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 * Cleanup the session daemon's data structures.
588 static void sessiond_cleanup(void)
591 struct ltt_session
*sess
, *stmp
;
594 DBG("Cleanup sessiond");
597 * Close the thread quit pipe. It has already done its job,
598 * since we are now called.
600 utils_close_pipe(thread_quit_pipe
);
603 * If opt_pidfile is undefined, the default file will be wiped when
604 * removing the rundir.
607 ret
= remove(opt_pidfile
);
609 PERROR("remove pidfile %s", opt_pidfile
);
613 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
616 snprintf(path
, PATH_MAX
,
618 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
619 DBG("Removing %s", path
);
622 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
623 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
);
624 DBG("Removing %s", path
);
628 snprintf(path
, PATH_MAX
,
629 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
631 DBG("Removing %s", path
);
634 snprintf(path
, PATH_MAX
,
635 DEFAULT_KCONSUMERD_PATH
,
637 DBG("Removing directory %s", path
);
640 /* ust consumerd 32 */
641 snprintf(path
, PATH_MAX
,
642 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
644 DBG("Removing %s", path
);
647 snprintf(path
, PATH_MAX
,
648 DEFAULT_USTCONSUMERD32_PATH
,
650 DBG("Removing directory %s", path
);
653 /* ust consumerd 64 */
654 snprintf(path
, PATH_MAX
,
655 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
657 DBG("Removing %s", path
);
660 snprintf(path
, PATH_MAX
,
661 DEFAULT_USTCONSUMERD64_PATH
,
663 DBG("Removing directory %s", path
);
666 DBG("Cleaning up all sessions");
668 /* Destroy session list mutex */
669 if (session_list_ptr
!= NULL
) {
670 pthread_mutex_destroy(&session_list_ptr
->lock
);
672 /* Cleanup ALL session */
673 cds_list_for_each_entry_safe(sess
, stmp
,
674 &session_list_ptr
->head
, list
) {
675 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
679 DBG("Cleaning up all agent apps");
680 agent_app_ht_clean();
682 DBG("Closing all UST sockets");
683 ust_app_clean_list();
684 buffer_reg_destroy_registries();
686 if (is_root
&& !opt_no_kernel
) {
687 DBG2("Closing kernel fd");
688 if (kernel_tracer_fd
>= 0) {
689 ret
= close(kernel_tracer_fd
);
694 DBG("Unloading kernel modules");
695 modprobe_remove_lttng_all();
699 close_consumer_sockets();
702 load_session_destroy_data(load_info
);
707 * Cleanup lock file by deleting it and finaly closing it which will
708 * release the file system lock.
710 if (lockfile_fd
>= 0) {
711 char lockfile_path
[PATH_MAX
];
713 ret
= generate_lock_file_path(lockfile_path
,
714 sizeof(lockfile_path
));
716 ret
= remove(lockfile_path
);
718 PERROR("remove lock file");
720 ret
= close(lockfile_fd
);
722 PERROR("close lock file");
728 * We do NOT rmdir rundir because there are other processes
729 * using it, for instance lttng-relayd, which can start in
730 * parallel with this teardown.
737 * Cleanup the daemon's option data structures.
739 static void sessiond_cleanup_options(void)
741 DBG("Cleaning up options");
744 * If the override option is set, the pointer points to a *non* const
745 * thus freeing it even though the variable type is set to const.
747 if (tracing_group_name_override
) {
748 free((void *) tracing_group_name
);
750 if (consumerd32_bin_override
) {
751 free((void *) consumerd32_bin
);
753 if (consumerd64_bin_override
) {
754 free((void *) consumerd64_bin
);
756 if (consumerd32_libdir_override
) {
757 free((void *) consumerd32_libdir
);
759 if (consumerd64_libdir_override
) {
760 free((void *) consumerd64_libdir
);
764 free(opt_load_session_path
);
765 free(kmod_probes_list
);
766 free(kmod_extra_probes_list
);
769 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
770 "Matthew, BEET driven development works!%c[%dm",
771 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
776 * Send data on a unix socket using the liblttsessiondcomm API.
778 * Return lttcomm error code.
780 static int send_unix_sock(int sock
, void *buf
, size_t len
)
782 /* Check valid length */
787 return lttcomm_send_unix_sock(sock
, buf
, len
);
791 * Free memory of a command context structure.
793 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
795 DBG("Clean command context structure");
797 if ((*cmd_ctx
)->llm
) {
798 free((*cmd_ctx
)->llm
);
800 if ((*cmd_ctx
)->lsm
) {
801 free((*cmd_ctx
)->lsm
);
809 * Notify UST applications using the shm mmap futex.
811 static int notify_ust_apps(int active
)
815 DBG("Notifying applications of session daemon state: %d", active
);
817 /* See shm.c for this call implying mmap, shm and futex calls */
818 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
819 if (wait_shm_mmap
== NULL
) {
823 /* Wake waiting process */
824 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
826 /* Apps notified successfully */
834 * Setup the outgoing data buffer for the response (llm) by allocating the
835 * right amount of memory and copying the original information from the lsm
838 * Return total size of the buffer pointed by buf.
840 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
846 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
847 if (cmd_ctx
->llm
== NULL
) {
853 /* Copy common data */
854 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
855 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
857 cmd_ctx
->llm
->data_size
= size
;
858 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
867 * Update the kernel poll set of all channel fd available over all tracing
868 * session. Add the wakeup pipe at the end of the set.
870 static int update_kernel_poll(struct lttng_poll_event
*events
)
873 struct ltt_session
*session
;
874 struct ltt_kernel_channel
*channel
;
876 DBG("Updating kernel poll set");
879 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
880 session_lock(session
);
881 if (session
->kernel_session
== NULL
) {
882 session_unlock(session
);
886 cds_list_for_each_entry(channel
,
887 &session
->kernel_session
->channel_list
.head
, list
) {
888 /* Add channel fd to the kernel poll set */
889 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
891 session_unlock(session
);
894 DBG("Channel fd %d added to kernel set", channel
->fd
);
896 session_unlock(session
);
898 session_unlock_list();
903 session_unlock_list();
908 * Find the channel fd from 'fd' over all tracing session. When found, check
909 * for new channel stream and send those stream fds to the kernel consumer.
911 * Useful for CPU hotplug feature.
913 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
916 struct ltt_session
*session
;
917 struct ltt_kernel_session
*ksess
;
918 struct ltt_kernel_channel
*channel
;
920 DBG("Updating kernel streams for channel fd %d", fd
);
923 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
924 session_lock(session
);
925 if (session
->kernel_session
== NULL
) {
926 session_unlock(session
);
929 ksess
= session
->kernel_session
;
931 cds_list_for_each_entry(channel
,
932 &ksess
->channel_list
.head
, list
) {
933 struct lttng_ht_iter iter
;
934 struct consumer_socket
*socket
;
936 if (channel
->fd
!= fd
) {
939 DBG("Channel found, updating kernel streams");
940 ret
= kernel_open_channel_stream(channel
);
944 /* Update the stream global counter */
945 ksess
->stream_count_global
+= ret
;
948 * Have we already sent fds to the consumer? If yes, it
949 * means that tracing is started so it is safe to send
950 * our updated stream fds.
952 if (ksess
->consumer_fds_sent
!= 1
953 || ksess
->consumer
== NULL
) {
959 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
960 &iter
.iter
, socket
, node
.node
) {
961 pthread_mutex_lock(socket
->lock
);
962 ret
= kernel_consumer_send_channel_stream(socket
,
964 session
->output_traces
? 1 : 0);
965 pthread_mutex_unlock(socket
->lock
);
973 session_unlock(session
);
975 session_unlock_list();
979 session_unlock(session
);
980 session_unlock_list();
985 * For each tracing session, update newly registered apps. The session list
986 * lock MUST be acquired before calling this.
988 static void update_ust_app(int app_sock
)
990 struct ltt_session
*sess
, *stmp
;
992 /* Consumer is in an ERROR state. Stop any application update. */
993 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
994 /* Stop the update process since the consumer is dead. */
998 /* For all tracing session(s) */
999 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
1000 struct ust_app
*app
;
1003 if (!sess
->ust_session
) {
1004 goto unlock_session
;
1008 assert(app_sock
>= 0);
1009 app
= ust_app_find_by_sock(app_sock
);
1012 * Application can be unregistered before so
1013 * this is possible hence simply stopping the
1016 DBG3("UST app update failed to find app sock %d",
1020 ust_app_global_update(sess
->ust_session
, app
);
1024 session_unlock(sess
);
1029 * This thread manage event coming from the kernel.
1031 * Features supported in this thread:
1034 static void *thread_manage_kernel(void *data
)
1036 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1037 uint32_t revents
, nb_fd
;
1039 struct lttng_poll_event events
;
1041 DBG("[thread] Thread manage kernel started");
1043 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1046 * This first step of the while is to clean this structure which could free
1047 * non NULL pointers so initialize it before the loop.
1049 lttng_poll_init(&events
);
1051 if (testpoint(sessiond_thread_manage_kernel
)) {
1052 goto error_testpoint
;
1055 health_code_update();
1057 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1058 goto error_testpoint
;
1062 health_code_update();
1064 if (update_poll_flag
== 1) {
1065 /* Clean events object. We are about to populate it again. */
1066 lttng_poll_clean(&events
);
1068 ret
= sessiond_set_thread_pollset(&events
, 2);
1070 goto error_poll_create
;
1073 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1078 /* This will add the available kernel channel if any. */
1079 ret
= update_kernel_poll(&events
);
1083 update_poll_flag
= 0;
1086 DBG("Thread kernel polling");
1088 /* Poll infinite value of time */
1090 health_poll_entry();
1091 ret
= lttng_poll_wait(&events
, -1);
1092 DBG("Thread kernel return from poll on %d fds",
1093 LTTNG_POLL_GETNB(&events
));
1097 * Restart interrupted system call.
1099 if (errno
== EINTR
) {
1103 } else if (ret
== 0) {
1104 /* Should not happen since timeout is infinite */
1105 ERR("Return value of poll is 0 with an infinite timeout.\n"
1106 "This should not have happened! Continuing...");
1112 for (i
= 0; i
< nb_fd
; i
++) {
1113 /* Fetch once the poll data */
1114 revents
= LTTNG_POLL_GETEV(&events
, i
);
1115 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1117 health_code_update();
1120 /* No activity for this FD (poll implementation). */
1124 /* Thread quit pipe has been closed. Killing thread. */
1125 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1131 /* Check for data on kernel pipe */
1132 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
1133 (void) lttng_read(kernel_poll_pipe
[0],
1136 * Ret value is useless here, if this pipe gets any actions an
1137 * update is required anyway.
1139 update_poll_flag
= 1;
1143 * New CPU detected by the kernel. Adding kernel stream to
1144 * kernel session and updating the kernel consumer
1146 if (revents
& LPOLLIN
) {
1147 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1153 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1154 * and unregister kernel stream at this point.
1163 lttng_poll_clean(&events
);
1166 utils_close_pipe(kernel_poll_pipe
);
1167 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1170 ERR("Health error occurred in %s", __func__
);
1171 WARN("Kernel thread died unexpectedly. "
1172 "Kernel tracing can continue but CPU hotplug is disabled.");
1174 health_unregister(health_sessiond
);
1175 DBG("Kernel thread dying");
1180 * Signal pthread condition of the consumer data that the thread.
1182 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1184 pthread_mutex_lock(&data
->cond_mutex
);
1187 * The state is set before signaling. It can be any value, it's the waiter
1188 * job to correctly interpret this condition variable associated to the
1189 * consumer pthread_cond.
1191 * A value of 0 means that the corresponding thread of the consumer data
1192 * was not started. 1 indicates that the thread has started and is ready
1193 * for action. A negative value means that there was an error during the
1196 data
->consumer_thread_is_ready
= state
;
1197 (void) pthread_cond_signal(&data
->cond
);
1199 pthread_mutex_unlock(&data
->cond_mutex
);
1203 * This thread manage the consumer error sent back to the session daemon.
1205 static void *thread_manage_consumer(void *data
)
1207 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1208 uint32_t revents
, nb_fd
;
1209 enum lttcomm_return_code code
;
1210 struct lttng_poll_event events
;
1211 struct consumer_data
*consumer_data
= data
;
1213 DBG("[thread] Manage consumer started");
1215 rcu_register_thread();
1216 rcu_thread_online();
1218 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1220 health_code_update();
1223 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1224 * metadata_sock. Nothing more will be added to this poll set.
1226 ret
= sessiond_set_thread_pollset(&events
, 3);
1232 * The error socket here is already in a listening state which was done
1233 * just before spawning this thread to avoid a race between the consumer
1234 * daemon exec trying to connect and the listen() call.
1236 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1241 health_code_update();
1243 /* Infinite blocking call, waiting for transmission */
1245 health_poll_entry();
1247 if (testpoint(sessiond_thread_manage_consumer
)) {
1251 ret
= lttng_poll_wait(&events
, -1);
1255 * Restart interrupted system call.
1257 if (errno
== EINTR
) {
1265 for (i
= 0; i
< nb_fd
; i
++) {
1266 /* Fetch once the poll data */
1267 revents
= LTTNG_POLL_GETEV(&events
, i
);
1268 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1270 health_code_update();
1273 /* No activity for this FD (poll implementation). */
1277 /* Thread quit pipe has been closed. Killing thread. */
1278 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1284 /* Event on the registration socket */
1285 if (pollfd
== consumer_data
->err_sock
) {
1286 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1287 ERR("consumer err socket poll error");
1293 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1299 * Set the CLOEXEC flag. Return code is useless because either way, the
1302 (void) utils_set_fd_cloexec(sock
);
1304 health_code_update();
1306 DBG2("Receiving code from consumer err_sock");
1308 /* Getting status code from kconsumerd */
1309 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1310 sizeof(enum lttcomm_return_code
));
1315 health_code_update();
1316 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1317 /* Connect both socket, command and metadata. */
1318 consumer_data
->cmd_sock
=
1319 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1320 consumer_data
->metadata_fd
=
1321 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1322 if (consumer_data
->cmd_sock
< 0
1323 || consumer_data
->metadata_fd
< 0) {
1324 PERROR("consumer connect cmd socket");
1325 /* On error, signal condition and quit. */
1326 signal_consumer_condition(consumer_data
, -1);
1329 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1330 /* Create metadata socket lock. */
1331 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1332 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1333 PERROR("zmalloc pthread mutex");
1337 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1339 signal_consumer_condition(consumer_data
, 1);
1340 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1341 DBG("Consumer metadata socket ready (fd: %d)",
1342 consumer_data
->metadata_fd
);
1344 ERR("consumer error when waiting for SOCK_READY : %s",
1345 lttcomm_get_readable_code(-code
));
1349 /* Remove the consumerd error sock since we've established a connexion */
1350 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1355 /* Add new accepted error socket. */
1356 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1361 /* Add metadata socket that is successfully connected. */
1362 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1363 LPOLLIN
| LPOLLRDHUP
);
1368 health_code_update();
1370 /* Infinite blocking call, waiting for transmission */
1373 health_code_update();
1375 /* Exit the thread because the thread quit pipe has been triggered. */
1377 /* Not a health error. */
1382 health_poll_entry();
1383 ret
= lttng_poll_wait(&events
, -1);
1387 * Restart interrupted system call.
1389 if (errno
== EINTR
) {
1397 for (i
= 0; i
< nb_fd
; i
++) {
1398 /* Fetch once the poll data */
1399 revents
= LTTNG_POLL_GETEV(&events
, i
);
1400 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1402 health_code_update();
1405 /* No activity for this FD (poll implementation). */
1410 * Thread quit pipe has been triggered, flag that we should stop
1411 * but continue the current loop to handle potential data from
1414 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1416 if (pollfd
== sock
) {
1417 /* Event on the consumerd socket */
1418 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1419 ERR("consumer err socket second poll error");
1422 health_code_update();
1423 /* Wait for any kconsumerd error */
1424 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1425 sizeof(enum lttcomm_return_code
));
1427 ERR("consumer closed the command socket");
1431 ERR("consumer return code : %s",
1432 lttcomm_get_readable_code(-code
));
1435 } else if (pollfd
== consumer_data
->metadata_fd
) {
1436 /* UST metadata requests */
1437 ret
= ust_consumer_metadata_request(
1438 &consumer_data
->metadata_sock
);
1440 ERR("Handling metadata request");
1444 /* No need for an else branch all FDs are tested prior. */
1446 health_code_update();
1452 * We lock here because we are about to close the sockets and some other
1453 * thread might be using them so get exclusive access which will abort all
1454 * other consumer command by other threads.
1456 pthread_mutex_lock(&consumer_data
->lock
);
1458 /* Immediately set the consumerd state to stopped */
1459 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1460 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1461 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1462 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1463 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1465 /* Code flow error... */
1469 if (consumer_data
->err_sock
>= 0) {
1470 ret
= close(consumer_data
->err_sock
);
1474 consumer_data
->err_sock
= -1;
1476 if (consumer_data
->cmd_sock
>= 0) {
1477 ret
= close(consumer_data
->cmd_sock
);
1481 consumer_data
->cmd_sock
= -1;
1483 if (consumer_data
->metadata_sock
.fd_ptr
&&
1484 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1485 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1497 unlink(consumer_data
->err_unix_sock_path
);
1498 unlink(consumer_data
->cmd_unix_sock_path
);
1499 consumer_data
->pid
= 0;
1500 pthread_mutex_unlock(&consumer_data
->lock
);
1502 /* Cleanup metadata socket mutex. */
1503 if (consumer_data
->metadata_sock
.lock
) {
1504 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1505 free(consumer_data
->metadata_sock
.lock
);
1507 lttng_poll_clean(&events
);
1511 ERR("Health error occurred in %s", __func__
);
1513 health_unregister(health_sessiond
);
1514 DBG("consumer thread cleanup completed");
1516 rcu_thread_offline();
1517 rcu_unregister_thread();
1523 * This thread manage application communication.
1525 static void *thread_manage_apps(void *data
)
1527 int i
, ret
, pollfd
, err
= -1;
1529 uint32_t revents
, nb_fd
;
1530 struct lttng_poll_event events
;
1532 DBG("[thread] Manage application started");
1534 rcu_register_thread();
1535 rcu_thread_online();
1537 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1539 if (testpoint(sessiond_thread_manage_apps
)) {
1540 goto error_testpoint
;
1543 health_code_update();
1545 ret
= sessiond_set_thread_pollset(&events
, 2);
1547 goto error_poll_create
;
1550 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1555 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1559 health_code_update();
1562 DBG("Apps thread polling");
1564 /* Inifinite blocking call, waiting for transmission */
1566 health_poll_entry();
1567 ret
= lttng_poll_wait(&events
, -1);
1568 DBG("Apps thread return from poll on %d fds",
1569 LTTNG_POLL_GETNB(&events
));
1573 * Restart interrupted system call.
1575 if (errno
== EINTR
) {
1583 for (i
= 0; i
< nb_fd
; i
++) {
1584 /* Fetch once the poll data */
1585 revents
= LTTNG_POLL_GETEV(&events
, i
);
1586 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1588 health_code_update();
1591 /* No activity for this FD (poll implementation). */
1595 /* Thread quit pipe has been closed. Killing thread. */
1596 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1602 /* Inspect the apps cmd pipe */
1603 if (pollfd
== apps_cmd_pipe
[0]) {
1604 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1605 ERR("Apps command pipe error");
1607 } else if (revents
& LPOLLIN
) {
1611 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1612 if (size_ret
< sizeof(sock
)) {
1613 PERROR("read apps cmd pipe");
1617 health_code_update();
1620 * We only monitor the error events of the socket. This
1621 * thread does not handle any incoming data from UST
1624 ret
= lttng_poll_add(&events
, sock
,
1625 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1630 DBG("Apps with sock %d added to poll set", sock
);
1634 * At this point, we know that a registered application made
1635 * the event at poll_wait.
1637 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1638 /* Removing from the poll set */
1639 ret
= lttng_poll_del(&events
, pollfd
);
1644 /* Socket closed on remote end. */
1645 ust_app_unregister(pollfd
);
1649 health_code_update();
1655 lttng_poll_clean(&events
);
1658 utils_close_pipe(apps_cmd_pipe
);
1659 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1662 * We don't clean the UST app hash table here since already registered
1663 * applications can still be controlled so let them be until the session
1664 * daemon dies or the applications stop.
1669 ERR("Health error occurred in %s", __func__
);
1671 health_unregister(health_sessiond
);
1672 DBG("Application communication apps thread cleanup complete");
1673 rcu_thread_offline();
1674 rcu_unregister_thread();
1679 * Send a socket to a thread This is called from the dispatch UST registration
1680 * thread once all sockets are set for the application.
1682 * The sock value can be invalid, we don't really care, the thread will handle
1683 * it and make the necessary cleanup if so.
1685 * On success, return 0 else a negative value being the errno message of the
1688 static int send_socket_to_thread(int fd
, int sock
)
1693 * It's possible that the FD is set as invalid with -1 concurrently just
1694 * before calling this function being a shutdown state of the thread.
1701 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1702 if (ret
< sizeof(sock
)) {
1703 PERROR("write apps pipe %d", fd
);
1710 /* All good. Don't send back the write positive ret value. */
1717 * Sanitize the wait queue of the dispatch registration thread meaning removing
1718 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1719 * notify socket is never received.
1721 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1723 int ret
, nb_fd
= 0, i
;
1724 unsigned int fd_added
= 0;
1725 struct lttng_poll_event events
;
1726 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1730 lttng_poll_init(&events
);
1732 /* Just skip everything for an empty queue. */
1733 if (!wait_queue
->count
) {
1737 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1742 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1743 &wait_queue
->head
, head
) {
1744 assert(wait_node
->app
);
1745 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1746 LPOLLHUP
| LPOLLERR
);
1759 * Poll but don't block so we can quickly identify the faulty events and
1760 * clean them afterwards from the wait queue.
1762 ret
= lttng_poll_wait(&events
, 0);
1768 for (i
= 0; i
< nb_fd
; i
++) {
1769 /* Get faulty FD. */
1770 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1771 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1774 /* No activity for this FD (poll implementation). */
1778 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1779 &wait_queue
->head
, head
) {
1780 if (pollfd
== wait_node
->app
->sock
&&
1781 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1782 cds_list_del(&wait_node
->head
);
1783 wait_queue
->count
--;
1784 ust_app_destroy(wait_node
->app
);
1792 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1796 lttng_poll_clean(&events
);
1800 lttng_poll_clean(&events
);
1802 ERR("Unable to sanitize wait queue");
1807 * Dispatch request from the registration threads to the application
1808 * communication thread.
1810 static void *thread_dispatch_ust_registration(void *data
)
1813 struct cds_wfcq_node
*node
;
1814 struct ust_command
*ust_cmd
= NULL
;
1815 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1816 struct ust_reg_wait_queue wait_queue
= {
1820 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1822 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1823 goto error_testpoint
;
1826 health_code_update();
1828 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1830 DBG("[thread] Dispatch UST command started");
1832 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1833 health_code_update();
1835 /* Atomically prepare the queue futex */
1836 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1839 struct ust_app
*app
= NULL
;
1843 * Make sure we don't have node(s) that have hung up before receiving
1844 * the notify socket. This is to clean the list in order to avoid
1845 * memory leaks from notify socket that are never seen.
1847 sanitize_wait_queue(&wait_queue
);
1849 health_code_update();
1850 /* Dequeue command for registration */
1851 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1853 DBG("Woken up but nothing in the UST command queue");
1854 /* Continue thread execution */
1858 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1860 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1861 " gid:%d sock:%d name:%s (version %d.%d)",
1862 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1863 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1864 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1865 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1867 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1868 wait_node
= zmalloc(sizeof(*wait_node
));
1870 PERROR("zmalloc wait_node dispatch");
1871 ret
= close(ust_cmd
->sock
);
1873 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1875 lttng_fd_put(LTTNG_FD_APPS
, 1);
1879 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1881 /* Create application object if socket is CMD. */
1882 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1884 if (!wait_node
->app
) {
1885 ret
= close(ust_cmd
->sock
);
1887 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1889 lttng_fd_put(LTTNG_FD_APPS
, 1);
1895 * Add application to the wait queue so we can set the notify
1896 * socket before putting this object in the global ht.
1898 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1903 * We have to continue here since we don't have the notify
1904 * socket and the application MUST be added to the hash table
1905 * only at that moment.
1910 * Look for the application in the local wait queue and set the
1911 * notify socket if found.
1913 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1914 &wait_queue
.head
, head
) {
1915 health_code_update();
1916 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1917 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1918 cds_list_del(&wait_node
->head
);
1920 app
= wait_node
->app
;
1922 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1928 * With no application at this stage the received socket is
1929 * basically useless so close it before we free the cmd data
1930 * structure for good.
1933 ret
= close(ust_cmd
->sock
);
1935 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1937 lttng_fd_put(LTTNG_FD_APPS
, 1);
1944 * @session_lock_list
1946 * Lock the global session list so from the register up to the
1947 * registration done message, no thread can see the application
1948 * and change its state.
1950 session_lock_list();
1954 * Add application to the global hash table. This needs to be
1955 * done before the update to the UST registry can locate the
1960 /* Set app version. This call will print an error if needed. */
1961 (void) ust_app_version(app
);
1963 /* Send notify socket through the notify pipe. */
1964 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1968 session_unlock_list();
1970 * No notify thread, stop the UST tracing. However, this is
1971 * not an internal error of the this thread thus setting
1972 * the health error code to a normal exit.
1979 * Update newly registered application with the tracing
1980 * registry info already enabled information.
1982 update_ust_app(app
->sock
);
1985 * Don't care about return value. Let the manage apps threads
1986 * handle app unregistration upon socket close.
1988 (void) ust_app_register_done(app
);
1991 * Even if the application socket has been closed, send the app
1992 * to the thread and unregistration will take place at that
1995 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1998 session_unlock_list();
2000 * No apps. thread, stop the UST tracing. However, this is
2001 * not an internal error of the this thread thus setting
2002 * the health error code to a normal exit.
2009 session_unlock_list();
2011 } while (node
!= NULL
);
2013 health_poll_entry();
2014 /* Futex wait on queue. Blocking call on futex() */
2015 futex_nto1_wait(&ust_cmd_queue
.futex
);
2018 /* Normal exit, no error */
2022 /* Clean up wait queue. */
2023 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2024 &wait_queue
.head
, head
) {
2025 cds_list_del(&wait_node
->head
);
2030 /* Empty command queue. */
2032 /* Dequeue command for registration */
2033 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2037 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2038 ret
= close(ust_cmd
->sock
);
2040 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2042 lttng_fd_put(LTTNG_FD_APPS
, 1);
2047 DBG("Dispatch thread dying");
2050 ERR("Health error occurred in %s", __func__
);
2052 health_unregister(health_sessiond
);
2057 * This thread manage application registration.
2059 static void *thread_registration_apps(void *data
)
2061 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2062 uint32_t revents
, nb_fd
;
2063 struct lttng_poll_event events
;
2065 * Get allocated in this thread, enqueued to a global queue, dequeued and
2066 * freed in the manage apps thread.
2068 struct ust_command
*ust_cmd
= NULL
;
2070 DBG("[thread] Manage application registration started");
2072 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2074 if (testpoint(sessiond_thread_registration_apps
)) {
2075 goto error_testpoint
;
2078 ret
= lttcomm_listen_unix_sock(apps_sock
);
2084 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2085 * more will be added to this poll set.
2087 ret
= sessiond_set_thread_pollset(&events
, 2);
2089 goto error_create_poll
;
2092 /* Add the application registration socket */
2093 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2095 goto error_poll_add
;
2098 /* Notify all applications to register */
2099 ret
= notify_ust_apps(1);
2101 ERR("Failed to notify applications or create the wait shared memory.\n"
2102 "Execution continues but there might be problem for already\n"
2103 "running applications that wishes to register.");
2107 DBG("Accepting application registration");
2109 /* Inifinite blocking call, waiting for transmission */
2111 health_poll_entry();
2112 ret
= lttng_poll_wait(&events
, -1);
2116 * Restart interrupted system call.
2118 if (errno
== EINTR
) {
2126 for (i
= 0; i
< nb_fd
; i
++) {
2127 health_code_update();
2129 /* Fetch once the poll data */
2130 revents
= LTTNG_POLL_GETEV(&events
, i
);
2131 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2134 /* No activity for this FD (poll implementation). */
2138 /* Thread quit pipe has been closed. Killing thread. */
2139 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2145 /* Event on the registration socket */
2146 if (pollfd
== apps_sock
) {
2147 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2148 ERR("Register apps socket poll error");
2150 } else if (revents
& LPOLLIN
) {
2151 sock
= lttcomm_accept_unix_sock(apps_sock
);
2157 * Set socket timeout for both receiving and ending.
2158 * app_socket_timeout is in seconds, whereas
2159 * lttcomm_setsockopt_rcv_timeout and
2160 * lttcomm_setsockopt_snd_timeout expect msec as
2163 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2164 app_socket_timeout
* 1000);
2165 (void) lttcomm_setsockopt_snd_timeout(sock
,
2166 app_socket_timeout
* 1000);
2169 * Set the CLOEXEC flag. Return code is useless because
2170 * either way, the show must go on.
2172 (void) utils_set_fd_cloexec(sock
);
2174 /* Create UST registration command for enqueuing */
2175 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2176 if (ust_cmd
== NULL
) {
2177 PERROR("ust command zmalloc");
2186 * Using message-based transmissions to ensure we don't
2187 * have to deal with partially received messages.
2189 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2191 ERR("Exhausted file descriptors allowed for applications.");
2201 health_code_update();
2202 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2205 /* Close socket of the application. */
2210 lttng_fd_put(LTTNG_FD_APPS
, 1);
2214 health_code_update();
2216 ust_cmd
->sock
= sock
;
2219 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2220 " gid:%d sock:%d name:%s (version %d.%d)",
2221 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2222 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2223 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2224 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2227 * Lock free enqueue the registration request. The red pill
2228 * has been taken! This apps will be part of the *system*.
2230 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2233 * Wake the registration queue futex. Implicit memory
2234 * barrier with the exchange in cds_wfcq_enqueue.
2236 futex_nto1_wake(&ust_cmd_queue
.futex
);
2244 /* Notify that the registration thread is gone */
2247 if (apps_sock
>= 0) {
2248 ret
= close(apps_sock
);
2258 lttng_fd_put(LTTNG_FD_APPS
, 1);
2260 unlink(apps_unix_sock_path
);
2263 lttng_poll_clean(&events
);
2267 DBG("UST Registration thread cleanup complete");
2270 ERR("Health error occurred in %s", __func__
);
2272 health_unregister(health_sessiond
);
2278 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2279 * exec or it will fails.
2281 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2284 struct timespec timeout
;
2286 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2287 consumer_data
->consumer_thread_is_ready
= 0;
2289 /* Setup pthread condition */
2290 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2293 PERROR("pthread_condattr_init consumer data");
2298 * Set the monotonic clock in order to make sure we DO NOT jump in time
2299 * between the clock_gettime() call and the timedwait call. See bug #324
2300 * for a more details and how we noticed it.
2302 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2305 PERROR("pthread_condattr_setclock consumer data");
2309 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2312 PERROR("pthread_cond_init consumer data");
2316 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2320 PERROR("pthread_create consumer");
2325 /* We are about to wait on a pthread condition */
2326 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2328 /* Get time for sem_timedwait absolute timeout */
2329 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2331 * Set the timeout for the condition timed wait even if the clock gettime
2332 * call fails since we might loop on that call and we want to avoid to
2333 * increment the timeout too many times.
2335 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2338 * The following loop COULD be skipped in some conditions so this is why we
2339 * set ret to 0 in order to make sure at least one round of the loop is
2345 * Loop until the condition is reached or when a timeout is reached. Note
2346 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2347 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2348 * possible. This loop does not take any chances and works with both of
2351 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2352 if (clock_ret
< 0) {
2353 PERROR("clock_gettime spawn consumer");
2354 /* Infinite wait for the consumerd thread to be ready */
2355 ret
= pthread_cond_wait(&consumer_data
->cond
,
2356 &consumer_data
->cond_mutex
);
2358 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2359 &consumer_data
->cond_mutex
, &timeout
);
2363 /* Release the pthread condition */
2364 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2368 if (ret
== ETIMEDOUT
) {
2372 * Call has timed out so we kill the kconsumerd_thread and return
2375 ERR("Condition timed out. The consumer thread was never ready."
2377 pth_ret
= pthread_cancel(consumer_data
->thread
);
2379 PERROR("pthread_cancel consumer thread");
2382 PERROR("pthread_cond_wait failed consumer thread");
2384 /* Caller is expecting a negative value on failure. */
2389 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2390 if (consumer_data
->pid
== 0) {
2391 ERR("Consumerd did not start");
2392 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2395 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2404 * Join consumer thread
2406 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2410 /* Consumer pid must be a real one. */
2411 if (consumer_data
->pid
> 0) {
2413 ret
= kill(consumer_data
->pid
, SIGTERM
);
2415 PERROR("Error killing consumer daemon");
2418 return pthread_join(consumer_data
->thread
, &status
);
2425 * Fork and exec a consumer daemon (consumerd).
2427 * Return pid if successful else -1.
2429 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2433 const char *consumer_to_use
;
2434 const char *verbosity
;
2437 DBG("Spawning consumerd");
2444 if (opt_verbose_consumer
) {
2445 verbosity
= "--verbose";
2446 } else if (lttng_opt_quiet
) {
2447 verbosity
= "--quiet";
2452 switch (consumer_data
->type
) {
2453 case LTTNG_CONSUMER_KERNEL
:
2455 * Find out which consumerd to execute. We will first try the
2456 * 64-bit path, then the sessiond's installation directory, and
2457 * fallback on the 32-bit one,
2459 DBG3("Looking for a kernel consumer at these locations:");
2460 DBG3(" 1) %s", consumerd64_bin
);
2461 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2462 DBG3(" 3) %s", consumerd32_bin
);
2463 if (stat(consumerd64_bin
, &st
) == 0) {
2464 DBG3("Found location #1");
2465 consumer_to_use
= consumerd64_bin
;
2466 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2467 DBG3("Found location #2");
2468 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2469 } else if (stat(consumerd32_bin
, &st
) == 0) {
2470 DBG3("Found location #3");
2471 consumer_to_use
= consumerd32_bin
;
2473 DBG("Could not find any valid consumerd executable");
2477 DBG("Using kernel consumer at: %s", consumer_to_use
);
2478 ret
= execl(consumer_to_use
,
2479 "lttng-consumerd", verbosity
, "-k",
2480 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2481 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2482 "--group", tracing_group_name
,
2485 case LTTNG_CONSUMER64_UST
:
2487 char *tmpnew
= NULL
;
2489 if (consumerd64_libdir
[0] != '\0') {
2493 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2497 tmplen
= strlen("LD_LIBRARY_PATH=")
2498 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2499 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2504 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2505 strcat(tmpnew
, consumerd64_libdir
);
2506 if (tmp
[0] != '\0') {
2507 strcat(tmpnew
, ":");
2508 strcat(tmpnew
, tmp
);
2510 ret
= putenv(tmpnew
);
2517 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2518 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2519 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2520 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2521 "--group", tracing_group_name
,
2523 if (consumerd64_libdir
[0] != '\0') {
2528 case LTTNG_CONSUMER32_UST
:
2530 char *tmpnew
= NULL
;
2532 if (consumerd32_libdir
[0] != '\0') {
2536 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2540 tmplen
= strlen("LD_LIBRARY_PATH=")
2541 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2542 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2547 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2548 strcat(tmpnew
, consumerd32_libdir
);
2549 if (tmp
[0] != '\0') {
2550 strcat(tmpnew
, ":");
2551 strcat(tmpnew
, tmp
);
2553 ret
= putenv(tmpnew
);
2560 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2561 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2562 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2563 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2564 "--group", tracing_group_name
,
2566 if (consumerd32_libdir
[0] != '\0') {
2572 PERROR("unknown consumer type");
2576 PERROR("Consumer execl()");
2578 /* Reaching this point, we got a failure on our execl(). */
2580 } else if (pid
> 0) {
2583 PERROR("start consumer fork");
2591 * Spawn the consumerd daemon and session daemon thread.
2593 static int start_consumerd(struct consumer_data
*consumer_data
)
2598 * Set the listen() state on the socket since there is a possible race
2599 * between the exec() of the consumer daemon and this call if place in the
2600 * consumer thread. See bug #366 for more details.
2602 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2607 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2608 if (consumer_data
->pid
!= 0) {
2609 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2613 ret
= spawn_consumerd(consumer_data
);
2615 ERR("Spawning consumerd failed");
2616 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2620 /* Setting up the consumer_data pid */
2621 consumer_data
->pid
= ret
;
2622 DBG2("Consumer pid %d", consumer_data
->pid
);
2623 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2625 DBG2("Spawning consumer control thread");
2626 ret
= spawn_consumer_thread(consumer_data
);
2628 ERR("Fatal error spawning consumer control thread");
2636 /* Cleanup already created sockets on error. */
2637 if (consumer_data
->err_sock
>= 0) {
2640 err
= close(consumer_data
->err_sock
);
2642 PERROR("close consumer data error socket");
2649 * Setup necessary data for kernel tracer action.
2651 static int init_kernel_tracer(void)
2655 /* Modprobe lttng kernel modules */
2656 ret
= modprobe_lttng_control();
2661 /* Open debugfs lttng */
2662 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2663 if (kernel_tracer_fd
< 0) {
2664 DBG("Failed to open %s", module_proc_lttng
);
2669 /* Validate kernel version */
2670 ret
= kernel_validate_version(kernel_tracer_fd
);
2675 ret
= modprobe_lttng_data();
2680 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2684 modprobe_remove_lttng_control();
2685 ret
= close(kernel_tracer_fd
);
2689 kernel_tracer_fd
= -1;
2690 return LTTNG_ERR_KERN_VERSION
;
2693 ret
= close(kernel_tracer_fd
);
2699 modprobe_remove_lttng_control();
2702 WARN("No kernel tracer available");
2703 kernel_tracer_fd
= -1;
2705 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2707 return LTTNG_ERR_KERN_NA
;
2713 * Copy consumer output from the tracing session to the domain session. The
2714 * function also applies the right modification on a per domain basis for the
2715 * trace files destination directory.
2717 * Should *NOT* be called with RCU read-side lock held.
2719 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2722 const char *dir_name
;
2723 struct consumer_output
*consumer
;
2726 assert(session
->consumer
);
2729 case LTTNG_DOMAIN_KERNEL
:
2730 DBG3("Copying tracing session consumer output in kernel session");
2732 * XXX: We should audit the session creation and what this function
2733 * does "extra" in order to avoid a destroy since this function is used
2734 * in the domain session creation (kernel and ust) only. Same for UST
2737 if (session
->kernel_session
->consumer
) {
2738 consumer_output_put(session
->kernel_session
->consumer
);
2740 session
->kernel_session
->consumer
=
2741 consumer_copy_output(session
->consumer
);
2742 /* Ease our life a bit for the next part */
2743 consumer
= session
->kernel_session
->consumer
;
2744 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2746 case LTTNG_DOMAIN_JUL
:
2747 case LTTNG_DOMAIN_LOG4J
:
2748 case LTTNG_DOMAIN_PYTHON
:
2749 case LTTNG_DOMAIN_UST
:
2750 DBG3("Copying tracing session consumer output in UST session");
2751 if (session
->ust_session
->consumer
) {
2752 consumer_output_put(session
->ust_session
->consumer
);
2754 session
->ust_session
->consumer
=
2755 consumer_copy_output(session
->consumer
);
2756 /* Ease our life a bit for the next part */
2757 consumer
= session
->ust_session
->consumer
;
2758 dir_name
= DEFAULT_UST_TRACE_DIR
;
2761 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2765 /* Append correct directory to subdir */
2766 strncat(consumer
->subdir
, dir_name
,
2767 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2768 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2777 * Create an UST session and add it to the session ust list.
2779 * Should *NOT* be called with RCU read-side lock held.
2781 static int create_ust_session(struct ltt_session
*session
,
2782 struct lttng_domain
*domain
)
2785 struct ltt_ust_session
*lus
= NULL
;
2789 assert(session
->consumer
);
2791 switch (domain
->type
) {
2792 case LTTNG_DOMAIN_JUL
:
2793 case LTTNG_DOMAIN_LOG4J
:
2794 case LTTNG_DOMAIN_PYTHON
:
2795 case LTTNG_DOMAIN_UST
:
2798 ERR("Unknown UST domain on create session %d", domain
->type
);
2799 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2803 DBG("Creating UST session");
2805 lus
= trace_ust_create_session(session
->id
);
2807 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2811 lus
->uid
= session
->uid
;
2812 lus
->gid
= session
->gid
;
2813 lus
->output_traces
= session
->output_traces
;
2814 lus
->snapshot_mode
= session
->snapshot_mode
;
2815 lus
->live_timer_interval
= session
->live_timer
;
2816 session
->ust_session
= lus
;
2817 if (session
->shm_path
[0]) {
2818 strncpy(lus
->root_shm_path
, session
->shm_path
,
2819 sizeof(lus
->root_shm_path
));
2820 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2821 strncpy(lus
->shm_path
, session
->shm_path
,
2822 sizeof(lus
->shm_path
));
2823 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2824 strncat(lus
->shm_path
, "/ust",
2825 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2827 /* Copy session output to the newly created UST session */
2828 ret
= copy_session_consumer(domain
->type
, session
);
2829 if (ret
!= LTTNG_OK
) {
2837 session
->ust_session
= NULL
;
2842 * Create a kernel tracer session then create the default channel.
2844 static int create_kernel_session(struct ltt_session
*session
)
2848 DBG("Creating kernel session");
2850 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2852 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2856 /* Code flow safety */
2857 assert(session
->kernel_session
);
2859 /* Copy session output to the newly created Kernel session */
2860 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2861 if (ret
!= LTTNG_OK
) {
2865 /* Create directory(ies) on local filesystem. */
2866 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2867 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2868 ret
= run_as_mkdir_recursive(
2869 session
->kernel_session
->consumer
->dst
.trace_path
,
2870 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2872 if (errno
!= EEXIST
) {
2873 ERR("Trace directory creation error");
2879 session
->kernel_session
->uid
= session
->uid
;
2880 session
->kernel_session
->gid
= session
->gid
;
2881 session
->kernel_session
->output_traces
= session
->output_traces
;
2882 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2887 trace_kernel_destroy_session(session
->kernel_session
);
2888 session
->kernel_session
= NULL
;
2893 * Count number of session permitted by uid/gid.
2895 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2898 struct ltt_session
*session
;
2900 DBG("Counting number of available session for UID %d GID %d",
2902 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2904 * Only list the sessions the user can control.
2906 if (!session_access_ok(session
, uid
, gid
)) {
2915 * Process the command requested by the lttng client within the command
2916 * context structure. This function make sure that the return structure (llm)
2917 * is set and ready for transmission before returning.
2919 * Return any error encountered or 0 for success.
2921 * "sock" is only used for special-case var. len data.
2923 * Should *NOT* be called with RCU read-side lock held.
2925 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2929 int need_tracing_session
= 1;
2932 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2936 switch (cmd_ctx
->lsm
->cmd_type
) {
2937 case LTTNG_CREATE_SESSION
:
2938 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2939 case LTTNG_CREATE_SESSION_LIVE
:
2940 case LTTNG_DESTROY_SESSION
:
2941 case LTTNG_LIST_SESSIONS
:
2942 case LTTNG_LIST_DOMAINS
:
2943 case LTTNG_START_TRACE
:
2944 case LTTNG_STOP_TRACE
:
2945 case LTTNG_DATA_PENDING
:
2946 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2947 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2948 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2949 case LTTNG_SNAPSHOT_RECORD
:
2950 case LTTNG_SAVE_SESSION
:
2951 case LTTNG_SET_SESSION_SHM_PATH
:
2958 if (opt_no_kernel
&& need_domain
2959 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2961 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2963 ret
= LTTNG_ERR_KERN_NA
;
2968 /* Deny register consumer if we already have a spawned consumer. */
2969 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2970 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2971 if (kconsumer_data
.pid
> 0) {
2972 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2973 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2976 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2980 * Check for command that don't needs to allocate a returned payload. We do
2981 * this here so we don't have to make the call for no payload at each
2984 switch(cmd_ctx
->lsm
->cmd_type
) {
2985 case LTTNG_LIST_SESSIONS
:
2986 case LTTNG_LIST_TRACEPOINTS
:
2987 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2988 case LTTNG_LIST_DOMAINS
:
2989 case LTTNG_LIST_CHANNELS
:
2990 case LTTNG_LIST_EVENTS
:
2991 case LTTNG_LIST_SYSCALLS
:
2992 case LTTNG_LIST_TRACKER_PIDS
:
2995 /* Setup lttng message with no payload */
2996 ret
= setup_lttng_msg(cmd_ctx
, 0);
2998 /* This label does not try to unlock the session */
2999 goto init_setup_error
;
3003 /* Commands that DO NOT need a session. */
3004 switch (cmd_ctx
->lsm
->cmd_type
) {
3005 case LTTNG_CREATE_SESSION
:
3006 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3007 case LTTNG_CREATE_SESSION_LIVE
:
3008 case LTTNG_CALIBRATE
:
3009 case LTTNG_LIST_SESSIONS
:
3010 case LTTNG_LIST_TRACEPOINTS
:
3011 case LTTNG_LIST_SYSCALLS
:
3012 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3013 case LTTNG_SAVE_SESSION
:
3014 need_tracing_session
= 0;
3017 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3019 * We keep the session list lock across _all_ commands
3020 * for now, because the per-session lock does not
3021 * handle teardown properly.
3023 session_lock_list();
3024 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3025 if (cmd_ctx
->session
== NULL
) {
3026 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3029 /* Acquire lock for the session */
3030 session_lock(cmd_ctx
->session
);
3036 * Commands that need a valid session but should NOT create one if none
3037 * exists. Instead of creating one and destroying it when the command is
3038 * handled, process that right before so we save some round trip in useless
3041 switch (cmd_ctx
->lsm
->cmd_type
) {
3042 case LTTNG_DISABLE_CHANNEL
:
3043 case LTTNG_DISABLE_EVENT
:
3044 switch (cmd_ctx
->lsm
->domain
.type
) {
3045 case LTTNG_DOMAIN_KERNEL
:
3046 if (!cmd_ctx
->session
->kernel_session
) {
3047 ret
= LTTNG_ERR_NO_CHANNEL
;
3051 case LTTNG_DOMAIN_JUL
:
3052 case LTTNG_DOMAIN_LOG4J
:
3053 case LTTNG_DOMAIN_PYTHON
:
3054 case LTTNG_DOMAIN_UST
:
3055 if (!cmd_ctx
->session
->ust_session
) {
3056 ret
= LTTNG_ERR_NO_CHANNEL
;
3061 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3073 * Check domain type for specific "pre-action".
3075 switch (cmd_ctx
->lsm
->domain
.type
) {
3076 case LTTNG_DOMAIN_KERNEL
:
3078 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3082 /* Kernel tracer check */
3083 if (kernel_tracer_fd
== -1) {
3084 /* Basically, load kernel tracer modules */
3085 ret
= init_kernel_tracer();
3091 /* Consumer is in an ERROR state. Report back to client */
3092 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3093 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3097 /* Need a session for kernel command */
3098 if (need_tracing_session
) {
3099 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3100 ret
= create_kernel_session(cmd_ctx
->session
);
3102 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3107 /* Start the kernel consumer daemon */
3108 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3109 if (kconsumer_data
.pid
== 0 &&
3110 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3111 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3112 ret
= start_consumerd(&kconsumer_data
);
3114 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3117 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3119 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3123 * The consumer was just spawned so we need to add the socket to
3124 * the consumer output of the session if exist.
3126 ret
= consumer_create_socket(&kconsumer_data
,
3127 cmd_ctx
->session
->kernel_session
->consumer
);
3134 case LTTNG_DOMAIN_JUL
:
3135 case LTTNG_DOMAIN_LOG4J
:
3136 case LTTNG_DOMAIN_PYTHON
:
3137 case LTTNG_DOMAIN_UST
:
3139 if (!ust_app_supported()) {
3140 ret
= LTTNG_ERR_NO_UST
;
3143 /* Consumer is in an ERROR state. Report back to client */
3144 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3145 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3149 if (need_tracing_session
) {
3150 /* Create UST session if none exist. */
3151 if (cmd_ctx
->session
->ust_session
== NULL
) {
3152 ret
= create_ust_session(cmd_ctx
->session
,
3153 &cmd_ctx
->lsm
->domain
);
3154 if (ret
!= LTTNG_OK
) {
3159 /* Start the UST consumer daemons */
3161 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3162 if (consumerd64_bin
[0] != '\0' &&
3163 ustconsumer64_data
.pid
== 0 &&
3164 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3165 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3166 ret
= start_consumerd(&ustconsumer64_data
);
3168 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3169 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3173 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3174 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3176 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3180 * Setup socket for consumer 64 bit. No need for atomic access
3181 * since it was set above and can ONLY be set in this thread.
3183 ret
= consumer_create_socket(&ustconsumer64_data
,
3184 cmd_ctx
->session
->ust_session
->consumer
);
3190 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3191 if (consumerd32_bin
[0] != '\0' &&
3192 ustconsumer32_data
.pid
== 0 &&
3193 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3194 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3195 ret
= start_consumerd(&ustconsumer32_data
);
3197 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3198 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3202 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3203 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3205 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3209 * Setup socket for consumer 64 bit. No need for atomic access
3210 * since it was set above and can ONLY be set in this thread.
3212 ret
= consumer_create_socket(&ustconsumer32_data
,
3213 cmd_ctx
->session
->ust_session
->consumer
);
3225 /* Validate consumer daemon state when start/stop trace command */
3226 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3227 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3228 switch (cmd_ctx
->lsm
->domain
.type
) {
3229 case LTTNG_DOMAIN_JUL
:
3230 case LTTNG_DOMAIN_LOG4J
:
3231 case LTTNG_DOMAIN_PYTHON
:
3232 case LTTNG_DOMAIN_UST
:
3233 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3234 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3238 case LTTNG_DOMAIN_KERNEL
:
3239 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3240 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3248 * Check that the UID or GID match that of the tracing session.
3249 * The root user can interact with all sessions.
3251 if (need_tracing_session
) {
3252 if (!session_access_ok(cmd_ctx
->session
,
3253 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3254 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3255 ret
= LTTNG_ERR_EPERM
;
3261 * Send relayd information to consumer as soon as we have a domain and a
3264 if (cmd_ctx
->session
&& need_domain
) {
3266 * Setup relayd if not done yet. If the relayd information was already
3267 * sent to the consumer, this call will gracefully return.
3269 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3270 if (ret
!= LTTNG_OK
) {
3275 /* Process by command type */
3276 switch (cmd_ctx
->lsm
->cmd_type
) {
3277 case LTTNG_ADD_CONTEXT
:
3279 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3280 cmd_ctx
->lsm
->u
.context
.channel_name
,
3281 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3284 case LTTNG_DISABLE_CHANNEL
:
3286 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3287 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3290 case LTTNG_DISABLE_EVENT
:
3294 * FIXME: handle filter; for now we just receive the filter's
3295 * bytecode along with the filter expression which are sent by
3296 * liblttng-ctl and discard them.
3298 * This fixes an issue where the client may block while sending
3299 * the filter payload and encounter an error because the session
3300 * daemon closes the socket without ever handling this data.
3302 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3303 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3306 char data
[LTTNG_FILTER_MAX_LEN
];
3308 DBG("Discarding disable event command payload of size %zu", count
);
3310 ret
= lttcomm_recv_unix_sock(sock
, data
,
3311 count
> sizeof(data
) ? sizeof(data
) : count
);
3316 count
-= (size_t) ret
;
3319 /* FIXME: passing packed structure to non-packed pointer */
3320 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3321 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3322 &cmd_ctx
->lsm
->u
.disable
.event
);
3325 case LTTNG_ENABLE_CHANNEL
:
3327 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3328 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3331 case LTTNG_TRACK_PID
:
3333 ret
= cmd_track_pid(cmd_ctx
->session
,
3334 cmd_ctx
->lsm
->domain
.type
,
3335 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3338 case LTTNG_UNTRACK_PID
:
3340 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3341 cmd_ctx
->lsm
->domain
.type
,
3342 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3345 case LTTNG_ENABLE_EVENT
:
3347 struct lttng_event_exclusion
*exclusion
= NULL
;
3348 struct lttng_filter_bytecode
*bytecode
= NULL
;
3349 char *filter_expression
= NULL
;
3351 /* Handle exclusion events and receive it from the client. */
3352 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3353 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3355 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3356 (count
* LTTNG_SYMBOL_NAME_LEN
));
3358 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3362 DBG("Receiving var len exclusion event list from client ...");
3363 exclusion
->count
= count
;
3364 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3365 count
* LTTNG_SYMBOL_NAME_LEN
);
3367 DBG("Nothing recv() from client var len data... continuing");
3370 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3375 /* Get filter expression from client. */
3376 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3377 size_t expression_len
=
3378 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3380 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3381 ret
= LTTNG_ERR_FILTER_INVAL
;
3386 filter_expression
= zmalloc(expression_len
);
3387 if (!filter_expression
) {
3389 ret
= LTTNG_ERR_FILTER_NOMEM
;
3393 /* Receive var. len. data */
3394 DBG("Receiving var len filter's expression from client ...");
3395 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3398 DBG("Nothing recv() from client car len data... continuing");
3400 free(filter_expression
);
3402 ret
= LTTNG_ERR_FILTER_INVAL
;
3407 /* Handle filter and get bytecode from client. */
3408 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3409 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3411 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3412 ret
= LTTNG_ERR_FILTER_INVAL
;
3413 free(filter_expression
);
3418 bytecode
= zmalloc(bytecode_len
);
3420 free(filter_expression
);
3422 ret
= LTTNG_ERR_FILTER_NOMEM
;
3426 /* Receive var. len. data */
3427 DBG("Receiving var len filter's bytecode from client ...");
3428 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3430 DBG("Nothing recv() from client car len data... continuing");
3432 free(filter_expression
);
3435 ret
= LTTNG_ERR_FILTER_INVAL
;
3439 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3440 free(filter_expression
);
3443 ret
= LTTNG_ERR_FILTER_INVAL
;
3448 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3449 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3450 &cmd_ctx
->lsm
->u
.enable
.event
,
3451 filter_expression
, bytecode
, exclusion
,
3452 kernel_poll_pipe
[1]);
3455 case LTTNG_LIST_TRACEPOINTS
:
3457 struct lttng_event
*events
;
3460 session_lock_list();
3461 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3462 session_unlock_list();
3463 if (nb_events
< 0) {
3464 /* Return value is a negative lttng_error_code. */
3470 * Setup lttng message with payload size set to the event list size in
3471 * bytes and then copy list into the llm payload.
3473 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3479 /* Copy event list into message payload */
3480 memcpy(cmd_ctx
->llm
->payload
, events
,
3481 sizeof(struct lttng_event
) * nb_events
);
3488 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3490 struct lttng_event_field
*fields
;
3493 session_lock_list();
3494 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3496 session_unlock_list();
3497 if (nb_fields
< 0) {
3498 /* Return value is a negative lttng_error_code. */
3504 * Setup lttng message with payload size set to the event list size in
3505 * bytes and then copy list into the llm payload.
3507 ret
= setup_lttng_msg(cmd_ctx
,
3508 sizeof(struct lttng_event_field
) * nb_fields
);
3514 /* Copy event list into message payload */
3515 memcpy(cmd_ctx
->llm
->payload
, fields
,
3516 sizeof(struct lttng_event_field
) * nb_fields
);
3523 case LTTNG_LIST_SYSCALLS
:
3525 struct lttng_event
*events
;
3528 nb_events
= cmd_list_syscalls(&events
);
3529 if (nb_events
< 0) {
3530 /* Return value is a negative lttng_error_code. */
3536 * Setup lttng message with payload size set to the event list size in
3537 * bytes and then copy list into the llm payload.
3539 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3545 /* Copy event list into message payload */
3546 memcpy(cmd_ctx
->llm
->payload
, events
,
3547 sizeof(struct lttng_event
) * nb_events
);
3554 case LTTNG_LIST_TRACKER_PIDS
:
3556 int32_t *pids
= NULL
;
3559 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3560 cmd_ctx
->lsm
->domain
.type
, &pids
);
3562 /* Return value is a negative lttng_error_code. */
3568 * Setup lttng message with payload size set to the event list size in
3569 * bytes and then copy list into the llm payload.
3571 ret
= setup_lttng_msg(cmd_ctx
, sizeof(int32_t) * nr_pids
);
3577 /* Copy event list into message payload */
3578 memcpy(cmd_ctx
->llm
->payload
, pids
,
3579 sizeof(int) * nr_pids
);
3586 case LTTNG_SET_CONSUMER_URI
:
3589 struct lttng_uri
*uris
;
3591 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3592 len
= nb_uri
* sizeof(struct lttng_uri
);
3595 ret
= LTTNG_ERR_INVALID
;
3599 uris
= zmalloc(len
);
3601 ret
= LTTNG_ERR_FATAL
;
3605 /* Receive variable len data */
3606 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3607 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3609 DBG("No URIs received from client... continuing");
3611 ret
= LTTNG_ERR_SESSION_FAIL
;
3616 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3618 if (ret
!= LTTNG_OK
) {
3625 case LTTNG_START_TRACE
:
3627 ret
= cmd_start_trace(cmd_ctx
->session
);
3630 case LTTNG_STOP_TRACE
:
3632 ret
= cmd_stop_trace(cmd_ctx
->session
);
3635 case LTTNG_CREATE_SESSION
:
3638 struct lttng_uri
*uris
= NULL
;
3640 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3641 len
= nb_uri
* sizeof(struct lttng_uri
);
3644 uris
= zmalloc(len
);
3646 ret
= LTTNG_ERR_FATAL
;
3650 /* Receive variable len data */
3651 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3652 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3654 DBG("No URIs received from client... continuing");
3656 ret
= LTTNG_ERR_SESSION_FAIL
;
3661 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3662 DBG("Creating session with ONE network URI is a bad call");
3663 ret
= LTTNG_ERR_SESSION_FAIL
;
3669 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3670 &cmd_ctx
->creds
, 0);
3676 case LTTNG_DESTROY_SESSION
:
3678 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3680 /* Set session to NULL so we do not unlock it after free. */
3681 cmd_ctx
->session
= NULL
;
3684 case LTTNG_LIST_DOMAINS
:
3687 struct lttng_domain
*domains
= NULL
;
3689 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3691 /* Return value is a negative lttng_error_code. */
3696 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3702 /* Copy event list into message payload */
3703 memcpy(cmd_ctx
->llm
->payload
, domains
,
3704 nb_dom
* sizeof(struct lttng_domain
));
3711 case LTTNG_LIST_CHANNELS
:
3714 struct lttng_channel
*channels
= NULL
;
3716 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3717 cmd_ctx
->session
, &channels
);
3719 /* Return value is a negative lttng_error_code. */
3724 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3730 /* Copy event list into message payload */
3731 memcpy(cmd_ctx
->llm
->payload
, channels
,
3732 nb_chan
* sizeof(struct lttng_channel
));
3739 case LTTNG_LIST_EVENTS
:
3742 struct lttng_event
*events
= NULL
;
3744 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3745 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3747 /* Return value is a negative lttng_error_code. */
3752 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3758 /* Copy event list into message payload */
3759 memcpy(cmd_ctx
->llm
->payload
, events
,
3760 nb_event
* sizeof(struct lttng_event
));
3767 case LTTNG_LIST_SESSIONS
:
3769 unsigned int nr_sessions
;
3771 session_lock_list();
3772 nr_sessions
= lttng_sessions_count(
3773 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3774 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3776 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3778 session_unlock_list();
3782 /* Filled the session array */
3783 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3784 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3785 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3787 session_unlock_list();
3792 case LTTNG_CALIBRATE
:
3794 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3795 &cmd_ctx
->lsm
->u
.calibrate
);
3798 case LTTNG_REGISTER_CONSUMER
:
3800 struct consumer_data
*cdata
;
3802 switch (cmd_ctx
->lsm
->domain
.type
) {
3803 case LTTNG_DOMAIN_KERNEL
:
3804 cdata
= &kconsumer_data
;
3807 ret
= LTTNG_ERR_UND
;
3811 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3812 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3815 case LTTNG_DATA_PENDING
:
3817 ret
= cmd_data_pending(cmd_ctx
->session
);
3820 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3822 struct lttcomm_lttng_output_id reply
;
3824 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3825 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3826 if (ret
!= LTTNG_OK
) {
3830 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3835 /* Copy output list into message payload */
3836 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3840 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3842 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3843 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3846 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3849 struct lttng_snapshot_output
*outputs
= NULL
;
3851 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3852 if (nb_output
< 0) {
3857 ret
= setup_lttng_msg(cmd_ctx
,
3858 nb_output
* sizeof(struct lttng_snapshot_output
));
3865 /* Copy output list into message payload */
3866 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3867 nb_output
* sizeof(struct lttng_snapshot_output
));
3874 case LTTNG_SNAPSHOT_RECORD
:
3876 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3877 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3878 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3881 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3884 struct lttng_uri
*uris
= NULL
;
3886 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3887 len
= nb_uri
* sizeof(struct lttng_uri
);
3890 uris
= zmalloc(len
);
3892 ret
= LTTNG_ERR_FATAL
;
3896 /* Receive variable len data */
3897 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3898 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3900 DBG("No URIs received from client... continuing");
3902 ret
= LTTNG_ERR_SESSION_FAIL
;
3907 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3908 DBG("Creating session with ONE network URI is a bad call");
3909 ret
= LTTNG_ERR_SESSION_FAIL
;
3915 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3916 nb_uri
, &cmd_ctx
->creds
);
3920 case LTTNG_CREATE_SESSION_LIVE
:
3923 struct lttng_uri
*uris
= NULL
;
3925 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3926 len
= nb_uri
* sizeof(struct lttng_uri
);
3929 uris
= zmalloc(len
);
3931 ret
= LTTNG_ERR_FATAL
;
3935 /* Receive variable len data */
3936 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3937 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3939 DBG("No URIs received from client... continuing");
3941 ret
= LTTNG_ERR_SESSION_FAIL
;
3946 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3947 DBG("Creating session with ONE network URI is a bad call");
3948 ret
= LTTNG_ERR_SESSION_FAIL
;
3954 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3955 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3959 case LTTNG_SAVE_SESSION
:
3961 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
3965 case LTTNG_SET_SESSION_SHM_PATH
:
3967 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
3968 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
3972 ret
= LTTNG_ERR_UND
;
3977 if (cmd_ctx
->llm
== NULL
) {
3978 DBG("Missing llm structure. Allocating one.");
3979 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3983 /* Set return code */
3984 cmd_ctx
->llm
->ret_code
= ret
;
3986 if (cmd_ctx
->session
) {
3987 session_unlock(cmd_ctx
->session
);
3989 if (need_tracing_session
) {
3990 session_unlock_list();
3997 * Thread managing health check socket.
3999 static void *thread_manage_health(void *data
)
4001 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4002 uint32_t revents
, nb_fd
;
4003 struct lttng_poll_event events
;
4004 struct health_comm_msg msg
;
4005 struct health_comm_reply reply
;
4007 DBG("[thread] Manage health check started");
4009 rcu_register_thread();
4011 /* We might hit an error path before this is created. */
4012 lttng_poll_init(&events
);
4014 /* Create unix socket */
4015 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4017 ERR("Unable to create health check Unix socket");
4023 /* lttng health client socket path permissions */
4024 ret
= chown(health_unix_sock_path
, 0,
4025 utils_get_group_id(tracing_group_name
));
4027 ERR("Unable to set group on %s", health_unix_sock_path
);
4033 ret
= chmod(health_unix_sock_path
,
4034 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4036 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4044 * Set the CLOEXEC flag. Return code is useless because either way, the
4047 (void) utils_set_fd_cloexec(sock
);
4049 ret
= lttcomm_listen_unix_sock(sock
);
4055 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4056 * more will be added to this poll set.
4058 ret
= sessiond_set_thread_pollset(&events
, 2);
4063 /* Add the application registration socket */
4064 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4069 sessiond_notify_ready();
4072 DBG("Health check ready");
4074 /* Inifinite blocking call, waiting for transmission */
4076 ret
= lttng_poll_wait(&events
, -1);
4079 * Restart interrupted system call.
4081 if (errno
== EINTR
) {
4089 for (i
= 0; i
< nb_fd
; i
++) {
4090 /* Fetch once the poll data */
4091 revents
= LTTNG_POLL_GETEV(&events
, i
);
4092 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4095 /* No activity for this FD (poll implementation). */
4099 /* Thread quit pipe has been closed. Killing thread. */
4100 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4106 /* Event on the registration socket */
4107 if (pollfd
== sock
) {
4108 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4109 ERR("Health socket poll error");
4115 new_sock
= lttcomm_accept_unix_sock(sock
);
4121 * Set the CLOEXEC flag. Return code is useless because either way, the
4124 (void) utils_set_fd_cloexec(new_sock
);
4126 DBG("Receiving data from client for health...");
4127 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4129 DBG("Nothing recv() from client... continuing");
4130 ret
= close(new_sock
);
4138 rcu_thread_online();
4140 memset(&reply
, 0, sizeof(reply
));
4141 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4143 * health_check_state returns 0 if health is
4146 if (!health_check_state(health_sessiond
, i
)) {
4147 reply
.ret_code
|= 1ULL << i
;
4151 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4153 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4155 ERR("Failed to send health data back to client");
4158 /* End of transmission */
4159 ret
= close(new_sock
);
4169 ERR("Health error occurred in %s", __func__
);
4171 DBG("Health check thread dying");
4172 unlink(health_unix_sock_path
);
4180 lttng_poll_clean(&events
);
4182 rcu_unregister_thread();
4187 * This thread manage all clients request using the unix client socket for
4190 static void *thread_manage_clients(void *data
)
4192 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4194 uint32_t revents
, nb_fd
;
4195 struct command_ctx
*cmd_ctx
= NULL
;
4196 struct lttng_poll_event events
;
4198 DBG("[thread] Manage client started");
4200 rcu_register_thread();
4202 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4204 health_code_update();
4206 ret
= lttcomm_listen_unix_sock(client_sock
);
4212 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4213 * more will be added to this poll set.
4215 ret
= sessiond_set_thread_pollset(&events
, 2);
4217 goto error_create_poll
;
4220 /* Add the application registration socket */
4221 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4226 sessiond_notify_ready();
4227 ret
= sem_post(&load_info
->message_thread_ready
);
4229 PERROR("sem_post message_thread_ready");
4233 /* This testpoint is after we signal readiness to the parent. */
4234 if (testpoint(sessiond_thread_manage_clients
)) {
4238 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4242 health_code_update();
4245 DBG("Accepting client command ...");
4247 /* Inifinite blocking call, waiting for transmission */
4249 health_poll_entry();
4250 ret
= lttng_poll_wait(&events
, -1);
4254 * Restart interrupted system call.
4256 if (errno
== EINTR
) {
4264 for (i
= 0; i
< nb_fd
; i
++) {
4265 /* Fetch once the poll data */
4266 revents
= LTTNG_POLL_GETEV(&events
, i
);
4267 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4269 health_code_update();
4272 /* No activity for this FD (poll implementation). */
4276 /* Thread quit pipe has been closed. Killing thread. */
4277 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4283 /* Event on the registration socket */
4284 if (pollfd
== client_sock
) {
4285 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4286 ERR("Client socket poll error");
4292 DBG("Wait for client response");
4294 health_code_update();
4296 sock
= lttcomm_accept_unix_sock(client_sock
);
4302 * Set the CLOEXEC flag. Return code is useless because either way, the
4305 (void) utils_set_fd_cloexec(sock
);
4307 /* Set socket option for credentials retrieval */
4308 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4313 /* Allocate context command to process the client request */
4314 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4315 if (cmd_ctx
== NULL
) {
4316 PERROR("zmalloc cmd_ctx");
4320 /* Allocate data buffer for reception */
4321 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4322 if (cmd_ctx
->lsm
== NULL
) {
4323 PERROR("zmalloc cmd_ctx->lsm");
4327 cmd_ctx
->llm
= NULL
;
4328 cmd_ctx
->session
= NULL
;
4330 health_code_update();
4333 * Data is received from the lttng client. The struct
4334 * lttcomm_session_msg (lsm) contains the command and data request of
4337 DBG("Receiving data from client ...");
4338 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4339 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4341 DBG("Nothing recv() from client... continuing");
4347 clean_command_ctx(&cmd_ctx
);
4351 health_code_update();
4353 // TODO: Validate cmd_ctx including sanity check for
4354 // security purpose.
4356 rcu_thread_online();
4358 * This function dispatch the work to the kernel or userspace tracer
4359 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4360 * informations for the client. The command context struct contains
4361 * everything this function may needs.
4363 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4364 rcu_thread_offline();
4372 * TODO: Inform client somehow of the fatal error. At
4373 * this point, ret < 0 means that a zmalloc failed
4374 * (ENOMEM). Error detected but still accept
4375 * command, unless a socket error has been
4378 clean_command_ctx(&cmd_ctx
);
4382 health_code_update();
4384 DBG("Sending response (size: %d, retcode: %s)",
4385 cmd_ctx
->lttng_msg_size
,
4386 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
4387 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4389 ERR("Failed to send data back to client");
4392 /* End of transmission */
4399 clean_command_ctx(&cmd_ctx
);
4401 health_code_update();
4413 lttng_poll_clean(&events
);
4414 clean_command_ctx(&cmd_ctx
);
4418 unlink(client_unix_sock_path
);
4419 if (client_sock
>= 0) {
4420 ret
= close(client_sock
);
4428 ERR("Health error occurred in %s", __func__
);
4431 health_unregister(health_sessiond
);
4433 DBG("Client thread dying");
4435 rcu_unregister_thread();
4438 * Since we are creating the consumer threads, we own them, so we need
4439 * to join them before our thread exits.
4441 ret
= join_consumer_thread(&kconsumer_data
);
4444 PERROR("join_consumer");
4447 ret
= join_consumer_thread(&ustconsumer32_data
);
4450 PERROR("join_consumer ust32");
4453 ret
= join_consumer_thread(&ustconsumer64_data
);
4456 PERROR("join_consumer ust64");
4463 * usage function on stderr
4465 static void usage(void)
4467 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4468 fprintf(stderr
, " -h, --help Display this usage.\n");
4469 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4470 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4471 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4472 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4473 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4474 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4475 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4476 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4477 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4478 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4479 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4480 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4481 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4482 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4483 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4484 fprintf(stderr
, " -V, --version Show version number.\n");
4485 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4486 fprintf(stderr
, " -q, --quiet No output at all.\n");
4487 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4488 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4489 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4490 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4491 fprintf(stderr
, " --agent-tcp-port Agent registration TCP port\n");
4492 fprintf(stderr
, " -f --config PATH Load daemon configuration file\n");
4493 fprintf(stderr
, " -l --load PATH Load session configuration\n");
4494 fprintf(stderr
, " --kmod-probes Specify kernel module probes to load\n");
4495 fprintf(stderr
, " --extra-kmod-probes Specify extra kernel module probes to load\n");
4498 static int string_match(const char *str1
, const char *str2
)
4500 return (str1
&& str2
) && !strcmp(str1
, str2
);
4504 * Take an option from the getopt output and set it in the right variable to be
4507 * Return 0 on success else a negative value.
4509 static int set_option(int opt
, const char *arg
, const char *optname
)
4513 if (arg
&& arg
[0] == '\0') {
4515 * This only happens if the value is read from daemon config
4516 * file. This means the option requires an argument and the
4517 * configuration file contains a line such as:
4524 if (string_match(optname
, "client-sock") || opt
== 'c') {
4525 if (lttng_is_setuid_setgid()) {
4526 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4527 "-c, --client-sock");
4529 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4531 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4532 if (lttng_is_setuid_setgid()) {
4533 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4536 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4538 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4540 } else if (string_match(optname
, "background") || opt
== 'b') {
4542 } else if (string_match(optname
, "group") || opt
== 'g') {
4543 if (lttng_is_setuid_setgid()) {
4544 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4548 * If the override option is set, the pointer points to a
4549 * *non* const thus freeing it even though the variable type is
4552 if (tracing_group_name_override
) {
4553 free((void *) tracing_group_name
);
4555 tracing_group_name
= strdup(arg
);
4556 if (!tracing_group_name
) {
4560 tracing_group_name_override
= 1;
4562 } else if (string_match(optname
, "help") || opt
== 'h') {
4565 } else if (string_match(optname
, "version") || opt
== 'V') {
4566 fprintf(stdout
, "%s\n", VERSION
);
4568 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4570 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4571 if (lttng_is_setuid_setgid()) {
4572 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4573 "--kconsumerd-err-sock");
4575 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4577 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4578 if (lttng_is_setuid_setgid()) {
4579 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4580 "--kconsumerd-cmd-sock");
4582 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4584 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4585 if (lttng_is_setuid_setgid()) {
4586 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4587 "--ustconsumerd64-err-sock");
4589 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4591 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4592 if (lttng_is_setuid_setgid()) {
4593 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4594 "--ustconsumerd64-cmd-sock");
4596 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4598 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4599 if (lttng_is_setuid_setgid()) {
4600 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4601 "--ustconsumerd32-err-sock");
4603 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4605 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4606 if (lttng_is_setuid_setgid()) {
4607 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4608 "--ustconsumerd32-cmd-sock");
4610 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4612 } else if (string_match(optname
, "no-kernel")) {
4614 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4615 lttng_opt_quiet
= 1;
4616 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4617 /* Verbose level can increase using multiple -v */
4619 /* Value obtained from config file */
4620 lttng_opt_verbose
= config_parse_value(arg
);
4622 /* -v used on command line */
4623 lttng_opt_verbose
++;
4625 /* Clamp value to [0, 3] */
4626 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4627 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4628 } else if (string_match(optname
, "verbose-consumer")) {
4630 opt_verbose_consumer
= config_parse_value(arg
);
4632 opt_verbose_consumer
+= 1;
4634 } else if (string_match(optname
, "consumerd32-path")) {
4635 if (lttng_is_setuid_setgid()) {
4636 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4637 "--consumerd32-path");
4639 if (consumerd32_bin_override
) {
4640 free((void *) consumerd32_bin
);
4642 consumerd32_bin
= strdup(arg
);
4643 if (!consumerd32_bin
) {
4647 consumerd32_bin_override
= 1;
4649 } else if (string_match(optname
, "consumerd32-libdir")) {
4650 if (lttng_is_setuid_setgid()) {
4651 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4652 "--consumerd32-libdir");
4654 if (consumerd32_libdir_override
) {
4655 free((void *) consumerd32_libdir
);
4657 consumerd32_libdir
= strdup(arg
);
4658 if (!consumerd32_libdir
) {
4662 consumerd32_libdir_override
= 1;
4664 } else if (string_match(optname
, "consumerd64-path")) {
4665 if (lttng_is_setuid_setgid()) {
4666 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4667 "--consumerd64-path");
4669 if (consumerd64_bin_override
) {
4670 free((void *) consumerd64_bin
);
4672 consumerd64_bin
= strdup(arg
);
4673 if (!consumerd64_bin
) {
4677 consumerd64_bin_override
= 1;
4679 } else if (string_match(optname
, "consumerd64-libdir")) {
4680 if (lttng_is_setuid_setgid()) {
4681 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4682 "--consumerd64-libdir");
4684 if (consumerd64_libdir_override
) {
4685 free((void *) consumerd64_libdir
);
4687 consumerd64_libdir
= strdup(arg
);
4688 if (!consumerd64_libdir
) {
4692 consumerd64_libdir_override
= 1;
4694 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4695 if (lttng_is_setuid_setgid()) {
4696 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4700 opt_pidfile
= strdup(arg
);
4706 } else if (string_match(optname
, "agent-tcp-port")) {
4707 if (lttng_is_setuid_setgid()) {
4708 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4709 "--agent-tcp-port");
4718 v
= strtoul(arg
, NULL
, 0);
4719 if (errno
!= 0 || !isdigit(arg
[0])) {
4720 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4723 if (v
== 0 || v
>= 65535) {
4724 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4727 agent_tcp_port
= (uint32_t) v
;
4728 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4730 } else if (string_match(optname
, "load") || opt
== 'l') {
4731 if (lttng_is_setuid_setgid()) {
4732 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4735 free(opt_load_session_path
);
4736 opt_load_session_path
= strdup(arg
);
4737 if (!opt_load_session_path
) {
4742 } else if (string_match(optname
, "kmod-probes")) {
4743 if (lttng_is_setuid_setgid()) {
4744 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4747 free(kmod_probes_list
);
4748 kmod_probes_list
= strdup(arg
);
4749 if (!kmod_probes_list
) {
4754 } else if (string_match(optname
, "extra-kmod-probes")) {
4755 if (lttng_is_setuid_setgid()) {
4756 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4757 "--extra-kmod-probes");
4759 free(kmod_extra_probes_list
);
4760 kmod_extra_probes_list
= strdup(arg
);
4761 if (!kmod_extra_probes_list
) {
4766 } else if (string_match(optname
, "config") || opt
== 'f') {
4767 /* This is handled in set_options() thus silent skip. */
4770 /* Unknown option or other error.
4771 * Error is printed by getopt, just return */
4776 if (ret
== -EINVAL
) {
4777 const char *opt_name
= "unknown";
4780 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4782 if (opt
== long_options
[i
].val
) {
4783 opt_name
= long_options
[i
].name
;
4788 WARN("Invalid argument provided for option \"%s\", using default value.",
4796 * config_entry_handler_cb used to handle options read from a config file.
4797 * See config_entry_handler_cb comment in common/config/config.h for the
4798 * return value conventions.
4800 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4804 if (!entry
|| !entry
->name
|| !entry
->value
) {
4809 /* Check if the option is to be ignored */
4810 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4811 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4816 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4819 /* Ignore if not fully matched. */
4820 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4825 * If the option takes no argument on the command line, we have to
4826 * check if the value is "true". We support non-zero numeric values,
4829 if (!long_options
[i
].has_arg
) {
4830 ret
= config_parse_value(entry
->value
);
4833 WARN("Invalid configuration value \"%s\" for option %s",
4834 entry
->value
, entry
->name
);
4836 /* False, skip boolean config option. */
4841 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4845 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4852 * daemon configuration loading and argument parsing
4854 static int set_options(int argc
, char **argv
)
4856 int ret
= 0, c
= 0, option_index
= 0;
4857 int orig_optopt
= optopt
, orig_optind
= optind
;
4859 const char *config_path
= NULL
;
4861 optstring
= utils_generate_optstring(long_options
,
4862 sizeof(long_options
) / sizeof(struct option
));
4868 /* Check for the --config option */
4869 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4870 &option_index
)) != -1) {
4874 } else if (c
!= 'f') {
4875 /* if not equal to --config option. */
4879 if (lttng_is_setuid_setgid()) {
4880 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4883 config_path
= utils_expand_path(optarg
);
4885 ERR("Failed to resolve path: %s", optarg
);
4890 ret
= config_get_section_entries(config_path
, config_section_name
,
4891 config_entry_handler
, NULL
);
4894 ERR("Invalid configuration option at line %i", ret
);
4900 /* Reset getopt's global state */
4901 optopt
= orig_optopt
;
4902 optind
= orig_optind
;
4906 * getopt_long() will not set option_index if it encounters a
4909 c
= getopt_long(argc
, argv
, optstring
, long_options
,
4916 * Pass NULL as the long option name if popt left the index
4919 ret
= set_option(c
, optarg
,
4920 option_index
< 0 ? NULL
:
4921 long_options
[option_index
].name
);
4933 * Creates the two needed socket by the daemon.
4934 * apps_sock - The communication socket for all UST apps.
4935 * client_sock - The communication of the cli tool (lttng).
4937 static int init_daemon_socket(void)
4942 old_umask
= umask(0);
4944 /* Create client tool unix socket */
4945 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4946 if (client_sock
< 0) {
4947 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4952 /* Set the cloexec flag */
4953 ret
= utils_set_fd_cloexec(client_sock
);
4955 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4956 "Continuing but note that the consumer daemon will have a "
4957 "reference to this socket on exec()", client_sock
);
4960 /* File permission MUST be 660 */
4961 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4963 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4968 /* Create the application unix socket */
4969 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4970 if (apps_sock
< 0) {
4971 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4976 /* Set the cloexec flag */
4977 ret
= utils_set_fd_cloexec(apps_sock
);
4979 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4980 "Continuing but note that the consumer daemon will have a "
4981 "reference to this socket on exec()", apps_sock
);
4984 /* File permission MUST be 666 */
4985 ret
= chmod(apps_unix_sock_path
,
4986 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4988 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4993 DBG3("Session daemon client socket %d and application socket %d created",
4994 client_sock
, apps_sock
);
5002 * Check if the global socket is available, and if a daemon is answering at the
5003 * other side. If yes, error is returned.
5005 static int check_existing_daemon(void)
5007 /* Is there anybody out there ? */
5008 if (lttng_session_daemon_alive()) {
5016 * Set the tracing group gid onto the client socket.
5018 * Race window between mkdir and chown is OK because we are going from more
5019 * permissive (root.root) to less permissive (root.tracing).
5021 static int set_permissions(char *rundir
)
5026 gid
= utils_get_group_id(tracing_group_name
);
5028 /* Set lttng run dir */
5029 ret
= chown(rundir
, 0, gid
);
5031 ERR("Unable to set group on %s", rundir
);
5036 * Ensure all applications and tracing group can search the run
5037 * dir. Allow everyone to read the directory, since it does not
5038 * buy us anything to hide its content.
5040 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5042 ERR("Unable to set permissions on %s", rundir
);
5046 /* lttng client socket path */
5047 ret
= chown(client_unix_sock_path
, 0, gid
);
5049 ERR("Unable to set group on %s", client_unix_sock_path
);
5053 /* kconsumer error socket path */
5054 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5056 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5060 /* 64-bit ustconsumer error socket path */
5061 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5063 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5067 /* 32-bit ustconsumer compat32 error socket path */
5068 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5070 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5074 DBG("All permissions are set");
5080 * Create the lttng run directory needed for all global sockets and pipe.
5082 static int create_lttng_rundir(const char *rundir
)
5086 DBG3("Creating LTTng run directory: %s", rundir
);
5088 ret
= mkdir(rundir
, S_IRWXU
);
5090 if (errno
!= EEXIST
) {
5091 ERR("Unable to create %s", rundir
);
5103 * Setup sockets and directory needed by the kconsumerd communication with the
5106 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5110 char path
[PATH_MAX
];
5112 switch (consumer_data
->type
) {
5113 case LTTNG_CONSUMER_KERNEL
:
5114 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5116 case LTTNG_CONSUMER64_UST
:
5117 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5119 case LTTNG_CONSUMER32_UST
:
5120 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5123 ERR("Consumer type unknown");
5128 DBG2("Creating consumer directory: %s", path
);
5130 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5132 if (errno
!= EEXIST
) {
5134 ERR("Failed to create %s", path
);
5140 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5142 ERR("Unable to set group on %s", path
);
5148 /* Create the kconsumerd error unix socket */
5149 consumer_data
->err_sock
=
5150 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5151 if (consumer_data
->err_sock
< 0) {
5152 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5158 * Set the CLOEXEC flag. Return code is useless because either way, the
5161 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5163 PERROR("utils_set_fd_cloexec");
5164 /* continue anyway */
5167 /* File permission MUST be 660 */
5168 ret
= chmod(consumer_data
->err_unix_sock_path
,
5169 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5171 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5181 * Signal handler for the daemon
5183 * Simply stop all worker threads, leaving main() return gracefully after
5184 * joining all threads and calling cleanup().
5186 static void sighandler(int sig
)
5190 DBG("SIGPIPE caught");
5193 DBG("SIGINT caught");
5197 DBG("SIGTERM caught");
5201 CMM_STORE_SHARED(recv_child_signal
, 1);
5209 * Setup signal handler for :
5210 * SIGINT, SIGTERM, SIGPIPE
5212 static int set_signal_handler(void)
5215 struct sigaction sa
;
5218 if ((ret
= sigemptyset(&sigset
)) < 0) {
5219 PERROR("sigemptyset");
5223 sa
.sa_handler
= sighandler
;
5224 sa
.sa_mask
= sigset
;
5226 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5227 PERROR("sigaction");
5231 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5232 PERROR("sigaction");
5236 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5237 PERROR("sigaction");
5241 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5242 PERROR("sigaction");
5246 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5252 * Set open files limit to unlimited. This daemon can open a large number of
5253 * file descriptors in order to consumer multiple kernel traces.
5255 static void set_ulimit(void)
5260 /* The kernel does not allowed an infinite limit for open files */
5261 lim
.rlim_cur
= 65535;
5262 lim
.rlim_max
= 65535;
5264 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5266 PERROR("failed to set open files limit");
5271 * Write pidfile using the rundir and opt_pidfile.
5273 static int write_pidfile(void)
5276 char pidfile_path
[PATH_MAX
];
5281 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
5283 /* Build pidfile path from rundir and opt_pidfile. */
5284 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5285 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5287 PERROR("snprintf pidfile path");
5293 * Create pid file in rundir.
5295 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5301 * Create lockfile using the rundir and return its fd.
5303 static int create_lockfile(void)
5306 char lockfile_path
[PATH_MAX
];
5308 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5313 ret
= utils_create_lock_file(lockfile_path
);
5319 * Write agent TCP port using the rundir.
5321 static int write_agent_port(void)
5324 char path
[PATH_MAX
];
5328 ret
= snprintf(path
, sizeof(path
), "%s/"
5329 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5331 PERROR("snprintf agent port path");
5336 * Create TCP agent port file in rundir.
5338 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5347 int main(int argc
, char **argv
)
5349 int ret
= 0, retval
= 0;
5351 const char *home_path
, *env_app_timeout
;
5353 /* Initialize agent apps ht global variable */
5354 agent_apps_ht_by_sock
= NULL
;
5356 init_kernel_workarounds();
5358 rcu_register_thread();
5360 if (set_signal_handler()) {
5362 goto exit_set_signal_handler
;
5365 setup_consumerd_path();
5367 page_size
= sysconf(_SC_PAGESIZE
);
5368 if (page_size
< 0) {
5369 PERROR("sysconf _SC_PAGESIZE");
5370 page_size
= LONG_MAX
;
5371 WARN("Fallback page size to %ld", page_size
);
5375 * Parse arguments and load the daemon configuration file.
5377 * We have an exit_options exit path to free memory reserved by
5378 * set_options. This is needed because the rest of sessiond_cleanup()
5379 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5380 * depends on set_options.
5383 if (set_options(argc
, argv
)) {
5389 if (opt_daemon
|| opt_background
) {
5392 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5400 * We are in the child. Make sure all other file descriptors are
5401 * closed, in case we are called with more opened file
5402 * descriptors than the standard ones.
5404 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5410 * Starting from here, we can create threads. This needs to be after
5411 * lttng_daemonize due to RCU.
5415 * Initialize the health check subsystem. This call should set the
5416 * appropriate time values.
5418 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5419 if (!health_sessiond
) {
5420 PERROR("health_app_create error");
5422 goto exit_health_sessiond_cleanup
;
5425 if (init_ht_cleanup_quit_pipe()) {
5427 goto exit_ht_cleanup_quit_pipe
;
5430 /* Setup the thread ht_cleanup communication pipe. */
5431 if (utils_create_pipe_cloexec(ht_cleanup_pipe
)) {
5433 goto exit_ht_cleanup_pipe
;
5436 /* Set up max poll set size */
5437 if (lttng_poll_set_max_size()) {
5439 goto exit_set_max_size
;
5442 /* Create thread to clean up RCU hash tables */
5443 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5444 thread_ht_cleanup
, (void *) NULL
);
5447 PERROR("pthread_create ht_cleanup");
5449 goto exit_ht_cleanup
;
5452 /* Create thread quit pipe */
5453 if (init_thread_quit_pipe()) {
5455 goto exit_init_data
;
5458 /* Check if daemon is UID = 0 */
5459 is_root
= !getuid();
5462 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5465 goto exit_init_data
;
5468 /* Create global run dir with root access */
5469 if (create_lttng_rundir(rundir
)) {
5471 goto exit_init_data
;
5474 if (strlen(apps_unix_sock_path
) == 0) {
5475 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5476 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5479 goto exit_init_data
;
5483 if (strlen(client_unix_sock_path
) == 0) {
5484 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5485 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5488 goto exit_init_data
;
5492 /* Set global SHM for ust */
5493 if (strlen(wait_shm_path
) == 0) {
5494 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5495 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5498 goto exit_init_data
;
5502 if (strlen(health_unix_sock_path
) == 0) {
5503 ret
= snprintf(health_unix_sock_path
,
5504 sizeof(health_unix_sock_path
),
5505 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5508 goto exit_init_data
;
5512 /* Setup kernel consumerd path */
5513 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5514 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5517 goto exit_init_data
;
5519 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5520 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5523 goto exit_init_data
;
5526 DBG2("Kernel consumer err path: %s",
5527 kconsumer_data
.err_unix_sock_path
);
5528 DBG2("Kernel consumer cmd path: %s",
5529 kconsumer_data
.cmd_unix_sock_path
);
5531 home_path
= utils_get_home_dir();
5532 if (home_path
== NULL
) {
5533 /* TODO: Add --socket PATH option */
5534 ERR("Can't get HOME directory for sockets creation.");
5536 goto exit_init_data
;
5540 * Create rundir from home path. This will create something like
5543 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5546 goto exit_init_data
;
5549 if (create_lttng_rundir(rundir
)) {
5551 goto exit_init_data
;
5554 if (strlen(apps_unix_sock_path
) == 0) {
5555 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5556 DEFAULT_HOME_APPS_UNIX_SOCK
,
5560 goto exit_init_data
;
5564 /* Set the cli tool unix socket path */
5565 if (strlen(client_unix_sock_path
) == 0) {
5566 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5567 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5571 goto exit_init_data
;
5575 /* Set global SHM for ust */
5576 if (strlen(wait_shm_path
) == 0) {
5577 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5578 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5582 goto exit_init_data
;
5586 /* Set health check Unix path */
5587 if (strlen(health_unix_sock_path
) == 0) {
5588 ret
= snprintf(health_unix_sock_path
,
5589 sizeof(health_unix_sock_path
),
5590 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5594 goto exit_init_data
;
5599 lockfile_fd
= create_lockfile();
5600 if (lockfile_fd
< 0) {
5602 goto exit_init_data
;
5605 /* Set consumer initial state */
5606 kernel_consumerd_state
= CONSUMER_STOPPED
;
5607 ust_consumerd_state
= CONSUMER_STOPPED
;
5609 DBG("Client socket path %s", client_unix_sock_path
);
5610 DBG("Application socket path %s", apps_unix_sock_path
);
5611 DBG("Application wait path %s", wait_shm_path
);
5612 DBG("LTTng run directory path: %s", rundir
);
5614 /* 32 bits consumerd path setup */
5615 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5616 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5618 PERROR("snprintf 32-bit consumer error socket path");
5620 goto exit_init_data
;
5622 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5623 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5625 PERROR("snprintf 32-bit consumer command socket path");
5627 goto exit_init_data
;
5630 DBG2("UST consumer 32 bits err path: %s",
5631 ustconsumer32_data
.err_unix_sock_path
);
5632 DBG2("UST consumer 32 bits cmd path: %s",
5633 ustconsumer32_data
.cmd_unix_sock_path
);
5635 /* 64 bits consumerd path setup */
5636 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5637 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5639 PERROR("snprintf 64-bit consumer error socket path");
5641 goto exit_init_data
;
5643 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5644 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5646 PERROR("snprintf 64-bit consumer command socket path");
5648 goto exit_init_data
;
5651 DBG2("UST consumer 64 bits err path: %s",
5652 ustconsumer64_data
.err_unix_sock_path
);
5653 DBG2("UST consumer 64 bits cmd path: %s",
5654 ustconsumer64_data
.cmd_unix_sock_path
);
5657 * See if daemon already exist.
5659 if (check_existing_daemon()) {
5660 ERR("Already running daemon.\n");
5662 * We do not goto exit because we must not cleanup()
5663 * because a daemon is already running.
5666 goto exit_init_data
;
5670 * Init UST app hash table. Alloc hash table before this point since
5671 * cleanup() can get called after that point.
5673 if (ust_app_ht_alloc()) {
5674 ERR("Failed to allocate UST app hash table");
5676 goto exit_init_data
;
5680 * Initialize agent app hash table. We allocate the hash table here
5681 * since cleanup() can get called after this point.
5683 if (agent_app_ht_alloc()) {
5684 ERR("Failed to allocate Agent app hash table");
5686 goto exit_init_data
;
5690 * These actions must be executed as root. We do that *after* setting up
5691 * the sockets path because we MUST make the check for another daemon using
5692 * those paths *before* trying to set the kernel consumer sockets and init
5696 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5698 goto exit_init_data
;
5701 /* Setup kernel tracer */
5702 if (!opt_no_kernel
) {
5703 init_kernel_tracer();
5704 if (kernel_tracer_fd
>= 0) {
5705 ret
= syscall_init_table();
5707 ERR("Unable to populate syscall table. "
5708 "Syscall tracing won't work "
5709 "for this session daemon.");
5714 /* Set ulimit for open files */
5717 /* init lttng_fd tracking must be done after set_ulimit. */
5720 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5722 goto exit_init_data
;
5725 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5727 goto exit_init_data
;
5730 /* Setup the needed unix socket */
5731 if (init_daemon_socket()) {
5733 goto exit_init_data
;
5736 /* Set credentials to socket */
5737 if (is_root
&& set_permissions(rundir
)) {
5739 goto exit_init_data
;
5742 /* Get parent pid if -S, --sig-parent is specified. */
5743 if (opt_sig_parent
) {
5747 /* Setup the kernel pipe for waking up the kernel thread */
5748 if (is_root
&& !opt_no_kernel
) {
5749 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5751 goto exit_init_data
;
5755 /* Setup the thread apps communication pipe. */
5756 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5758 goto exit_init_data
;
5761 /* Setup the thread apps notify communication pipe. */
5762 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5764 goto exit_init_data
;
5767 /* Initialize global buffer per UID and PID registry. */
5768 buffer_reg_init_uid_registry();
5769 buffer_reg_init_pid_registry();
5771 /* Init UST command queue. */
5772 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5775 * Get session list pointer. This pointer MUST NOT be free'd. This list
5776 * is statically declared in session.c
5778 session_list_ptr
= session_get_list();
5782 /* Check for the application socket timeout env variable. */
5783 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5784 if (env_app_timeout
) {
5785 app_socket_timeout
= atoi(env_app_timeout
);
5787 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5790 ret
= write_pidfile();
5792 ERR("Error in write_pidfile");
5794 goto exit_init_data
;
5796 ret
= write_agent_port();
5798 ERR("Error in write_agent_port");
5800 goto exit_init_data
;
5803 /* Initialize communication library */
5805 /* Initialize TCP timeout values */
5806 lttcomm_inet_init();
5808 if (load_session_init_data(&load_info
) < 0) {
5810 goto exit_init_data
;
5812 load_info
->path
= opt_load_session_path
;
5814 /* Create health-check thread */
5815 ret
= pthread_create(&health_thread
, NULL
,
5816 thread_manage_health
, (void *) NULL
);
5819 PERROR("pthread_create health");
5824 /* Create thread to manage the client socket */
5825 ret
= pthread_create(&client_thread
, NULL
,
5826 thread_manage_clients
, (void *) NULL
);
5829 PERROR("pthread_create clients");
5834 /* Create thread to dispatch registration */
5835 ret
= pthread_create(&dispatch_thread
, NULL
,
5836 thread_dispatch_ust_registration
, (void *) NULL
);
5839 PERROR("pthread_create dispatch");
5844 /* Create thread to manage application registration. */
5845 ret
= pthread_create(®_apps_thread
, NULL
,
5846 thread_registration_apps
, (void *) NULL
);
5849 PERROR("pthread_create registration");
5854 /* Create thread to manage application socket */
5855 ret
= pthread_create(&apps_thread
, NULL
,
5856 thread_manage_apps
, (void *) NULL
);
5859 PERROR("pthread_create apps");
5864 /* Create thread to manage application notify socket */
5865 ret
= pthread_create(&apps_notify_thread
, NULL
,
5866 ust_thread_manage_notify
, (void *) NULL
);
5869 PERROR("pthread_create notify");
5871 goto exit_apps_notify
;
5874 /* Create agent registration thread. */
5875 ret
= pthread_create(&agent_reg_thread
, NULL
,
5876 agent_thread_manage_registration
, (void *) NULL
);
5879 PERROR("pthread_create agent");
5881 goto exit_agent_reg
;
5884 /* Don't start this thread if kernel tracing is not requested nor root */
5885 if (is_root
&& !opt_no_kernel
) {
5886 /* Create kernel thread to manage kernel event */
5887 ret
= pthread_create(&kernel_thread
, NULL
,
5888 thread_manage_kernel
, (void *) NULL
);
5891 PERROR("pthread_create kernel");
5897 /* Create session loading thread. */
5898 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
5902 PERROR("pthread_create load_session_thread");
5904 goto exit_load_session
;
5908 * This is where we start awaiting program completion (e.g. through
5909 * signal that asks threads to teardown).
5912 ret
= pthread_join(load_session_thread
, &status
);
5915 PERROR("pthread_join load_session_thread");
5920 if (is_root
&& !opt_no_kernel
) {
5921 ret
= pthread_join(kernel_thread
, &status
);
5924 PERROR("pthread_join");
5930 ret
= pthread_join(agent_reg_thread
, &status
);
5933 PERROR("pthread_join agent");
5938 ret
= pthread_join(apps_notify_thread
, &status
);
5941 PERROR("pthread_join apps notify");
5946 ret
= pthread_join(apps_thread
, &status
);
5949 PERROR("pthread_join apps");
5954 ret
= pthread_join(reg_apps_thread
, &status
);
5957 PERROR("pthread_join");
5963 * Join dispatch thread after joining reg_apps_thread to ensure
5964 * we don't leak applications in the queue.
5966 ret
= pthread_join(dispatch_thread
, &status
);
5969 PERROR("pthread_join");
5974 ret
= pthread_join(client_thread
, &status
);
5977 PERROR("pthread_join");
5982 ret
= pthread_join(health_thread
, &status
);
5985 PERROR("pthread_join health thread");
5992 * sessiond_cleanup() is called when no other thread is running, except
5993 * the ht_cleanup thread, which is needed to destroy the hash tables.
5995 rcu_thread_online();
5997 rcu_thread_offline();
5998 rcu_unregister_thread();
6000 ret
= notify_thread_pipe(ht_cleanup_quit_pipe
[1]);
6002 ERR("write error on ht_cleanup quit pipe");
6006 ret
= pthread_join(ht_cleanup_thread
, &status
);
6009 PERROR("pthread_join ht cleanup thread");
6015 utils_close_pipe(ht_cleanup_pipe
);
6016 exit_ht_cleanup_pipe
:
6019 * Close the ht_cleanup quit pipe.
6021 utils_close_pipe(ht_cleanup_quit_pipe
);
6022 exit_ht_cleanup_quit_pipe
:
6024 health_app_destroy(health_sessiond
);
6025 exit_health_sessiond_cleanup
:
6028 sessiond_cleanup_options();
6030 exit_set_signal_handler
:
6031 /* Ensure all prior call_rcu are done. */