2 * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
33 #include <sys/mount.h>
34 #include <sys/resource.h>
35 #include <sys/socket.h>
37 #include <sys/types.h>
39 #include <urcu/uatomic.h>
43 #include <common/common.h>
44 #include <common/compat/socket.h>
45 #include <common/compat/getenv.h>
46 #include <common/defaults.h>
47 #include <common/kernel-consumer/kernel-consumer.h>
48 #include <common/futex.h>
49 #include <common/relayd/relayd.h>
50 #include <common/utils.h>
51 #include <common/daemonize.h>
52 #include <common/config/config.h>
54 #include "lttng-sessiond.h"
55 #include "buffer-registry.h"
62 #include "kernel-consumer.h"
66 #include "ust-consumer.h"
69 #include "health-sessiond.h"
70 #include "testpoint.h"
71 #include "ust-thread.h"
72 #include "agent-thread.h"
74 #include "load-session-thread.h"
77 #define CONSUMERD_FILE "lttng-consumerd"
80 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
81 static int tracing_group_name_override
;
82 static char *opt_pidfile
;
83 static int opt_sig_parent
;
84 static int opt_verbose_consumer
;
85 static int opt_daemon
, opt_background
;
86 static int opt_no_kernel
;
87 static char *opt_load_session_path
;
88 static pid_t ppid
; /* Parent PID for --sig-parent option */
89 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
91 static int lockfile_fd
= -1;
93 /* Set to 1 when a SIGUSR1 signal is received. */
94 static int recv_child_signal
;
97 * Consumer daemon specific control data. Every value not initialized here is
98 * set to 0 by the static definition.
100 static struct consumer_data kconsumer_data
= {
101 .type
= LTTNG_CONSUMER_KERNEL
,
102 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
103 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
106 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 .lock
= PTHREAD_MUTEX_INITIALIZER
,
108 .cond
= PTHREAD_COND_INITIALIZER
,
109 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
111 static struct consumer_data ustconsumer64_data
= {
112 .type
= LTTNG_CONSUMER64_UST
,
113 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
114 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
117 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
118 .lock
= PTHREAD_MUTEX_INITIALIZER
,
119 .cond
= PTHREAD_COND_INITIALIZER
,
120 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
122 static struct consumer_data ustconsumer32_data
= {
123 .type
= LTTNG_CONSUMER32_UST
,
124 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
125 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
128 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
129 .lock
= PTHREAD_MUTEX_INITIALIZER
,
130 .cond
= PTHREAD_COND_INITIALIZER
,
131 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
134 /* Command line options */
135 static const struct option long_options
[] = {
136 { "client-sock", required_argument
, 0, 'c' },
137 { "apps-sock", required_argument
, 0, 'a' },
138 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
139 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
140 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
141 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
142 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
143 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
144 { "consumerd32-path", required_argument
, 0, '\0' },
145 { "consumerd32-libdir", required_argument
, 0, '\0' },
146 { "consumerd64-path", required_argument
, 0, '\0' },
147 { "consumerd64-libdir", required_argument
, 0, '\0' },
148 { "daemonize", no_argument
, 0, 'd' },
149 { "background", no_argument
, 0, 'b' },
150 { "sig-parent", no_argument
, 0, 'S' },
151 { "help", no_argument
, 0, 'h' },
152 { "group", required_argument
, 0, 'g' },
153 { "version", no_argument
, 0, 'V' },
154 { "quiet", no_argument
, 0, 'q' },
155 { "verbose", no_argument
, 0, 'v' },
156 { "verbose-consumer", no_argument
, 0, '\0' },
157 { "no-kernel", no_argument
, 0, '\0' },
158 { "pidfile", required_argument
, 0, 'p' },
159 { "agent-tcp-port", required_argument
, 0, '\0' },
160 { "config", required_argument
, 0, 'f' },
161 { "load", required_argument
, 0, 'l' },
162 { "kmod-probes", required_argument
, 0, '\0' },
163 { "extra-kmod-probes", required_argument
, 0, '\0' },
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options
[] = { "help", "version", "config" };
170 /* Shared between threads */
171 static int dispatch_thread_exit
;
173 /* Global application Unix socket path */
174 static char apps_unix_sock_path
[PATH_MAX
];
175 /* Global client Unix socket path */
176 static char client_unix_sock_path
[PATH_MAX
];
177 /* global wait shm path for UST */
178 static char wait_shm_path
[PATH_MAX
];
179 /* Global health check unix path */
180 static char health_unix_sock_path
[PATH_MAX
];
182 /* Sockets and FDs */
183 static int client_sock
= -1;
184 static int apps_sock
= -1;
185 int kernel_tracer_fd
= -1;
186 static int kernel_poll_pipe
[2] = { -1, -1 };
189 * Quit pipe for all threads. This permits a single cancellation point
190 * for all threads when receiving an event on the pipe.
192 static int thread_quit_pipe
[2] = { -1, -1 };
193 static int ht_cleanup_quit_pipe
[2] = { -1, -1 };
196 * This pipe is used to inform the thread managing application communication
197 * that a command is queued and ready to be processed.
199 static int apps_cmd_pipe
[2] = { -1, -1 };
201 int apps_cmd_notify_pipe
[2] = { -1, -1 };
203 /* Pthread, Mutexes and Semaphores */
204 static pthread_t apps_thread
;
205 static pthread_t apps_notify_thread
;
206 static pthread_t reg_apps_thread
;
207 static pthread_t client_thread
;
208 static pthread_t kernel_thread
;
209 static pthread_t dispatch_thread
;
210 static pthread_t health_thread
;
211 static pthread_t ht_cleanup_thread
;
212 static pthread_t agent_reg_thread
;
213 static pthread_t load_session_thread
;
216 * UST registration command queue. This queue is tied with a futex and uses a N
217 * wakers / 1 waiter implemented and detailed in futex.c/.h
219 * The thread_registration_apps and thread_dispatch_ust_registration uses this
220 * queue along with the wait/wake scheme. The thread_manage_apps receives down
221 * the line new application socket and monitors it for any I/O error or clean
222 * close that triggers an unregistration of the application.
224 static struct ust_cmd_queue ust_cmd_queue
;
227 * Pointer initialized before thread creation.
229 * This points to the tracing session list containing the session count and a
230 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
231 * MUST NOT be taken if you call a public function in session.c.
233 * The lock is nested inside the structure: session_list_ptr->lock. Please use
234 * session_lock_list and session_unlock_list for lock acquisition.
236 static struct ltt_session_list
*session_list_ptr
;
238 int ust_consumerd64_fd
= -1;
239 int ust_consumerd32_fd
= -1;
241 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
242 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
243 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
244 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
245 static int consumerd32_bin_override
;
246 static int consumerd64_bin_override
;
247 static int consumerd32_libdir_override
;
248 static int consumerd64_libdir_override
;
250 static const char *module_proc_lttng
= "/proc/lttng";
253 * Consumer daemon state which is changed when spawning it, killing it or in
254 * case of a fatal error.
256 enum consumerd_state
{
257 CONSUMER_STARTED
= 1,
258 CONSUMER_STOPPED
= 2,
263 * This consumer daemon state is used to validate if a client command will be
264 * able to reach the consumer. If not, the client is informed. For instance,
265 * doing a "lttng start" when the consumer state is set to ERROR will return an
266 * error to the client.
268 * The following example shows a possible race condition of this scheme:
270 * consumer thread error happens
272 * client cmd checks state -> still OK
273 * consumer thread exit, sets error
274 * client cmd try to talk to consumer
277 * However, since the consumer is a different daemon, we have no way of making
278 * sure the command will reach it safely even with this state flag. This is why
279 * we consider that up to the state validation during command processing, the
280 * command is safe. After that, we can not guarantee the correctness of the
281 * client request vis-a-vis the consumer.
283 static enum consumerd_state ust_consumerd_state
;
284 static enum consumerd_state kernel_consumerd_state
;
287 * Socket timeout for receiving and sending in seconds.
289 static int app_socket_timeout
;
291 /* Set in main() with the current page size. */
294 /* Application health monitoring */
295 struct health_app
*health_sessiond
;
297 /* Agent TCP port for registration. Used by the agent thread. */
298 unsigned int agent_tcp_port
= DEFAULT_AGENT_TCP_PORT
;
300 /* Am I root or not. */
301 int is_root
; /* Set to 1 if the daemon is running as root */
303 const char * const config_section_name
= "sessiond";
305 /* Load session thread information to operate. */
306 struct load_session_thread_data
*load_info
;
309 * Whether sessiond is ready for commands/health check requests.
310 * NR_LTTNG_SESSIOND_READY must match the number of calls to
311 * sessiond_notify_ready().
313 #define NR_LTTNG_SESSIOND_READY 3
314 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
316 /* Notify parents that we are ready for cmd and health check */
318 void sessiond_notify_ready(void)
320 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
322 * Notify parent pid that we are ready to accept command
323 * for client side. This ppid is the one from the
324 * external process that spawned us.
326 if (opt_sig_parent
) {
331 * Notify the parent of the fork() process that we are
334 if (opt_daemon
|| opt_background
) {
335 kill(child_ppid
, SIGUSR1
);
341 void setup_consumerd_path(void)
343 const char *bin
, *libdir
;
346 * Allow INSTALL_BIN_PATH to be used as a target path for the
347 * native architecture size consumer if CONFIG_CONSUMER*_PATH
348 * has not been defined.
350 #if (CAA_BITS_PER_LONG == 32)
351 if (!consumerd32_bin
[0]) {
352 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
354 if (!consumerd32_libdir
[0]) {
355 consumerd32_libdir
= INSTALL_LIB_PATH
;
357 #elif (CAA_BITS_PER_LONG == 64)
358 if (!consumerd64_bin
[0]) {
359 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
361 if (!consumerd64_libdir
[0]) {
362 consumerd64_libdir
= INSTALL_LIB_PATH
;
365 #error "Unknown bitness"
369 * runtime env. var. overrides the build default.
371 bin
= lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
373 consumerd32_bin
= bin
;
375 bin
= lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
377 consumerd64_bin
= bin
;
379 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
381 consumerd32_libdir
= libdir
;
383 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
385 consumerd64_libdir
= libdir
;
390 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
397 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
403 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
415 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
417 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
419 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
423 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
425 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event
*events
,
428 return __sessiond_set_thread_pollset(events
, size
,
429 ht_cleanup_quit_pipe
);
433 int __sessiond_check_thread_quit_pipe(int fd
, uint32_t events
, int a_pipe
)
435 if (fd
== a_pipe
&& (events
& LPOLLIN
)) {
442 * Check if the thread quit pipe was triggered.
444 * Return 1 if it was triggered else 0;
446 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
448 return __sessiond_check_thread_quit_pipe(fd
, events
,
449 thread_quit_pipe
[0]);
453 * Check if the ht_cleanup thread quit pipe was triggered.
455 * Return 1 if it was triggered else 0;
457 int sessiond_check_ht_cleanup_quit(int fd
, uint32_t events
)
459 return __sessiond_check_thread_quit_pipe(fd
, events
,
460 ht_cleanup_quit_pipe
[0]);
464 * Init thread quit pipe.
466 * Return -1 on error or 0 if all pipes are created.
468 static int __init_thread_quit_pipe(int *a_pipe
)
474 PERROR("thread quit pipe");
478 for (i
= 0; i
< 2; i
++) {
479 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
490 static int init_thread_quit_pipe(void)
492 return __init_thread_quit_pipe(thread_quit_pipe
);
495 static int init_ht_cleanup_quit_pipe(void)
497 return __init_thread_quit_pipe(ht_cleanup_quit_pipe
);
501 * Stop all threads by closing the thread quit pipe.
503 static void stop_threads(void)
507 /* Stopping all threads */
508 DBG("Terminating all threads");
509 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
511 ERR("write error on thread quit pipe");
514 /* Dispatch thread */
515 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
516 futex_nto1_wake(&ust_cmd_queue
.futex
);
520 * Close every consumer sockets.
522 static void close_consumer_sockets(void)
526 if (kconsumer_data
.err_sock
>= 0) {
527 ret
= close(kconsumer_data
.err_sock
);
529 PERROR("kernel consumer err_sock close");
532 if (ustconsumer32_data
.err_sock
>= 0) {
533 ret
= close(ustconsumer32_data
.err_sock
);
535 PERROR("UST consumerd32 err_sock close");
538 if (ustconsumer64_data
.err_sock
>= 0) {
539 ret
= close(ustconsumer64_data
.err_sock
);
541 PERROR("UST consumerd64 err_sock close");
544 if (kconsumer_data
.cmd_sock
>= 0) {
545 ret
= close(kconsumer_data
.cmd_sock
);
547 PERROR("kernel consumer cmd_sock close");
550 if (ustconsumer32_data
.cmd_sock
>= 0) {
551 ret
= close(ustconsumer32_data
.cmd_sock
);
553 PERROR("UST consumerd32 cmd_sock close");
556 if (ustconsumer64_data
.cmd_sock
>= 0) {
557 ret
= close(ustconsumer64_data
.cmd_sock
);
559 PERROR("UST consumerd64 cmd_sock close");
565 * Generate the full lock file path using the rundir.
567 * Return the snprintf() return value thus a negative value is an error.
569 static int generate_lock_file_path(char *path
, size_t len
)
576 /* Build lockfile path from rundir. */
577 ret
= snprintf(path
, len
, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE
, rundir
);
579 PERROR("snprintf lockfile path");
586 * Wait on consumer process termination.
588 * Need to be called with the consumer data lock held or from a context
589 * ensuring no concurrent access to data (e.g: cleanup).
591 static void wait_consumer(struct consumer_data
*consumer_data
)
596 if (consumer_data
->pid
<= 0) {
600 DBG("Waiting for complete teardown of consumerd (PID: %d)",
602 ret
= waitpid(consumer_data
->pid
, &status
, 0);
604 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
606 if (!WIFEXITED(status
)) {
607 ERR("consumerd termination with error: %d",
610 consumer_data
->pid
= 0;
614 * Cleanup the session daemon's data structures.
616 static void sessiond_cleanup(void)
619 struct ltt_session
*sess
, *stmp
;
622 DBG("Cleanup sessiond");
625 * Close the thread quit pipe. It has already done its job,
626 * since we are now called.
628 utils_close_pipe(thread_quit_pipe
);
631 * If opt_pidfile is undefined, the default file will be wiped when
632 * removing the rundir.
635 ret
= remove(opt_pidfile
);
637 PERROR("remove pidfile %s", opt_pidfile
);
641 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
644 snprintf(path
, PATH_MAX
,
646 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
647 DBG("Removing %s", path
);
650 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
651 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
);
652 DBG("Removing %s", path
);
656 snprintf(path
, PATH_MAX
,
657 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
659 DBG("Removing %s", path
);
662 snprintf(path
, PATH_MAX
,
663 DEFAULT_KCONSUMERD_PATH
,
665 DBG("Removing directory %s", path
);
668 /* ust consumerd 32 */
669 snprintf(path
, PATH_MAX
,
670 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
672 DBG("Removing %s", path
);
675 snprintf(path
, PATH_MAX
,
676 DEFAULT_USTCONSUMERD32_PATH
,
678 DBG("Removing directory %s", path
);
681 /* ust consumerd 64 */
682 snprintf(path
, PATH_MAX
,
683 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
685 DBG("Removing %s", path
);
688 snprintf(path
, PATH_MAX
,
689 DEFAULT_USTCONSUMERD64_PATH
,
691 DBG("Removing directory %s", path
);
694 DBG("Cleaning up all sessions");
696 /* Destroy session list mutex */
697 if (session_list_ptr
!= NULL
) {
698 pthread_mutex_destroy(&session_list_ptr
->lock
);
700 /* Cleanup ALL session */
701 cds_list_for_each_entry_safe(sess
, stmp
,
702 &session_list_ptr
->head
, list
) {
703 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
707 wait_consumer(&kconsumer_data
);
708 wait_consumer(&ustconsumer64_data
);
709 wait_consumer(&ustconsumer32_data
);
711 DBG("Cleaning up all agent apps");
712 agent_app_ht_clean();
714 DBG("Closing all UST sockets");
715 ust_app_clean_list();
716 buffer_reg_destroy_registries();
718 if (is_root
&& !opt_no_kernel
) {
719 DBG2("Closing kernel fd");
720 if (kernel_tracer_fd
>= 0) {
721 ret
= close(kernel_tracer_fd
);
726 DBG("Unloading kernel modules");
727 modprobe_remove_lttng_all();
731 close_consumer_sockets();
734 load_session_destroy_data(load_info
);
739 * Cleanup lock file by deleting it and finaly closing it which will
740 * release the file system lock.
742 if (lockfile_fd
>= 0) {
743 char lockfile_path
[PATH_MAX
];
745 ret
= generate_lock_file_path(lockfile_path
,
746 sizeof(lockfile_path
));
748 ret
= remove(lockfile_path
);
750 PERROR("remove lock file");
752 ret
= close(lockfile_fd
);
754 PERROR("close lock file");
760 * We do NOT rmdir rundir because there are other processes
761 * using it, for instance lttng-relayd, which can start in
762 * parallel with this teardown.
769 * Cleanup the daemon's option data structures.
771 static void sessiond_cleanup_options(void)
773 DBG("Cleaning up options");
776 * If the override option is set, the pointer points to a *non* const
777 * thus freeing it even though the variable type is set to const.
779 if (tracing_group_name_override
) {
780 free((void *) tracing_group_name
);
782 if (consumerd32_bin_override
) {
783 free((void *) consumerd32_bin
);
785 if (consumerd64_bin_override
) {
786 free((void *) consumerd64_bin
);
788 if (consumerd32_libdir_override
) {
789 free((void *) consumerd32_libdir
);
791 if (consumerd64_libdir_override
) {
792 free((void *) consumerd64_libdir
);
796 free(opt_load_session_path
);
797 free(kmod_probes_list
);
798 free(kmod_extra_probes_list
);
801 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
802 "Matthew, BEET driven development works!%c[%dm",
803 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
808 * Send data on a unix socket using the liblttsessiondcomm API.
810 * Return lttcomm error code.
812 static int send_unix_sock(int sock
, void *buf
, size_t len
)
814 /* Check valid length */
819 return lttcomm_send_unix_sock(sock
, buf
, len
);
823 * Free memory of a command context structure.
825 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
827 DBG("Clean command context structure");
829 if ((*cmd_ctx
)->llm
) {
830 free((*cmd_ctx
)->llm
);
832 if ((*cmd_ctx
)->lsm
) {
833 free((*cmd_ctx
)->lsm
);
841 * Notify UST applications using the shm mmap futex.
843 static int notify_ust_apps(int active
)
847 DBG("Notifying applications of session daemon state: %d", active
);
849 /* See shm.c for this call implying mmap, shm and futex calls */
850 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
851 if (wait_shm_mmap
== NULL
) {
855 /* Wake waiting process */
856 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
858 /* Apps notified successfully */
866 * Setup the outgoing data buffer for the response (llm) by allocating the
867 * right amount of memory and copying the original information from the lsm
870 * Return total size of the buffer pointed by buf.
872 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
878 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
879 if (cmd_ctx
->llm
== NULL
) {
885 /* Copy common data */
886 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
887 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
889 cmd_ctx
->llm
->data_size
= size
;
890 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
899 * Update the kernel poll set of all channel fd available over all tracing
900 * session. Add the wakeup pipe at the end of the set.
902 static int update_kernel_poll(struct lttng_poll_event
*events
)
905 struct ltt_session
*session
;
906 struct ltt_kernel_channel
*channel
;
908 DBG("Updating kernel poll set");
911 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
912 session_lock(session
);
913 if (session
->kernel_session
== NULL
) {
914 session_unlock(session
);
918 cds_list_for_each_entry(channel
,
919 &session
->kernel_session
->channel_list
.head
, list
) {
920 /* Add channel fd to the kernel poll set */
921 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
923 session_unlock(session
);
926 DBG("Channel fd %d added to kernel set", channel
->fd
);
928 session_unlock(session
);
930 session_unlock_list();
935 session_unlock_list();
940 * Find the channel fd from 'fd' over all tracing session. When found, check
941 * for new channel stream and send those stream fds to the kernel consumer.
943 * Useful for CPU hotplug feature.
945 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
948 struct ltt_session
*session
;
949 struct ltt_kernel_session
*ksess
;
950 struct ltt_kernel_channel
*channel
;
952 DBG("Updating kernel streams for channel fd %d", fd
);
955 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
956 session_lock(session
);
957 if (session
->kernel_session
== NULL
) {
958 session_unlock(session
);
961 ksess
= session
->kernel_session
;
963 cds_list_for_each_entry(channel
,
964 &ksess
->channel_list
.head
, list
) {
965 struct lttng_ht_iter iter
;
966 struct consumer_socket
*socket
;
968 if (channel
->fd
!= fd
) {
971 DBG("Channel found, updating kernel streams");
972 ret
= kernel_open_channel_stream(channel
);
976 /* Update the stream global counter */
977 ksess
->stream_count_global
+= ret
;
980 * Have we already sent fds to the consumer? If yes, it
981 * means that tracing is started so it is safe to send
982 * our updated stream fds.
984 if (ksess
->consumer_fds_sent
!= 1
985 || ksess
->consumer
== NULL
) {
991 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
992 &iter
.iter
, socket
, node
.node
) {
993 pthread_mutex_lock(socket
->lock
);
994 ret
= kernel_consumer_send_channel_stream(socket
,
996 session
->output_traces
? 1 : 0);
997 pthread_mutex_unlock(socket
->lock
);
1005 session_unlock(session
);
1007 session_unlock_list();
1011 session_unlock(session
);
1012 session_unlock_list();
1017 * For each tracing session, update newly registered apps. The session list
1018 * lock MUST be acquired before calling this.
1020 static void update_ust_app(int app_sock
)
1022 struct ltt_session
*sess
, *stmp
;
1024 /* Consumer is in an ERROR state. Stop any application update. */
1025 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
1026 /* Stop the update process since the consumer is dead. */
1030 /* For all tracing session(s) */
1031 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
1032 struct ust_app
*app
;
1035 if (!sess
->ust_session
) {
1036 goto unlock_session
;
1040 assert(app_sock
>= 0);
1041 app
= ust_app_find_by_sock(app_sock
);
1044 * Application can be unregistered before so
1045 * this is possible hence simply stopping the
1048 DBG3("UST app update failed to find app sock %d",
1052 ust_app_global_update(sess
->ust_session
, app
);
1056 session_unlock(sess
);
1061 * This thread manage event coming from the kernel.
1063 * Features supported in this thread:
1066 static void *thread_manage_kernel(void *data
)
1068 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1069 uint32_t revents
, nb_fd
;
1071 struct lttng_poll_event events
;
1073 DBG("[thread] Thread manage kernel started");
1075 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1078 * This first step of the while is to clean this structure which could free
1079 * non NULL pointers so initialize it before the loop.
1081 lttng_poll_init(&events
);
1083 if (testpoint(sessiond_thread_manage_kernel
)) {
1084 goto error_testpoint
;
1087 health_code_update();
1089 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1090 goto error_testpoint
;
1094 health_code_update();
1096 if (update_poll_flag
== 1) {
1097 /* Clean events object. We are about to populate it again. */
1098 lttng_poll_clean(&events
);
1100 ret
= sessiond_set_thread_pollset(&events
, 2);
1102 goto error_poll_create
;
1105 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1110 /* This will add the available kernel channel if any. */
1111 ret
= update_kernel_poll(&events
);
1115 update_poll_flag
= 0;
1118 DBG("Thread kernel polling");
1120 /* Poll infinite value of time */
1122 health_poll_entry();
1123 ret
= lttng_poll_wait(&events
, -1);
1124 DBG("Thread kernel return from poll on %d fds",
1125 LTTNG_POLL_GETNB(&events
));
1129 * Restart interrupted system call.
1131 if (errno
== EINTR
) {
1135 } else if (ret
== 0) {
1136 /* Should not happen since timeout is infinite */
1137 ERR("Return value of poll is 0 with an infinite timeout.\n"
1138 "This should not have happened! Continuing...");
1144 for (i
= 0; i
< nb_fd
; i
++) {
1145 /* Fetch once the poll data */
1146 revents
= LTTNG_POLL_GETEV(&events
, i
);
1147 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1149 health_code_update();
1152 /* No activity for this FD (poll implementation). */
1156 /* Thread quit pipe has been closed. Killing thread. */
1157 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1163 /* Check for data on kernel pipe */
1164 if (revents
& LPOLLIN
) {
1165 if (pollfd
== kernel_poll_pipe
[0]) {
1166 (void) lttng_read(kernel_poll_pipe
[0],
1169 * Ret value is useless here, if this pipe gets any actions an
1170 * update is required anyway.
1172 update_poll_flag
= 1;
1176 * New CPU detected by the kernel. Adding kernel stream to
1177 * kernel session and updating the kernel consumer
1179 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1185 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1186 update_poll_flag
= 1;
1189 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1197 lttng_poll_clean(&events
);
1200 utils_close_pipe(kernel_poll_pipe
);
1201 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1204 ERR("Health error occurred in %s", __func__
);
1205 WARN("Kernel thread died unexpectedly. "
1206 "Kernel tracing can continue but CPU hotplug is disabled.");
1208 health_unregister(health_sessiond
);
1209 DBG("Kernel thread dying");
1214 * Signal pthread condition of the consumer data that the thread.
1216 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1218 pthread_mutex_lock(&data
->cond_mutex
);
1221 * The state is set before signaling. It can be any value, it's the waiter
1222 * job to correctly interpret this condition variable associated to the
1223 * consumer pthread_cond.
1225 * A value of 0 means that the corresponding thread of the consumer data
1226 * was not started. 1 indicates that the thread has started and is ready
1227 * for action. A negative value means that there was an error during the
1230 data
->consumer_thread_is_ready
= state
;
1231 (void) pthread_cond_signal(&data
->cond
);
1233 pthread_mutex_unlock(&data
->cond_mutex
);
1237 * This thread manage the consumer error sent back to the session daemon.
1239 static void *thread_manage_consumer(void *data
)
1241 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1242 uint32_t revents
, nb_fd
;
1243 enum lttcomm_return_code code
;
1244 struct lttng_poll_event events
;
1245 struct consumer_data
*consumer_data
= data
;
1247 DBG("[thread] Manage consumer started");
1249 rcu_register_thread();
1250 rcu_thread_online();
1252 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1254 health_code_update();
1257 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1258 * metadata_sock. Nothing more will be added to this poll set.
1260 ret
= sessiond_set_thread_pollset(&events
, 3);
1266 * The error socket here is already in a listening state which was done
1267 * just before spawning this thread to avoid a race between the consumer
1268 * daemon exec trying to connect and the listen() call.
1270 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1275 health_code_update();
1277 /* Infinite blocking call, waiting for transmission */
1279 health_poll_entry();
1281 if (testpoint(sessiond_thread_manage_consumer
)) {
1285 ret
= lttng_poll_wait(&events
, -1);
1289 * Restart interrupted system call.
1291 if (errno
== EINTR
) {
1299 for (i
= 0; i
< nb_fd
; i
++) {
1300 /* Fetch once the poll data */
1301 revents
= LTTNG_POLL_GETEV(&events
, i
);
1302 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1304 health_code_update();
1307 /* No activity for this FD (poll implementation). */
1311 /* Thread quit pipe has been closed. Killing thread. */
1312 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1318 /* Event on the registration socket */
1319 if (pollfd
== consumer_data
->err_sock
) {
1320 if (revents
& LPOLLIN
) {
1322 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1323 ERR("consumer err socket poll error");
1326 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1332 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1338 * Set the CLOEXEC flag. Return code is useless because either way, the
1341 (void) utils_set_fd_cloexec(sock
);
1343 health_code_update();
1345 DBG2("Receiving code from consumer err_sock");
1347 /* Getting status code from kconsumerd */
1348 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1349 sizeof(enum lttcomm_return_code
));
1354 health_code_update();
1355 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1356 /* Connect both socket, command and metadata. */
1357 consumer_data
->cmd_sock
=
1358 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1359 consumer_data
->metadata_fd
=
1360 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1361 if (consumer_data
->cmd_sock
< 0
1362 || consumer_data
->metadata_fd
< 0) {
1363 PERROR("consumer connect cmd socket");
1364 /* On error, signal condition and quit. */
1365 signal_consumer_condition(consumer_data
, -1);
1368 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1369 /* Create metadata socket lock. */
1370 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1371 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1372 PERROR("zmalloc pthread mutex");
1376 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1378 signal_consumer_condition(consumer_data
, 1);
1379 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1380 DBG("Consumer metadata socket ready (fd: %d)",
1381 consumer_data
->metadata_fd
);
1383 ERR("consumer error when waiting for SOCK_READY : %s",
1384 lttcomm_get_readable_code(-code
));
1388 /* Remove the consumerd error sock since we've established a connexion */
1389 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1394 /* Add new accepted error socket. */
1395 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1400 /* Add metadata socket that is successfully connected. */
1401 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1402 LPOLLIN
| LPOLLRDHUP
);
1407 health_code_update();
1409 /* Infinite blocking call, waiting for transmission */
1412 health_code_update();
1414 /* Exit the thread because the thread quit pipe has been triggered. */
1416 /* Not a health error. */
1421 health_poll_entry();
1422 ret
= lttng_poll_wait(&events
, -1);
1426 * Restart interrupted system call.
1428 if (errno
== EINTR
) {
1436 for (i
= 0; i
< nb_fd
; i
++) {
1437 /* Fetch once the poll data */
1438 revents
= LTTNG_POLL_GETEV(&events
, i
);
1439 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1441 health_code_update();
1444 /* No activity for this FD (poll implementation). */
1449 * Thread quit pipe has been triggered, flag that we should stop
1450 * but continue the current loop to handle potential data from
1453 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1455 if (pollfd
== sock
) {
1456 /* Event on the consumerd socket */
1457 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1458 && !(revents
& LPOLLIN
)) {
1459 ERR("consumer err socket second poll error");
1462 health_code_update();
1463 /* Wait for any kconsumerd error */
1464 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1465 sizeof(enum lttcomm_return_code
));
1467 ERR("consumer closed the command socket");
1471 ERR("consumer return code : %s",
1472 lttcomm_get_readable_code(-code
));
1475 } else if (pollfd
== consumer_data
->metadata_fd
) {
1476 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1477 && !(revents
& LPOLLIN
)) {
1478 ERR("consumer err metadata socket second poll error");
1481 /* UST metadata requests */
1482 ret
= ust_consumer_metadata_request(
1483 &consumer_data
->metadata_sock
);
1485 ERR("Handling metadata request");
1489 /* No need for an else branch all FDs are tested prior. */
1491 health_code_update();
1497 * We lock here because we are about to close the sockets and some other
1498 * thread might be using them so get exclusive access which will abort all
1499 * other consumer command by other threads.
1501 pthread_mutex_lock(&consumer_data
->lock
);
1503 /* Immediately set the consumerd state to stopped */
1504 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1505 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1506 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1507 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1508 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1510 /* Code flow error... */
1514 if (consumer_data
->err_sock
>= 0) {
1515 ret
= close(consumer_data
->err_sock
);
1519 consumer_data
->err_sock
= -1;
1521 if (consumer_data
->cmd_sock
>= 0) {
1522 ret
= close(consumer_data
->cmd_sock
);
1526 consumer_data
->cmd_sock
= -1;
1528 if (consumer_data
->metadata_sock
.fd_ptr
&&
1529 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1530 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1542 unlink(consumer_data
->err_unix_sock_path
);
1543 unlink(consumer_data
->cmd_unix_sock_path
);
1544 pthread_mutex_unlock(&consumer_data
->lock
);
1546 /* Cleanup metadata socket mutex. */
1547 if (consumer_data
->metadata_sock
.lock
) {
1548 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1549 free(consumer_data
->metadata_sock
.lock
);
1551 lttng_poll_clean(&events
);
1555 ERR("Health error occurred in %s", __func__
);
1557 health_unregister(health_sessiond
);
1558 DBG("consumer thread cleanup completed");
1560 rcu_thread_offline();
1561 rcu_unregister_thread();
1567 * This thread manage application communication.
1569 static void *thread_manage_apps(void *data
)
1571 int i
, ret
, pollfd
, err
= -1;
1573 uint32_t revents
, nb_fd
;
1574 struct lttng_poll_event events
;
1576 DBG("[thread] Manage application started");
1578 rcu_register_thread();
1579 rcu_thread_online();
1581 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1583 if (testpoint(sessiond_thread_manage_apps
)) {
1584 goto error_testpoint
;
1587 health_code_update();
1589 ret
= sessiond_set_thread_pollset(&events
, 2);
1591 goto error_poll_create
;
1594 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1599 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1603 health_code_update();
1606 DBG("Apps thread polling");
1608 /* Inifinite blocking call, waiting for transmission */
1610 health_poll_entry();
1611 ret
= lttng_poll_wait(&events
, -1);
1612 DBG("Apps thread return from poll on %d fds",
1613 LTTNG_POLL_GETNB(&events
));
1617 * Restart interrupted system call.
1619 if (errno
== EINTR
) {
1627 for (i
= 0; i
< nb_fd
; i
++) {
1628 /* Fetch once the poll data */
1629 revents
= LTTNG_POLL_GETEV(&events
, i
);
1630 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1632 health_code_update();
1635 /* No activity for this FD (poll implementation). */
1639 /* Thread quit pipe has been closed. Killing thread. */
1640 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1646 /* Inspect the apps cmd pipe */
1647 if (pollfd
== apps_cmd_pipe
[0]) {
1648 if (revents
& LPOLLIN
) {
1652 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1653 if (size_ret
< sizeof(sock
)) {
1654 PERROR("read apps cmd pipe");
1658 health_code_update();
1661 * Since this is a command socket (write then read),
1662 * we only monitor the error events of the socket.
1664 ret
= lttng_poll_add(&events
, sock
,
1665 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1670 DBG("Apps with sock %d added to poll set", sock
);
1671 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1672 ERR("Apps command pipe error");
1675 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1680 * At this point, we know that a registered application made
1681 * the event at poll_wait.
1683 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1684 /* Removing from the poll set */
1685 ret
= lttng_poll_del(&events
, pollfd
);
1690 /* Socket closed on remote end. */
1691 ust_app_unregister(pollfd
);
1693 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1698 health_code_update();
1704 lttng_poll_clean(&events
);
1707 utils_close_pipe(apps_cmd_pipe
);
1708 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1711 * We don't clean the UST app hash table here since already registered
1712 * applications can still be controlled so let them be until the session
1713 * daemon dies or the applications stop.
1718 ERR("Health error occurred in %s", __func__
);
1720 health_unregister(health_sessiond
);
1721 DBG("Application communication apps thread cleanup complete");
1722 rcu_thread_offline();
1723 rcu_unregister_thread();
1728 * Send a socket to a thread This is called from the dispatch UST registration
1729 * thread once all sockets are set for the application.
1731 * The sock value can be invalid, we don't really care, the thread will handle
1732 * it and make the necessary cleanup if so.
1734 * On success, return 0 else a negative value being the errno message of the
1737 static int send_socket_to_thread(int fd
, int sock
)
1742 * It's possible that the FD is set as invalid with -1 concurrently just
1743 * before calling this function being a shutdown state of the thread.
1750 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1751 if (ret
< sizeof(sock
)) {
1752 PERROR("write apps pipe %d", fd
);
1759 /* All good. Don't send back the write positive ret value. */
1766 * Sanitize the wait queue of the dispatch registration thread meaning removing
1767 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1768 * notify socket is never received.
1770 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1772 int ret
, nb_fd
= 0, i
;
1773 unsigned int fd_added
= 0;
1774 struct lttng_poll_event events
;
1775 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1779 lttng_poll_init(&events
);
1781 /* Just skip everything for an empty queue. */
1782 if (!wait_queue
->count
) {
1786 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1791 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1792 &wait_queue
->head
, head
) {
1793 assert(wait_node
->app
);
1794 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1795 LPOLLHUP
| LPOLLERR
);
1808 * Poll but don't block so we can quickly identify the faulty events and
1809 * clean them afterwards from the wait queue.
1811 ret
= lttng_poll_wait(&events
, 0);
1817 for (i
= 0; i
< nb_fd
; i
++) {
1818 /* Get faulty FD. */
1819 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1820 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1823 /* No activity for this FD (poll implementation). */
1827 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1828 &wait_queue
->head
, head
) {
1829 if (pollfd
== wait_node
->app
->sock
&&
1830 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1831 cds_list_del(&wait_node
->head
);
1832 wait_queue
->count
--;
1833 ust_app_destroy(wait_node
->app
);
1837 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1844 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1848 lttng_poll_clean(&events
);
1852 lttng_poll_clean(&events
);
1854 ERR("Unable to sanitize wait queue");
1859 * Dispatch request from the registration threads to the application
1860 * communication thread.
1862 static void *thread_dispatch_ust_registration(void *data
)
1865 struct cds_wfcq_node
*node
;
1866 struct ust_command
*ust_cmd
= NULL
;
1867 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1868 struct ust_reg_wait_queue wait_queue
= {
1872 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1874 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1875 goto error_testpoint
;
1878 health_code_update();
1880 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1882 DBG("[thread] Dispatch UST command started");
1884 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1885 health_code_update();
1887 /* Atomically prepare the queue futex */
1888 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1891 struct ust_app
*app
= NULL
;
1895 * Make sure we don't have node(s) that have hung up before receiving
1896 * the notify socket. This is to clean the list in order to avoid
1897 * memory leaks from notify socket that are never seen.
1899 sanitize_wait_queue(&wait_queue
);
1901 health_code_update();
1902 /* Dequeue command for registration */
1903 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1905 DBG("Woken up but nothing in the UST command queue");
1906 /* Continue thread execution */
1910 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1912 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1913 " gid:%d sock:%d name:%s (version %d.%d)",
1914 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1915 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1916 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1917 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1919 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1920 wait_node
= zmalloc(sizeof(*wait_node
));
1922 PERROR("zmalloc wait_node dispatch");
1923 ret
= close(ust_cmd
->sock
);
1925 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1927 lttng_fd_put(LTTNG_FD_APPS
, 1);
1931 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1933 /* Create application object if socket is CMD. */
1934 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1936 if (!wait_node
->app
) {
1937 ret
= close(ust_cmd
->sock
);
1939 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1941 lttng_fd_put(LTTNG_FD_APPS
, 1);
1947 * Add application to the wait queue so we can set the notify
1948 * socket before putting this object in the global ht.
1950 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1955 * We have to continue here since we don't have the notify
1956 * socket and the application MUST be added to the hash table
1957 * only at that moment.
1962 * Look for the application in the local wait queue and set the
1963 * notify socket if found.
1965 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1966 &wait_queue
.head
, head
) {
1967 health_code_update();
1968 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1969 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1970 cds_list_del(&wait_node
->head
);
1972 app
= wait_node
->app
;
1974 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1980 * With no application at this stage the received socket is
1981 * basically useless so close it before we free the cmd data
1982 * structure for good.
1985 ret
= close(ust_cmd
->sock
);
1987 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1989 lttng_fd_put(LTTNG_FD_APPS
, 1);
1996 * @session_lock_list
1998 * Lock the global session list so from the register up to the
1999 * registration done message, no thread can see the application
2000 * and change its state.
2002 session_lock_list();
2006 * Add application to the global hash table. This needs to be
2007 * done before the update to the UST registry can locate the
2012 /* Set app version. This call will print an error if needed. */
2013 (void) ust_app_version(app
);
2015 /* Send notify socket through the notify pipe. */
2016 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
2020 session_unlock_list();
2022 * No notify thread, stop the UST tracing. However, this is
2023 * not an internal error of the this thread thus setting
2024 * the health error code to a normal exit.
2031 * Update newly registered application with the tracing
2032 * registry info already enabled information.
2034 update_ust_app(app
->sock
);
2037 * Don't care about return value. Let the manage apps threads
2038 * handle app unregistration upon socket close.
2040 (void) ust_app_register_done(app
);
2043 * Even if the application socket has been closed, send the app
2044 * to the thread and unregistration will take place at that
2047 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2050 session_unlock_list();
2052 * No apps. thread, stop the UST tracing. However, this is
2053 * not an internal error of the this thread thus setting
2054 * the health error code to a normal exit.
2061 session_unlock_list();
2063 } while (node
!= NULL
);
2065 health_poll_entry();
2066 /* Futex wait on queue. Blocking call on futex() */
2067 futex_nto1_wait(&ust_cmd_queue
.futex
);
2070 /* Normal exit, no error */
2074 /* Clean up wait queue. */
2075 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2076 &wait_queue
.head
, head
) {
2077 cds_list_del(&wait_node
->head
);
2082 /* Empty command queue. */
2084 /* Dequeue command for registration */
2085 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2089 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2090 ret
= close(ust_cmd
->sock
);
2092 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2094 lttng_fd_put(LTTNG_FD_APPS
, 1);
2099 DBG("Dispatch thread dying");
2102 ERR("Health error occurred in %s", __func__
);
2104 health_unregister(health_sessiond
);
2109 * This thread manage application registration.
2111 static void *thread_registration_apps(void *data
)
2113 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2114 uint32_t revents
, nb_fd
;
2115 struct lttng_poll_event events
;
2117 * Get allocated in this thread, enqueued to a global queue, dequeued and
2118 * freed in the manage apps thread.
2120 struct ust_command
*ust_cmd
= NULL
;
2122 DBG("[thread] Manage application registration started");
2124 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2126 if (testpoint(sessiond_thread_registration_apps
)) {
2127 goto error_testpoint
;
2130 ret
= lttcomm_listen_unix_sock(apps_sock
);
2136 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2137 * more will be added to this poll set.
2139 ret
= sessiond_set_thread_pollset(&events
, 2);
2141 goto error_create_poll
;
2144 /* Add the application registration socket */
2145 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2147 goto error_poll_add
;
2150 /* Notify all applications to register */
2151 ret
= notify_ust_apps(1);
2153 ERR("Failed to notify applications or create the wait shared memory.\n"
2154 "Execution continues but there might be problem for already\n"
2155 "running applications that wishes to register.");
2159 DBG("Accepting application registration");
2161 /* Inifinite blocking call, waiting for transmission */
2163 health_poll_entry();
2164 ret
= lttng_poll_wait(&events
, -1);
2168 * Restart interrupted system call.
2170 if (errno
== EINTR
) {
2178 for (i
= 0; i
< nb_fd
; i
++) {
2179 health_code_update();
2181 /* Fetch once the poll data */
2182 revents
= LTTNG_POLL_GETEV(&events
, i
);
2183 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2186 /* No activity for this FD (poll implementation). */
2190 /* Thread quit pipe has been closed. Killing thread. */
2191 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2197 /* Event on the registration socket */
2198 if (pollfd
== apps_sock
) {
2199 if (revents
& LPOLLIN
) {
2200 sock
= lttcomm_accept_unix_sock(apps_sock
);
2206 * Set socket timeout for both receiving and ending.
2207 * app_socket_timeout is in seconds, whereas
2208 * lttcomm_setsockopt_rcv_timeout and
2209 * lttcomm_setsockopt_snd_timeout expect msec as
2212 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2213 app_socket_timeout
* 1000);
2214 (void) lttcomm_setsockopt_snd_timeout(sock
,
2215 app_socket_timeout
* 1000);
2218 * Set the CLOEXEC flag. Return code is useless because
2219 * either way, the show must go on.
2221 (void) utils_set_fd_cloexec(sock
);
2223 /* Create UST registration command for enqueuing */
2224 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2225 if (ust_cmd
== NULL
) {
2226 PERROR("ust command zmalloc");
2235 * Using message-based transmissions to ensure we don't
2236 * have to deal with partially received messages.
2238 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2240 ERR("Exhausted file descriptors allowed for applications.");
2250 health_code_update();
2251 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2254 /* Close socket of the application. */
2259 lttng_fd_put(LTTNG_FD_APPS
, 1);
2263 health_code_update();
2265 ust_cmd
->sock
= sock
;
2268 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2269 " gid:%d sock:%d name:%s (version %d.%d)",
2270 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2271 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2272 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2273 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2276 * Lock free enqueue the registration request. The red pill
2277 * has been taken! This apps will be part of the *system*.
2279 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2282 * Wake the registration queue futex. Implicit memory
2283 * barrier with the exchange in cds_wfcq_enqueue.
2285 futex_nto1_wake(&ust_cmd_queue
.futex
);
2286 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2287 ERR("Register apps socket poll error");
2290 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2299 /* Notify that the registration thread is gone */
2302 if (apps_sock
>= 0) {
2303 ret
= close(apps_sock
);
2313 lttng_fd_put(LTTNG_FD_APPS
, 1);
2315 unlink(apps_unix_sock_path
);
2318 lttng_poll_clean(&events
);
2322 DBG("UST Registration thread cleanup complete");
2325 ERR("Health error occurred in %s", __func__
);
2327 health_unregister(health_sessiond
);
2333 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2334 * exec or it will fails.
2336 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2339 struct timespec timeout
;
2341 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2342 consumer_data
->consumer_thread_is_ready
= 0;
2344 /* Setup pthread condition */
2345 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2348 PERROR("pthread_condattr_init consumer data");
2353 * Set the monotonic clock in order to make sure we DO NOT jump in time
2354 * between the clock_gettime() call and the timedwait call. See bug #324
2355 * for a more details and how we noticed it.
2357 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2360 PERROR("pthread_condattr_setclock consumer data");
2364 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2367 PERROR("pthread_cond_init consumer data");
2371 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2375 PERROR("pthread_create consumer");
2380 /* We are about to wait on a pthread condition */
2381 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2383 /* Get time for sem_timedwait absolute timeout */
2384 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2386 * Set the timeout for the condition timed wait even if the clock gettime
2387 * call fails since we might loop on that call and we want to avoid to
2388 * increment the timeout too many times.
2390 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2393 * The following loop COULD be skipped in some conditions so this is why we
2394 * set ret to 0 in order to make sure at least one round of the loop is
2400 * Loop until the condition is reached or when a timeout is reached. Note
2401 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2402 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2403 * possible. This loop does not take any chances and works with both of
2406 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2407 if (clock_ret
< 0) {
2408 PERROR("clock_gettime spawn consumer");
2409 /* Infinite wait for the consumerd thread to be ready */
2410 ret
= pthread_cond_wait(&consumer_data
->cond
,
2411 &consumer_data
->cond_mutex
);
2413 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2414 &consumer_data
->cond_mutex
, &timeout
);
2418 /* Release the pthread condition */
2419 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2423 if (ret
== ETIMEDOUT
) {
2427 * Call has timed out so we kill the kconsumerd_thread and return
2430 ERR("Condition timed out. The consumer thread was never ready."
2432 pth_ret
= pthread_cancel(consumer_data
->thread
);
2434 PERROR("pthread_cancel consumer thread");
2437 PERROR("pthread_cond_wait failed consumer thread");
2439 /* Caller is expecting a negative value on failure. */
2444 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2445 if (consumer_data
->pid
== 0) {
2446 ERR("Consumerd did not start");
2447 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2450 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2459 * Join consumer thread
2461 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2465 /* Consumer pid must be a real one. */
2466 if (consumer_data
->pid
> 0) {
2468 ret
= kill(consumer_data
->pid
, SIGTERM
);
2470 PERROR("Error killing consumer daemon");
2473 return pthread_join(consumer_data
->thread
, &status
);
2480 * Fork and exec a consumer daemon (consumerd).
2482 * Return pid if successful else -1.
2484 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2488 const char *consumer_to_use
;
2489 const char *verbosity
;
2492 DBG("Spawning consumerd");
2499 if (opt_verbose_consumer
) {
2500 verbosity
= "--verbose";
2501 } else if (lttng_opt_quiet
) {
2502 verbosity
= "--quiet";
2507 switch (consumer_data
->type
) {
2508 case LTTNG_CONSUMER_KERNEL
:
2510 * Find out which consumerd to execute. We will first try the
2511 * 64-bit path, then the sessiond's installation directory, and
2512 * fallback on the 32-bit one,
2514 DBG3("Looking for a kernel consumer at these locations:");
2515 DBG3(" 1) %s", consumerd64_bin
);
2516 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2517 DBG3(" 3) %s", consumerd32_bin
);
2518 if (stat(consumerd64_bin
, &st
) == 0) {
2519 DBG3("Found location #1");
2520 consumer_to_use
= consumerd64_bin
;
2521 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2522 DBG3("Found location #2");
2523 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2524 } else if (stat(consumerd32_bin
, &st
) == 0) {
2525 DBG3("Found location #3");
2526 consumer_to_use
= consumerd32_bin
;
2528 DBG("Could not find any valid consumerd executable");
2532 DBG("Using kernel consumer at: %s", consumer_to_use
);
2533 ret
= execl(consumer_to_use
,
2534 "lttng-consumerd", verbosity
, "-k",
2535 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2536 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2537 "--group", tracing_group_name
,
2540 case LTTNG_CONSUMER64_UST
:
2542 char *tmpnew
= NULL
;
2544 if (consumerd64_libdir
[0] != '\0') {
2548 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2552 tmplen
= strlen("LD_LIBRARY_PATH=")
2553 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2554 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2559 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2560 strcat(tmpnew
, consumerd64_libdir
);
2561 if (tmp
[0] != '\0') {
2562 strcat(tmpnew
, ":");
2563 strcat(tmpnew
, tmp
);
2565 ret
= putenv(tmpnew
);
2572 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2573 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2574 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2575 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2576 "--group", tracing_group_name
,
2578 if (consumerd64_libdir
[0] != '\0') {
2583 case LTTNG_CONSUMER32_UST
:
2585 char *tmpnew
= NULL
;
2587 if (consumerd32_libdir
[0] != '\0') {
2591 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2595 tmplen
= strlen("LD_LIBRARY_PATH=")
2596 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2597 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2602 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2603 strcat(tmpnew
, consumerd32_libdir
);
2604 if (tmp
[0] != '\0') {
2605 strcat(tmpnew
, ":");
2606 strcat(tmpnew
, tmp
);
2608 ret
= putenv(tmpnew
);
2615 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2616 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2617 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2618 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2619 "--group", tracing_group_name
,
2621 if (consumerd32_libdir
[0] != '\0') {
2627 PERROR("unknown consumer type");
2631 PERROR("Consumer execl()");
2633 /* Reaching this point, we got a failure on our execl(). */
2635 } else if (pid
> 0) {
2638 PERROR("start consumer fork");
2646 * Spawn the consumerd daemon and session daemon thread.
2648 static int start_consumerd(struct consumer_data
*consumer_data
)
2653 * Set the listen() state on the socket since there is a possible race
2654 * between the exec() of the consumer daemon and this call if place in the
2655 * consumer thread. See bug #366 for more details.
2657 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2662 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2663 if (consumer_data
->pid
!= 0) {
2664 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2668 ret
= spawn_consumerd(consumer_data
);
2670 ERR("Spawning consumerd failed");
2671 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2675 /* Setting up the consumer_data pid */
2676 consumer_data
->pid
= ret
;
2677 DBG2("Consumer pid %d", consumer_data
->pid
);
2678 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2680 DBG2("Spawning consumer control thread");
2681 ret
= spawn_consumer_thread(consumer_data
);
2683 ERR("Fatal error spawning consumer control thread");
2691 /* Cleanup already created sockets on error. */
2692 if (consumer_data
->err_sock
>= 0) {
2695 err
= close(consumer_data
->err_sock
);
2697 PERROR("close consumer data error socket");
2704 * Setup necessary data for kernel tracer action.
2706 static int init_kernel_tracer(void)
2710 /* Modprobe lttng kernel modules */
2711 ret
= modprobe_lttng_control();
2716 /* Open debugfs lttng */
2717 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2718 if (kernel_tracer_fd
< 0) {
2719 DBG("Failed to open %s", module_proc_lttng
);
2724 /* Validate kernel version */
2725 ret
= kernel_validate_version(kernel_tracer_fd
);
2730 ret
= modprobe_lttng_data();
2735 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2739 modprobe_remove_lttng_control();
2740 ret
= close(kernel_tracer_fd
);
2744 kernel_tracer_fd
= -1;
2745 return LTTNG_ERR_KERN_VERSION
;
2748 ret
= close(kernel_tracer_fd
);
2754 modprobe_remove_lttng_control();
2757 WARN("No kernel tracer available");
2758 kernel_tracer_fd
= -1;
2760 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2762 return LTTNG_ERR_KERN_NA
;
2768 * Copy consumer output from the tracing session to the domain session. The
2769 * function also applies the right modification on a per domain basis for the
2770 * trace files destination directory.
2772 * Should *NOT* be called with RCU read-side lock held.
2774 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2777 const char *dir_name
;
2778 struct consumer_output
*consumer
;
2781 assert(session
->consumer
);
2784 case LTTNG_DOMAIN_KERNEL
:
2785 DBG3("Copying tracing session consumer output in kernel session");
2787 * XXX: We should audit the session creation and what this function
2788 * does "extra" in order to avoid a destroy since this function is used
2789 * in the domain session creation (kernel and ust) only. Same for UST
2792 if (session
->kernel_session
->consumer
) {
2793 consumer_output_put(session
->kernel_session
->consumer
);
2795 session
->kernel_session
->consumer
=
2796 consumer_copy_output(session
->consumer
);
2797 /* Ease our life a bit for the next part */
2798 consumer
= session
->kernel_session
->consumer
;
2799 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2801 case LTTNG_DOMAIN_JUL
:
2802 case LTTNG_DOMAIN_LOG4J
:
2803 case LTTNG_DOMAIN_PYTHON
:
2804 case LTTNG_DOMAIN_UST
:
2805 DBG3("Copying tracing session consumer output in UST session");
2806 if (session
->ust_session
->consumer
) {
2807 consumer_output_put(session
->ust_session
->consumer
);
2809 session
->ust_session
->consumer
=
2810 consumer_copy_output(session
->consumer
);
2811 /* Ease our life a bit for the next part */
2812 consumer
= session
->ust_session
->consumer
;
2813 dir_name
= DEFAULT_UST_TRACE_DIR
;
2816 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2820 /* Append correct directory to subdir */
2821 strncat(consumer
->subdir
, dir_name
,
2822 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2823 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2832 * Create an UST session and add it to the session ust list.
2834 * Should *NOT* be called with RCU read-side lock held.
2836 static int create_ust_session(struct ltt_session
*session
,
2837 struct lttng_domain
*domain
)
2840 struct ltt_ust_session
*lus
= NULL
;
2844 assert(session
->consumer
);
2846 switch (domain
->type
) {
2847 case LTTNG_DOMAIN_JUL
:
2848 case LTTNG_DOMAIN_LOG4J
:
2849 case LTTNG_DOMAIN_PYTHON
:
2850 case LTTNG_DOMAIN_UST
:
2853 ERR("Unknown UST domain on create session %d", domain
->type
);
2854 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2858 DBG("Creating UST session");
2860 lus
= trace_ust_create_session(session
->id
);
2862 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2866 lus
->uid
= session
->uid
;
2867 lus
->gid
= session
->gid
;
2868 lus
->output_traces
= session
->output_traces
;
2869 lus
->snapshot_mode
= session
->snapshot_mode
;
2870 lus
->live_timer_interval
= session
->live_timer
;
2871 session
->ust_session
= lus
;
2872 if (session
->shm_path
[0]) {
2873 strncpy(lus
->root_shm_path
, session
->shm_path
,
2874 sizeof(lus
->root_shm_path
));
2875 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2876 strncpy(lus
->shm_path
, session
->shm_path
,
2877 sizeof(lus
->shm_path
));
2878 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2879 strncat(lus
->shm_path
, "/ust",
2880 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2882 /* Copy session output to the newly created UST session */
2883 ret
= copy_session_consumer(domain
->type
, session
);
2884 if (ret
!= LTTNG_OK
) {
2892 session
->ust_session
= NULL
;
2897 * Create a kernel tracer session then create the default channel.
2899 static int create_kernel_session(struct ltt_session
*session
)
2903 DBG("Creating kernel session");
2905 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2907 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2911 /* Code flow safety */
2912 assert(session
->kernel_session
);
2914 /* Copy session output to the newly created Kernel session */
2915 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2916 if (ret
!= LTTNG_OK
) {
2920 /* Create directory(ies) on local filesystem. */
2921 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2922 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2923 ret
= run_as_mkdir_recursive(
2924 session
->kernel_session
->consumer
->dst
.trace_path
,
2925 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2927 if (errno
!= EEXIST
) {
2928 ERR("Trace directory creation error");
2934 session
->kernel_session
->uid
= session
->uid
;
2935 session
->kernel_session
->gid
= session
->gid
;
2936 session
->kernel_session
->output_traces
= session
->output_traces
;
2937 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2942 trace_kernel_destroy_session(session
->kernel_session
);
2943 session
->kernel_session
= NULL
;
2948 * Count number of session permitted by uid/gid.
2950 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2953 struct ltt_session
*session
;
2955 DBG("Counting number of available session for UID %d GID %d",
2957 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2959 * Only list the sessions the user can control.
2961 if (!session_access_ok(session
, uid
, gid
)) {
2970 * Process the command requested by the lttng client within the command
2971 * context structure. This function make sure that the return structure (llm)
2972 * is set and ready for transmission before returning.
2974 * Return any error encountered or 0 for success.
2976 * "sock" is only used for special-case var. len data.
2978 * Should *NOT* be called with RCU read-side lock held.
2980 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2984 int need_tracing_session
= 1;
2987 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2989 assert(!rcu_read_ongoing());
2993 switch (cmd_ctx
->lsm
->cmd_type
) {
2994 case LTTNG_CREATE_SESSION
:
2995 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2996 case LTTNG_CREATE_SESSION_LIVE
:
2997 case LTTNG_DESTROY_SESSION
:
2998 case LTTNG_LIST_SESSIONS
:
2999 case LTTNG_LIST_DOMAINS
:
3000 case LTTNG_START_TRACE
:
3001 case LTTNG_STOP_TRACE
:
3002 case LTTNG_DATA_PENDING
:
3003 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3004 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3005 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3006 case LTTNG_SNAPSHOT_RECORD
:
3007 case LTTNG_SAVE_SESSION
:
3008 case LTTNG_SET_SESSION_SHM_PATH
:
3015 if (opt_no_kernel
&& need_domain
3016 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3018 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3020 ret
= LTTNG_ERR_KERN_NA
;
3025 /* Deny register consumer if we already have a spawned consumer. */
3026 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3027 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3028 if (kconsumer_data
.pid
> 0) {
3029 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3030 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3033 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3037 * Check for command that don't needs to allocate a returned payload. We do
3038 * this here so we don't have to make the call for no payload at each
3041 switch(cmd_ctx
->lsm
->cmd_type
) {
3042 case LTTNG_LIST_SESSIONS
:
3043 case LTTNG_LIST_TRACEPOINTS
:
3044 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3045 case LTTNG_LIST_DOMAINS
:
3046 case LTTNG_LIST_CHANNELS
:
3047 case LTTNG_LIST_EVENTS
:
3048 case LTTNG_LIST_SYSCALLS
:
3049 case LTTNG_LIST_TRACKER_PIDS
:
3052 /* Setup lttng message with no payload */
3053 ret
= setup_lttng_msg(cmd_ctx
, 0);
3055 /* This label does not try to unlock the session */
3056 goto init_setup_error
;
3060 /* Commands that DO NOT need a session. */
3061 switch (cmd_ctx
->lsm
->cmd_type
) {
3062 case LTTNG_CREATE_SESSION
:
3063 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3064 case LTTNG_CREATE_SESSION_LIVE
:
3065 case LTTNG_CALIBRATE
:
3066 case LTTNG_LIST_SESSIONS
:
3067 case LTTNG_LIST_TRACEPOINTS
:
3068 case LTTNG_LIST_SYSCALLS
:
3069 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3070 case LTTNG_SAVE_SESSION
:
3071 need_tracing_session
= 0;
3074 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3076 * We keep the session list lock across _all_ commands
3077 * for now, because the per-session lock does not
3078 * handle teardown properly.
3080 session_lock_list();
3081 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3082 if (cmd_ctx
->session
== NULL
) {
3083 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3086 /* Acquire lock for the session */
3087 session_lock(cmd_ctx
->session
);
3093 * Commands that need a valid session but should NOT create one if none
3094 * exists. Instead of creating one and destroying it when the command is
3095 * handled, process that right before so we save some round trip in useless
3098 switch (cmd_ctx
->lsm
->cmd_type
) {
3099 case LTTNG_DISABLE_CHANNEL
:
3100 case LTTNG_DISABLE_EVENT
:
3101 switch (cmd_ctx
->lsm
->domain
.type
) {
3102 case LTTNG_DOMAIN_KERNEL
:
3103 if (!cmd_ctx
->session
->kernel_session
) {
3104 ret
= LTTNG_ERR_NO_CHANNEL
;
3108 case LTTNG_DOMAIN_JUL
:
3109 case LTTNG_DOMAIN_LOG4J
:
3110 case LTTNG_DOMAIN_PYTHON
:
3111 case LTTNG_DOMAIN_UST
:
3112 if (!cmd_ctx
->session
->ust_session
) {
3113 ret
= LTTNG_ERR_NO_CHANNEL
;
3118 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3130 * Check domain type for specific "pre-action".
3132 switch (cmd_ctx
->lsm
->domain
.type
) {
3133 case LTTNG_DOMAIN_KERNEL
:
3135 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3139 /* Kernel tracer check */
3140 if (kernel_tracer_fd
== -1) {
3141 /* Basically, load kernel tracer modules */
3142 ret
= init_kernel_tracer();
3148 /* Consumer is in an ERROR state. Report back to client */
3149 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3150 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3154 /* Need a session for kernel command */
3155 if (need_tracing_session
) {
3156 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3157 ret
= create_kernel_session(cmd_ctx
->session
);
3159 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3164 /* Start the kernel consumer daemon */
3165 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3166 if (kconsumer_data
.pid
== 0 &&
3167 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3168 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3169 ret
= start_consumerd(&kconsumer_data
);
3171 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3174 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3176 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3180 * The consumer was just spawned so we need to add the socket to
3181 * the consumer output of the session if exist.
3183 ret
= consumer_create_socket(&kconsumer_data
,
3184 cmd_ctx
->session
->kernel_session
->consumer
);
3191 case LTTNG_DOMAIN_JUL
:
3192 case LTTNG_DOMAIN_LOG4J
:
3193 case LTTNG_DOMAIN_PYTHON
:
3194 case LTTNG_DOMAIN_UST
:
3196 if (!ust_app_supported()) {
3197 ret
= LTTNG_ERR_NO_UST
;
3200 /* Consumer is in an ERROR state. Report back to client */
3201 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3202 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3206 if (need_tracing_session
) {
3207 /* Create UST session if none exist. */
3208 if (cmd_ctx
->session
->ust_session
== NULL
) {
3209 ret
= create_ust_session(cmd_ctx
->session
,
3210 &cmd_ctx
->lsm
->domain
);
3211 if (ret
!= LTTNG_OK
) {
3216 /* Start the UST consumer daemons */
3218 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3219 if (consumerd64_bin
[0] != '\0' &&
3220 ustconsumer64_data
.pid
== 0 &&
3221 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3222 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3223 ret
= start_consumerd(&ustconsumer64_data
);
3225 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3226 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3230 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3231 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3233 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3237 * Setup socket for consumer 64 bit. No need for atomic access
3238 * since it was set above and can ONLY be set in this thread.
3240 ret
= consumer_create_socket(&ustconsumer64_data
,
3241 cmd_ctx
->session
->ust_session
->consumer
);
3247 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3248 if (consumerd32_bin
[0] != '\0' &&
3249 ustconsumer32_data
.pid
== 0 &&
3250 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3251 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3252 ret
= start_consumerd(&ustconsumer32_data
);
3254 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3255 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3259 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3260 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3262 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3266 * Setup socket for consumer 64 bit. No need for atomic access
3267 * since it was set above and can ONLY be set in this thread.
3269 ret
= consumer_create_socket(&ustconsumer32_data
,
3270 cmd_ctx
->session
->ust_session
->consumer
);
3282 /* Validate consumer daemon state when start/stop trace command */
3283 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3284 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3285 switch (cmd_ctx
->lsm
->domain
.type
) {
3286 case LTTNG_DOMAIN_JUL
:
3287 case LTTNG_DOMAIN_LOG4J
:
3288 case LTTNG_DOMAIN_PYTHON
:
3289 case LTTNG_DOMAIN_UST
:
3290 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3291 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3295 case LTTNG_DOMAIN_KERNEL
:
3296 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3297 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3305 * Check that the UID or GID match that of the tracing session.
3306 * The root user can interact with all sessions.
3308 if (need_tracing_session
) {
3309 if (!session_access_ok(cmd_ctx
->session
,
3310 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3311 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3312 ret
= LTTNG_ERR_EPERM
;
3318 * Send relayd information to consumer as soon as we have a domain and a
3321 if (cmd_ctx
->session
&& need_domain
) {
3323 * Setup relayd if not done yet. If the relayd information was already
3324 * sent to the consumer, this call will gracefully return.
3326 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3327 if (ret
!= LTTNG_OK
) {
3332 /* Process by command type */
3333 switch (cmd_ctx
->lsm
->cmd_type
) {
3334 case LTTNG_ADD_CONTEXT
:
3336 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3337 cmd_ctx
->lsm
->u
.context
.channel_name
,
3338 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3341 case LTTNG_DISABLE_CHANNEL
:
3343 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3344 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3347 case LTTNG_DISABLE_EVENT
:
3351 * FIXME: handle filter; for now we just receive the filter's
3352 * bytecode along with the filter expression which are sent by
3353 * liblttng-ctl and discard them.
3355 * This fixes an issue where the client may block while sending
3356 * the filter payload and encounter an error because the session
3357 * daemon closes the socket without ever handling this data.
3359 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3360 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3363 char data
[LTTNG_FILTER_MAX_LEN
];
3365 DBG("Discarding disable event command payload of size %zu", count
);
3367 ret
= lttcomm_recv_unix_sock(sock
, data
,
3368 count
> sizeof(data
) ? sizeof(data
) : count
);
3373 count
-= (size_t) ret
;
3376 /* FIXME: passing packed structure to non-packed pointer */
3377 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3378 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3379 &cmd_ctx
->lsm
->u
.disable
.event
);
3382 case LTTNG_ENABLE_CHANNEL
:
3384 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3385 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3388 case LTTNG_TRACK_PID
:
3390 ret
= cmd_track_pid(cmd_ctx
->session
,
3391 cmd_ctx
->lsm
->domain
.type
,
3392 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3395 case LTTNG_UNTRACK_PID
:
3397 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3398 cmd_ctx
->lsm
->domain
.type
,
3399 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3402 case LTTNG_ENABLE_EVENT
:
3404 struct lttng_event_exclusion
*exclusion
= NULL
;
3405 struct lttng_filter_bytecode
*bytecode
= NULL
;
3406 char *filter_expression
= NULL
;
3408 /* Handle exclusion events and receive it from the client. */
3409 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3410 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3412 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3413 (count
* LTTNG_SYMBOL_NAME_LEN
));
3415 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3419 DBG("Receiving var len exclusion event list from client ...");
3420 exclusion
->count
= count
;
3421 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3422 count
* LTTNG_SYMBOL_NAME_LEN
);
3424 DBG("Nothing recv() from client var len data... continuing");
3427 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3432 /* Get filter expression from client. */
3433 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3434 size_t expression_len
=
3435 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3437 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3438 ret
= LTTNG_ERR_FILTER_INVAL
;
3443 filter_expression
= zmalloc(expression_len
);
3444 if (!filter_expression
) {
3446 ret
= LTTNG_ERR_FILTER_NOMEM
;
3450 /* Receive var. len. data */
3451 DBG("Receiving var len filter's expression from client ...");
3452 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3455 DBG("Nothing recv() from client car len data... continuing");
3457 free(filter_expression
);
3459 ret
= LTTNG_ERR_FILTER_INVAL
;
3464 /* Handle filter and get bytecode from client. */
3465 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3466 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3468 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3469 ret
= LTTNG_ERR_FILTER_INVAL
;
3470 free(filter_expression
);
3475 bytecode
= zmalloc(bytecode_len
);
3477 free(filter_expression
);
3479 ret
= LTTNG_ERR_FILTER_NOMEM
;
3483 /* Receive var. len. data */
3484 DBG("Receiving var len filter's bytecode from client ...");
3485 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3487 DBG("Nothing recv() from client car len data... continuing");
3489 free(filter_expression
);
3492 ret
= LTTNG_ERR_FILTER_INVAL
;
3496 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3497 free(filter_expression
);
3500 ret
= LTTNG_ERR_FILTER_INVAL
;
3505 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3506 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3507 &cmd_ctx
->lsm
->u
.enable
.event
,
3508 filter_expression
, bytecode
, exclusion
,
3509 kernel_poll_pipe
[1]);
3512 case LTTNG_LIST_TRACEPOINTS
:
3514 struct lttng_event
*events
;
3517 session_lock_list();
3518 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3519 session_unlock_list();
3520 if (nb_events
< 0) {
3521 /* Return value is a negative lttng_error_code. */
3527 * Setup lttng message with payload size set to the event list size in
3528 * bytes and then copy list into the llm payload.
3530 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3536 /* Copy event list into message payload */
3537 memcpy(cmd_ctx
->llm
->payload
, events
,
3538 sizeof(struct lttng_event
) * nb_events
);
3545 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3547 struct lttng_event_field
*fields
;
3550 session_lock_list();
3551 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3553 session_unlock_list();
3554 if (nb_fields
< 0) {
3555 /* Return value is a negative lttng_error_code. */
3561 * Setup lttng message with payload size set to the event list size in
3562 * bytes and then copy list into the llm payload.
3564 ret
= setup_lttng_msg(cmd_ctx
,
3565 sizeof(struct lttng_event_field
) * nb_fields
);
3571 /* Copy event list into message payload */
3572 memcpy(cmd_ctx
->llm
->payload
, fields
,
3573 sizeof(struct lttng_event_field
) * nb_fields
);
3580 case LTTNG_LIST_SYSCALLS
:
3582 struct lttng_event
*events
;
3585 nb_events
= cmd_list_syscalls(&events
);
3586 if (nb_events
< 0) {
3587 /* Return value is a negative lttng_error_code. */
3593 * Setup lttng message with payload size set to the event list size in
3594 * bytes and then copy list into the llm payload.
3596 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3602 /* Copy event list into message payload */
3603 memcpy(cmd_ctx
->llm
->payload
, events
,
3604 sizeof(struct lttng_event
) * nb_events
);
3611 case LTTNG_LIST_TRACKER_PIDS
:
3613 int32_t *pids
= NULL
;
3616 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3617 cmd_ctx
->lsm
->domain
.type
, &pids
);
3619 /* Return value is a negative lttng_error_code. */
3625 * Setup lttng message with payload size set to the event list size in
3626 * bytes and then copy list into the llm payload.
3628 ret
= setup_lttng_msg(cmd_ctx
, sizeof(int32_t) * nr_pids
);
3634 /* Copy event list into message payload */
3635 memcpy(cmd_ctx
->llm
->payload
, pids
,
3636 sizeof(int) * nr_pids
);
3643 case LTTNG_SET_CONSUMER_URI
:
3646 struct lttng_uri
*uris
;
3648 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3649 len
= nb_uri
* sizeof(struct lttng_uri
);
3652 ret
= LTTNG_ERR_INVALID
;
3656 uris
= zmalloc(len
);
3658 ret
= LTTNG_ERR_FATAL
;
3662 /* Receive variable len data */
3663 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3664 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3666 DBG("No URIs received from client... continuing");
3668 ret
= LTTNG_ERR_SESSION_FAIL
;
3673 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3675 if (ret
!= LTTNG_OK
) {
3682 case LTTNG_START_TRACE
:
3684 ret
= cmd_start_trace(cmd_ctx
->session
);
3687 case LTTNG_STOP_TRACE
:
3689 ret
= cmd_stop_trace(cmd_ctx
->session
);
3692 case LTTNG_CREATE_SESSION
:
3695 struct lttng_uri
*uris
= NULL
;
3697 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3698 len
= nb_uri
* sizeof(struct lttng_uri
);
3701 uris
= zmalloc(len
);
3703 ret
= LTTNG_ERR_FATAL
;
3707 /* Receive variable len data */
3708 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3709 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3711 DBG("No URIs received from client... continuing");
3713 ret
= LTTNG_ERR_SESSION_FAIL
;
3718 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3719 DBG("Creating session with ONE network URI is a bad call");
3720 ret
= LTTNG_ERR_SESSION_FAIL
;
3726 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3727 &cmd_ctx
->creds
, 0);
3733 case LTTNG_DESTROY_SESSION
:
3735 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3737 /* Set session to NULL so we do not unlock it after free. */
3738 cmd_ctx
->session
= NULL
;
3741 case LTTNG_LIST_DOMAINS
:
3744 struct lttng_domain
*domains
= NULL
;
3746 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3748 /* Return value is a negative lttng_error_code. */
3753 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3759 /* Copy event list into message payload */
3760 memcpy(cmd_ctx
->llm
->payload
, domains
,
3761 nb_dom
* sizeof(struct lttng_domain
));
3768 case LTTNG_LIST_CHANNELS
:
3771 struct lttng_channel
*channels
= NULL
;
3773 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3774 cmd_ctx
->session
, &channels
);
3776 /* Return value is a negative lttng_error_code. */
3781 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3787 /* Copy event list into message payload */
3788 memcpy(cmd_ctx
->llm
->payload
, channels
,
3789 nb_chan
* sizeof(struct lttng_channel
));
3796 case LTTNG_LIST_EVENTS
:
3799 struct lttng_event
*events
= NULL
;
3801 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3802 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3804 /* Return value is a negative lttng_error_code. */
3809 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3815 /* Copy event list into message payload */
3816 memcpy(cmd_ctx
->llm
->payload
, events
,
3817 nb_event
* sizeof(struct lttng_event
));
3824 case LTTNG_LIST_SESSIONS
:
3826 unsigned int nr_sessions
;
3828 session_lock_list();
3829 nr_sessions
= lttng_sessions_count(
3830 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3831 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3833 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3835 session_unlock_list();
3839 /* Filled the session array */
3840 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3841 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3842 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3844 session_unlock_list();
3849 case LTTNG_CALIBRATE
:
3851 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3852 &cmd_ctx
->lsm
->u
.calibrate
);
3855 case LTTNG_REGISTER_CONSUMER
:
3857 struct consumer_data
*cdata
;
3859 switch (cmd_ctx
->lsm
->domain
.type
) {
3860 case LTTNG_DOMAIN_KERNEL
:
3861 cdata
= &kconsumer_data
;
3864 ret
= LTTNG_ERR_UND
;
3868 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3869 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3872 case LTTNG_DATA_PENDING
:
3874 ret
= cmd_data_pending(cmd_ctx
->session
);
3877 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3879 struct lttcomm_lttng_output_id reply
;
3881 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3882 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3883 if (ret
!= LTTNG_OK
) {
3887 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3892 /* Copy output list into message payload */
3893 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3897 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3899 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3900 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3903 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3906 struct lttng_snapshot_output
*outputs
= NULL
;
3908 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3909 if (nb_output
< 0) {
3914 ret
= setup_lttng_msg(cmd_ctx
,
3915 nb_output
* sizeof(struct lttng_snapshot_output
));
3922 /* Copy output list into message payload */
3923 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3924 nb_output
* sizeof(struct lttng_snapshot_output
));
3931 case LTTNG_SNAPSHOT_RECORD
:
3933 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3934 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3935 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3938 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3941 struct lttng_uri
*uris
= NULL
;
3943 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3944 len
= nb_uri
* sizeof(struct lttng_uri
);
3947 uris
= zmalloc(len
);
3949 ret
= LTTNG_ERR_FATAL
;
3953 /* Receive variable len data */
3954 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3955 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3957 DBG("No URIs received from client... continuing");
3959 ret
= LTTNG_ERR_SESSION_FAIL
;
3964 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3965 DBG("Creating session with ONE network URI is a bad call");
3966 ret
= LTTNG_ERR_SESSION_FAIL
;
3972 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3973 nb_uri
, &cmd_ctx
->creds
);
3977 case LTTNG_CREATE_SESSION_LIVE
:
3980 struct lttng_uri
*uris
= NULL
;
3982 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3983 len
= nb_uri
* sizeof(struct lttng_uri
);
3986 uris
= zmalloc(len
);
3988 ret
= LTTNG_ERR_FATAL
;
3992 /* Receive variable len data */
3993 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3994 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3996 DBG("No URIs received from client... continuing");
3998 ret
= LTTNG_ERR_SESSION_FAIL
;
4003 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4004 DBG("Creating session with ONE network URI is a bad call");
4005 ret
= LTTNG_ERR_SESSION_FAIL
;
4011 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4012 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4016 case LTTNG_SAVE_SESSION
:
4018 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4022 case LTTNG_SET_SESSION_SHM_PATH
:
4024 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4025 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4029 ret
= LTTNG_ERR_UND
;
4034 if (cmd_ctx
->llm
== NULL
) {
4035 DBG("Missing llm structure. Allocating one.");
4036 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
4040 /* Set return code */
4041 cmd_ctx
->llm
->ret_code
= ret
;
4043 if (cmd_ctx
->session
) {
4044 session_unlock(cmd_ctx
->session
);
4046 if (need_tracing_session
) {
4047 session_unlock_list();
4050 assert(!rcu_read_ongoing());
4055 * Thread managing health check socket.
4057 static void *thread_manage_health(void *data
)
4059 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4060 uint32_t revents
, nb_fd
;
4061 struct lttng_poll_event events
;
4062 struct health_comm_msg msg
;
4063 struct health_comm_reply reply
;
4065 DBG("[thread] Manage health check started");
4067 rcu_register_thread();
4069 /* We might hit an error path before this is created. */
4070 lttng_poll_init(&events
);
4072 /* Create unix socket */
4073 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4075 ERR("Unable to create health check Unix socket");
4081 /* lttng health client socket path permissions */
4082 ret
= chown(health_unix_sock_path
, 0,
4083 utils_get_group_id(tracing_group_name
));
4085 ERR("Unable to set group on %s", health_unix_sock_path
);
4091 ret
= chmod(health_unix_sock_path
,
4092 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4094 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4102 * Set the CLOEXEC flag. Return code is useless because either way, the
4105 (void) utils_set_fd_cloexec(sock
);
4107 ret
= lttcomm_listen_unix_sock(sock
);
4113 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4114 * more will be added to this poll set.
4116 ret
= sessiond_set_thread_pollset(&events
, 2);
4121 /* Add the application registration socket */
4122 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4127 sessiond_notify_ready();
4130 DBG("Health check ready");
4132 /* Inifinite blocking call, waiting for transmission */
4134 ret
= lttng_poll_wait(&events
, -1);
4137 * Restart interrupted system call.
4139 if (errno
== EINTR
) {
4147 for (i
= 0; i
< nb_fd
; i
++) {
4148 /* Fetch once the poll data */
4149 revents
= LTTNG_POLL_GETEV(&events
, i
);
4150 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4153 /* No activity for this FD (poll implementation). */
4157 /* Thread quit pipe has been closed. Killing thread. */
4158 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4164 /* Event on the registration socket */
4165 if (pollfd
== sock
) {
4166 if (revents
& LPOLLIN
) {
4168 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4169 ERR("Health socket poll error");
4172 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4178 new_sock
= lttcomm_accept_unix_sock(sock
);
4184 * Set the CLOEXEC flag. Return code is useless because either way, the
4187 (void) utils_set_fd_cloexec(new_sock
);
4189 DBG("Receiving data from client for health...");
4190 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4192 DBG("Nothing recv() from client... continuing");
4193 ret
= close(new_sock
);
4201 rcu_thread_online();
4203 memset(&reply
, 0, sizeof(reply
));
4204 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4206 * health_check_state returns 0 if health is
4209 if (!health_check_state(health_sessiond
, i
)) {
4210 reply
.ret_code
|= 1ULL << i
;
4214 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4216 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4218 ERR("Failed to send health data back to client");
4221 /* End of transmission */
4222 ret
= close(new_sock
);
4232 ERR("Health error occurred in %s", __func__
);
4234 DBG("Health check thread dying");
4235 unlink(health_unix_sock_path
);
4243 lttng_poll_clean(&events
);
4245 rcu_unregister_thread();
4250 * This thread manage all clients request using the unix client socket for
4253 static void *thread_manage_clients(void *data
)
4255 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4257 uint32_t revents
, nb_fd
;
4258 struct command_ctx
*cmd_ctx
= NULL
;
4259 struct lttng_poll_event events
;
4261 DBG("[thread] Manage client started");
4263 rcu_register_thread();
4265 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4267 health_code_update();
4269 ret
= lttcomm_listen_unix_sock(client_sock
);
4275 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4276 * more will be added to this poll set.
4278 ret
= sessiond_set_thread_pollset(&events
, 2);
4280 goto error_create_poll
;
4283 /* Add the application registration socket */
4284 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4289 sessiond_notify_ready();
4290 ret
= sem_post(&load_info
->message_thread_ready
);
4292 PERROR("sem_post message_thread_ready");
4296 /* This testpoint is after we signal readiness to the parent. */
4297 if (testpoint(sessiond_thread_manage_clients
)) {
4301 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4305 health_code_update();
4308 DBG("Accepting client command ...");
4310 /* Inifinite blocking call, waiting for transmission */
4312 health_poll_entry();
4313 ret
= lttng_poll_wait(&events
, -1);
4317 * Restart interrupted system call.
4319 if (errno
== EINTR
) {
4327 for (i
= 0; i
< nb_fd
; i
++) {
4328 /* Fetch once the poll data */
4329 revents
= LTTNG_POLL_GETEV(&events
, i
);
4330 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4332 health_code_update();
4335 /* No activity for this FD (poll implementation). */
4339 /* Thread quit pipe has been closed. Killing thread. */
4340 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4346 /* Event on the registration socket */
4347 if (pollfd
== client_sock
) {
4348 if (revents
& LPOLLIN
) {
4350 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4351 ERR("Client socket poll error");
4354 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4360 DBG("Wait for client response");
4362 health_code_update();
4364 sock
= lttcomm_accept_unix_sock(client_sock
);
4370 * Set the CLOEXEC flag. Return code is useless because either way, the
4373 (void) utils_set_fd_cloexec(sock
);
4375 /* Set socket option for credentials retrieval */
4376 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4381 /* Allocate context command to process the client request */
4382 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4383 if (cmd_ctx
== NULL
) {
4384 PERROR("zmalloc cmd_ctx");
4388 /* Allocate data buffer for reception */
4389 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4390 if (cmd_ctx
->lsm
== NULL
) {
4391 PERROR("zmalloc cmd_ctx->lsm");
4395 cmd_ctx
->llm
= NULL
;
4396 cmd_ctx
->session
= NULL
;
4398 health_code_update();
4401 * Data is received from the lttng client. The struct
4402 * lttcomm_session_msg (lsm) contains the command and data request of
4405 DBG("Receiving data from client ...");
4406 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4407 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4409 DBG("Nothing recv() from client... continuing");
4415 clean_command_ctx(&cmd_ctx
);
4419 health_code_update();
4421 // TODO: Validate cmd_ctx including sanity check for
4422 // security purpose.
4424 rcu_thread_online();
4426 * This function dispatch the work to the kernel or userspace tracer
4427 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4428 * informations for the client. The command context struct contains
4429 * everything this function may needs.
4431 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4432 rcu_thread_offline();
4440 * TODO: Inform client somehow of the fatal error. At
4441 * this point, ret < 0 means that a zmalloc failed
4442 * (ENOMEM). Error detected but still accept
4443 * command, unless a socket error has been
4446 clean_command_ctx(&cmd_ctx
);
4450 health_code_update();
4452 DBG("Sending response (size: %d, retcode: %s)",
4453 cmd_ctx
->lttng_msg_size
,
4454 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
4455 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4457 ERR("Failed to send data back to client");
4460 /* End of transmission */
4467 clean_command_ctx(&cmd_ctx
);
4469 health_code_update();
4481 lttng_poll_clean(&events
);
4482 clean_command_ctx(&cmd_ctx
);
4486 unlink(client_unix_sock_path
);
4487 if (client_sock
>= 0) {
4488 ret
= close(client_sock
);
4496 ERR("Health error occurred in %s", __func__
);
4499 health_unregister(health_sessiond
);
4501 DBG("Client thread dying");
4503 rcu_unregister_thread();
4506 * Since we are creating the consumer threads, we own them, so we need
4507 * to join them before our thread exits.
4509 ret
= join_consumer_thread(&kconsumer_data
);
4512 PERROR("join_consumer");
4515 ret
= join_consumer_thread(&ustconsumer32_data
);
4518 PERROR("join_consumer ust32");
4521 ret
= join_consumer_thread(&ustconsumer64_data
);
4524 PERROR("join_consumer ust64");
4531 * usage function on stderr
4533 static void usage(void)
4535 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4536 fprintf(stderr
, " -h, --help Display this usage.\n");
4537 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4538 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4539 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4540 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4541 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4542 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4543 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4544 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4545 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4546 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4547 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4548 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4549 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4550 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4551 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4552 fprintf(stderr
, " -V, --version Show version number.\n");
4553 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4554 fprintf(stderr
, " -q, --quiet No output at all.\n");
4555 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4556 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4557 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4558 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4559 fprintf(stderr
, " --agent-tcp-port Agent registration TCP port\n");
4560 fprintf(stderr
, " -f --config PATH Load daemon configuration file\n");
4561 fprintf(stderr
, " -l --load PATH Load session configuration\n");
4562 fprintf(stderr
, " --kmod-probes Specify kernel module probes to load\n");
4563 fprintf(stderr
, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4566 static int string_match(const char *str1
, const char *str2
)
4568 return (str1
&& str2
) && !strcmp(str1
, str2
);
4572 * Take an option from the getopt output and set it in the right variable to be
4575 * Return 0 on success else a negative value.
4577 static int set_option(int opt
, const char *arg
, const char *optname
)
4581 if (arg
&& arg
[0] == '\0') {
4583 * This only happens if the value is read from daemon config
4584 * file. This means the option requires an argument and the
4585 * configuration file contains a line such as:
4592 if (string_match(optname
, "client-sock") || opt
== 'c') {
4593 if (lttng_is_setuid_setgid()) {
4594 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4595 "-c, --client-sock");
4597 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4599 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4600 if (lttng_is_setuid_setgid()) {
4601 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4604 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4606 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4608 } else if (string_match(optname
, "background") || opt
== 'b') {
4610 } else if (string_match(optname
, "group") || opt
== 'g') {
4611 if (lttng_is_setuid_setgid()) {
4612 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4616 * If the override option is set, the pointer points to a
4617 * *non* const thus freeing it even though the variable type is
4620 if (tracing_group_name_override
) {
4621 free((void *) tracing_group_name
);
4623 tracing_group_name
= strdup(arg
);
4624 if (!tracing_group_name
) {
4628 tracing_group_name_override
= 1;
4630 } else if (string_match(optname
, "help") || opt
== 'h') {
4633 } else if (string_match(optname
, "version") || opt
== 'V') {
4634 fprintf(stdout
, "%s\n", VERSION
);
4636 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4638 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4639 if (lttng_is_setuid_setgid()) {
4640 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4641 "--kconsumerd-err-sock");
4643 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4645 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4646 if (lttng_is_setuid_setgid()) {
4647 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4648 "--kconsumerd-cmd-sock");
4650 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4652 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4653 if (lttng_is_setuid_setgid()) {
4654 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4655 "--ustconsumerd64-err-sock");
4657 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4659 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4660 if (lttng_is_setuid_setgid()) {
4661 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4662 "--ustconsumerd64-cmd-sock");
4664 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4666 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4667 if (lttng_is_setuid_setgid()) {
4668 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4669 "--ustconsumerd32-err-sock");
4671 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4673 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4674 if (lttng_is_setuid_setgid()) {
4675 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4676 "--ustconsumerd32-cmd-sock");
4678 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4680 } else if (string_match(optname
, "no-kernel")) {
4682 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4683 lttng_opt_quiet
= 1;
4684 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4685 /* Verbose level can increase using multiple -v */
4687 /* Value obtained from config file */
4688 lttng_opt_verbose
= config_parse_value(arg
);
4690 /* -v used on command line */
4691 lttng_opt_verbose
++;
4693 /* Clamp value to [0, 3] */
4694 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4695 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4696 } else if (string_match(optname
, "verbose-consumer")) {
4698 opt_verbose_consumer
= config_parse_value(arg
);
4700 opt_verbose_consumer
+= 1;
4702 } else if (string_match(optname
, "consumerd32-path")) {
4703 if (lttng_is_setuid_setgid()) {
4704 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4705 "--consumerd32-path");
4707 if (consumerd32_bin_override
) {
4708 free((void *) consumerd32_bin
);
4710 consumerd32_bin
= strdup(arg
);
4711 if (!consumerd32_bin
) {
4715 consumerd32_bin_override
= 1;
4717 } else if (string_match(optname
, "consumerd32-libdir")) {
4718 if (lttng_is_setuid_setgid()) {
4719 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4720 "--consumerd32-libdir");
4722 if (consumerd32_libdir_override
) {
4723 free((void *) consumerd32_libdir
);
4725 consumerd32_libdir
= strdup(arg
);
4726 if (!consumerd32_libdir
) {
4730 consumerd32_libdir_override
= 1;
4732 } else if (string_match(optname
, "consumerd64-path")) {
4733 if (lttng_is_setuid_setgid()) {
4734 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4735 "--consumerd64-path");
4737 if (consumerd64_bin_override
) {
4738 free((void *) consumerd64_bin
);
4740 consumerd64_bin
= strdup(arg
);
4741 if (!consumerd64_bin
) {
4745 consumerd64_bin_override
= 1;
4747 } else if (string_match(optname
, "consumerd64-libdir")) {
4748 if (lttng_is_setuid_setgid()) {
4749 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4750 "--consumerd64-libdir");
4752 if (consumerd64_libdir_override
) {
4753 free((void *) consumerd64_libdir
);
4755 consumerd64_libdir
= strdup(arg
);
4756 if (!consumerd64_libdir
) {
4760 consumerd64_libdir_override
= 1;
4762 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4763 if (lttng_is_setuid_setgid()) {
4764 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4768 opt_pidfile
= strdup(arg
);
4774 } else if (string_match(optname
, "agent-tcp-port")) {
4775 if (lttng_is_setuid_setgid()) {
4776 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4777 "--agent-tcp-port");
4786 v
= strtoul(arg
, NULL
, 0);
4787 if (errno
!= 0 || !isdigit(arg
[0])) {
4788 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4791 if (v
== 0 || v
>= 65535) {
4792 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4795 agent_tcp_port
= (uint32_t) v
;
4796 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4798 } else if (string_match(optname
, "load") || opt
== 'l') {
4799 if (lttng_is_setuid_setgid()) {
4800 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4803 free(opt_load_session_path
);
4804 opt_load_session_path
= strdup(arg
);
4805 if (!opt_load_session_path
) {
4810 } else if (string_match(optname
, "kmod-probes")) {
4811 if (lttng_is_setuid_setgid()) {
4812 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4815 free(kmod_probes_list
);
4816 kmod_probes_list
= strdup(arg
);
4817 if (!kmod_probes_list
) {
4822 } else if (string_match(optname
, "extra-kmod-probes")) {
4823 if (lttng_is_setuid_setgid()) {
4824 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4825 "--extra-kmod-probes");
4827 free(kmod_extra_probes_list
);
4828 kmod_extra_probes_list
= strdup(arg
);
4829 if (!kmod_extra_probes_list
) {
4834 } else if (string_match(optname
, "config") || opt
== 'f') {
4835 /* This is handled in set_options() thus silent skip. */
4838 /* Unknown option or other error.
4839 * Error is printed by getopt, just return */
4844 if (ret
== -EINVAL
) {
4845 const char *opt_name
= "unknown";
4848 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4850 if (opt
== long_options
[i
].val
) {
4851 opt_name
= long_options
[i
].name
;
4856 WARN("Invalid argument provided for option \"%s\", using default value.",
4864 * config_entry_handler_cb used to handle options read from a config file.
4865 * See config_entry_handler_cb comment in common/config/config.h for the
4866 * return value conventions.
4868 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4872 if (!entry
|| !entry
->name
|| !entry
->value
) {
4877 /* Check if the option is to be ignored */
4878 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4879 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4884 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4887 /* Ignore if not fully matched. */
4888 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4893 * If the option takes no argument on the command line, we have to
4894 * check if the value is "true". We support non-zero numeric values,
4897 if (!long_options
[i
].has_arg
) {
4898 ret
= config_parse_value(entry
->value
);
4901 WARN("Invalid configuration value \"%s\" for option %s",
4902 entry
->value
, entry
->name
);
4904 /* False, skip boolean config option. */
4909 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4913 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4920 * daemon configuration loading and argument parsing
4922 static int set_options(int argc
, char **argv
)
4924 int ret
= 0, c
= 0, option_index
= 0;
4925 int orig_optopt
= optopt
, orig_optind
= optind
;
4927 const char *config_path
= NULL
;
4929 optstring
= utils_generate_optstring(long_options
,
4930 sizeof(long_options
) / sizeof(struct option
));
4936 /* Check for the --config option */
4937 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4938 &option_index
)) != -1) {
4942 } else if (c
!= 'f') {
4943 /* if not equal to --config option. */
4947 if (lttng_is_setuid_setgid()) {
4948 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4951 config_path
= utils_expand_path(optarg
);
4953 ERR("Failed to resolve path: %s", optarg
);
4958 ret
= config_get_section_entries(config_path
, config_section_name
,
4959 config_entry_handler
, NULL
);
4962 ERR("Invalid configuration option at line %i", ret
);
4968 /* Reset getopt's global state */
4969 optopt
= orig_optopt
;
4970 optind
= orig_optind
;
4974 * getopt_long() will not set option_index if it encounters a
4977 c
= getopt_long(argc
, argv
, optstring
, long_options
,
4984 * Pass NULL as the long option name if popt left the index
4987 ret
= set_option(c
, optarg
,
4988 option_index
< 0 ? NULL
:
4989 long_options
[option_index
].name
);
5001 * Creates the two needed socket by the daemon.
5002 * apps_sock - The communication socket for all UST apps.
5003 * client_sock - The communication of the cli tool (lttng).
5005 static int init_daemon_socket(void)
5010 old_umask
= umask(0);
5012 /* Create client tool unix socket */
5013 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
5014 if (client_sock
< 0) {
5015 ERR("Create unix sock failed: %s", client_unix_sock_path
);
5020 /* Set the cloexec flag */
5021 ret
= utils_set_fd_cloexec(client_sock
);
5023 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5024 "Continuing but note that the consumer daemon will have a "
5025 "reference to this socket on exec()", client_sock
);
5028 /* File permission MUST be 660 */
5029 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5031 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5036 /* Create the application unix socket */
5037 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5038 if (apps_sock
< 0) {
5039 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5044 /* Set the cloexec flag */
5045 ret
= utils_set_fd_cloexec(apps_sock
);
5047 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5048 "Continuing but note that the consumer daemon will have a "
5049 "reference to this socket on exec()", apps_sock
);
5052 /* File permission MUST be 666 */
5053 ret
= chmod(apps_unix_sock_path
,
5054 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5056 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5061 DBG3("Session daemon client socket %d and application socket %d created",
5062 client_sock
, apps_sock
);
5070 * Check if the global socket is available, and if a daemon is answering at the
5071 * other side. If yes, error is returned.
5073 static int check_existing_daemon(void)
5075 /* Is there anybody out there ? */
5076 if (lttng_session_daemon_alive()) {
5084 * Set the tracing group gid onto the client socket.
5086 * Race window between mkdir and chown is OK because we are going from more
5087 * permissive (root.root) to less permissive (root.tracing).
5089 static int set_permissions(char *rundir
)
5094 gid
= utils_get_group_id(tracing_group_name
);
5096 /* Set lttng run dir */
5097 ret
= chown(rundir
, 0, gid
);
5099 ERR("Unable to set group on %s", rundir
);
5104 * Ensure all applications and tracing group can search the run
5105 * dir. Allow everyone to read the directory, since it does not
5106 * buy us anything to hide its content.
5108 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5110 ERR("Unable to set permissions on %s", rundir
);
5114 /* lttng client socket path */
5115 ret
= chown(client_unix_sock_path
, 0, gid
);
5117 ERR("Unable to set group on %s", client_unix_sock_path
);
5121 /* kconsumer error socket path */
5122 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5124 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5128 /* 64-bit ustconsumer error socket path */
5129 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5131 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5135 /* 32-bit ustconsumer compat32 error socket path */
5136 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5138 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5142 DBG("All permissions are set");
5148 * Create the lttng run directory needed for all global sockets and pipe.
5150 static int create_lttng_rundir(const char *rundir
)
5154 DBG3("Creating LTTng run directory: %s", rundir
);
5156 ret
= mkdir(rundir
, S_IRWXU
);
5158 if (errno
!= EEXIST
) {
5159 ERR("Unable to create %s", rundir
);
5171 * Setup sockets and directory needed by the kconsumerd communication with the
5174 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5178 char path
[PATH_MAX
];
5180 switch (consumer_data
->type
) {
5181 case LTTNG_CONSUMER_KERNEL
:
5182 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5184 case LTTNG_CONSUMER64_UST
:
5185 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5187 case LTTNG_CONSUMER32_UST
:
5188 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5191 ERR("Consumer type unknown");
5196 DBG2("Creating consumer directory: %s", path
);
5198 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5200 if (errno
!= EEXIST
) {
5202 ERR("Failed to create %s", path
);
5208 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5210 ERR("Unable to set group on %s", path
);
5216 /* Create the kconsumerd error unix socket */
5217 consumer_data
->err_sock
=
5218 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5219 if (consumer_data
->err_sock
< 0) {
5220 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5226 * Set the CLOEXEC flag. Return code is useless because either way, the
5229 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5231 PERROR("utils_set_fd_cloexec");
5232 /* continue anyway */
5235 /* File permission MUST be 660 */
5236 ret
= chmod(consumer_data
->err_unix_sock_path
,
5237 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5239 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5249 * Signal handler for the daemon
5251 * Simply stop all worker threads, leaving main() return gracefully after
5252 * joining all threads and calling cleanup().
5254 static void sighandler(int sig
)
5258 DBG("SIGPIPE caught");
5261 DBG("SIGINT caught");
5265 DBG("SIGTERM caught");
5269 CMM_STORE_SHARED(recv_child_signal
, 1);
5277 * Setup signal handler for :
5278 * SIGINT, SIGTERM, SIGPIPE
5280 static int set_signal_handler(void)
5283 struct sigaction sa
;
5286 if ((ret
= sigemptyset(&sigset
)) < 0) {
5287 PERROR("sigemptyset");
5291 sa
.sa_handler
= sighandler
;
5292 sa
.sa_mask
= sigset
;
5294 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5295 PERROR("sigaction");
5299 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5300 PERROR("sigaction");
5304 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5305 PERROR("sigaction");
5309 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5310 PERROR("sigaction");
5314 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5320 * Set open files limit to unlimited. This daemon can open a large number of
5321 * file descriptors in order to consumer multiple kernel traces.
5323 static void set_ulimit(void)
5328 /* The kernel does not allowed an infinite limit for open files */
5329 lim
.rlim_cur
= 65535;
5330 lim
.rlim_max
= 65535;
5332 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5334 PERROR("failed to set open files limit");
5339 * Write pidfile using the rundir and opt_pidfile.
5341 static int write_pidfile(void)
5344 char pidfile_path
[PATH_MAX
];
5349 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
5351 /* Build pidfile path from rundir and opt_pidfile. */
5352 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5353 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5355 PERROR("snprintf pidfile path");
5361 * Create pid file in rundir.
5363 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5369 * Create lockfile using the rundir and return its fd.
5371 static int create_lockfile(void)
5374 char lockfile_path
[PATH_MAX
];
5376 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5381 ret
= utils_create_lock_file(lockfile_path
);
5387 * Write agent TCP port using the rundir.
5389 static int write_agent_port(void)
5392 char path
[PATH_MAX
];
5396 ret
= snprintf(path
, sizeof(path
), "%s/"
5397 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5399 PERROR("snprintf agent port path");
5404 * Create TCP agent port file in rundir.
5406 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5415 int main(int argc
, char **argv
)
5417 int ret
= 0, retval
= 0;
5419 const char *home_path
, *env_app_timeout
;
5421 /* Initialize agent apps ht global variable */
5422 agent_apps_ht_by_sock
= NULL
;
5424 init_kernel_workarounds();
5426 rcu_register_thread();
5428 if (set_signal_handler()) {
5430 goto exit_set_signal_handler
;
5433 setup_consumerd_path();
5435 page_size
= sysconf(_SC_PAGESIZE
);
5436 if (page_size
< 0) {
5437 PERROR("sysconf _SC_PAGESIZE");
5438 page_size
= LONG_MAX
;
5439 WARN("Fallback page size to %ld", page_size
);
5443 * Parse arguments and load the daemon configuration file.
5445 * We have an exit_options exit path to free memory reserved by
5446 * set_options. This is needed because the rest of sessiond_cleanup()
5447 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5448 * depends on set_options.
5451 if (set_options(argc
, argv
)) {
5457 if (opt_daemon
|| opt_background
) {
5460 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5468 * We are in the child. Make sure all other file descriptors are
5469 * closed, in case we are called with more opened file
5470 * descriptors than the standard ones.
5472 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5478 * Starting from here, we can create threads. This needs to be after
5479 * lttng_daemonize due to RCU.
5483 * Initialize the health check subsystem. This call should set the
5484 * appropriate time values.
5486 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5487 if (!health_sessiond
) {
5488 PERROR("health_app_create error");
5490 goto exit_health_sessiond_cleanup
;
5493 if (init_ht_cleanup_quit_pipe()) {
5495 goto exit_ht_cleanup_quit_pipe
;
5498 /* Setup the thread ht_cleanup communication pipe. */
5499 if (utils_create_pipe_cloexec(ht_cleanup_pipe
)) {
5501 goto exit_ht_cleanup_pipe
;
5504 /* Set up max poll set size */
5505 if (lttng_poll_set_max_size()) {
5507 goto exit_set_max_size
;
5510 /* Create thread to clean up RCU hash tables */
5511 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5512 thread_ht_cleanup
, (void *) NULL
);
5515 PERROR("pthread_create ht_cleanup");
5517 goto exit_ht_cleanup
;
5520 /* Create thread quit pipe */
5521 if (init_thread_quit_pipe()) {
5523 goto exit_init_data
;
5526 /* Check if daemon is UID = 0 */
5527 is_root
= !getuid();
5530 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5533 goto exit_init_data
;
5536 /* Create global run dir with root access */
5537 if (create_lttng_rundir(rundir
)) {
5539 goto exit_init_data
;
5542 if (strlen(apps_unix_sock_path
) == 0) {
5543 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5544 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5547 goto exit_init_data
;
5551 if (strlen(client_unix_sock_path
) == 0) {
5552 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5553 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5556 goto exit_init_data
;
5560 /* Set global SHM for ust */
5561 if (strlen(wait_shm_path
) == 0) {
5562 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5563 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5566 goto exit_init_data
;
5570 if (strlen(health_unix_sock_path
) == 0) {
5571 ret
= snprintf(health_unix_sock_path
,
5572 sizeof(health_unix_sock_path
),
5573 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5576 goto exit_init_data
;
5580 /* Setup kernel consumerd path */
5581 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5582 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5585 goto exit_init_data
;
5587 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5588 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5591 goto exit_init_data
;
5594 DBG2("Kernel consumer err path: %s",
5595 kconsumer_data
.err_unix_sock_path
);
5596 DBG2("Kernel consumer cmd path: %s",
5597 kconsumer_data
.cmd_unix_sock_path
);
5599 home_path
= utils_get_home_dir();
5600 if (home_path
== NULL
) {
5601 /* TODO: Add --socket PATH option */
5602 ERR("Can't get HOME directory for sockets creation.");
5604 goto exit_init_data
;
5608 * Create rundir from home path. This will create something like
5611 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5614 goto exit_init_data
;
5617 if (create_lttng_rundir(rundir
)) {
5619 goto exit_init_data
;
5622 if (strlen(apps_unix_sock_path
) == 0) {
5623 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5624 DEFAULT_HOME_APPS_UNIX_SOCK
,
5628 goto exit_init_data
;
5632 /* Set the cli tool unix socket path */
5633 if (strlen(client_unix_sock_path
) == 0) {
5634 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5635 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5639 goto exit_init_data
;
5643 /* Set global SHM for ust */
5644 if (strlen(wait_shm_path
) == 0) {
5645 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5646 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5650 goto exit_init_data
;
5654 /* Set health check Unix path */
5655 if (strlen(health_unix_sock_path
) == 0) {
5656 ret
= snprintf(health_unix_sock_path
,
5657 sizeof(health_unix_sock_path
),
5658 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5662 goto exit_init_data
;
5667 lockfile_fd
= create_lockfile();
5668 if (lockfile_fd
< 0) {
5670 goto exit_init_data
;
5673 /* Set consumer initial state */
5674 kernel_consumerd_state
= CONSUMER_STOPPED
;
5675 ust_consumerd_state
= CONSUMER_STOPPED
;
5677 DBG("Client socket path %s", client_unix_sock_path
);
5678 DBG("Application socket path %s", apps_unix_sock_path
);
5679 DBG("Application wait path %s", wait_shm_path
);
5680 DBG("LTTng run directory path: %s", rundir
);
5682 /* 32 bits consumerd path setup */
5683 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5684 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5686 PERROR("snprintf 32-bit consumer error socket path");
5688 goto exit_init_data
;
5690 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5691 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5693 PERROR("snprintf 32-bit consumer command socket path");
5695 goto exit_init_data
;
5698 DBG2("UST consumer 32 bits err path: %s",
5699 ustconsumer32_data
.err_unix_sock_path
);
5700 DBG2("UST consumer 32 bits cmd path: %s",
5701 ustconsumer32_data
.cmd_unix_sock_path
);
5703 /* 64 bits consumerd path setup */
5704 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5705 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5707 PERROR("snprintf 64-bit consumer error socket path");
5709 goto exit_init_data
;
5711 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5712 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5714 PERROR("snprintf 64-bit consumer command socket path");
5716 goto exit_init_data
;
5719 DBG2("UST consumer 64 bits err path: %s",
5720 ustconsumer64_data
.err_unix_sock_path
);
5721 DBG2("UST consumer 64 bits cmd path: %s",
5722 ustconsumer64_data
.cmd_unix_sock_path
);
5725 * See if daemon already exist.
5727 if (check_existing_daemon()) {
5728 ERR("Already running daemon.\n");
5730 * We do not goto exit because we must not cleanup()
5731 * because a daemon is already running.
5734 goto exit_init_data
;
5738 * Init UST app hash table. Alloc hash table before this point since
5739 * cleanup() can get called after that point.
5741 if (ust_app_ht_alloc()) {
5742 ERR("Failed to allocate UST app hash table");
5744 goto exit_init_data
;
5748 * Initialize agent app hash table. We allocate the hash table here
5749 * since cleanup() can get called after this point.
5751 if (agent_app_ht_alloc()) {
5752 ERR("Failed to allocate Agent app hash table");
5754 goto exit_init_data
;
5758 * These actions must be executed as root. We do that *after* setting up
5759 * the sockets path because we MUST make the check for another daemon using
5760 * those paths *before* trying to set the kernel consumer sockets and init
5764 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5766 goto exit_init_data
;
5769 /* Setup kernel tracer */
5770 if (!opt_no_kernel
) {
5771 init_kernel_tracer();
5772 if (kernel_tracer_fd
>= 0) {
5773 ret
= syscall_init_table();
5775 ERR("Unable to populate syscall table. "
5776 "Syscall tracing won't work "
5777 "for this session daemon.");
5782 /* Set ulimit for open files */
5785 /* init lttng_fd tracking must be done after set_ulimit. */
5788 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5790 goto exit_init_data
;
5793 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5795 goto exit_init_data
;
5798 /* Setup the needed unix socket */
5799 if (init_daemon_socket()) {
5801 goto exit_init_data
;
5804 /* Set credentials to socket */
5805 if (is_root
&& set_permissions(rundir
)) {
5807 goto exit_init_data
;
5810 /* Get parent pid if -S, --sig-parent is specified. */
5811 if (opt_sig_parent
) {
5815 /* Setup the kernel pipe for waking up the kernel thread */
5816 if (is_root
&& !opt_no_kernel
) {
5817 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5819 goto exit_init_data
;
5823 /* Setup the thread apps communication pipe. */
5824 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5826 goto exit_init_data
;
5829 /* Setup the thread apps notify communication pipe. */
5830 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5832 goto exit_init_data
;
5835 /* Initialize global buffer per UID and PID registry. */
5836 buffer_reg_init_uid_registry();
5837 buffer_reg_init_pid_registry();
5839 /* Init UST command queue. */
5840 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5843 * Get session list pointer. This pointer MUST NOT be free'd. This list
5844 * is statically declared in session.c
5846 session_list_ptr
= session_get_list();
5850 /* Check for the application socket timeout env variable. */
5851 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5852 if (env_app_timeout
) {
5853 app_socket_timeout
= atoi(env_app_timeout
);
5855 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5858 ret
= write_pidfile();
5860 ERR("Error in write_pidfile");
5862 goto exit_init_data
;
5864 ret
= write_agent_port();
5866 ERR("Error in write_agent_port");
5868 goto exit_init_data
;
5871 /* Initialize communication library */
5873 /* Initialize TCP timeout values */
5874 lttcomm_inet_init();
5876 if (load_session_init_data(&load_info
) < 0) {
5878 goto exit_init_data
;
5880 load_info
->path
= opt_load_session_path
;
5882 /* Create health-check thread */
5883 ret
= pthread_create(&health_thread
, NULL
,
5884 thread_manage_health
, (void *) NULL
);
5887 PERROR("pthread_create health");
5892 /* Create thread to manage the client socket */
5893 ret
= pthread_create(&client_thread
, NULL
,
5894 thread_manage_clients
, (void *) NULL
);
5897 PERROR("pthread_create clients");
5902 /* Create thread to dispatch registration */
5903 ret
= pthread_create(&dispatch_thread
, NULL
,
5904 thread_dispatch_ust_registration
, (void *) NULL
);
5907 PERROR("pthread_create dispatch");
5912 /* Create thread to manage application registration. */
5913 ret
= pthread_create(®_apps_thread
, NULL
,
5914 thread_registration_apps
, (void *) NULL
);
5917 PERROR("pthread_create registration");
5922 /* Create thread to manage application socket */
5923 ret
= pthread_create(&apps_thread
, NULL
,
5924 thread_manage_apps
, (void *) NULL
);
5927 PERROR("pthread_create apps");
5932 /* Create thread to manage application notify socket */
5933 ret
= pthread_create(&apps_notify_thread
, NULL
,
5934 ust_thread_manage_notify
, (void *) NULL
);
5937 PERROR("pthread_create notify");
5939 goto exit_apps_notify
;
5942 /* Create agent registration thread. */
5943 ret
= pthread_create(&agent_reg_thread
, NULL
,
5944 agent_thread_manage_registration
, (void *) NULL
);
5947 PERROR("pthread_create agent");
5949 goto exit_agent_reg
;
5952 /* Don't start this thread if kernel tracing is not requested nor root */
5953 if (is_root
&& !opt_no_kernel
) {
5954 /* Create kernel thread to manage kernel event */
5955 ret
= pthread_create(&kernel_thread
, NULL
,
5956 thread_manage_kernel
, (void *) NULL
);
5959 PERROR("pthread_create kernel");
5965 /* Create session loading thread. */
5966 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
5970 PERROR("pthread_create load_session_thread");
5972 goto exit_load_session
;
5976 * This is where we start awaiting program completion (e.g. through
5977 * signal that asks threads to teardown).
5980 ret
= pthread_join(load_session_thread
, &status
);
5983 PERROR("pthread_join load_session_thread");
5988 if (is_root
&& !opt_no_kernel
) {
5989 ret
= pthread_join(kernel_thread
, &status
);
5992 PERROR("pthread_join");
5998 ret
= pthread_join(agent_reg_thread
, &status
);
6001 PERROR("pthread_join agent");
6006 ret
= pthread_join(apps_notify_thread
, &status
);
6009 PERROR("pthread_join apps notify");
6014 ret
= pthread_join(apps_thread
, &status
);
6017 PERROR("pthread_join apps");
6022 ret
= pthread_join(reg_apps_thread
, &status
);
6025 PERROR("pthread_join");
6031 * Join dispatch thread after joining reg_apps_thread to ensure
6032 * we don't leak applications in the queue.
6034 ret
= pthread_join(dispatch_thread
, &status
);
6037 PERROR("pthread_join");
6042 ret
= pthread_join(client_thread
, &status
);
6045 PERROR("pthread_join");
6050 ret
= pthread_join(health_thread
, &status
);
6053 PERROR("pthread_join health thread");
6060 * sessiond_cleanup() is called when no other thread is running, except
6061 * the ht_cleanup thread, which is needed to destroy the hash tables.
6063 rcu_thread_online();
6065 rcu_thread_offline();
6066 rcu_unregister_thread();
6068 ret
= notify_thread_pipe(ht_cleanup_quit_pipe
[1]);
6070 ERR("write error on ht_cleanup quit pipe");
6074 ret
= pthread_join(ht_cleanup_thread
, &status
);
6077 PERROR("pthread_join ht cleanup thread");
6083 utils_close_pipe(ht_cleanup_pipe
);
6084 exit_ht_cleanup_pipe
:
6087 * Close the ht_cleanup quit pipe.
6089 utils_close_pipe(ht_cleanup_quit_pipe
);
6090 exit_ht_cleanup_quit_pipe
:
6092 health_app_destroy(health_sessiond
);
6093 exit_health_sessiond_cleanup
:
6096 sessiond_cleanup_options();
6098 exit_set_signal_handler
:
6099 /* Ensure all prior call_rcu are done. */