= cmd_list_syscalls(&events
);
3562 if (nb_events
< 0) {
3563 /* Return value is a negative lttng_error_code. */
3569 * Setup lttng message with payload size set to the event list size in
3570 * bytes and then copy list into the llm payload.
3572 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3578 /* Copy event list into message payload */
3579 memcpy(cmd_ctx
->llm
->payload
, events
,
3580 sizeof(struct lttng_event
) * nb_events
);
3587 case LTTNG_LIST_TRACKER_PIDS
:
3589 int32_t *pids
= NULL
;
3592 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3593 cmd_ctx
->lsm
->domain
.type
, &pids
);
3595 /* Return value is a negative lttng_error_code. */
3601 * Setup lttng message with payload size set to the event list size in
3602 * bytes and then copy list into the llm payload.
3604 ret
= setup_lttng_msg(cmd_ctx
, sizeof(int32_t) * nr_pids
);
3610 /* Copy event list into message payload */
3611 memcpy(cmd_ctx
->llm
->payload
, pids
,
3612 sizeof(int) * nr_pids
);
3619 case LTTNG_SET_CONSUMER_URI
:
3622 struct lttng_uri
*uris
;
3624 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3625 len
= nb_uri
* sizeof(struct lttng_uri
);
3628 ret
= LTTNG_ERR_INVALID
;
3632 uris
= zmalloc(len
);
3634 ret
= LTTNG_ERR_FATAL
;
3638 /* Receive variable len data */
3639 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3640 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3642 DBG("No URIs received from client... continuing");
3644 ret
= LTTNG_ERR_SESSION_FAIL
;
3649 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3651 if (ret
!= LTTNG_OK
) {
3658 case LTTNG_START_TRACE
:
3660 ret
= cmd_start_trace(cmd_ctx
->session
);
3663 case LTTNG_STOP_TRACE
:
3665 ret
= cmd_stop_trace(cmd_ctx
->session
);
3668 case LTTNG_CREATE_SESSION
:
3671 struct lttng_uri
*uris
= NULL
;
3673 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3674 len
= nb_uri
* sizeof(struct lttng_uri
);
3677 uris
= zmalloc(len
);
3679 ret
= LTTNG_ERR_FATAL
;
3683 /* Receive variable len data */
3684 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3685 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3687 DBG("No URIs received from client... continuing");
3689 ret
= LTTNG_ERR_SESSION_FAIL
;
3694 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3695 DBG("Creating session with ONE network URI is a bad call");
3696 ret
= LTTNG_ERR_SESSION_FAIL
;
3702 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3703 &cmd_ctx
->creds
, 0);
3709 case LTTNG_DESTROY_SESSION
:
3711 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3713 /* Set session to NULL so we do not unlock it after free. */
3714 cmd_ctx
->session
= NULL
;
3717 case LTTNG_LIST_DOMAINS
:
3720 struct lttng_domain
*domains
= NULL
;
3722 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3724 /* Return value is a negative lttng_error_code. */
3729 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3735 /* Copy event list into message payload */
3736 memcpy(cmd_ctx
->llm
->payload
, domains
,
3737 nb_dom
* sizeof(struct lttng_domain
));
3744 case LTTNG_LIST_CHANNELS
:
3747 struct lttng_channel
*channels
= NULL
;
3749 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3750 cmd_ctx
->session
, &channels
);
3752 /* Return value is a negative lttng_error_code. */
3757 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3763 /* Copy event list into message payload */
3764 memcpy(cmd_ctx
->llm
->payload
, channels
,
3765 nb_chan
* sizeof(struct lttng_channel
));
3772 case LTTNG_LIST_EVENTS
:
3775 struct lttng_event
*events
= NULL
;
3777 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3778 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3780 /* Return value is a negative lttng_error_code. */
3785 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3791 /* Copy event list into message payload */
3792 memcpy(cmd_ctx
->llm
->payload
, events
,
3793 nb_event
* sizeof(struct lttng_event
));
3800 case LTTNG_LIST_SESSIONS
:
3802 unsigned int nr_sessions
;
3804 session_lock_list();
3805 nr_sessions
= lttng_sessions_count(
3806 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3807 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3809 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3811 session_unlock_list();
3815 /* Filled the session array */
3816 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3817 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3818 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3820 session_unlock_list();
3825 case LTTNG_CALIBRATE
:
3827 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3828 &cmd_ctx
->lsm
->u
.calibrate
);
3831 case LTTNG_REGISTER_CONSUMER
:
3833 struct consumer_data
*cdata
;
3835 switch (cmd_ctx
->lsm
->domain
.type
) {
3836 case LTTNG_DOMAIN_KERNEL
:
3837 cdata
= &kconsumer_data
;
3840 ret
= LTTNG_ERR_UND
;
3844 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3845 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3848 case LTTNG_DATA_PENDING
:
3850 ret
= cmd_data_pending(cmd_ctx
->session
);
3853 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3855 struct lttcomm_lttng_output_id reply
;
3857 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3858 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3859 if (ret
!= LTTNG_OK
) {
3863 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3868 /* Copy output list into message payload */
3869 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3873 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3875 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3876 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3879 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3882 struct lttng_snapshot_output
*outputs
= NULL
;
3884 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3885 if (nb_output
< 0) {
3890 ret
= setup_lttng_msg(cmd_ctx
,
3891 nb_output
* sizeof(struct lttng_snapshot_output
));
3898 /* Copy output list into message payload */
3899 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3900 nb_output
* sizeof(struct lttng_snapshot_output
));
3907 case LTTNG_SNAPSHOT_RECORD
:
3909 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3910 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3911 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3914 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3917 struct lttng_uri
*uris
= NULL
;
3919 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3920 len
= nb_uri
* sizeof(struct lttng_uri
);
3923 uris
= zmalloc(len
);
3925 ret
= LTTNG_ERR_FATAL
;
3929 /* Receive variable len data */
3930 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3931 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3933 DBG("No URIs received from client... continuing");
3935 ret
= LTTNG_ERR_SESSION_FAIL
;
3940 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3941 DBG("Creating session with ONE network URI is a bad call");
3942 ret
= LTTNG_ERR_SESSION_FAIL
;
3948 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3949 nb_uri
, &cmd_ctx
->creds
);
3953 case LTTNG_CREATE_SESSION_LIVE
:
3956 struct lttng_uri
*uris
= NULL
;
3958 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3959 len
= nb_uri
* sizeof(struct lttng_uri
);
3962 uris
= zmalloc(len
);
3964 ret
= LTTNG_ERR_FATAL
;
3968 /* Receive variable len data */
3969 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3970 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3972 DBG("No URIs received from client... continuing");
3974 ret
= LTTNG_ERR_SESSION_FAIL
;
3979 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3980 DBG("Creating session with ONE network URI is a bad call");
3981 ret
= LTTNG_ERR_SESSION_FAIL
;
3987 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3988 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3992 case LTTNG_SAVE_SESSION
:
3994 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
3998 case LTTNG_SET_SESSION_SHM_PATH
:
4000 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4001 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4005 ret
= LTTNG_ERR_UND
;
4010 if (cmd_ctx
->llm
== NULL
) {
4011 DBG("Missing llm structure. Allocating one.");
4012 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
4016 /* Set return code */
4017 cmd_ctx
->llm
->ret_code
= ret
;
4019 if (cmd_ctx
->session
) {
4020 session_unlock(cmd_ctx
->session
);
4022 if (need_tracing_session
) {
4023 session_unlock_list();
4026 assert(!rcu_read_ongoing());
4031 * Thread managing health check socket.
4033 static void *thread_manage_health(void *data
)
4035 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4036 uint32_t revents
, nb_fd
;
4037 struct lttng_poll_event events
;
4038 struct health_comm_msg msg
;
4039 struct health_comm_reply reply
;
4041 DBG("[thread] Manage health check started");
4043 rcu_register_thread();
4045 /* We might hit an error path before this is created. */
4046 lttng_poll_init(&events
);
4048 /* Create unix socket */
4049 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4051 ERR("Unable to create health check Unix socket");
4057 /* lttng health client socket path permissions */
4058 ret
= chown(health_unix_sock_path
, 0,
4059 utils_get_group_id(tracing_group_name
));
4061 ERR("Unable to set group on %s", health_unix_sock_path
);
4067 ret
= chmod(health_unix_sock_path
,
4068 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4070 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4078 * Set the CLOEXEC flag. Return code is useless because either way, the
4081 (void) utils_set_fd_cloexec(sock
);
4083 ret
= lttcomm_listen_unix_sock(sock
);
4089 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4090 * more will be added to this poll set.
4092 ret
= sessiond_set_thread_pollset(&events
, 2);
4097 /* Add the application registration socket */
4098 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4103 sessiond_notify_ready();
4106 DBG("Health check ready");
4108 /* Inifinite blocking call, waiting for transmission */
4110 ret
= lttng_poll_wait(&events
, -1);
4113 * Restart interrupted system call.
4115 if (errno
== EINTR
) {
4123 for (i
= 0; i
< nb_fd
; i
++) {
4124 /* Fetch once the poll data */
4125 revents
= LTTNG_POLL_GETEV(&events
, i
);
4126 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4129 /* No activity for this FD (poll implementation). */
4133 /* Thread quit pipe has been closed. Killing thread. */
4134 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4140 /* Event on the registration socket */
4141 if (pollfd
== sock
) {
4142 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4143 ERR("Health socket poll error");
4149 new_sock
= lttcomm_accept_unix_sock(sock
);
4155 * Set the CLOEXEC flag. Return code is useless because either way, the
4158 (void) utils_set_fd_cloexec(new_sock
);
4160 DBG("Receiving data from client for health...");
4161 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4163 DBG("Nothing recv() from client... continuing");
4164 ret
= close(new_sock
);
4172 rcu_thread_online();
4174 memset(&reply
, 0, sizeof(reply
));
4175 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4177 * health_check_state returns 0 if health is
4180 if (!health_check_state(health_sessiond
, i
)) {
4181 reply
.ret_code
|= 1ULL << i
;
4185 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4187 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4189 ERR("Failed to send health data back to client");
4192 /* End of transmission */
4193 ret
= close(new_sock
);
4203 ERR("Health error occurred in %s", __func__
);
4205 DBG("Health check thread dying");
4206 unlink(health_unix_sock_path
);
4214 lttng_poll_clean(&events
);
4216 rcu_unregister_thread();
4221 * This thread manage all clients request using the unix client socket for
4224 static void *thread_manage_clients(void *data
)
4226 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4228 uint32_t revents
, nb_fd
;
4229 struct command_ctx
*cmd_ctx
= NULL
;
4230 struct lttng_poll_event events
;
4232 DBG("[thread] Manage client started");
4234 rcu_register_thread();
4236 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4238 health_code_update();
4240 ret
= lttcomm_listen_unix_sock(client_sock
);
4246 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4247 * more will be added to this poll set.
4249 ret
= sessiond_set_thread_pollset(&events
, 2);
4251 goto error_create_poll
;
4254 /* Add the application registration socket */
4255 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4260 sessiond_notify_ready();
4261 ret
= sem_post(&load_info
->message_thread_ready
);
4263 PERROR("sem_post message_thread_ready");
4267 /* This testpoint is after we signal readiness to the parent. */
4268 if (testpoint(sessiond_thread_manage_clients
)) {
4272 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4276 health_code_update();
4279 DBG("Accepting client command ...");
4281 /* Inifinite blocking call, waiting for transmission */
4283 health_poll_entry();
4284 ret
= lttng_poll_wait(&events
, -1);
4288 * Restart interrupted system call.
4290 if (errno
== EINTR
) {
4298 for (i
= 0; i
< nb_fd
; i
++) {
4299 /* Fetch once the poll data */
4300 revents
= LTTNG_POLL_GETEV(&events
, i
);
4301 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4303 health_code_update();
4306 /* No activity for this FD (poll implementation). */
4310 /* Thread quit pipe has been closed. Killing thread. */
4311 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4317 /* Event on the registration socket */
4318 if (pollfd
== client_sock
) {
4319 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4320 ERR("Client socket poll error");
4326 DBG("Wait for client response");
4328 health_code_update();
4330 sock
= lttcomm_accept_unix_sock(client_sock
);
4336 * Set the CLOEXEC flag. Return code is useless because either way, the
4339 (void) utils_set_fd_cloexec(sock
);
4341 /* Set socket option for credentials retrieval */
4342 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4347 /* Allocate context command to process the client request */
4348 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4349 if (cmd_ctx
== NULL
) {
4350 PERROR("zmalloc cmd_ctx");
4354 /* Allocate data buffer for reception */
4355 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4356 if (cmd_ctx
->lsm
== NULL
) {
4357 PERROR("zmalloc cmd_ctx->lsm");
4361 cmd_ctx
->llm
= NULL
;
4362 cmd_ctx
->session
= NULL
;
4364 health_code_update();
4367 * Data is received from the lttng client. The struct
4368 * lttcomm_session_msg (lsm) contains the command and data request of
4371 DBG("Receiving data from client ...");
4372 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4373 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4375 DBG("Nothing recv() from client... continuing");
4381 clean_command_ctx(&cmd_ctx
);
4385 health_code_update();
4387 // TODO: Validate cmd_ctx including sanity check for
4388 // security purpose.
4390 rcu_thread_online();
4392 * This function dispatch the work to the kernel or userspace tracer
4393 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4394 * informations for the client. The command context struct contains
4395 * everything this function may needs.
4397 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4398 rcu_thread_offline();
4406 * TODO: Inform client somehow of the fatal error. At
4407 * this point, ret < 0 means that a zmalloc failed
4408 * (ENOMEM). Error detected but still accept
4409 * command, unless a socket error has been
4412 clean_command_ctx(&cmd_ctx
);
4416 health_code_update();
4418 DBG("Sending response (size: %d, retcode: %s)",
4419 cmd_ctx
->lttng_msg_size
,
4420 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
4421 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4423 ERR("Failed to send data back to client");
4426 /* End of transmission */
4433 clean_command_ctx(&cmd_ctx
);
4435 health_code_update();
4447 lttng_poll_clean(&events
);
4448 clean_command_ctx(&cmd_ctx
);
4452 unlink(client_unix_sock_path
);
4453 if (client_sock
>= 0) {
4454 ret
= close(client_sock
);
4462 ERR("Health error occurred in %s", __func__
);
4465 health_unregister(health_sessiond
);
4467 DBG("Client thread dying");
4469 rcu_unregister_thread();
4472 * Since we are creating the consumer threads, we own them, so we need
4473 * to join them before our thread exits.
4475 ret
= join_consumer_thread(&kconsumer_data
);
4478 PERROR("join_consumer");
4481 ret
= join_consumer_thread(&ustconsumer32_data
);
4484 PERROR("join_consumer ust32");
4487 ret
= join_consumer_thread(&ustconsumer64_data
);
4490 PERROR("join_consumer ust64");
4497 * usage function on stderr
4499 static void usage(void)
4501 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4502 fprintf(stderr
, " -h, --help Display this usage.\n");
4503 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4504 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4505 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4506 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4507 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4508 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4509 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4510 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4511 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4512 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4513 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4514 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4515 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4516 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4517 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4518 fprintf(stderr
, " -V, --version Show version number.\n");
4519 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4520 fprintf(stderr
, " -q, --quiet No output at all.\n");
4521 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4522 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4523 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4524 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4525 fprintf(stderr
, " --agent-tcp-port Agent registration TCP port\n");
4526 fprintf(stderr
, " -f --config PATH Load daemon configuration file\n");
4527 fprintf(stderr
, " -l --load PATH Load session configuration\n");
4528 fprintf(stderr
, " --kmod-probes Specify kernel module probes to load\n");
4529 fprintf(stderr
, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4532 static int string_match(const char *str1
, const char *str2
)
4534 return (str1
&& str2
) && !strcmp(str1
, str2
);
4538 * Take an option from the getopt output and set it in the right variable to be
4541 * Return 0 on success else a negative value.
4543 static int set_option(int opt
, const char *arg
, const char *optname
)
4547 if (arg
&& arg
[0] == '\0') {
4549 * This only happens if the value is read from daemon config
4550 * file. This means the option requires an argument and the
4551 * configuration file contains a line such as:
4558 if (string_match(optname
, "client-sock") || opt
== 'c') {
4559 if (lttng_is_setuid_setgid()) {
4560 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4561 "-c, --client-sock");
4563 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4565 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4566 if (lttng_is_setuid_setgid()) {
4567 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4570 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4572 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4574 } else if (string_match(optname
, "background") || opt
== 'b') {
4576 } else if (string_match(optname
, "group") || opt
== 'g') {
4577 if (lttng_is_setuid_setgid()) {
4578 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4582 * If the override option is set, the pointer points to a
4583 * *non* const thus freeing it even though the variable type is
4586 if (tracing_group_name_override
) {
4587 free((void *) tracing_group_name
);
4589 tracing_group_name
= strdup(arg
);
4590 if (!tracing_group_name
) {
4594 tracing_group_name_override
= 1;
4596 } else if (string_match(optname
, "help") || opt
== 'h') {
4599 } else if (string_match(optname
, "version") || opt
== 'V') {
4600 fprintf(stdout
, "%s\n", VERSION
);
4602 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4604 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4605 if (lttng_is_setuid_setgid()) {
4606 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4607 "--kconsumerd-err-sock");
4609 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4611 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4612 if (lttng_is_setuid_setgid()) {
4613 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4614 "--kconsumerd-cmd-sock");
4616 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4618 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4619 if (lttng_is_setuid_setgid()) {
4620 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4621 "--ustconsumerd64-err-sock");
4623 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4625 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4626 if (lttng_is_setuid_setgid()) {
4627 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4628 "--ustconsumerd64-cmd-sock");
4630 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4632 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4633 if (lttng_is_setuid_setgid()) {
4634 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4635 "--ustconsumerd32-err-sock");
4637 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4639 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4640 if (lttng_is_setuid_setgid()) {
4641 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4642 "--ustconsumerd32-cmd-sock");
4644 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4646 } else if (string_match(optname
, "no-kernel")) {
4648 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4649 lttng_opt_quiet
= 1;
4650 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4651 /* Verbose level can increase using multiple -v */
4653 /* Value obtained from config file */
4654 lttng_opt_verbose
= config_parse_value(arg
);
4656 /* -v used on command line */
4657 lttng_opt_verbose
++;
4659 /* Clamp value to [0, 3] */
4660 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4661 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4662 } else if (string_match(optname
, "verbose-consumer")) {
4664 opt_verbose_consumer
= config_parse_value(arg
);
4666 opt_verbose_consumer
+= 1;
4668 } else if (string_match(optname
, "consumerd32-path")) {
4669 if (lttng_is_setuid_setgid()) {
4670 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4671 "--consumerd32-path");
4673 if (consumerd32_bin_override
) {
4674 free((void *) consumerd32_bin
);
4676 consumerd32_bin
= strdup(arg
);
4677 if (!consumerd32_bin
) {
4681 consumerd32_bin_override
= 1;
4683 } else if (string_match(optname
, "consumerd32-libdir")) {
4684 if (lttng_is_setuid_setgid()) {
4685 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4686 "--consumerd32-libdir");
4688 if (consumerd32_libdir_override
) {
4689 free((void *) consumerd32_libdir
);
4691 consumerd32_libdir
= strdup(arg
);
4692 if (!consumerd32_libdir
) {
4696 consumerd32_libdir_override
= 1;
4698 } else if (string_match(optname
, "consumerd64-path")) {
4699 if (lttng_is_setuid_setgid()) {
4700 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4701 "--consumerd64-path");
4703 if (consumerd64_bin_override
) {
4704 free((void *) consumerd64_bin
);
4706 consumerd64_bin
= strdup(arg
);
4707 if (!consumerd64_bin
) {
4711 consumerd64_bin_override
= 1;
4713 } else if (string_match(optname
, "consumerd64-libdir")) {
4714 if (lttng_is_setuid_setgid()) {
4715 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4716 "--consumerd64-libdir");
4718 if (consumerd64_libdir_override
) {
4719 free((void *) consumerd64_libdir
);
4721 consumerd64_libdir
= strdup(arg
);
4722 if (!consumerd64_libdir
) {
4726 consumerd64_libdir_override
= 1;
4728 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4729 if (lttng_is_setuid_setgid()) {
4730 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4734 opt_pidfile
= strdup(arg
);
4740 } else if (string_match(optname
, "agent-tcp-port")) {
4741 if (lttng_is_setuid_setgid()) {
4742 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4743 "--agent-tcp-port");
4752 v
= strtoul(arg
, NULL
, 0);
4753 if (errno
!= 0 || !isdigit(arg
[0])) {
4754 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4757 if (v
== 0 || v
>= 65535) {
4758 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4761 agent_tcp_port
= (uint32_t) v
;
4762 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4764 } else if (string_match(optname
, "load") || opt
== 'l') {
4765 if (lttng_is_setuid_setgid()) {
4766 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4769 free(opt_load_session_path
);
4770 opt_load_session_path
= strdup(arg
);
4771 if (!opt_load_session_path
) {
4776 } else if (string_match(optname
, "kmod-probes")) {
4777 if (lttng_is_setuid_setgid()) {
4778 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4781 free(kmod_probes_list
);
4782 kmod_probes_list
= strdup(arg
);
4783 if (!kmod_probes_list
) {
4788 } else if (string_match(optname
, "extra-kmod-probes")) {
4789 if (lttng_is_setuid_setgid()) {
4790 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4791 "--extra-kmod-probes");
4793 free(kmod_extra_probes_list
);
4794 kmod_extra_probes_list
= strdup(arg
);
4795 if (!kmod_extra_probes_list
) {
4800 } else if (string_match(optname
, "config") || opt
== 'f') {
4801 /* This is handled in set_options() thus silent skip. */
4804 /* Unknown option or other error.
4805 * Error is printed by getopt, just return */
4810 if (ret
== -EINVAL
) {
4811 const char *opt_name
= "unknown";
4814 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4816 if (opt
== long_options
[i
].val
) {
4817 opt_name
= long_options
[i
].name
;
4822 WARN("Invalid argument provided for option \"%s\", using default value.",
4830 * config_entry_handler_cb used to handle options read from a config file.
4831 * See config_entry_handler_cb comment in common/config/config.h for the
4832 * return value conventions.
4834 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4838 if (!entry
|| !entry
->name
|| !entry
->value
) {
4843 /* Check if the option is to be ignored */
4844 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4845 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4850 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4853 /* Ignore if not fully matched. */
4854 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4859 * If the option takes no argument on the command line, we have to
4860 * check if the value is "true". We support non-zero numeric values,
4863 if (!long_options
[i
].has_arg
) {
4864 ret
= config_parse_value(entry
->value
);
4867 WARN("Invalid configuration value \"%s\" for option %s",
4868 entry
->value
, entry
->name
);
4870 /* False, skip boolean config option. */
4875 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4879 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4886 * daemon configuration loading and argument parsing
4888 static int set_options(int argc
, char **argv
)
4890 int ret
= 0, c
= 0, option_index
= 0;
4891 int orig_optopt
= optopt
, orig_optind
= optind
;
4893 const char *config_path
= NULL
;
4895 optstring
= utils_generate_optstring(long_options
,
4896 sizeof(long_options
) / sizeof(struct option
));
4902 /* Check for the --config option */
4903 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4904 &option_index
)) != -1) {
4908 } else if (c
!= 'f') {
4909 /* if not equal to --config option. */
4913 if (lttng_is_setuid_setgid()) {
4914 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4917 config_path
= utils_expand_path(optarg
);
4919 ERR("Failed to resolve path: %s", optarg
);
4924 ret
= config_get_section_entries(config_path
, config_section_name
,
4925 config_entry_handler
, NULL
);
4928 ERR("Invalid configuration option at line %i", ret
);
4934 /* Reset getopt's global state */
4935 optopt
= orig_optopt
;
4936 optind
= orig_optind
;
4940 * getopt_long() will not set option_index if it encounters a
4943 c
= getopt_long(argc
, argv
, optstring
, long_options
,
4950 * Pass NULL as the long option name if popt left the index
4953 ret
= set_option(c
, optarg
,
4954 option_index
< 0 ? NULL
:
4955 long_options
[option_index
].name
);
4967 * Creates the two needed socket by the daemon.
4968 * apps_sock - The communication socket for all UST apps.
4969 * client_sock - The communication of the cli tool (lttng).
4971 static int init_daemon_socket(void)
4976 old_umask
= umask(0);
4978 /* Create client tool unix socket */
4979 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4980 if (client_sock
< 0) {
4981 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4986 /* Set the cloexec flag */
4987 ret
= utils_set_fd_cloexec(client_sock
);
4989 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4990 "Continuing but note that the consumer daemon will have a "
4991 "reference to this socket on exec()", client_sock
);
4994 /* File permission MUST be 660 */
4995 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4997 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5002 /* Create the application unix socket */
5003 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5004 if (apps_sock
< 0) {
5005 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5010 /* Set the cloexec flag */
5011 ret
= utils_set_fd_cloexec(apps_sock
);
5013 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5014 "Continuing but note that the consumer daemon will have a "
5015 "reference to this socket on exec()", apps_sock
);
5018 /* File permission MUST be 666 */
5019 ret
= chmod(apps_unix_sock_path
,
5020 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5022 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5027 DBG3("Session daemon client socket %d and application socket %d created",
5028 client_sock
, apps_sock
);
5036 * Check if the global socket is available, and if a daemon is answering at the
5037 * other side. If yes, error is returned.
5039 static int check_existing_daemon(void)
5041 /* Is there anybody out there ? */
5042 if (lttng_session_daemon_alive()) {
5050 * Set the tracing group gid onto the client socket.
5052 * Race window between mkdir and chown is OK because we are going from more
5053 * permissive (root.root) to less permissive (root.tracing).
5055 static int set_permissions(char *rundir
)
5060 gid
= utils_get_group_id(tracing_group_name
);
5062 /* Set lttng run dir */
5063 ret
= chown(rundir
, 0, gid
);
5065 ERR("Unable to set group on %s", rundir
);
5070 * Ensure all applications and tracing group can search the run
5071 * dir. Allow everyone to read the directory, since it does not
5072 * buy us anything to hide its content.
5074 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5076 ERR("Unable to set permissions on %s", rundir
);
5080 /* lttng client socket path */
5081 ret
= chown(client_unix_sock_path
, 0, gid
);
5083 ERR("Unable to set group on %s", client_unix_sock_path
);
5087 /* kconsumer error socket path */
5088 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5090 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5094 /* 64-bit ustconsumer error socket path */
5095 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5097 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5101 /* 32-bit ustconsumer compat32 error socket path */
5102 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5104 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5108 DBG("All permissions are set");
5114 * Create the lttng run directory needed for all global sockets and pipe.
5116 static int create_lttng_rundir(const char *rundir
)
5120 DBG3("Creating LTTng run directory: %s", rundir
);
5122 ret
= mkdir(rundir
, S_IRWXU
);
5124 if (errno
!= EEXIST
) {
5125 ERR("Unable to create %s", rundir
);
5137 * Setup sockets and directory needed by the kconsumerd communication with the
5140 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5144 char path
[PATH_MAX
];
5146 switch (consumer_data
->type
) {
5147 case LTTNG_CONSUMER_KERNEL
:
5148 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5150 case LTTNG_CONSUMER64_UST
:
5151 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5153 case LTTNG_CONSUMER32_UST
:
5154 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5157 ERR("Consumer type unknown");
5162 DBG2("Creating consumer directory: %s", path
);
5164 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5166 if (errno
!= EEXIST
) {
5168 ERR("Failed to create %s", path
);
5174 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5176 ERR("Unable to set group on %s", path
);
5182 /* Create the kconsumerd error unix socket */
5183 consumer_data
->err_sock
=
5184 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5185 if (consumer_data
->err_sock
< 0) {
5186 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5192 * Set the CLOEXEC flag. Return code is useless because either way, the
5195 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5197 PERROR("utils_set_fd_cloexec");
5198 /* continue anyway */
5201 /* File permission MUST be 660 */
5202 ret
= chmod(consumer_data
->err_unix_sock_path
,
5203 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5205 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5215 * Signal handler for the daemon
5217 * Simply stop all worker threads, leaving main() return gracefully after
5218 * joining all threads and calling cleanup().
5220 static void sighandler(int sig
)
5224 DBG("SIGPIPE caught");
5227 DBG("SIGINT caught");
5231 DBG("SIGTERM caught");
5235 CMM_STORE_SHARED(recv_child_signal
, 1);
5243 * Setup signal handler for :
5244 * SIGINT, SIGTERM, SIGPIPE
5246 static int set_signal_handler(void)
5249 struct sigaction sa
;
5252 if ((ret
= sigemptyset(&sigset
)) < 0) {
5253 PERROR("sigemptyset");
5257 sa
.sa_handler
= sighandler
;
5258 sa
.sa_mask
= sigset
;
5260 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5261 PERROR("sigaction");
5265 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5266 PERROR("sigaction");
5270 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5271 PERROR("sigaction");
5275 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5276 PERROR("sigaction");
5280 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5286 * Set open files limit to unlimited. This daemon can open a large number of
5287 * file descriptors in order to consumer multiple kernel traces.
5289 static void set_ulimit(void)
5294 /* The kernel does not allowed an infinite limit for open files */
5295 lim
.rlim_cur
= 65535;
5296 lim
.rlim_max
= 65535;
5298 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5300 PERROR("failed to set open files limit");
5305 * Write pidfile using the rundir and opt_pidfile.
5307 static int write_pidfile(void)
5310 char pidfile_path
[PATH_MAX
];
5315 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
5317 /* Build pidfile path from rundir and opt_pidfile. */
5318 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5319 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5321 PERROR("snprintf pidfile path");
5327 * Create pid file in rundir.
5329 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5335 * Create lockfile using the rundir and return its fd.
5337 static int create_lockfile(void)
5340 char lockfile_path
[PATH_MAX
];
5342 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5347 ret
= utils_create_lock_file(lockfile_path
);
5353 * Write agent TCP port using the rundir.
5355 static int write_agent_port(void)
5358 char path
[PATH_MAX
];
5362 ret
= snprintf(path
, sizeof(path
), "%s/"
5363 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5365 PERROR("snprintf agent port path");
5370 * Create TCP agent port file in rundir.
5372 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5381 int main(int argc
, char **argv
)
5383 int ret
= 0, retval
= 0;
5385 const char *home_path
, *env_app_timeout
;
5387 /* Initialize agent apps ht global variable */
5388 agent_apps_ht_by_sock
= NULL
;
5390 init_kernel_workarounds();
5392 rcu_register_thread();
5394 if (set_signal_handler()) {
5396 goto exit_set_signal_handler
;
5399 setup_consumerd_path();
5401 page_size
= sysconf(_SC_PAGESIZE
);
5402 if (page_size
< 0) {
5403 PERROR("sysconf _SC_PAGESIZE");
5404 page_size
= LONG_MAX
;
5405 WARN("Fallback page size to %ld", page_size
);
5409 * Parse arguments and load the daemon configuration file.
5411 * We have an exit_options exit path to free memory reserved by
5412 * set_options. This is needed because the rest of sessiond_cleanup()
5413 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5414 * depends on set_options.
5417 if (set_options(argc
, argv
)) {
5423 if (opt_daemon
|| opt_background
) {
5426 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5434 * We are in the child. Make sure all other file descriptors are
5435 * closed, in case we are called with more opened file
5436 * descriptors than the standard ones.
5438 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5444 * Starting from here, we can create threads. This needs to be after
5445 * lttng_daemonize due to RCU.
5449 * Initialize the health check subsystem. This call should set the
5450 * appropriate time values.
5452 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5453 if (!health_sessiond
) {
5454 PERROR("health_app_create error");
5456 goto exit_health_sessiond_cleanup
;
5459 if (init_ht_cleanup_quit_pipe()) {
5461 goto exit_ht_cleanup_quit_pipe
;
5464 /* Setup the thread ht_cleanup communication pipe. */
5465 if (utils_create_pipe_cloexec(ht_cleanup_pipe
)) {
5467 goto exit_ht_cleanup_pipe
;
5470 /* Set up max poll set size */
5471 if (lttng_poll_set_max_size()) {
5473 goto exit_set_max_size
;
5476 /* Create thread to clean up RCU hash tables */
5477 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5478 thread_ht_cleanup
, (void *) NULL
);
5481 PERROR("pthread_create ht_cleanup");
5483 goto exit_ht_cleanup
;
5486 /* Create thread quit pipe */
5487 if (init_thread_quit_pipe()) {
5489 goto exit_init_data
;
5492 /* Check if daemon is UID = 0 */
5493 is_root
= !getuid();
5496 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5499 goto exit_init_data
;
5502 /* Create global run dir with root access */
5503 if (create_lttng_rundir(rundir
)) {
5505 goto exit_init_data
;
5508 if (strlen(apps_unix_sock_path
) == 0) {
5509 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5510 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5513 goto exit_init_data
;
5517 if (strlen(client_unix_sock_path
) == 0) {
5518 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5519 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5522 goto exit_init_data
;
5526 /* Set global SHM for ust */
5527 if (strlen(wait_shm_path
) == 0) {
5528 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5529 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5532 goto exit_init_data
;
5536 if (strlen(health_unix_sock_path
) == 0) {
5537 ret
= snprintf(health_unix_sock_path
,
5538 sizeof(health_unix_sock_path
),
5539 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5542 goto exit_init_data
;
5546 /* Setup kernel consumerd path */
5547 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5548 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5551 goto exit_init_data
;
5553 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5554 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5557 goto exit_init_data
;
5560 DBG2("Kernel consumer err path: %s",
5561 kconsumer_data
.err_unix_sock_path
);
5562 DBG2("Kernel consumer cmd path: %s",
5563 kconsumer_data
.cmd_unix_sock_path
);
5565 home_path
= utils_get_home_dir();
5566 if (home_path
== NULL
) {
5567 /* TODO: Add --socket PATH option */
5568 ERR("Can't get HOME directory for sockets creation.");
5570 goto exit_init_data
;
5574 * Create rundir from home path. This will create something like
5577 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5580 goto exit_init_data
;
5583 if (create_lttng_rundir(rundir
)) {
5585 goto exit_init_data
;
5588 if (strlen(apps_unix_sock_path
) == 0) {
5589 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5590 DEFAULT_HOME_APPS_UNIX_SOCK
,
5594 goto exit_init_data
;
5598 /* Set the cli tool unix socket path */
5599 if (strlen(client_unix_sock_path
) == 0) {
5600 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5601 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5605 goto exit_init_data
;
5609 /* Set global SHM for ust */
5610 if (strlen(wait_shm_path
) == 0) {
5611 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5612 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5616 goto exit_init_data
;
5620 /* Set health check Unix path */
5621 if (strlen(health_unix_sock_path
) == 0) {
5622 ret
= snprintf(health_unix_sock_path
,
5623 sizeof(health_unix_sock_path
),
5624 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5628 goto exit_init_data
;
5633 lockfile_fd
= create_lockfile();
5634 if (lockfile_fd
< 0) {
5636 goto exit_init_data
;
5639 /* Set consumer initial state */
5640 kernel_consumerd_state
= CONSUMER_STOPPED
;
5641 ust_consumerd_state
= CONSUMER_STOPPED
;
5643 DBG("Client socket path %s", client_unix_sock_path
);
5644 DBG("Application socket path %s", apps_unix_sock_path
);
5645 DBG("Application wait path %s", wait_shm_path
);
5646 DBG("LTTng run directory path: %s", rundir
);
5648 /* 32 bits consumerd path setup */
5649 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5650 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5652 PERROR("snprintf 32-bit consumer error socket path");
5654 goto exit_init_data
;
5656 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5657 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5659 PERROR("snprintf 32-bit consumer command socket path");
5661 goto exit_init_data
;
5664 DBG2("UST consumer 32 bits err path: %s",
5665 ustconsumer32_data
.err_unix_sock_path
);
5666 DBG2("UST consumer 32 bits cmd path: %s",
5667 ustconsumer32_data
.cmd_unix_sock_path
);
5669 /* 64 bits consumerd path setup */
5670 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5671 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5673 PERROR("snprintf 64-bit consumer error socket path");
5675 goto exit_init_data
;
5677 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5678 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5680 PERROR("snprintf 64-bit consumer command socket path");
5682 goto exit_init_data
;
5685 DBG2("UST consumer 64 bits err path: %s",
5686 ustconsumer64_data
.err_unix_sock_path
);
5687 DBG2("UST consumer 64 bits cmd path: %s",
5688 ustconsumer64_data
.cmd_unix_sock_path
);
5691 * See if daemon already exist.
5693 if (check_existing_daemon()) {
5694 ERR("Already running daemon.\n");
5696 * We do not goto exit because we must not cleanup()
5697 * because a daemon is already running.
5700 goto exit_init_data
;
5704 * Init UST app hash table. Alloc hash table before this point since
5705 * cleanup() can get called after that point.
5707 if (ust_app_ht_alloc()) {
5708 ERR("Failed to allocate UST app hash table");
5710 goto exit_init_data
;
5714 * Initialize agent app hash table. We allocate the hash table here
5715 * since cleanup() can get called after this point.
5717 if (agent_app_ht_alloc()) {
5718 ERR("Failed to allocate Agent app hash table");
5720 goto exit_init_data
;
5724 * These actions must be executed as root. We do that *after* setting up
5725 * the sockets path because we MUST make the check for another daemon using
5726 * those paths *before* trying to set the kernel consumer sockets and init
5730 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5732 goto exit_init_data
;
5735 /* Setup kernel tracer */
5736 if (!opt_no_kernel
) {
5737 init_kernel_tracer();
5738 if (kernel_tracer_fd
>= 0) {
5739 ret
= syscall_init_table();
5741 ERR("Unable to populate syscall table. "
5742 "Syscall tracing won't work "
5743 "for this session daemon.");
5748 /* Set ulimit for open files */
5751 /* init lttng_fd tracking must be done after set_ulimit. */
5754 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5756 goto exit_init_data
;
5759 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5761 goto exit_init_data
;
5764 /* Setup the needed unix socket */
5765 if (init_daemon_socket()) {
5767 goto exit_init_data
;
5770 /* Set credentials to socket */
5771 if (is_root
&& set_permissions(rundir
)) {
5773 goto exit_init_data
;
5776 /* Get parent pid if -S, --sig-parent is specified. */
5777 if (opt_sig_parent
) {
5781 /* Setup the kernel pipe for waking up the kernel thread */
5782 if (is_root
&& !opt_no_kernel
) {
5783 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5785 goto exit_init_data
;
5789 /* Setup the thread apps communication pipe. */
5790 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5792 goto exit_init_data
;
5795 /* Setup the thread apps notify communication pipe. */
5796 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5798 goto exit_init_data
;
5801 /* Initialize global buffer per UID and PID registry. */
5802 buffer_reg_init_uid_registry();
5803 buffer_reg_init_pid_registry();
5805 /* Init UST command queue. */
5806 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5809 * Get session list pointer. This pointer MUST NOT be free'd. This list
5810 * is statically declared in session.c
5812 session_list_ptr
= session_get_list();
5816 /* Check for the application socket timeout env variable. */
5817 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5818 if (env_app_timeout
) {
5819 app_socket_timeout
= atoi(env_app_timeout
);
5821 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5824 ret
= write_pidfile();
5826 ERR("Error in write_pidfile");
5828 goto exit_init_data
;
5830 ret
= write_agent_port();
5832 ERR("Error in write_agent_port");
5834 goto exit_init_data
;
5837 /* Initialize communication library */
5839 /* Initialize TCP timeout values */
5840 lttcomm_inet_init();
5842 if (load_session_init_data(&load_info
) < 0) {
5844 goto exit_init_data
;
5846 load_info
->path
= opt_load_session_path
;
5848 /* Create health-check thread */
5849 ret
= pthread_create(&health_thread
, NULL
,
5850 thread_manage_health
, (void *) NULL
);
5853 PERROR("pthread_create health");
5858 /* Create thread to manage the client socket */
5859 ret
= pthread_create(&client_thread
, NULL
,
5860 thread_manage_clients
, (void *) NULL
);
5863 PERROR("pthread_create clients");
5868 /* Create thread to dispatch registration */
5869 ret
= pthread_create(&dispatch_thread
, NULL
,
5870 thread_dispatch_ust_registration
, (void *) NULL
);
5873 PERROR("pthread_create dispatch");
5878 /* Create thread to manage application registration. */
5879 ret
= pthread_create(®_apps_thread
, NULL
,
5880 thread_registration_apps
, (void *) NULL
);
5883 PERROR("pthread_create registration");
5888 /* Create thread to manage application socket */
5889 ret
= pthread_create(&apps_thread
, NULL
,
5890 thread_manage_apps
, (void *) NULL
);
5893 PERROR("pthread_create apps");
5898 /* Create thread to manage application notify socket */
5899 ret
= pthread_create(&apps_notify_thread
, NULL
,
5900 ust_thread_manage_notify
, (void *) NULL
);
5903 PERROR("pthread_create notify");
5905 goto exit_apps_notify
;
5908 /* Create agent registration thread. */
5909 ret
= pthread_create(&agent_reg_thread
, NULL
,
5910 agent_thread_manage_registration
, (void *) NULL
);
5913 PERROR("pthread_create agent");
5915 goto exit_agent_reg
;
5918 /* Don't start this thread if kernel tracing is not requested nor root */
5919 if (is_root
&& !opt_no_kernel
) {
5920 /* Create kernel thread to manage kernel event */
5921 ret
= pthread_create(&kernel_thread
, NULL
,
5922 thread_manage_kernel
, (void *) NULL
);
5925 PERROR("pthread_create kernel");
5931 /* Create session loading thread. */
5932 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
5936 PERROR("pthread_create load_session_thread");
5938 goto exit_load_session
;
5942 * This is where we start awaiting program completion (e.g. through
5943 * signal that asks threads to teardown).
5946 ret
= pthread_join(load_session_thread
, &status
);
5949 PERROR("pthread_join load_session_thread");
5954 if (is_root
&& !opt_no_kernel
) {
5955 ret
= pthread_join(kernel_thread
, &status
);
5958 PERROR("pthread_join");
5964 ret
= pthread_join(agent_reg_thread
, &status
);
5967 PERROR("pthread_join agent");
5972 ret
= pthread_join(apps_notify_thread
, &status
);
5975 PERROR("pthread_join apps notify");
5980 ret
= pthread_join(apps_thread
, &status
);
5983 PERROR("pthread_join apps");
5988 ret
= pthread_join(reg_apps_thread
, &status
);
5991 PERROR("pthread_join");
5997 * Join dispatch thread after joining reg_apps_thread to ensure
5998 * we don't leak applications in the queue.
6000 ret
= pthread_join(dispatch_thread
, &status
);
6003 PERROR("pthread_join");
6008 ret
= pthread_join(client_thread
, &status
);
6011 PERROR("pthread_join");
6016 ret
= pthread_join(health_thread
, &status
);
6019 PERROR("pthread_join health thread");
6026 * sessiond_cleanup() is called when no other thread is running, except
6027 * the ht_cleanup thread, which is needed to destroy the hash tables.
6029 rcu_thread_online();
6031 rcu_thread_offline();
6032 rcu_unregister_thread();
6034 ret
= notify_thread_pipe(ht_cleanup_quit_pipe
[1]);
6036 ERR("write error on ht_cleanup quit pipe");
6040 ret
= pthread_join(ht_cleanup_thread
, &status
);
6043 PERROR("pthread_join ht cleanup thread");
6049 utils_close_pipe(ht_cleanup_pipe
);
6050 exit_ht_cleanup_pipe
:
6053 * Close the ht_cleanup quit pipe.
6055 utils_close_pipe(ht_cleanup_quit_pipe
);
6056 exit_ht_cleanup_quit_pipe
:
6058 health_app_destroy(health_sessiond
);
6059 exit_health_sessiond_cleanup
:
6062 sessiond_cleanup_options();
6064 exit_set_signal_handler
:
6065 /* Ensure all prior call_rcu are done. */