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
);
800 run_as_destroy_worker();
803 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
804 "Matthew, BEET driven development works!%c[%dm",
805 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
810 * Send data on a unix socket using the liblttsessiondcomm API.
812 * Return lttcomm error code.
814 static int send_unix_sock(int sock
, void *buf
, size_t len
)
816 /* Check valid length */
821 return lttcomm_send_unix_sock(sock
, buf
, len
);
825 * Free memory of a command context structure.
827 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
829 DBG("Clean command context structure");
831 if ((*cmd_ctx
)->llm
) {
832 free((*cmd_ctx
)->llm
);
834 if ((*cmd_ctx
)->lsm
) {
835 free((*cmd_ctx
)->lsm
);
843 * Notify UST applications using the shm mmap futex.
845 static int notify_ust_apps(int active
)
849 DBG("Notifying applications of session daemon state: %d", active
);
851 /* See shm.c for this call implying mmap, shm and futex calls */
852 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
853 if (wait_shm_mmap
== NULL
) {
857 /* Wake waiting process */
858 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
860 /* Apps notified successfully */
868 * Setup the outgoing data buffer for the response (llm) by allocating the
869 * right amount of memory and copying the original information from the lsm
872 * Return total size of the buffer pointed by buf.
874 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
880 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
881 if (cmd_ctx
->llm
== NULL
) {
887 /* Copy common data */
888 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
889 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
891 cmd_ctx
->llm
->data_size
= size
;
892 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
901 * Update the kernel poll set of all channel fd available over all tracing
902 * session. Add the wakeup pipe at the end of the set.
904 static int update_kernel_poll(struct lttng_poll_event
*events
)
907 struct ltt_session
*session
;
908 struct ltt_kernel_channel
*channel
;
910 DBG("Updating kernel poll set");
913 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
914 session_lock(session
);
915 if (session
->kernel_session
== NULL
) {
916 session_unlock(session
);
920 cds_list_for_each_entry(channel
,
921 &session
->kernel_session
->channel_list
.head
, list
) {
922 /* Add channel fd to the kernel poll set */
923 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
925 session_unlock(session
);
928 DBG("Channel fd %d added to kernel set", channel
->fd
);
930 session_unlock(session
);
932 session_unlock_list();
937 session_unlock_list();
942 * Find the channel fd from 'fd' over all tracing session. When found, check
943 * for new channel stream and send those stream fds to the kernel consumer.
945 * Useful for CPU hotplug feature.
947 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
950 struct ltt_session
*session
;
951 struct ltt_kernel_session
*ksess
;
952 struct ltt_kernel_channel
*channel
;
954 DBG("Updating kernel streams for channel fd %d", fd
);
957 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
958 session_lock(session
);
959 if (session
->kernel_session
== NULL
) {
960 session_unlock(session
);
963 ksess
= session
->kernel_session
;
965 cds_list_for_each_entry(channel
,
966 &ksess
->channel_list
.head
, list
) {
967 struct lttng_ht_iter iter
;
968 struct consumer_socket
*socket
;
970 if (channel
->fd
!= fd
) {
973 DBG("Channel found, updating kernel streams");
974 ret
= kernel_open_channel_stream(channel
);
978 /* Update the stream global counter */
979 ksess
->stream_count_global
+= ret
;
982 * Have we already sent fds to the consumer? If yes, it
983 * means that tracing is started so it is safe to send
984 * our updated stream fds.
986 if (ksess
->consumer_fds_sent
!= 1
987 || ksess
->consumer
== NULL
) {
993 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
994 &iter
.iter
, socket
, node
.node
) {
995 pthread_mutex_lock(socket
->lock
);
996 ret
= kernel_consumer_send_channel_stream(socket
,
998 session
->output_traces
? 1 : 0);
999 pthread_mutex_unlock(socket
->lock
);
1007 session_unlock(session
);
1009 session_unlock_list();
1013 session_unlock(session
);
1014 session_unlock_list();
1019 * For each tracing session, update newly registered apps. The session list
1020 * lock MUST be acquired before calling this.
1022 static void update_ust_app(int app_sock
)
1024 struct ltt_session
*sess
, *stmp
;
1026 /* Consumer is in an ERROR state. Stop any application update. */
1027 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
1028 /* Stop the update process since the consumer is dead. */
1032 /* For all tracing session(s) */
1033 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
1034 struct ust_app
*app
;
1037 if (!sess
->ust_session
) {
1038 goto unlock_session
;
1042 assert(app_sock
>= 0);
1043 app
= ust_app_find_by_sock(app_sock
);
1046 * Application can be unregistered before so
1047 * this is possible hence simply stopping the
1050 DBG3("UST app update failed to find app sock %d",
1054 ust_app_global_update(sess
->ust_session
, app
);
1058 session_unlock(sess
);
1063 * This thread manage event coming from the kernel.
1065 * Features supported in this thread:
1068 static void *thread_manage_kernel(void *data
)
1070 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1071 uint32_t revents
, nb_fd
;
1073 struct lttng_poll_event events
;
1075 DBG("[thread] Thread manage kernel started");
1077 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1080 * This first step of the while is to clean this structure which could free
1081 * non NULL pointers so initialize it before the loop.
1083 lttng_poll_init(&events
);
1085 if (testpoint(sessiond_thread_manage_kernel
)) {
1086 goto error_testpoint
;
1089 health_code_update();
1091 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1092 goto error_testpoint
;
1096 health_code_update();
1098 if (update_poll_flag
== 1) {
1099 /* Clean events object. We are about to populate it again. */
1100 lttng_poll_clean(&events
);
1102 ret
= sessiond_set_thread_pollset(&events
, 2);
1104 goto error_poll_create
;
1107 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1112 /* This will add the available kernel channel if any. */
1113 ret
= update_kernel_poll(&events
);
1117 update_poll_flag
= 0;
1120 DBG("Thread kernel polling");
1122 /* Poll infinite value of time */
1124 health_poll_entry();
1125 ret
= lttng_poll_wait(&events
, -1);
1126 DBG("Thread kernel return from poll on %d fds",
1127 LTTNG_POLL_GETNB(&events
));
1131 * Restart interrupted system call.
1133 if (errno
== EINTR
) {
1137 } else if (ret
== 0) {
1138 /* Should not happen since timeout is infinite */
1139 ERR("Return value of poll is 0 with an infinite timeout.\n"
1140 "This should not have happened! Continuing...");
1146 for (i
= 0; i
< nb_fd
; i
++) {
1147 /* Fetch once the poll data */
1148 revents
= LTTNG_POLL_GETEV(&events
, i
);
1149 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1151 health_code_update();
1154 /* No activity for this FD (poll implementation). */
1158 /* Thread quit pipe has been closed. Killing thread. */
1159 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1165 /* Check for data on kernel pipe */
1166 if (revents
& LPOLLIN
) {
1167 if (pollfd
== kernel_poll_pipe
[0]) {
1168 (void) lttng_read(kernel_poll_pipe
[0],
1171 * Ret value is useless here, if this pipe gets any actions an
1172 * update is required anyway.
1174 update_poll_flag
= 1;
1178 * New CPU detected by the kernel. Adding kernel stream to
1179 * kernel session and updating the kernel consumer
1181 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1187 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1188 update_poll_flag
= 1;
1191 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1199 lttng_poll_clean(&events
);
1202 utils_close_pipe(kernel_poll_pipe
);
1203 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1206 ERR("Health error occurred in %s", __func__
);
1207 WARN("Kernel thread died unexpectedly. "
1208 "Kernel tracing can continue but CPU hotplug is disabled.");
1210 health_unregister(health_sessiond
);
1211 DBG("Kernel thread dying");
1216 * Signal pthread condition of the consumer data that the thread.
1218 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1220 pthread_mutex_lock(&data
->cond_mutex
);
1223 * The state is set before signaling. It can be any value, it's the waiter
1224 * job to correctly interpret this condition variable associated to the
1225 * consumer pthread_cond.
1227 * A value of 0 means that the corresponding thread of the consumer data
1228 * was not started. 1 indicates that the thread has started and is ready
1229 * for action. A negative value means that there was an error during the
1232 data
->consumer_thread_is_ready
= state
;
1233 (void) pthread_cond_signal(&data
->cond
);
1235 pthread_mutex_unlock(&data
->cond_mutex
);
1239 * This thread manage the consumer error sent back to the session daemon.
1241 static void *thread_manage_consumer(void *data
)
1243 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1244 uint32_t revents
, nb_fd
;
1245 enum lttcomm_return_code code
;
1246 struct lttng_poll_event events
;
1247 struct consumer_data
*consumer_data
= data
;
1249 DBG("[thread] Manage consumer started");
1251 rcu_register_thread();
1252 rcu_thread_online();
1254 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1256 health_code_update();
1259 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1260 * metadata_sock. Nothing more will be added to this poll set.
1262 ret
= sessiond_set_thread_pollset(&events
, 3);
1268 * The error socket here is already in a listening state which was done
1269 * just before spawning this thread to avoid a race between the consumer
1270 * daemon exec trying to connect and the listen() call.
1272 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1277 health_code_update();
1279 /* Infinite blocking call, waiting for transmission */
1281 health_poll_entry();
1283 if (testpoint(sessiond_thread_manage_consumer
)) {
1287 ret
= lttng_poll_wait(&events
, -1);
1291 * Restart interrupted system call.
1293 if (errno
== EINTR
) {
1301 for (i
= 0; i
< nb_fd
; i
++) {
1302 /* Fetch once the poll data */
1303 revents
= LTTNG_POLL_GETEV(&events
, i
);
1304 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1306 health_code_update();
1309 /* No activity for this FD (poll implementation). */
1313 /* Thread quit pipe has been closed. Killing thread. */
1314 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1320 /* Event on the registration socket */
1321 if (pollfd
== consumer_data
->err_sock
) {
1322 if (revents
& LPOLLIN
) {
1324 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1325 ERR("consumer err socket poll error");
1328 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1334 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1340 * Set the CLOEXEC flag. Return code is useless because either way, the
1343 (void) utils_set_fd_cloexec(sock
);
1345 health_code_update();
1347 DBG2("Receiving code from consumer err_sock");
1349 /* Getting status code from kconsumerd */
1350 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1351 sizeof(enum lttcomm_return_code
));
1356 health_code_update();
1357 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1358 /* Connect both socket, command and metadata. */
1359 consumer_data
->cmd_sock
=
1360 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1361 consumer_data
->metadata_fd
=
1362 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1363 if (consumer_data
->cmd_sock
< 0
1364 || consumer_data
->metadata_fd
< 0) {
1365 PERROR("consumer connect cmd socket");
1366 /* On error, signal condition and quit. */
1367 signal_consumer_condition(consumer_data
, -1);
1370 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1371 /* Create metadata socket lock. */
1372 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1373 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1374 PERROR("zmalloc pthread mutex");
1378 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1380 signal_consumer_condition(consumer_data
, 1);
1381 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1382 DBG("Consumer metadata socket ready (fd: %d)",
1383 consumer_data
->metadata_fd
);
1385 ERR("consumer error when waiting for SOCK_READY : %s",
1386 lttcomm_get_readable_code(-code
));
1390 /* Remove the consumerd error sock since we've established a connexion */
1391 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1396 /* Add new accepted error socket. */
1397 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1402 /* Add metadata socket that is successfully connected. */
1403 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1404 LPOLLIN
| LPOLLRDHUP
);
1409 health_code_update();
1411 /* Infinite blocking call, waiting for transmission */
1414 health_code_update();
1416 /* Exit the thread because the thread quit pipe has been triggered. */
1418 /* Not a health error. */
1423 health_poll_entry();
1424 ret
= lttng_poll_wait(&events
, -1);
1428 * Restart interrupted system call.
1430 if (errno
== EINTR
) {
1438 for (i
= 0; i
< nb_fd
; i
++) {
1439 /* Fetch once the poll data */
1440 revents
= LTTNG_POLL_GETEV(&events
, i
);
1441 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1443 health_code_update();
1446 /* No activity for this FD (poll implementation). */
1451 * Thread quit pipe has been triggered, flag that we should stop
1452 * but continue the current loop to handle potential data from
1455 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1457 if (pollfd
== sock
) {
1458 /* Event on the consumerd socket */
1459 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1460 && !(revents
& LPOLLIN
)) {
1461 ERR("consumer err socket second poll error");
1464 health_code_update();
1465 /* Wait for any kconsumerd error */
1466 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1467 sizeof(enum lttcomm_return_code
));
1469 ERR("consumer closed the command socket");
1473 ERR("consumer return code : %s",
1474 lttcomm_get_readable_code(-code
));
1477 } else if (pollfd
== consumer_data
->metadata_fd
) {
1478 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1479 && !(revents
& LPOLLIN
)) {
1480 ERR("consumer err metadata socket second poll error");
1483 /* UST metadata requests */
1484 ret
= ust_consumer_metadata_request(
1485 &consumer_data
->metadata_sock
);
1487 ERR("Handling metadata request");
1491 /* No need for an else branch all FDs are tested prior. */
1493 health_code_update();
1499 * We lock here because we are about to close the sockets and some other
1500 * thread might be using them so get exclusive access which will abort all
1501 * other consumer command by other threads.
1503 pthread_mutex_lock(&consumer_data
->lock
);
1505 /* Immediately set the consumerd state to stopped */
1506 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1507 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1508 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1509 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1510 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1512 /* Code flow error... */
1516 if (consumer_data
->err_sock
>= 0) {
1517 ret
= close(consumer_data
->err_sock
);
1521 consumer_data
->err_sock
= -1;
1523 if (consumer_data
->cmd_sock
>= 0) {
1524 ret
= close(consumer_data
->cmd_sock
);
1528 consumer_data
->cmd_sock
= -1;
1530 if (consumer_data
->metadata_sock
.fd_ptr
&&
1531 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1532 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1544 unlink(consumer_data
->err_unix_sock_path
);
1545 unlink(consumer_data
->cmd_unix_sock_path
);
1546 pthread_mutex_unlock(&consumer_data
->lock
);
1548 /* Cleanup metadata socket mutex. */
1549 if (consumer_data
->metadata_sock
.lock
) {
1550 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1551 free(consumer_data
->metadata_sock
.lock
);
1553 lttng_poll_clean(&events
);
1557 ERR("Health error occurred in %s", __func__
);
1559 health_unregister(health_sessiond
);
1560 DBG("consumer thread cleanup completed");
1562 rcu_thread_offline();
1563 rcu_unregister_thread();
1569 * This thread manage application communication.
1571 static void *thread_manage_apps(void *data
)
1573 int i
, ret
, pollfd
, err
= -1;
1575 uint32_t revents
, nb_fd
;
1576 struct lttng_poll_event events
;
1578 DBG("[thread] Manage application started");
1580 rcu_register_thread();
1581 rcu_thread_online();
1583 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1585 if (testpoint(sessiond_thread_manage_apps
)) {
1586 goto error_testpoint
;
1589 health_code_update();
1591 ret
= sessiond_set_thread_pollset(&events
, 2);
1593 goto error_poll_create
;
1596 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1601 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1605 health_code_update();
1608 DBG("Apps thread polling");
1610 /* Inifinite blocking call, waiting for transmission */
1612 health_poll_entry();
1613 ret
= lttng_poll_wait(&events
, -1);
1614 DBG("Apps thread return from poll on %d fds",
1615 LTTNG_POLL_GETNB(&events
));
1619 * Restart interrupted system call.
1621 if (errno
== EINTR
) {
1629 for (i
= 0; i
< nb_fd
; i
++) {
1630 /* Fetch once the poll data */
1631 revents
= LTTNG_POLL_GETEV(&events
, i
);
1632 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1634 health_code_update();
1637 /* No activity for this FD (poll implementation). */
1641 /* Thread quit pipe has been closed. Killing thread. */
1642 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1648 /* Inspect the apps cmd pipe */
1649 if (pollfd
== apps_cmd_pipe
[0]) {
1650 if (revents
& LPOLLIN
) {
1654 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1655 if (size_ret
< sizeof(sock
)) {
1656 PERROR("read apps cmd pipe");
1660 health_code_update();
1663 * Since this is a command socket (write then read),
1664 * we only monitor the error events of the socket.
1666 ret
= lttng_poll_add(&events
, sock
,
1667 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1672 DBG("Apps with sock %d added to poll set", sock
);
1673 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1674 ERR("Apps command pipe error");
1677 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1682 * At this point, we know that a registered application made
1683 * the event at poll_wait.
1685 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1686 /* Removing from the poll set */
1687 ret
= lttng_poll_del(&events
, pollfd
);
1692 /* Socket closed on remote end. */
1693 ust_app_unregister(pollfd
);
1695 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1700 health_code_update();
1706 lttng_poll_clean(&events
);
1709 utils_close_pipe(apps_cmd_pipe
);
1710 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1713 * We don't clean the UST app hash table here since already registered
1714 * applications can still be controlled so let them be until the session
1715 * daemon dies or the applications stop.
1720 ERR("Health error occurred in %s", __func__
);
1722 health_unregister(health_sessiond
);
1723 DBG("Application communication apps thread cleanup complete");
1724 rcu_thread_offline();
1725 rcu_unregister_thread();
1730 * Send a socket to a thread This is called from the dispatch UST registration
1731 * thread once all sockets are set for the application.
1733 * The sock value can be invalid, we don't really care, the thread will handle
1734 * it and make the necessary cleanup if so.
1736 * On success, return 0 else a negative value being the errno message of the
1739 static int send_socket_to_thread(int fd
, int sock
)
1744 * It's possible that the FD is set as invalid with -1 concurrently just
1745 * before calling this function being a shutdown state of the thread.
1752 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1753 if (ret
< sizeof(sock
)) {
1754 PERROR("write apps pipe %d", fd
);
1761 /* All good. Don't send back the write positive ret value. */
1768 * Sanitize the wait queue of the dispatch registration thread meaning removing
1769 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1770 * notify socket is never received.
1772 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1774 int ret
, nb_fd
= 0, i
;
1775 unsigned int fd_added
= 0;
1776 struct lttng_poll_event events
;
1777 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1781 lttng_poll_init(&events
);
1783 /* Just skip everything for an empty queue. */
1784 if (!wait_queue
->count
) {
1788 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1793 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1794 &wait_queue
->head
, head
) {
1795 assert(wait_node
->app
);
1796 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1797 LPOLLHUP
| LPOLLERR
);
1810 * Poll but don't block so we can quickly identify the faulty events and
1811 * clean them afterwards from the wait queue.
1813 ret
= lttng_poll_wait(&events
, 0);
1819 for (i
= 0; i
< nb_fd
; i
++) {
1820 /* Get faulty FD. */
1821 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1822 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1825 /* No activity for this FD (poll implementation). */
1829 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1830 &wait_queue
->head
, head
) {
1831 if (pollfd
== wait_node
->app
->sock
&&
1832 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1833 cds_list_del(&wait_node
->head
);
1834 wait_queue
->count
--;
1835 ust_app_destroy(wait_node
->app
);
1839 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1846 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1850 lttng_poll_clean(&events
);
1854 lttng_poll_clean(&events
);
1856 ERR("Unable to sanitize wait queue");
1861 * Dispatch request from the registration threads to the application
1862 * communication thread.
1864 static void *thread_dispatch_ust_registration(void *data
)
1867 struct cds_wfcq_node
*node
;
1868 struct ust_command
*ust_cmd
= NULL
;
1869 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1870 struct ust_reg_wait_queue wait_queue
= {
1874 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1876 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1877 goto error_testpoint
;
1880 health_code_update();
1882 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1884 DBG("[thread] Dispatch UST command started");
1886 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1887 health_code_update();
1889 /* Atomically prepare the queue futex */
1890 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1893 struct ust_app
*app
= NULL
;
1897 * Make sure we don't have node(s) that have hung up before receiving
1898 * the notify socket. This is to clean the list in order to avoid
1899 * memory leaks from notify socket that are never seen.
1901 sanitize_wait_queue(&wait_queue
);
1903 health_code_update();
1904 /* Dequeue command for registration */
1905 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1907 DBG("Woken up but nothing in the UST command queue");
1908 /* Continue thread execution */
1912 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1914 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1915 " gid:%d sock:%d name:%s (version %d.%d)",
1916 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1917 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1918 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1919 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1921 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1922 wait_node
= zmalloc(sizeof(*wait_node
));
1924 PERROR("zmalloc wait_node dispatch");
1925 ret
= close(ust_cmd
->sock
);
1927 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1929 lttng_fd_put(LTTNG_FD_APPS
, 1);
1933 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1935 /* Create application object if socket is CMD. */
1936 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1938 if (!wait_node
->app
) {
1939 ret
= close(ust_cmd
->sock
);
1941 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1943 lttng_fd_put(LTTNG_FD_APPS
, 1);
1949 * Add application to the wait queue so we can set the notify
1950 * socket before putting this object in the global ht.
1952 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1957 * We have to continue here since we don't have the notify
1958 * socket and the application MUST be added to the hash table
1959 * only at that moment.
1964 * Look for the application in the local wait queue and set the
1965 * notify socket if found.
1967 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1968 &wait_queue
.head
, head
) {
1969 health_code_update();
1970 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1971 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1972 cds_list_del(&wait_node
->head
);
1974 app
= wait_node
->app
;
1976 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1982 * With no application at this stage the received socket is
1983 * basically useless so close it before we free the cmd data
1984 * structure for good.
1987 ret
= close(ust_cmd
->sock
);
1989 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1991 lttng_fd_put(LTTNG_FD_APPS
, 1);
1998 * @session_lock_list
2000 * Lock the global session list so from the register up to the
2001 * registration done message, no thread can see the application
2002 * and change its state.
2004 session_lock_list();
2008 * Add application to the global hash table. This needs to be
2009 * done before the update to the UST registry can locate the
2014 /* Set app version. This call will print an error if needed. */
2015 (void) ust_app_version(app
);
2017 /* Send notify socket through the notify pipe. */
2018 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
2022 session_unlock_list();
2024 * No notify thread, stop the UST tracing. However, this is
2025 * not an internal error of the this thread thus setting
2026 * the health error code to a normal exit.
2033 * Update newly registered application with the tracing
2034 * registry info already enabled information.
2036 update_ust_app(app
->sock
);
2039 * Don't care about return value. Let the manage apps threads
2040 * handle app unregistration upon socket close.
2042 (void) ust_app_register_done(app
);
2045 * Even if the application socket has been closed, send the app
2046 * to the thread and unregistration will take place at that
2049 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2052 session_unlock_list();
2054 * No apps. thread, stop the UST tracing. However, this is
2055 * not an internal error of the this thread thus setting
2056 * the health error code to a normal exit.
2063 session_unlock_list();
2065 } while (node
!= NULL
);
2067 health_poll_entry();
2068 /* Futex wait on queue. Blocking call on futex() */
2069 futex_nto1_wait(&ust_cmd_queue
.futex
);
2072 /* Normal exit, no error */
2076 /* Clean up wait queue. */
2077 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2078 &wait_queue
.head
, head
) {
2079 cds_list_del(&wait_node
->head
);
2084 /* Empty command queue. */
2086 /* Dequeue command for registration */
2087 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2091 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2092 ret
= close(ust_cmd
->sock
);
2094 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2096 lttng_fd_put(LTTNG_FD_APPS
, 1);
2101 DBG("Dispatch thread dying");
2104 ERR("Health error occurred in %s", __func__
);
2106 health_unregister(health_sessiond
);
2111 * This thread manage application registration.
2113 static void *thread_registration_apps(void *data
)
2115 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2116 uint32_t revents
, nb_fd
;
2117 struct lttng_poll_event events
;
2119 * Get allocated in this thread, enqueued to a global queue, dequeued and
2120 * freed in the manage apps thread.
2122 struct ust_command
*ust_cmd
= NULL
;
2124 DBG("[thread] Manage application registration started");
2126 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2128 if (testpoint(sessiond_thread_registration_apps
)) {
2129 goto error_testpoint
;
2132 ret
= lttcomm_listen_unix_sock(apps_sock
);
2138 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2139 * more will be added to this poll set.
2141 ret
= sessiond_set_thread_pollset(&events
, 2);
2143 goto error_create_poll
;
2146 /* Add the application registration socket */
2147 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2149 goto error_poll_add
;
2152 /* Notify all applications to register */
2153 ret
= notify_ust_apps(1);
2155 ERR("Failed to notify applications or create the wait shared memory.\n"
2156 "Execution continues but there might be problem for already\n"
2157 "running applications that wishes to register.");
2161 DBG("Accepting application registration");
2163 /* Inifinite blocking call, waiting for transmission */
2165 health_poll_entry();
2166 ret
= lttng_poll_wait(&events
, -1);
2170 * Restart interrupted system call.
2172 if (errno
== EINTR
) {
2180 for (i
= 0; i
< nb_fd
; i
++) {
2181 health_code_update();
2183 /* Fetch once the poll data */
2184 revents
= LTTNG_POLL_GETEV(&events
, i
);
2185 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2188 /* No activity for this FD (poll implementation). */
2192 /* Thread quit pipe has been closed. Killing thread. */
2193 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2199 /* Event on the registration socket */
2200 if (pollfd
== apps_sock
) {
2201 if (revents
& LPOLLIN
) {
2202 sock
= lttcomm_accept_unix_sock(apps_sock
);
2208 * Set socket timeout for both receiving and ending.
2209 * app_socket_timeout is in seconds, whereas
2210 * lttcomm_setsockopt_rcv_timeout and
2211 * lttcomm_setsockopt_snd_timeout expect msec as
2214 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2215 app_socket_timeout
* 1000);
2216 (void) lttcomm_setsockopt_snd_timeout(sock
,
2217 app_socket_timeout
* 1000);
2220 * Set the CLOEXEC flag. Return code is useless because
2221 * either way, the show must go on.
2223 (void) utils_set_fd_cloexec(sock
);
2225 /* Create UST registration command for enqueuing */
2226 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2227 if (ust_cmd
== NULL
) {
2228 PERROR("ust command zmalloc");
2237 * Using message-based transmissions to ensure we don't
2238 * have to deal with partially received messages.
2240 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2242 ERR("Exhausted file descriptors allowed for applications.");
2252 health_code_update();
2253 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2256 /* Close socket of the application. */
2261 lttng_fd_put(LTTNG_FD_APPS
, 1);
2265 health_code_update();
2267 ust_cmd
->sock
= sock
;
2270 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2271 " gid:%d sock:%d name:%s (version %d.%d)",
2272 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2273 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2274 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2275 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2278 * Lock free enqueue the registration request. The red pill
2279 * has been taken! This apps will be part of the *system*.
2281 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2284 * Wake the registration queue futex. Implicit memory
2285 * barrier with the exchange in cds_wfcq_enqueue.
2287 futex_nto1_wake(&ust_cmd_queue
.futex
);
2288 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2289 ERR("Register apps socket poll error");
2292 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2301 /* Notify that the registration thread is gone */
2304 if (apps_sock
>= 0) {
2305 ret
= close(apps_sock
);
2315 lttng_fd_put(LTTNG_FD_APPS
, 1);
2317 unlink(apps_unix_sock_path
);
2320 lttng_poll_clean(&events
);
2324 DBG("UST Registration thread cleanup complete");
2327 ERR("Health error occurred in %s", __func__
);
2329 health_unregister(health_sessiond
);
2335 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2336 * exec or it will fails.
2338 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2341 struct timespec timeout
;
2343 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2344 consumer_data
->consumer_thread_is_ready
= 0;
2346 /* Setup pthread condition */
2347 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2350 PERROR("pthread_condattr_init consumer data");
2355 * Set the monotonic clock in order to make sure we DO NOT jump in time
2356 * between the clock_gettime() call and the timedwait call. See bug #324
2357 * for a more details and how we noticed it.
2359 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2362 PERROR("pthread_condattr_setclock consumer data");
2366 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2369 PERROR("pthread_cond_init consumer data");
2373 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2377 PERROR("pthread_create consumer");
2382 /* We are about to wait on a pthread condition */
2383 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2385 /* Get time for sem_timedwait absolute timeout */
2386 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2388 * Set the timeout for the condition timed wait even if the clock gettime
2389 * call fails since we might loop on that call and we want to avoid to
2390 * increment the timeout too many times.
2392 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2395 * The following loop COULD be skipped in some conditions so this is why we
2396 * set ret to 0 in order to make sure at least one round of the loop is
2402 * Loop until the condition is reached or when a timeout is reached. Note
2403 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2404 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2405 * possible. This loop does not take any chances and works with both of
2408 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2409 if (clock_ret
< 0) {
2410 PERROR("clock_gettime spawn consumer");
2411 /* Infinite wait for the consumerd thread to be ready */
2412 ret
= pthread_cond_wait(&consumer_data
->cond
,
2413 &consumer_data
->cond_mutex
);
2415 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2416 &consumer_data
->cond_mutex
, &timeout
);
2420 /* Release the pthread condition */
2421 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2425 if (ret
== ETIMEDOUT
) {
2429 * Call has timed out so we kill the kconsumerd_thread and return
2432 ERR("Condition timed out. The consumer thread was never ready."
2434 pth_ret
= pthread_cancel(consumer_data
->thread
);
2436 PERROR("pthread_cancel consumer thread");
2439 PERROR("pthread_cond_wait failed consumer thread");
2441 /* Caller is expecting a negative value on failure. */
2446 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2447 if (consumer_data
->pid
== 0) {
2448 ERR("Consumerd did not start");
2449 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2452 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2461 * Join consumer thread
2463 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2467 /* Consumer pid must be a real one. */
2468 if (consumer_data
->pid
> 0) {
2470 ret
= kill(consumer_data
->pid
, SIGTERM
);
2472 PERROR("Error killing consumer daemon");
2475 return pthread_join(consumer_data
->thread
, &status
);
2482 * Fork and exec a consumer daemon (consumerd).
2484 * Return pid if successful else -1.
2486 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2490 const char *consumer_to_use
;
2491 const char *verbosity
;
2494 DBG("Spawning consumerd");
2501 if (opt_verbose_consumer
) {
2502 verbosity
= "--verbose";
2503 } else if (lttng_opt_quiet
) {
2504 verbosity
= "--quiet";
2509 switch (consumer_data
->type
) {
2510 case LTTNG_CONSUMER_KERNEL
:
2512 * Find out which consumerd to execute. We will first try the
2513 * 64-bit path, then the sessiond's installation directory, and
2514 * fallback on the 32-bit one,
2516 DBG3("Looking for a kernel consumer at these locations:");
2517 DBG3(" 1) %s", consumerd64_bin
);
2518 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2519 DBG3(" 3) %s", consumerd32_bin
);
2520 if (stat(consumerd64_bin
, &st
) == 0) {
2521 DBG3("Found location #1");
2522 consumer_to_use
= consumerd64_bin
;
2523 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2524 DBG3("Found location #2");
2525 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2526 } else if (stat(consumerd32_bin
, &st
) == 0) {
2527 DBG3("Found location #3");
2528 consumer_to_use
= consumerd32_bin
;
2530 DBG("Could not find any valid consumerd executable");
2534 DBG("Using kernel consumer at: %s", consumer_to_use
);
2535 ret
= execl(consumer_to_use
,
2536 "lttng-consumerd", verbosity
, "-k",
2537 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2538 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2539 "--group", tracing_group_name
,
2542 case LTTNG_CONSUMER64_UST
:
2544 char *tmpnew
= NULL
;
2546 if (consumerd64_libdir
[0] != '\0') {
2550 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2554 tmplen
= strlen("LD_LIBRARY_PATH=")
2555 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2556 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2561 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2562 strcat(tmpnew
, consumerd64_libdir
);
2563 if (tmp
[0] != '\0') {
2564 strcat(tmpnew
, ":");
2565 strcat(tmpnew
, tmp
);
2567 ret
= putenv(tmpnew
);
2574 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2575 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2576 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2577 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2578 "--group", tracing_group_name
,
2580 if (consumerd64_libdir
[0] != '\0') {
2585 case LTTNG_CONSUMER32_UST
:
2587 char *tmpnew
= NULL
;
2589 if (consumerd32_libdir
[0] != '\0') {
2593 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2597 tmplen
= strlen("LD_LIBRARY_PATH=")
2598 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2599 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2604 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2605 strcat(tmpnew
, consumerd32_libdir
);
2606 if (tmp
[0] != '\0') {
2607 strcat(tmpnew
, ":");
2608 strcat(tmpnew
, tmp
);
2610 ret
= putenv(tmpnew
);
2617 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2618 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2619 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2620 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2621 "--group", tracing_group_name
,
2623 if (consumerd32_libdir
[0] != '\0') {
2629 PERROR("unknown consumer type");
2633 PERROR("Consumer execl()");
2635 /* Reaching this point, we got a failure on our execl(). */
2637 } else if (pid
> 0) {
2640 PERROR("start consumer fork");
2648 * Spawn the consumerd daemon and session daemon thread.
2650 static int start_consumerd(struct consumer_data
*consumer_data
)
2655 * Set the listen() state on the socket since there is a possible race
2656 * between the exec() of the consumer daemon and this call if place in the
2657 * consumer thread. See bug #366 for more details.
2659 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2664 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2665 if (consumer_data
->pid
!= 0) {
2666 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2670 ret
= spawn_consumerd(consumer_data
);
2672 ERR("Spawning consumerd failed");
2673 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2677 /* Setting up the consumer_data pid */
2678 consumer_data
->pid
= ret
;
2679 DBG2("Consumer pid %d", consumer_data
->pid
);
2680 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2682 DBG2("Spawning consumer control thread");
2683 ret
= spawn_consumer_thread(consumer_data
);
2685 ERR("Fatal error spawning consumer control thread");
2693 /* Cleanup already created sockets on error. */
2694 if (consumer_data
->err_sock
>= 0) {
2697 err
= close(consumer_data
->err_sock
);
2699 PERROR("close consumer data error socket");
2706 * Setup necessary data for kernel tracer action.
2708 static int init_kernel_tracer(void)
2712 /* Modprobe lttng kernel modules */
2713 ret
= modprobe_lttng_control();
2718 /* Open debugfs lttng */
2719 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2720 if (kernel_tracer_fd
< 0) {
2721 DBG("Failed to open %s", module_proc_lttng
);
2726 /* Validate kernel version */
2727 ret
= kernel_validate_version(kernel_tracer_fd
);
2732 ret
= modprobe_lttng_data();
2737 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2741 modprobe_remove_lttng_control();
2742 ret
= close(kernel_tracer_fd
);
2746 kernel_tracer_fd
= -1;
2747 return LTTNG_ERR_KERN_VERSION
;
2750 ret
= close(kernel_tracer_fd
);
2756 modprobe_remove_lttng_control();
2759 WARN("No kernel tracer available");
2760 kernel_tracer_fd
= -1;
2762 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2764 return LTTNG_ERR_KERN_NA
;
2770 * Copy consumer output from the tracing session to the domain session. The
2771 * function also applies the right modification on a per domain basis for the
2772 * trace files destination directory.
2774 * Should *NOT* be called with RCU read-side lock held.
2776 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2779 const char *dir_name
;
2780 struct consumer_output
*consumer
;
2783 assert(session
->consumer
);
2786 case LTTNG_DOMAIN_KERNEL
:
2787 DBG3("Copying tracing session consumer output in kernel session");
2789 * XXX: We should audit the session creation and what this function
2790 * does "extra" in order to avoid a destroy since this function is used
2791 * in the domain session creation (kernel and ust) only. Same for UST
2794 if (session
->kernel_session
->consumer
) {
2795 consumer_output_put(session
->kernel_session
->consumer
);
2797 session
->kernel_session
->consumer
=
2798 consumer_copy_output(session
->consumer
);
2799 /* Ease our life a bit for the next part */
2800 consumer
= session
->kernel_session
->consumer
;
2801 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2803 case LTTNG_DOMAIN_JUL
:
2804 case LTTNG_DOMAIN_LOG4J
:
2805 case LTTNG_DOMAIN_PYTHON
:
2806 case LTTNG_DOMAIN_UST
:
2807 DBG3("Copying tracing session consumer output in UST session");
2808 if (session
->ust_session
->consumer
) {
2809 consumer_output_put(session
->ust_session
->consumer
);
2811 session
->ust_session
->consumer
=
2812 consumer_copy_output(session
->consumer
);
2813 /* Ease our life a bit for the next part */
2814 consumer
= session
->ust_session
->consumer
;
2815 dir_name
= DEFAULT_UST_TRACE_DIR
;
2818 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2822 /* Append correct directory to subdir */
2823 strncat(consumer
->subdir
, dir_name
,
2824 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2825 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2834 * Create an UST session and add it to the session ust list.
2836 * Should *NOT* be called with RCU read-side lock held.
2838 static int create_ust_session(struct ltt_session
*session
,
2839 struct lttng_domain
*domain
)
2842 struct ltt_ust_session
*lus
= NULL
;
2846 assert(session
->consumer
);
2848 switch (domain
->type
) {
2849 case LTTNG_DOMAIN_JUL
:
2850 case LTTNG_DOMAIN_LOG4J
:
2851 case LTTNG_DOMAIN_PYTHON
:
2852 case LTTNG_DOMAIN_UST
:
2855 ERR("Unknown UST domain on create session %d", domain
->type
);
2856 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2860 DBG("Creating UST session");
2862 lus
= trace_ust_create_session(session
->id
);
2864 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2868 lus
->uid
= session
->uid
;
2869 lus
->gid
= session
->gid
;
2870 lus
->output_traces
= session
->output_traces
;
2871 lus
->snapshot_mode
= session
->snapshot_mode
;
2872 lus
->live_timer_interval
= session
->live_timer
;
2873 session
->ust_session
= lus
;
2874 if (session
->shm_path
[0]) {
2875 strncpy(lus
->root_shm_path
, session
->shm_path
,
2876 sizeof(lus
->root_shm_path
));
2877 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2878 strncpy(lus
->shm_path
, session
->shm_path
,
2879 sizeof(lus
->shm_path
));
2880 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2881 strncat(lus
->shm_path
, "/ust",
2882 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2884 /* Copy session output to the newly created UST session */
2885 ret
= copy_session_consumer(domain
->type
, session
);
2886 if (ret
!= LTTNG_OK
) {
2894 session
->ust_session
= NULL
;
2899 * Create a kernel tracer session then create the default channel.
2901 static int create_kernel_session(struct ltt_session
*session
)
2905 DBG("Creating kernel session");
2907 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2909 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2913 /* Code flow safety */
2914 assert(session
->kernel_session
);
2916 /* Copy session output to the newly created Kernel session */
2917 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2918 if (ret
!= LTTNG_OK
) {
2922 /* Create directory(ies) on local filesystem. */
2923 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2924 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2925 ret
= run_as_mkdir_recursive(
2926 session
->kernel_session
->consumer
->dst
.trace_path
,
2927 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2929 if (errno
!= EEXIST
) {
2930 ERR("Trace directory creation error");
2936 session
->kernel_session
->uid
= session
->uid
;
2937 session
->kernel_session
->gid
= session
->gid
;
2938 session
->kernel_session
->output_traces
= session
->output_traces
;
2939 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2944 trace_kernel_destroy_session(session
->kernel_session
);
2945 session
->kernel_session
= NULL
;
2950 * Count number of session permitted by uid/gid.
2952 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2955 struct ltt_session
*session
;
2957 DBG("Counting number of available session for UID %d GID %d",
2959 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2961 * Only list the sessions the user can control.
2963 if (!session_access_ok(session
, uid
, gid
)) {
2972 * Process the command requested by the lttng client within the command
2973 * context structure. This function make sure that the return structure (llm)
2974 * is set and ready for transmission before returning.
2976 * Return any error encountered or 0 for success.
2978 * "sock" is only used for special-case var. len data.
2980 * Should *NOT* be called with RCU read-side lock held.
2982 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2986 int need_tracing_session
= 1;
2989 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2991 assert(!rcu_read_ongoing());
2995 switch (cmd_ctx
->lsm
->cmd_type
) {
2996 case LTTNG_CREATE_SESSION
:
2997 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2998 case LTTNG_CREATE_SESSION_LIVE
:
2999 case LTTNG_DESTROY_SESSION
:
3000 case LTTNG_LIST_SESSIONS
:
3001 case LTTNG_LIST_DOMAINS
:
3002 case LTTNG_START_TRACE
:
3003 case LTTNG_STOP_TRACE
:
3004 case LTTNG_DATA_PENDING
:
3005 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3006 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3007 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3008 case LTTNG_SNAPSHOT_RECORD
:
3009 case LTTNG_SAVE_SESSION
:
3010 case LTTNG_SET_SESSION_SHM_PATH
:
3017 if (opt_no_kernel
&& need_domain
3018 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3020 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3022 ret
= LTTNG_ERR_KERN_NA
;
3027 /* Deny register consumer if we already have a spawned consumer. */
3028 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3029 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3030 if (kconsumer_data
.pid
> 0) {
3031 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3032 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3035 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3039 * Check for command that don't needs to allocate a returned payload. We do
3040 * this here so we don't have to make the call for no payload at each
3043 switch(cmd_ctx
->lsm
->cmd_type
) {
3044 case LTTNG_LIST_SESSIONS
:
3045 case LTTNG_LIST_TRACEPOINTS
:
3046 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3047 case LTTNG_LIST_DOMAINS
:
3048 case LTTNG_LIST_CHANNELS
:
3049 case LTTNG_LIST_EVENTS
:
3050 case LTTNG_LIST_SYSCALLS
:
3051 case LTTNG_LIST_TRACKER_PIDS
:
3054 /* Setup lttng message with no payload */
3055 ret
= setup_lttng_msg(cmd_ctx
, 0);
3057 /* This label does not try to unlock the session */
3058 goto init_setup_error
;
3062 /* Commands that DO NOT need a session. */
3063 switch (cmd_ctx
->lsm
->cmd_type
) {
3064 case LTTNG_CREATE_SESSION
:
3065 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3066 case LTTNG_CREATE_SESSION_LIVE
:
3067 case LTTNG_CALIBRATE
:
3068 case LTTNG_LIST_SESSIONS
:
3069 case LTTNG_LIST_TRACEPOINTS
:
3070 case LTTNG_LIST_SYSCALLS
:
3071 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3072 case LTTNG_SAVE_SESSION
:
3073 need_tracing_session
= 0;
3076 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3078 * We keep the session list lock across _all_ commands
3079 * for now, because the per-session lock does not
3080 * handle teardown properly.
3082 session_lock_list();
3083 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3084 if (cmd_ctx
->session
== NULL
) {
3085 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3088 /* Acquire lock for the session */
3089 session_lock(cmd_ctx
->session
);
3095 * Commands that need a valid session but should NOT create one if none
3096 * exists. Instead of creating one and destroying it when the command is
3097 * handled, process that right before so we save some round trip in useless
3100 switch (cmd_ctx
->lsm
->cmd_type
) {
3101 case LTTNG_DISABLE_CHANNEL
:
3102 case LTTNG_DISABLE_EVENT
:
3103 switch (cmd_ctx
->lsm
->domain
.type
) {
3104 case LTTNG_DOMAIN_KERNEL
:
3105 if (!cmd_ctx
->session
->kernel_session
) {
3106 ret
= LTTNG_ERR_NO_CHANNEL
;
3110 case LTTNG_DOMAIN_JUL
:
3111 case LTTNG_DOMAIN_LOG4J
:
3112 case LTTNG_DOMAIN_PYTHON
:
3113 case LTTNG_DOMAIN_UST
:
3114 if (!cmd_ctx
->session
->ust_session
) {
3115 ret
= LTTNG_ERR_NO_CHANNEL
;
3120 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3132 * Check domain type for specific "pre-action".
3134 switch (cmd_ctx
->lsm
->domain
.type
) {
3135 case LTTNG_DOMAIN_KERNEL
:
3137 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3141 /* Kernel tracer check */
3142 if (kernel_tracer_fd
== -1) {
3143 /* Basically, load kernel tracer modules */
3144 ret
= init_kernel_tracer();
3150 /* Consumer is in an ERROR state. Report back to client */
3151 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3152 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3156 /* Need a session for kernel command */
3157 if (need_tracing_session
) {
3158 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3159 ret
= create_kernel_session(cmd_ctx
->session
);
3161 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3166 /* Start the kernel consumer daemon */
3167 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3168 if (kconsumer_data
.pid
== 0 &&
3169 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3170 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3171 ret
= start_consumerd(&kconsumer_data
);
3173 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3176 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3178 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3182 * The consumer was just spawned so we need to add the socket to
3183 * the consumer output of the session if exist.
3185 ret
= consumer_create_socket(&kconsumer_data
,
3186 cmd_ctx
->session
->kernel_session
->consumer
);
3193 case LTTNG_DOMAIN_JUL
:
3194 case LTTNG_DOMAIN_LOG4J
:
3195 case LTTNG_DOMAIN_PYTHON
:
3196 case LTTNG_DOMAIN_UST
:
3198 if (!ust_app_supported()) {
3199 ret
= LTTNG_ERR_NO_UST
;
3202 /* Consumer is in an ERROR state. Report back to client */
3203 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3204 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3208 if (need_tracing_session
) {
3209 /* Create UST session if none exist. */
3210 if (cmd_ctx
->session
->ust_session
== NULL
) {
3211 ret
= create_ust_session(cmd_ctx
->session
,
3212 &cmd_ctx
->lsm
->domain
);
3213 if (ret
!= LTTNG_OK
) {
3218 /* Start the UST consumer daemons */
3220 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3221 if (consumerd64_bin
[0] != '\0' &&
3222 ustconsumer64_data
.pid
== 0 &&
3223 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3224 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3225 ret
= start_consumerd(&ustconsumer64_data
);
3227 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3228 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3232 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3233 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3235 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3239 * Setup socket for consumer 64 bit. No need for atomic access
3240 * since it was set above and can ONLY be set in this thread.
3242 ret
= consumer_create_socket(&ustconsumer64_data
,
3243 cmd_ctx
->session
->ust_session
->consumer
);
3249 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3250 if (consumerd32_bin
[0] != '\0' &&
3251 ustconsumer32_data
.pid
== 0 &&
3252 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3253 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3254 ret
= start_consumerd(&ustconsumer32_data
);
3256 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3257 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3261 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3262 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3264 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3268 * Setup socket for consumer 64 bit. No need for atomic access
3269 * since it was set above and can ONLY be set in this thread.
3271 ret
= consumer_create_socket(&ustconsumer32_data
,
3272 cmd_ctx
->session
->ust_session
->consumer
);
3284 /* Validate consumer daemon state when start/stop trace command */
3285 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3286 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3287 switch (cmd_ctx
->lsm
->domain
.type
) {
3288 case LTTNG_DOMAIN_JUL
:
3289 case LTTNG_DOMAIN_LOG4J
:
3290 case LTTNG_DOMAIN_PYTHON
:
3291 case LTTNG_DOMAIN_UST
:
3292 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3293 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3297 case LTTNG_DOMAIN_KERNEL
:
3298 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3299 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3307 * Check that the UID or GID match that of the tracing session.
3308 * The root user can interact with all sessions.
3310 if (need_tracing_session
) {
3311 if (!session_access_ok(cmd_ctx
->session
,
3312 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3313 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3314 ret
= LTTNG_ERR_EPERM
;
3320 * Send relayd information to consumer as soon as we have a domain and a
3323 if (cmd_ctx
->session
&& need_domain
) {
3325 * Setup relayd if not done yet. If the relayd information was already
3326 * sent to the consumer, this call will gracefully return.
3328 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3329 if (ret
!= LTTNG_OK
) {
3334 /* Process by command type */
3335 switch (cmd_ctx
->lsm
->cmd_type
) {
3336 case LTTNG_ADD_CONTEXT
:
3338 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3339 cmd_ctx
->lsm
->u
.context
.channel_name
,
3340 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3343 case LTTNG_DISABLE_CHANNEL
:
3345 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3346 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3349 case LTTNG_DISABLE_EVENT
:
3353 * FIXME: handle filter; for now we just receive the filter's
3354 * bytecode along with the filter expression which are sent by
3355 * liblttng-ctl and discard them.
3357 * This fixes an issue where the client may block while sending
3358 * the filter payload and encounter an error because the session
3359 * daemon closes the socket without ever handling this data.
3361 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3362 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3365 char data
[LTTNG_FILTER_MAX_LEN
];
3367 DBG("Discarding disable event command payload of size %zu", count
);
3369 ret
= lttcomm_recv_unix_sock(sock
, data
,
3370 count
> sizeof(data
) ? sizeof(data
) : count
);
3375 count
-= (size_t) ret
;
3378 /* FIXME: passing packed structure to non-packed pointer */
3379 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3380 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3381 &cmd_ctx
->lsm
->u
.disable
.event
);
3384 case LTTNG_ENABLE_CHANNEL
:
3386 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3387 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3390 case LTTNG_TRACK_PID
:
3392 ret
= cmd_track_pid(cmd_ctx
->session
,
3393 cmd_ctx
->lsm
->domain
.type
,
3394 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3397 case LTTNG_UNTRACK_PID
:
3399 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3400 cmd_ctx
->lsm
->domain
.type
,
3401 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3404 case LTTNG_ENABLE_EVENT
:
3406 struct lttng_event_exclusion
*exclusion
= NULL
;
3407 struct lttng_filter_bytecode
*bytecode
= NULL
;
3408 char *filter_expression
= NULL
;
3410 /* Handle exclusion events and receive it from the client. */
3411 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3412 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3414 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3415 (count
* LTTNG_SYMBOL_NAME_LEN
));
3417 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3421 DBG("Receiving var len exclusion event list from client ...");
3422 exclusion
->count
= count
;
3423 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3424 count
* LTTNG_SYMBOL_NAME_LEN
);
3426 DBG("Nothing recv() from client var len data... continuing");
3429 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3434 /* Get filter expression from client. */
3435 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3436 size_t expression_len
=
3437 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3439 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3440 ret
= LTTNG_ERR_FILTER_INVAL
;
3445 filter_expression
= zmalloc(expression_len
);
3446 if (!filter_expression
) {
3448 ret
= LTTNG_ERR_FILTER_NOMEM
;
3452 /* Receive var. len. data */
3453 DBG("Receiving var len filter's expression from client ...");
3454 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3457 DBG("Nothing recv() from client car len data... continuing");
3459 free(filter_expression
);
3461 ret
= LTTNG_ERR_FILTER_INVAL
;
3466 /* Handle filter and get bytecode from client. */
3467 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3468 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3470 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3471 ret
= LTTNG_ERR_FILTER_INVAL
;
3472 free(filter_expression
);
3477 bytecode
= zmalloc(bytecode_len
);
3479 free(filter_expression
);
3481 ret
= LTTNG_ERR_FILTER_NOMEM
;
3485 /* Receive var. len. data */
3486 DBG("Receiving var len filter's bytecode from client ...");
3487 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3489 DBG("Nothing recv() from client car len data... continuing");
3491 free(filter_expression
);
3494 ret
= LTTNG_ERR_FILTER_INVAL
;
3498 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3499 free(filter_expression
);
3502 ret
= LTTNG_ERR_FILTER_INVAL
;
3507 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3508 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3509 &cmd_ctx
->lsm
->u
.enable
.event
,
3510 filter_expression
, bytecode
, exclusion
,
3511 kernel_poll_pipe
[1]);
3514 case LTTNG_LIST_TRACEPOINTS
:
3516 struct lttng_event
*events
;
3519 session_lock_list();
3520 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3521 session_unlock_list();
3522 if (nb_events
< 0) {
3523 /* Return value is a negative lttng_error_code. */
3529 * Setup lttng message with payload size set to the event list size in
3530 * bytes and then copy list into the llm payload.
3532 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3538 /* Copy event list into message payload */
3539 memcpy(cmd_ctx
->llm
->payload
, events
,
3540 sizeof(struct lttng_event
) * nb_events
);
3547 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3549 struct lttng_event_field
*fields
;
3552 session_lock_list();
3553 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3555 session_unlock_list();
3556 if (nb_fields
< 0) {
3557 /* Return value is a negative lttng_error_code. */
3563 * Setup lttng message with payload size set to the event list size in
3564 * bytes and then copy list into the llm payload.
3566 ret
= setup_lttng_msg(cmd_ctx
,
3567 sizeof(struct lttng_event_field
) * nb_fields
);
3573 /* Copy event list into message payload */
3574 memcpy(cmd_ctx
->llm
->payload
, fields
,
3575 sizeof(struct lttng_event_field
) * nb_fields
);
3582 case LTTNG_LIST_SYSCALLS
:
3584 struct lttng_event
*events
;
3587 nb_events
= cmd_list_syscalls(&events
);
3588 if (nb_events
< 0) {
3589 /* Return value is a negative lttng_error_code. */
3595 * Setup lttng message with payload size set to the event list size in
3596 * bytes and then copy list into the llm payload.
3598 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3604 /* Copy event list into message payload */
3605 memcpy(cmd_ctx
->llm
->payload
, events
,
3606 sizeof(struct lttng_event
) * nb_events
);
3613 case LTTNG_LIST_TRACKER_PIDS
:
3615 int32_t *pids
= NULL
;
3618 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3619 cmd_ctx
->lsm
->domain
.type
, &pids
);
3621 /* Return value is a negative lttng_error_code. */
3627 * Setup lttng message with payload size set to the event list size in
3628 * bytes and then copy list into the llm payload.
3630 ret
= setup_lttng_msg(cmd_ctx
, sizeof(int32_t) * nr_pids
);
3636 /* Copy event list into message payload */
3637 memcpy(cmd_ctx
->llm
->payload
, pids
,
3638 sizeof(int) * nr_pids
);
3645 case LTTNG_SET_CONSUMER_URI
:
3648 struct lttng_uri
*uris
;
3650 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3651 len
= nb_uri
* sizeof(struct lttng_uri
);
3654 ret
= LTTNG_ERR_INVALID
;
3658 uris
= zmalloc(len
);
3660 ret
= LTTNG_ERR_FATAL
;
3664 /* Receive variable len data */
3665 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3666 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3668 DBG("No URIs received from client... continuing");
3670 ret
= LTTNG_ERR_SESSION_FAIL
;
3675 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3677 if (ret
!= LTTNG_OK
) {
3684 case LTTNG_START_TRACE
:
3686 ret
= cmd_start_trace(cmd_ctx
->session
);
3689 case LTTNG_STOP_TRACE
:
3691 ret
= cmd_stop_trace(cmd_ctx
->session
);
3694 case LTTNG_CREATE_SESSION
:
3697 struct lttng_uri
*uris
= NULL
;
3699 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3700 len
= nb_uri
* sizeof(struct lttng_uri
);
3703 uris
= zmalloc(len
);
3705 ret
= LTTNG_ERR_FATAL
;
3709 /* Receive variable len data */
3710 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3711 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3713 DBG("No URIs received from client... continuing");
3715 ret
= LTTNG_ERR_SESSION_FAIL
;
3720 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3721 DBG("Creating session with ONE network URI is a bad call");
3722 ret
= LTTNG_ERR_SESSION_FAIL
;
3728 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3729 &cmd_ctx
->creds
, 0);
3735 case LTTNG_DESTROY_SESSION
:
3737 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3739 /* Set session to NULL so we do not unlock it after free. */
3740 cmd_ctx
->session
= NULL
;
3743 case LTTNG_LIST_DOMAINS
:
3746 struct lttng_domain
*domains
= NULL
;
3748 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3750 /* Return value is a negative lttng_error_code. */
3755 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3761 /* Copy event list into message payload */
3762 memcpy(cmd_ctx
->llm
->payload
, domains
,
3763 nb_dom
* sizeof(struct lttng_domain
));
3770 case LTTNG_LIST_CHANNELS
:
3773 struct lttng_channel
*channels
= NULL
;
3775 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3776 cmd_ctx
->session
, &channels
);
3778 /* Return value is a negative lttng_error_code. */
3783 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3789 /* Copy event list into message payload */
3790 memcpy(cmd_ctx
->llm
->payload
, channels
,
3791 nb_chan
* sizeof(struct lttng_channel
));
3798 case LTTNG_LIST_EVENTS
:
3801 struct lttng_event
*events
= NULL
;
3803 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3804 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3806 /* Return value is a negative lttng_error_code. */
3811 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3817 /* Copy event list into message payload */
3818 memcpy(cmd_ctx
->llm
->payload
, events
,
3819 nb_event
* sizeof(struct lttng_event
));
3826 case LTTNG_LIST_SESSIONS
:
3828 unsigned int nr_sessions
;
3830 session_lock_list();
3831 nr_sessions
= lttng_sessions_count(
3832 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3833 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3835 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3837 session_unlock_list();
3841 /* Filled the session array */
3842 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3843 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3844 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3846 session_unlock_list();
3851 case LTTNG_CALIBRATE
:
3853 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3854 &cmd_ctx
->lsm
->u
.calibrate
);
3857 case LTTNG_REGISTER_CONSUMER
:
3859 struct consumer_data
*cdata
;
3861 switch (cmd_ctx
->lsm
->domain
.type
) {
3862 case LTTNG_DOMAIN_KERNEL
:
3863 cdata
= &kconsumer_data
;
3866 ret
= LTTNG_ERR_UND
;
3870 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3871 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3874 case LTTNG_DATA_PENDING
:
3876 ret
= cmd_data_pending(cmd_ctx
->session
);
3879 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3881 struct lttcomm_lttng_output_id reply
;
3883 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3884 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3885 if (ret
!= LTTNG_OK
) {
3889 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3894 /* Copy output list into message payload */
3895 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3899 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3901 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3902 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3905 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3908 struct lttng_snapshot_output
*outputs
= NULL
;
3910 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3911 if (nb_output
< 0) {
3916 ret
= setup_lttng_msg(cmd_ctx
,
3917 nb_output
* sizeof(struct lttng_snapshot_output
));
3924 /* Copy output list into message payload */
3925 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3926 nb_output
* sizeof(struct lttng_snapshot_output
));
3933 case LTTNG_SNAPSHOT_RECORD
:
3935 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3936 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3937 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3940 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3943 struct lttng_uri
*uris
= NULL
;
3945 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3946 len
= nb_uri
* sizeof(struct lttng_uri
);
3949 uris
= zmalloc(len
);
3951 ret
= LTTNG_ERR_FATAL
;
3955 /* Receive variable len data */
3956 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3957 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3959 DBG("No URIs received from client... continuing");
3961 ret
= LTTNG_ERR_SESSION_FAIL
;
3966 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3967 DBG("Creating session with ONE network URI is a bad call");
3968 ret
= LTTNG_ERR_SESSION_FAIL
;
3974 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3975 nb_uri
, &cmd_ctx
->creds
);
3979 case LTTNG_CREATE_SESSION_LIVE
:
3982 struct lttng_uri
*uris
= NULL
;
3984 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3985 len
= nb_uri
* sizeof(struct lttng_uri
);
3988 uris
= zmalloc(len
);
3990 ret
= LTTNG_ERR_FATAL
;
3994 /* Receive variable len data */
3995 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3996 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3998 DBG("No URIs received from client... continuing");
4000 ret
= LTTNG_ERR_SESSION_FAIL
;
4005 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4006 DBG("Creating session with ONE network URI is a bad call");
4007 ret
= LTTNG_ERR_SESSION_FAIL
;
4013 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4014 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4018 case LTTNG_SAVE_SESSION
:
4020 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4024 case LTTNG_SET_SESSION_SHM_PATH
:
4026 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4027 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4031 ret
= LTTNG_ERR_UND
;
4036 if (cmd_ctx
->llm
== NULL
) {
4037 DBG("Missing llm structure. Allocating one.");
4038 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
4042 /* Set return code */
4043 cmd_ctx
->llm
->ret_code
= ret
;
4045 if (cmd_ctx
->session
) {
4046 session_unlock(cmd_ctx
->session
);
4048 if (need_tracing_session
) {
4049 session_unlock_list();
4052 assert(!rcu_read_ongoing());
4057 * Thread managing health check socket.
4059 static void *thread_manage_health(void *data
)
4061 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4062 uint32_t revents
, nb_fd
;
4063 struct lttng_poll_event events
;
4064 struct health_comm_msg msg
;
4065 struct health_comm_reply reply
;
4067 DBG("[thread] Manage health check started");
4069 rcu_register_thread();
4071 /* We might hit an error path before this is created. */
4072 lttng_poll_init(&events
);
4074 /* Create unix socket */
4075 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4077 ERR("Unable to create health check Unix socket");
4083 /* lttng health client socket path permissions */
4084 ret
= chown(health_unix_sock_path
, 0,
4085 utils_get_group_id(tracing_group_name
));
4087 ERR("Unable to set group on %s", health_unix_sock_path
);
4093 ret
= chmod(health_unix_sock_path
,
4094 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4096 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4104 * Set the CLOEXEC flag. Return code is useless because either way, the
4107 (void) utils_set_fd_cloexec(sock
);
4109 ret
= lttcomm_listen_unix_sock(sock
);
4115 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4116 * more will be added to this poll set.
4118 ret
= sessiond_set_thread_pollset(&events
, 2);
4123 /* Add the application registration socket */
4124 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4129 sessiond_notify_ready();
4132 DBG("Health check ready");
4134 /* Inifinite blocking call, waiting for transmission */
4136 ret
= lttng_poll_wait(&events
, -1);
4139 * Restart interrupted system call.
4141 if (errno
== EINTR
) {
4149 for (i
= 0; i
< nb_fd
; i
++) {
4150 /* Fetch once the poll data */
4151 revents
= LTTNG_POLL_GETEV(&events
, i
);
4152 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4155 /* No activity for this FD (poll implementation). */
4159 /* Thread quit pipe has been closed. Killing thread. */
4160 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4166 /* Event on the registration socket */
4167 if (pollfd
== sock
) {
4168 if (revents
& LPOLLIN
) {
4170 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4171 ERR("Health socket poll error");
4174 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4180 new_sock
= lttcomm_accept_unix_sock(sock
);
4186 * Set the CLOEXEC flag. Return code is useless because either way, the
4189 (void) utils_set_fd_cloexec(new_sock
);
4191 DBG("Receiving data from client for health...");
4192 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4194 DBG("Nothing recv() from client... continuing");
4195 ret
= close(new_sock
);
4203 rcu_thread_online();
4205 memset(&reply
, 0, sizeof(reply
));
4206 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4208 * health_check_state returns 0 if health is
4211 if (!health_check_state(health_sessiond
, i
)) {
4212 reply
.ret_code
|= 1ULL << i
;
4216 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4218 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4220 ERR("Failed to send health data back to client");
4223 /* End of transmission */
4224 ret
= close(new_sock
);
4234 ERR("Health error occurred in %s", __func__
);
4236 DBG("Health check thread dying");
4237 unlink(health_unix_sock_path
);
4245 lttng_poll_clean(&events
);
4247 rcu_unregister_thread();
4252 * This thread manage all clients request using the unix client socket for
4255 static void *thread_manage_clients(void *data
)
4257 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4259 uint32_t revents
, nb_fd
;
4260 struct command_ctx
*cmd_ctx
= NULL
;
4261 struct lttng_poll_event events
;
4263 DBG("[thread] Manage client started");
4265 rcu_register_thread();
4267 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4269 health_code_update();
4271 ret
= lttcomm_listen_unix_sock(client_sock
);
4277 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4278 * more will be added to this poll set.
4280 ret
= sessiond_set_thread_pollset(&events
, 2);
4282 goto error_create_poll
;
4285 /* Add the application registration socket */
4286 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4291 sessiond_notify_ready();
4292 ret
= sem_post(&load_info
->message_thread_ready
);
4294 PERROR("sem_post message_thread_ready");
4298 /* This testpoint is after we signal readiness to the parent. */
4299 if (testpoint(sessiond_thread_manage_clients
)) {
4303 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4307 health_code_update();
4310 DBG("Accepting client command ...");
4312 /* Inifinite blocking call, waiting for transmission */
4314 health_poll_entry();
4315 ret
= lttng_poll_wait(&events
, -1);
4319 * Restart interrupted system call.
4321 if (errno
== EINTR
) {
4329 for (i
= 0; i
< nb_fd
; i
++) {
4330 /* Fetch once the poll data */
4331 revents
= LTTNG_POLL_GETEV(&events
, i
);
4332 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4334 health_code_update();
4337 /* No activity for this FD (poll implementation). */
4341 /* Thread quit pipe has been closed. Killing thread. */
4342 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4348 /* Event on the registration socket */
4349 if (pollfd
== client_sock
) {
4350 if (revents
& LPOLLIN
) {
4352 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4353 ERR("Client socket poll error");
4356 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4362 DBG("Wait for client response");
4364 health_code_update();
4366 sock
= lttcomm_accept_unix_sock(client_sock
);
4372 * Set the CLOEXEC flag. Return code is useless because either way, the
4375 (void) utils_set_fd_cloexec(sock
);
4377 /* Set socket option for credentials retrieval */
4378 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4383 /* Allocate context command to process the client request */
4384 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4385 if (cmd_ctx
== NULL
) {
4386 PERROR("zmalloc cmd_ctx");
4390 /* Allocate data buffer for reception */
4391 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4392 if (cmd_ctx
->lsm
== NULL
) {
4393 PERROR("zmalloc cmd_ctx->lsm");
4397 cmd_ctx
->llm
= NULL
;
4398 cmd_ctx
->session
= NULL
;
4400 health_code_update();
4403 * Data is received from the lttng client. The struct
4404 * lttcomm_session_msg (lsm) contains the command and data request of
4407 DBG("Receiving data from client ...");
4408 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4409 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4411 DBG("Nothing recv() from client... continuing");
4417 clean_command_ctx(&cmd_ctx
);
4421 health_code_update();
4423 // TODO: Validate cmd_ctx including sanity check for
4424 // security purpose.
4426 rcu_thread_online();
4428 * This function dispatch the work to the kernel or userspace tracer
4429 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4430 * informations for the client. The command context struct contains
4431 * everything this function may needs.
4433 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4434 rcu_thread_offline();
4442 * TODO: Inform client somehow of the fatal error. At
4443 * this point, ret < 0 means that a zmalloc failed
4444 * (ENOMEM). Error detected but still accept
4445 * command, unless a socket error has been
4448 clean_command_ctx(&cmd_ctx
);
4452 health_code_update();
4454 DBG("Sending response (size: %d, retcode: %s)",
4455 cmd_ctx
->lttng_msg_size
,
4456 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
4457 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4459 ERR("Failed to send data back to client");
4462 /* End of transmission */
4469 clean_command_ctx(&cmd_ctx
);
4471 health_code_update();
4483 lttng_poll_clean(&events
);
4484 clean_command_ctx(&cmd_ctx
);
4488 unlink(client_unix_sock_path
);
4489 if (client_sock
>= 0) {
4490 ret
= close(client_sock
);
4498 ERR("Health error occurred in %s", __func__
);
4501 health_unregister(health_sessiond
);
4503 DBG("Client thread dying");
4505 rcu_unregister_thread();
4508 * Since we are creating the consumer threads, we own them, so we need
4509 * to join them before our thread exits.
4511 ret
= join_consumer_thread(&kconsumer_data
);
4514 PERROR("join_consumer");
4517 ret
= join_consumer_thread(&ustconsumer32_data
);
4520 PERROR("join_consumer ust32");
4523 ret
= join_consumer_thread(&ustconsumer64_data
);
4526 PERROR("join_consumer ust64");
4533 * usage function on stderr
4535 static void usage(void)
4537 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4538 fprintf(stderr
, " -h, --help Display this usage.\n");
4539 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4540 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4541 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4542 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4543 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4544 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4545 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4546 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4547 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4548 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4549 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4550 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4551 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4552 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4553 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4554 fprintf(stderr
, " -V, --version Show version number.\n");
4555 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4556 fprintf(stderr
, " -q, --quiet No output at all.\n");
4557 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4558 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4559 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4560 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4561 fprintf(stderr
, " --agent-tcp-port Agent registration TCP port\n");
4562 fprintf(stderr
, " -f --config PATH Load daemon configuration file\n");
4563 fprintf(stderr
, " -l --load PATH Load session configuration\n");
4564 fprintf(stderr
, " --kmod-probes Specify kernel module probes to load\n");
4565 fprintf(stderr
, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4568 static int string_match(const char *str1
, const char *str2
)
4570 return (str1
&& str2
) && !strcmp(str1
, str2
);
4574 * Take an option from the getopt output and set it in the right variable to be
4577 * Return 0 on success else a negative value.
4579 static int set_option(int opt
, const char *arg
, const char *optname
)
4583 if (string_match(optname
, "client-sock") || opt
== 'c') {
4584 if (!arg
|| *arg
== '\0') {
4588 if (lttng_is_setuid_setgid()) {
4589 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4590 "-c, --client-sock");
4592 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4594 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4595 if (!arg
|| *arg
== '\0') {
4599 if (lttng_is_setuid_setgid()) {
4600 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4603 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4605 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4607 } else if (string_match(optname
, "background") || opt
== 'b') {
4609 } else if (string_match(optname
, "group") || opt
== 'g') {
4610 if (!arg
|| *arg
== '\0') {
4614 if (lttng_is_setuid_setgid()) {
4615 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4619 * If the override option is set, the pointer points to a
4620 * *non* const thus freeing it even though the variable type is
4623 if (tracing_group_name_override
) {
4624 free((void *) tracing_group_name
);
4626 tracing_group_name
= strdup(arg
);
4627 if (!tracing_group_name
) {
4631 tracing_group_name_override
= 1;
4633 } else if (string_match(optname
, "help") || opt
== 'h') {
4636 } else if (string_match(optname
, "version") || opt
== 'V') {
4637 fprintf(stdout
, "%s\n", VERSION
);
4639 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4641 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4642 if (!arg
|| *arg
== '\0') {
4646 if (lttng_is_setuid_setgid()) {
4647 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4648 "--kconsumerd-err-sock");
4650 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4652 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4653 if (!arg
|| *arg
== '\0') {
4657 if (lttng_is_setuid_setgid()) {
4658 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4659 "--kconsumerd-cmd-sock");
4661 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4663 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4664 if (!arg
|| *arg
== '\0') {
4668 if (lttng_is_setuid_setgid()) {
4669 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4670 "--ustconsumerd64-err-sock");
4672 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4674 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4675 if (!arg
|| *arg
== '\0') {
4679 if (lttng_is_setuid_setgid()) {
4680 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4681 "--ustconsumerd64-cmd-sock");
4683 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4685 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4686 if (!arg
|| *arg
== '\0') {
4690 if (lttng_is_setuid_setgid()) {
4691 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4692 "--ustconsumerd32-err-sock");
4694 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4696 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4697 if (!arg
|| *arg
== '\0') {
4701 if (lttng_is_setuid_setgid()) {
4702 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4703 "--ustconsumerd32-cmd-sock");
4705 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4707 } else if (string_match(optname
, "no-kernel")) {
4709 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4710 lttng_opt_quiet
= 1;
4711 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4712 /* Verbose level can increase using multiple -v */
4714 /* Value obtained from config file */
4715 lttng_opt_verbose
= config_parse_value(arg
);
4717 /* -v used on command line */
4718 lttng_opt_verbose
++;
4720 /* Clamp value to [0, 3] */
4721 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4722 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4723 } else if (string_match(optname
, "verbose-consumer")) {
4725 opt_verbose_consumer
= config_parse_value(arg
);
4727 opt_verbose_consumer
+= 1;
4729 } else if (string_match(optname
, "consumerd32-path")) {
4730 if (!arg
|| *arg
== '\0') {
4734 if (lttng_is_setuid_setgid()) {
4735 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4736 "--consumerd32-path");
4738 if (consumerd32_bin_override
) {
4739 free((void *) consumerd32_bin
);
4741 consumerd32_bin
= strdup(arg
);
4742 if (!consumerd32_bin
) {
4746 consumerd32_bin_override
= 1;
4748 } else if (string_match(optname
, "consumerd32-libdir")) {
4749 if (!arg
|| *arg
== '\0') {
4753 if (lttng_is_setuid_setgid()) {
4754 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4755 "--consumerd32-libdir");
4757 if (consumerd32_libdir_override
) {
4758 free((void *) consumerd32_libdir
);
4760 consumerd32_libdir
= strdup(arg
);
4761 if (!consumerd32_libdir
) {
4765 consumerd32_libdir_override
= 1;
4767 } else if (string_match(optname
, "consumerd64-path")) {
4768 if (!arg
|| *arg
== '\0') {
4772 if (lttng_is_setuid_setgid()) {
4773 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4774 "--consumerd64-path");
4776 if (consumerd64_bin_override
) {
4777 free((void *) consumerd64_bin
);
4779 consumerd64_bin
= strdup(arg
);
4780 if (!consumerd64_bin
) {
4784 consumerd64_bin_override
= 1;
4786 } else if (string_match(optname
, "consumerd64-libdir")) {
4787 if (!arg
|| *arg
== '\0') {
4791 if (lttng_is_setuid_setgid()) {
4792 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4793 "--consumerd64-libdir");
4795 if (consumerd64_libdir_override
) {
4796 free((void *) consumerd64_libdir
);
4798 consumerd64_libdir
= strdup(arg
);
4799 if (!consumerd64_libdir
) {
4803 consumerd64_libdir_override
= 1;
4805 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4806 if (!arg
|| *arg
== '\0') {
4810 if (lttng_is_setuid_setgid()) {
4811 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4815 opt_pidfile
= strdup(arg
);
4821 } else if (string_match(optname
, "agent-tcp-port")) {
4822 if (!arg
|| *arg
== '\0') {
4826 if (lttng_is_setuid_setgid()) {
4827 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4828 "--agent-tcp-port");
4837 v
= strtoul(arg
, NULL
, 0);
4838 if (errno
!= 0 || !isdigit(arg
[0])) {
4839 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4842 if (v
== 0 || v
>= 65535) {
4843 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4846 agent_tcp_port
= (uint32_t) v
;
4847 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4849 } else if (string_match(optname
, "load") || opt
== 'l') {
4850 if (!arg
|| *arg
== '\0') {
4854 if (lttng_is_setuid_setgid()) {
4855 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4858 free(opt_load_session_path
);
4859 opt_load_session_path
= strdup(arg
);
4860 if (!opt_load_session_path
) {
4865 } else if (string_match(optname
, "kmod-probes")) {
4866 if (!arg
|| *arg
== '\0') {
4870 if (lttng_is_setuid_setgid()) {
4871 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4874 free(kmod_probes_list
);
4875 kmod_probes_list
= strdup(arg
);
4876 if (!kmod_probes_list
) {
4881 } else if (string_match(optname
, "extra-kmod-probes")) {
4882 if (!arg
|| *arg
== '\0') {
4886 if (lttng_is_setuid_setgid()) {
4887 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4888 "--extra-kmod-probes");
4890 free(kmod_extra_probes_list
);
4891 kmod_extra_probes_list
= strdup(arg
);
4892 if (!kmod_extra_probes_list
) {
4897 } else if (string_match(optname
, "config") || opt
== 'f') {
4898 /* This is handled in set_options() thus silent skip. */
4901 /* Unknown option or other error.
4902 * Error is printed by getopt, just return */
4907 if (ret
== -EINVAL
) {
4908 const char *opt_name
= "unknown";
4911 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4913 if (opt
== long_options
[i
].val
) {
4914 opt_name
= long_options
[i
].name
;
4919 WARN("Invalid argument provided for option \"%s\", using default value.",
4927 * config_entry_handler_cb used to handle options read from a config file.
4928 * See config_entry_handler_cb comment in common/config/config.h for the
4929 * return value conventions.
4931 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4935 if (!entry
|| !entry
->name
|| !entry
->value
) {
4940 /* Check if the option is to be ignored */
4941 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4942 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4947 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4950 /* Ignore if not fully matched. */
4951 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4956 * If the option takes no argument on the command line, we have to
4957 * check if the value is "true". We support non-zero numeric values,
4960 if (!long_options
[i
].has_arg
) {
4961 ret
= config_parse_value(entry
->value
);
4964 WARN("Invalid configuration value \"%s\" for option %s",
4965 entry
->value
, entry
->name
);
4967 /* False, skip boolean config option. */
4972 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4976 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4983 * daemon configuration loading and argument parsing
4985 static int set_options(int argc
, char **argv
)
4987 int ret
= 0, c
= 0, option_index
= 0;
4988 int orig_optopt
= optopt
, orig_optind
= optind
;
4990 const char *config_path
= NULL
;
4992 optstring
= utils_generate_optstring(long_options
,
4993 sizeof(long_options
) / sizeof(struct option
));
4999 /* Check for the --config option */
5000 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5001 &option_index
)) != -1) {
5005 } else if (c
!= 'f') {
5006 /* if not equal to --config option. */
5010 if (lttng_is_setuid_setgid()) {
5011 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5014 config_path
= utils_expand_path(optarg
);
5016 ERR("Failed to resolve path: %s", optarg
);
5021 ret
= config_get_section_entries(config_path
, config_section_name
,
5022 config_entry_handler
, NULL
);
5025 ERR("Invalid configuration option at line %i", ret
);
5031 /* Reset getopt's global state */
5032 optopt
= orig_optopt
;
5033 optind
= orig_optind
;
5037 * getopt_long() will not set option_index if it encounters a
5040 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5047 * Pass NULL as the long option name if popt left the index
5050 ret
= set_option(c
, optarg
,
5051 option_index
< 0 ? NULL
:
5052 long_options
[option_index
].name
);
5064 * Creates the two needed socket by the daemon.
5065 * apps_sock - The communication socket for all UST apps.
5066 * client_sock - The communication of the cli tool (lttng).
5068 static int init_daemon_socket(void)
5073 old_umask
= umask(0);
5075 /* Create client tool unix socket */
5076 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
5077 if (client_sock
< 0) {
5078 ERR("Create unix sock failed: %s", client_unix_sock_path
);
5083 /* Set the cloexec flag */
5084 ret
= utils_set_fd_cloexec(client_sock
);
5086 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5087 "Continuing but note that the consumer daemon will have a "
5088 "reference to this socket on exec()", client_sock
);
5091 /* File permission MUST be 660 */
5092 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5094 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5099 /* Create the application unix socket */
5100 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5101 if (apps_sock
< 0) {
5102 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5107 /* Set the cloexec flag */
5108 ret
= utils_set_fd_cloexec(apps_sock
);
5110 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5111 "Continuing but note that the consumer daemon will have a "
5112 "reference to this socket on exec()", apps_sock
);
5115 /* File permission MUST be 666 */
5116 ret
= chmod(apps_unix_sock_path
,
5117 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5119 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5124 DBG3("Session daemon client socket %d and application socket %d created",
5125 client_sock
, apps_sock
);
5133 * Check if the global socket is available, and if a daemon is answering at the
5134 * other side. If yes, error is returned.
5136 static int check_existing_daemon(void)
5138 /* Is there anybody out there ? */
5139 if (lttng_session_daemon_alive()) {
5147 * Set the tracing group gid onto the client socket.
5149 * Race window between mkdir and chown is OK because we are going from more
5150 * permissive (root.root) to less permissive (root.tracing).
5152 static int set_permissions(char *rundir
)
5157 gid
= utils_get_group_id(tracing_group_name
);
5159 /* Set lttng run dir */
5160 ret
= chown(rundir
, 0, gid
);
5162 ERR("Unable to set group on %s", rundir
);
5167 * Ensure all applications and tracing group can search the run
5168 * dir. Allow everyone to read the directory, since it does not
5169 * buy us anything to hide its content.
5171 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5173 ERR("Unable to set permissions on %s", rundir
);
5177 /* lttng client socket path */
5178 ret
= chown(client_unix_sock_path
, 0, gid
);
5180 ERR("Unable to set group on %s", client_unix_sock_path
);
5184 /* kconsumer error socket path */
5185 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5187 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5191 /* 64-bit ustconsumer error socket path */
5192 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5194 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5198 /* 32-bit ustconsumer compat32 error socket path */
5199 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5201 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5205 DBG("All permissions are set");
5211 * Create the lttng run directory needed for all global sockets and pipe.
5213 static int create_lttng_rundir(const char *rundir
)
5217 DBG3("Creating LTTng run directory: %s", rundir
);
5219 ret
= mkdir(rundir
, S_IRWXU
);
5221 if (errno
!= EEXIST
) {
5222 ERR("Unable to create %s", rundir
);
5234 * Setup sockets and directory needed by the kconsumerd communication with the
5237 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5241 char path
[PATH_MAX
];
5243 switch (consumer_data
->type
) {
5244 case LTTNG_CONSUMER_KERNEL
:
5245 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5247 case LTTNG_CONSUMER64_UST
:
5248 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5250 case LTTNG_CONSUMER32_UST
:
5251 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5254 ERR("Consumer type unknown");
5259 DBG2("Creating consumer directory: %s", path
);
5261 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5263 if (errno
!= EEXIST
) {
5265 ERR("Failed to create %s", path
);
5271 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5273 ERR("Unable to set group on %s", path
);
5279 /* Create the kconsumerd error unix socket */
5280 consumer_data
->err_sock
=
5281 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5282 if (consumer_data
->err_sock
< 0) {
5283 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5289 * Set the CLOEXEC flag. Return code is useless because either way, the
5292 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5294 PERROR("utils_set_fd_cloexec");
5295 /* continue anyway */
5298 /* File permission MUST be 660 */
5299 ret
= chmod(consumer_data
->err_unix_sock_path
,
5300 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5302 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5312 * Signal handler for the daemon
5314 * Simply stop all worker threads, leaving main() return gracefully after
5315 * joining all threads and calling cleanup().
5317 static void sighandler(int sig
)
5321 DBG("SIGPIPE caught");
5324 DBG("SIGINT caught");
5328 DBG("SIGTERM caught");
5332 CMM_STORE_SHARED(recv_child_signal
, 1);
5340 * Setup signal handler for :
5341 * SIGINT, SIGTERM, SIGPIPE
5343 static int set_signal_handler(void)
5346 struct sigaction sa
;
5349 if ((ret
= sigemptyset(&sigset
)) < 0) {
5350 PERROR("sigemptyset");
5354 sa
.sa_handler
= sighandler
;
5355 sa
.sa_mask
= sigset
;
5357 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5358 PERROR("sigaction");
5362 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5363 PERROR("sigaction");
5367 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5368 PERROR("sigaction");
5372 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5373 PERROR("sigaction");
5377 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5383 * Set open files limit to unlimited. This daemon can open a large number of
5384 * file descriptors in order to consumer multiple kernel traces.
5386 static void set_ulimit(void)
5391 /* The kernel does not allowed an infinite limit for open files */
5392 lim
.rlim_cur
= 65535;
5393 lim
.rlim_max
= 65535;
5395 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5397 PERROR("failed to set open files limit");
5402 * Write pidfile using the rundir and opt_pidfile.
5404 static int write_pidfile(void)
5407 char pidfile_path
[PATH_MAX
];
5412 if (lttng_strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
))) {
5417 /* Build pidfile path from rundir and opt_pidfile. */
5418 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5419 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5421 PERROR("snprintf pidfile path");
5427 * Create pid file in rundir.
5429 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5435 * Create lockfile using the rundir and return its fd.
5437 static int create_lockfile(void)
5440 char lockfile_path
[PATH_MAX
];
5442 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5447 ret
= utils_create_lock_file(lockfile_path
);
5453 * Write agent TCP port using the rundir.
5455 static int write_agent_port(void)
5458 char path
[PATH_MAX
];
5462 ret
= snprintf(path
, sizeof(path
), "%s/"
5463 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5465 PERROR("snprintf agent port path");
5470 * Create TCP agent port file in rundir.
5472 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5481 int main(int argc
, char **argv
)
5483 int ret
= 0, retval
= 0;
5485 const char *home_path
, *env_app_timeout
;
5487 /* Initialize agent apps ht global variable */
5488 agent_apps_ht_by_sock
= NULL
;
5490 init_kernel_workarounds();
5492 rcu_register_thread();
5494 if (set_signal_handler()) {
5496 goto exit_set_signal_handler
;
5499 setup_consumerd_path();
5501 page_size
= sysconf(_SC_PAGESIZE
);
5502 if (page_size
< 0) {
5503 PERROR("sysconf _SC_PAGESIZE");
5504 page_size
= LONG_MAX
;
5505 WARN("Fallback page size to %ld", page_size
);
5509 * Parse arguments and load the daemon configuration file.
5511 * We have an exit_options exit path to free memory reserved by
5512 * set_options. This is needed because the rest of sessiond_cleanup()
5513 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5514 * depends on set_options.
5517 if (set_options(argc
, argv
)) {
5523 if (opt_daemon
|| opt_background
) {
5526 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5534 * We are in the child. Make sure all other file descriptors are
5535 * closed, in case we are called with more opened file
5536 * descriptors than the standard ones.
5538 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5543 if (run_as_create_worker(argv
[0]) < 0) {
5544 goto exit_create_run_as_worker_cleanup
;
5548 * Starting from here, we can create threads. This needs to be after
5549 * lttng_daemonize due to RCU.
5553 * Initialize the health check subsystem. This call should set the
5554 * appropriate time values.
5556 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5557 if (!health_sessiond
) {
5558 PERROR("health_app_create error");
5560 goto exit_health_sessiond_cleanup
;
5563 if (init_ht_cleanup_quit_pipe()) {
5565 goto exit_ht_cleanup_quit_pipe
;
5568 /* Setup the thread ht_cleanup communication pipe. */
5569 if (utils_create_pipe_cloexec(ht_cleanup_pipe
)) {
5571 goto exit_ht_cleanup_pipe
;
5574 /* Set up max poll set size */
5575 if (lttng_poll_set_max_size()) {
5577 goto exit_set_max_size
;
5580 /* Create thread to clean up RCU hash tables */
5581 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5582 thread_ht_cleanup
, (void *) NULL
);
5585 PERROR("pthread_create ht_cleanup");
5587 goto exit_ht_cleanup
;
5590 /* Create thread quit pipe */
5591 if (init_thread_quit_pipe()) {
5593 goto exit_init_data
;
5596 /* Check if daemon is UID = 0 */
5597 is_root
= !getuid();
5600 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5603 goto exit_init_data
;
5606 /* Create global run dir with root access */
5607 if (create_lttng_rundir(rundir
)) {
5609 goto exit_init_data
;
5612 if (strlen(apps_unix_sock_path
) == 0) {
5613 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5614 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5617 goto exit_init_data
;
5621 if (strlen(client_unix_sock_path
) == 0) {
5622 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5623 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5626 goto exit_init_data
;
5630 /* Set global SHM for ust */
5631 if (strlen(wait_shm_path
) == 0) {
5632 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5633 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5636 goto exit_init_data
;
5640 if (strlen(health_unix_sock_path
) == 0) {
5641 ret
= snprintf(health_unix_sock_path
,
5642 sizeof(health_unix_sock_path
),
5643 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5646 goto exit_init_data
;
5650 /* Setup kernel consumerd path */
5651 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5652 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5655 goto exit_init_data
;
5657 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5658 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5661 goto exit_init_data
;
5664 DBG2("Kernel consumer err path: %s",
5665 kconsumer_data
.err_unix_sock_path
);
5666 DBG2("Kernel consumer cmd path: %s",
5667 kconsumer_data
.cmd_unix_sock_path
);
5669 home_path
= utils_get_home_dir();
5670 if (home_path
== NULL
) {
5671 /* TODO: Add --socket PATH option */
5672 ERR("Can't get HOME directory for sockets creation.");
5674 goto exit_init_data
;
5678 * Create rundir from home path. This will create something like
5681 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5684 goto exit_init_data
;
5687 if (create_lttng_rundir(rundir
)) {
5689 goto exit_init_data
;
5692 if (strlen(apps_unix_sock_path
) == 0) {
5693 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5694 DEFAULT_HOME_APPS_UNIX_SOCK
,
5698 goto exit_init_data
;
5702 /* Set the cli tool unix socket path */
5703 if (strlen(client_unix_sock_path
) == 0) {
5704 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5705 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5709 goto exit_init_data
;
5713 /* Set global SHM for ust */
5714 if (strlen(wait_shm_path
) == 0) {
5715 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5716 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5720 goto exit_init_data
;
5724 /* Set health check Unix path */
5725 if (strlen(health_unix_sock_path
) == 0) {
5726 ret
= snprintf(health_unix_sock_path
,
5727 sizeof(health_unix_sock_path
),
5728 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5732 goto exit_init_data
;
5737 lockfile_fd
= create_lockfile();
5738 if (lockfile_fd
< 0) {
5740 goto exit_init_data
;
5743 /* Set consumer initial state */
5744 kernel_consumerd_state
= CONSUMER_STOPPED
;
5745 ust_consumerd_state
= CONSUMER_STOPPED
;
5747 DBG("Client socket path %s", client_unix_sock_path
);
5748 DBG("Application socket path %s", apps_unix_sock_path
);
5749 DBG("Application wait path %s", wait_shm_path
);
5750 DBG("LTTng run directory path: %s", rundir
);
5752 /* 32 bits consumerd path setup */
5753 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5754 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5756 PERROR("snprintf 32-bit consumer error socket path");
5758 goto exit_init_data
;
5760 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5761 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5763 PERROR("snprintf 32-bit consumer command socket path");
5765 goto exit_init_data
;
5768 DBG2("UST consumer 32 bits err path: %s",
5769 ustconsumer32_data
.err_unix_sock_path
);
5770 DBG2("UST consumer 32 bits cmd path: %s",
5771 ustconsumer32_data
.cmd_unix_sock_path
);
5773 /* 64 bits consumerd path setup */
5774 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5775 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5777 PERROR("snprintf 64-bit consumer error socket path");
5779 goto exit_init_data
;
5781 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5782 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5784 PERROR("snprintf 64-bit consumer command socket path");
5786 goto exit_init_data
;
5789 DBG2("UST consumer 64 bits err path: %s",
5790 ustconsumer64_data
.err_unix_sock_path
);
5791 DBG2("UST consumer 64 bits cmd path: %s",
5792 ustconsumer64_data
.cmd_unix_sock_path
);
5795 * See if daemon already exist.
5797 if (check_existing_daemon()) {
5798 ERR("Already running daemon.\n");
5800 * We do not goto exit because we must not cleanup()
5801 * because a daemon is already running.
5804 goto exit_init_data
;
5808 * Init UST app hash table. Alloc hash table before this point since
5809 * cleanup() can get called after that point.
5811 if (ust_app_ht_alloc()) {
5812 ERR("Failed to allocate UST app hash table");
5814 goto exit_init_data
;
5818 * Initialize agent app hash table. We allocate the hash table here
5819 * since cleanup() can get called after this point.
5821 if (agent_app_ht_alloc()) {
5822 ERR("Failed to allocate Agent app hash table");
5824 goto exit_init_data
;
5828 * These actions must be executed as root. We do that *after* setting up
5829 * the sockets path because we MUST make the check for another daemon using
5830 * those paths *before* trying to set the kernel consumer sockets and init
5834 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5836 goto exit_init_data
;
5839 /* Setup kernel tracer */
5840 if (!opt_no_kernel
) {
5841 init_kernel_tracer();
5842 if (kernel_tracer_fd
>= 0) {
5843 ret
= syscall_init_table();
5845 ERR("Unable to populate syscall table. "
5846 "Syscall tracing won't work "
5847 "for this session daemon.");
5852 /* Set ulimit for open files */
5855 /* init lttng_fd tracking must be done after set_ulimit. */
5858 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5860 goto exit_init_data
;
5863 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5865 goto exit_init_data
;
5868 /* Setup the needed unix socket */
5869 if (init_daemon_socket()) {
5871 goto exit_init_data
;
5874 /* Set credentials to socket */
5875 if (is_root
&& set_permissions(rundir
)) {
5877 goto exit_init_data
;
5880 /* Get parent pid if -S, --sig-parent is specified. */
5881 if (opt_sig_parent
) {
5885 /* Setup the kernel pipe for waking up the kernel thread */
5886 if (is_root
&& !opt_no_kernel
) {
5887 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5889 goto exit_init_data
;
5893 /* Setup the thread apps communication pipe. */
5894 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5896 goto exit_init_data
;
5899 /* Setup the thread apps notify communication pipe. */
5900 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5902 goto exit_init_data
;
5905 /* Initialize global buffer per UID and PID registry. */
5906 buffer_reg_init_uid_registry();
5907 buffer_reg_init_pid_registry();
5909 /* Init UST command queue. */
5910 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5913 * Get session list pointer. This pointer MUST NOT be free'd. This list
5914 * is statically declared in session.c
5916 session_list_ptr
= session_get_list();
5920 /* Check for the application socket timeout env variable. */
5921 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5922 if (env_app_timeout
) {
5923 app_socket_timeout
= atoi(env_app_timeout
);
5925 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5928 ret
= write_pidfile();
5930 ERR("Error in write_pidfile");
5932 goto exit_init_data
;
5934 ret
= write_agent_port();
5936 ERR("Error in write_agent_port");
5938 goto exit_init_data
;
5941 /* Initialize communication library */
5943 /* Initialize TCP timeout values */
5944 lttcomm_inet_init();
5946 if (load_session_init_data(&load_info
) < 0) {
5948 goto exit_init_data
;
5950 load_info
->path
= opt_load_session_path
;
5952 /* Create health-check thread */
5953 ret
= pthread_create(&health_thread
, NULL
,
5954 thread_manage_health
, (void *) NULL
);
5957 PERROR("pthread_create health");
5962 /* Create thread to manage the client socket */
5963 ret
= pthread_create(&client_thread
, NULL
,
5964 thread_manage_clients
, (void *) NULL
);
5967 PERROR("pthread_create clients");
5972 /* Create thread to dispatch registration */
5973 ret
= pthread_create(&dispatch_thread
, NULL
,
5974 thread_dispatch_ust_registration
, (void *) NULL
);
5977 PERROR("pthread_create dispatch");
5982 /* Create thread to manage application registration. */
5983 ret
= pthread_create(®_apps_thread
, NULL
,
5984 thread_registration_apps
, (void *) NULL
);
5987 PERROR("pthread_create registration");
5992 /* Create thread to manage application socket */
5993 ret
= pthread_create(&apps_thread
, NULL
,
5994 thread_manage_apps
, (void *) NULL
);
5997 PERROR("pthread_create apps");
6002 /* Create thread to manage application notify socket */
6003 ret
= pthread_create(&apps_notify_thread
, NULL
,
6004 ust_thread_manage_notify
, (void *) NULL
);
6007 PERROR("pthread_create notify");
6009 goto exit_apps_notify
;
6012 /* Create agent registration thread. */
6013 ret
= pthread_create(&agent_reg_thread
, NULL
,
6014 agent_thread_manage_registration
, (void *) NULL
);
6017 PERROR("pthread_create agent");
6019 goto exit_agent_reg
;
6022 /* Don't start this thread if kernel tracing is not requested nor root */
6023 if (is_root
&& !opt_no_kernel
) {
6024 /* Create kernel thread to manage kernel event */
6025 ret
= pthread_create(&kernel_thread
, NULL
,
6026 thread_manage_kernel
, (void *) NULL
);
6029 PERROR("pthread_create kernel");
6035 /* Create session loading thread. */
6036 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
6040 PERROR("pthread_create load_session_thread");
6042 goto exit_load_session
;
6046 * This is where we start awaiting program completion (e.g. through
6047 * signal that asks threads to teardown).
6050 ret
= pthread_join(load_session_thread
, &status
);
6053 PERROR("pthread_join load_session_thread");
6058 if (is_root
&& !opt_no_kernel
) {
6059 ret
= pthread_join(kernel_thread
, &status
);
6062 PERROR("pthread_join");
6068 ret
= pthread_join(agent_reg_thread
, &status
);
6071 PERROR("pthread_join agent");
6076 ret
= pthread_join(apps_notify_thread
, &status
);
6079 PERROR("pthread_join apps notify");
6084 ret
= pthread_join(apps_thread
, &status
);
6087 PERROR("pthread_join apps");
6092 ret
= pthread_join(reg_apps_thread
, &status
);
6095 PERROR("pthread_join");
6101 * Join dispatch thread after joining reg_apps_thread to ensure
6102 * we don't leak applications in the queue.
6104 ret
= pthread_join(dispatch_thread
, &status
);
6107 PERROR("pthread_join");
6112 ret
= pthread_join(client_thread
, &status
);
6115 PERROR("pthread_join");
6120 ret
= pthread_join(health_thread
, &status
);
6123 PERROR("pthread_join health thread");
6130 * sessiond_cleanup() is called when no other thread is running, except
6131 * the ht_cleanup thread, which is needed to destroy the hash tables.
6133 rcu_thread_online();
6135 rcu_thread_offline();
6136 rcu_unregister_thread();
6138 ret
= notify_thread_pipe(ht_cleanup_quit_pipe
[1]);
6140 ERR("write error on ht_cleanup quit pipe");
6144 ret
= pthread_join(ht_cleanup_thread
, &status
);
6147 PERROR("pthread_join ht cleanup thread");
6153 utils_close_pipe(ht_cleanup_pipe
);
6154 exit_ht_cleanup_pipe
:
6157 * Close the ht_cleanup quit pipe.
6159 utils_close_pipe(ht_cleanup_quit_pipe
);
6160 exit_ht_cleanup_quit_pipe
:
6162 health_app_destroy(health_sessiond
);
6163 exit_health_sessiond_cleanup
:
6164 exit_create_run_as_worker_cleanup
:
6167 /* Ensure all prior call_rcu are done. */
6170 sessiond_cleanup_options();
6172 exit_set_signal_handler
: