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.
32 #include <sys/mount.h>
33 #include <sys/resource.h>
34 #include <sys/socket.h>
36 #include <sys/types.h>
38 #include <urcu/uatomic.h>
42 #include <common/common.h>
43 #include <common/compat/socket.h>
44 #include <common/defaults.h>
45 #include <common/kernel-consumer/kernel-consumer.h>
46 #include <common/futex.h>
47 #include <common/relayd/relayd.h>
48 #include <common/utils.h>
49 #include <common/daemonize.h>
50 #include <common/config/config.h>
52 #include "lttng-sessiond.h"
53 #include "buffer-registry.h"
60 #include "kernel-consumer.h"
64 #include "ust-consumer.h"
67 #include "health-sessiond.h"
68 #include "testpoint.h"
69 #include "ust-thread.h"
70 #include "jul-thread.h"
72 #include "load-session-thread.h"
74 #define CONSUMERD_FILE "lttng-consumerd"
77 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
78 static int tracing_group_name_override
;
79 static char *opt_pidfile
;
80 static int opt_sig_parent
;
81 static int opt_verbose_consumer
;
82 static int opt_daemon
, opt_background
;
83 static int opt_no_kernel
;
84 static char *opt_load_session_path
;
85 static pid_t ppid
; /* Parent PID for --sig-parent option */
86 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
88 static int lockfile_fd
= -1;
90 /* Set to 1 when a SIGUSR1 signal is received. */
91 static int recv_child_signal
;
94 * Consumer daemon specific control data. Every value not initialized here is
95 * set to 0 by the static definition.
97 static struct consumer_data kconsumer_data
= {
98 .type
= LTTNG_CONSUMER_KERNEL
,
99 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
100 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
103 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
104 .lock
= PTHREAD_MUTEX_INITIALIZER
,
105 .cond
= PTHREAD_COND_INITIALIZER
,
106 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
108 static struct consumer_data ustconsumer64_data
= {
109 .type
= LTTNG_CONSUMER64_UST
,
110 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
111 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
114 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
115 .lock
= PTHREAD_MUTEX_INITIALIZER
,
116 .cond
= PTHREAD_COND_INITIALIZER
,
117 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
119 static struct consumer_data ustconsumer32_data
= {
120 .type
= LTTNG_CONSUMER32_UST
,
121 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
122 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
125 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
126 .lock
= PTHREAD_MUTEX_INITIALIZER
,
127 .cond
= PTHREAD_COND_INITIALIZER
,
128 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
131 /* Command line options */
132 static const struct option long_options
[] = {
133 { "client-sock", 1, 0, 'c' },
134 { "apps-sock", 1, 0, 'a' },
135 { "kconsumerd-cmd-sock", 1, 0, 'C' },
136 { "kconsumerd-err-sock", 1, 0, 'E' },
137 { "ustconsumerd32-cmd-sock", 1, 0, 'G' },
138 { "ustconsumerd32-err-sock", 1, 0, 'H' },
139 { "ustconsumerd64-cmd-sock", 1, 0, 'D' },
140 { "ustconsumerd64-err-sock", 1, 0, 'F' },
141 { "consumerd32-path", 1, 0, 'u' },
142 { "consumerd32-libdir", 1, 0, 'U' },
143 { "consumerd64-path", 1, 0, 't' },
144 { "consumerd64-libdir", 1, 0, 'T' },
145 { "daemonize", 0, 0, 'd' },
146 { "background", 0, 0, 'b' },
147 { "sig-parent", 0, 0, 'S' },
148 { "help", 0, 0, 'h' },
149 { "group", 1, 0, 'g' },
150 { "version", 0, 0, 'V' },
151 { "quiet", 0, 0, 'q' },
152 { "verbose", 0, 0, 'v' },
153 { "verbose-consumer", 0, 0, 'Z' },
154 { "no-kernel", 0, 0, 'N' },
155 { "pidfile", 1, 0, 'p' },
156 { "jul-tcp-port", 1, 0, 'J' },
157 { "config", 1, 0, 'f' },
158 { "load", 1, 0, 'l' },
159 { "kmod-probes", 1, 0, 'P' },
163 /* Command line options to ignore from configuration file */
164 static const char *config_ignore_options
[] = { "help", "version", "config" };
166 /* Shared between threads */
167 static int dispatch_thread_exit
;
169 /* Global application Unix socket path */
170 static char apps_unix_sock_path
[PATH_MAX
];
171 /* Global client Unix socket path */
172 static char client_unix_sock_path
[PATH_MAX
];
173 /* global wait shm path for UST */
174 static char wait_shm_path
[PATH_MAX
];
175 /* Global health check unix path */
176 static char health_unix_sock_path
[PATH_MAX
];
178 /* Sockets and FDs */
179 static int client_sock
= -1;
180 static int apps_sock
= -1;
181 int kernel_tracer_fd
= -1;
182 static int kernel_poll_pipe
[2] = { -1, -1 };
185 * Quit pipe for all threads. This permits a single cancellation point
186 * for all threads when receiving an event on the pipe.
188 static int thread_quit_pipe
[2] = { -1, -1 };
191 * This pipe is used to inform the thread managing application communication
192 * that a command is queued and ready to be processed.
194 static int apps_cmd_pipe
[2] = { -1, -1 };
196 int apps_cmd_notify_pipe
[2] = { -1, -1 };
198 /* Pthread, Mutexes and Semaphores */
199 static pthread_t apps_thread
;
200 static pthread_t apps_notify_thread
;
201 static pthread_t reg_apps_thread
;
202 static pthread_t client_thread
;
203 static pthread_t kernel_thread
;
204 static pthread_t dispatch_thread
;
205 static pthread_t health_thread
;
206 static pthread_t ht_cleanup_thread
;
207 static pthread_t jul_reg_thread
;
208 static pthread_t load_session_thread
;
211 * UST registration command queue. This queue is tied with a futex and uses a N
212 * wakers / 1 waiter implemented and detailed in futex.c/.h
214 * The thread_registration_apps and thread_dispatch_ust_registration uses this
215 * queue along with the wait/wake scheme. The thread_manage_apps receives down
216 * the line new application socket and monitors it for any I/O error or clean
217 * close that triggers an unregistration of the application.
219 static struct ust_cmd_queue ust_cmd_queue
;
222 * Pointer initialized before thread creation.
224 * This points to the tracing session list containing the session count and a
225 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
226 * MUST NOT be taken if you call a public function in session.c.
228 * The lock is nested inside the structure: session_list_ptr->lock. Please use
229 * session_lock_list and session_unlock_list for lock acquisition.
231 static struct ltt_session_list
*session_list_ptr
;
233 int ust_consumerd64_fd
= -1;
234 int ust_consumerd32_fd
= -1;
236 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
237 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
238 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
239 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
240 static int consumerd32_bin_override
;
241 static int consumerd64_bin_override
;
242 static int consumerd32_libdir_override
;
243 static int consumerd64_libdir_override
;
245 static const char *module_proc_lttng
= "/proc/lttng";
248 * Consumer daemon state which is changed when spawning it, killing it or in
249 * case of a fatal error.
251 enum consumerd_state
{
252 CONSUMER_STARTED
= 1,
253 CONSUMER_STOPPED
= 2,
258 * This consumer daemon state is used to validate if a client command will be
259 * able to reach the consumer. If not, the client is informed. For instance,
260 * doing a "lttng start" when the consumer state is set to ERROR will return an
261 * error to the client.
263 * The following example shows a possible race condition of this scheme:
265 * consumer thread error happens
267 * client cmd checks state -> still OK
268 * consumer thread exit, sets error
269 * client cmd try to talk to consumer
272 * However, since the consumer is a different daemon, we have no way of making
273 * sure the command will reach it safely even with this state flag. This is why
274 * we consider that up to the state validation during command processing, the
275 * command is safe. After that, we can not guarantee the correctness of the
276 * client request vis-a-vis the consumer.
278 static enum consumerd_state ust_consumerd_state
;
279 static enum consumerd_state kernel_consumerd_state
;
282 * Socket timeout for receiving and sending in seconds.
284 static int app_socket_timeout
;
286 /* Set in main() with the current page size. */
289 /* Application health monitoring */
290 struct health_app
*health_sessiond
;
292 /* JUL TCP port for registration. Used by the JUL thread. */
293 unsigned int jul_tcp_port
= DEFAULT_JUL_TCP_PORT
;
295 /* Am I root or not. */
296 int is_root
; /* Set to 1 if the daemon is running as root */
298 const char * const config_section_name
= "sessiond";
300 /* Load session thread information to operate. */
301 struct load_session_thread_data
*load_info
;
304 * Whether sessiond is ready for commands/health check requests.
305 * NR_LTTNG_SESSIOND_READY must match the number of calls to
306 * sessiond_notify_ready().
308 #define NR_LTTNG_SESSIOND_READY 3
309 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
311 /* Notify parents that we are ready for cmd and health check */
313 void sessiond_notify_ready(void)
315 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
317 * Notify parent pid that we are ready to accept command
318 * for client side. This ppid is the one from the
319 * external process that spawned us.
321 if (opt_sig_parent
) {
326 * Notify the parent of the fork() process that we are
329 if (opt_daemon
|| opt_background
) {
330 kill(child_ppid
, SIGUSR1
);
336 void setup_consumerd_path(void)
338 const char *bin
, *libdir
;
341 * Allow INSTALL_BIN_PATH to be used as a target path for the
342 * native architecture size consumer if CONFIG_CONSUMER*_PATH
343 * has not been defined.
345 #if (CAA_BITS_PER_LONG == 32)
346 if (!consumerd32_bin
[0]) {
347 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
349 if (!consumerd32_libdir
[0]) {
350 consumerd32_libdir
= INSTALL_LIB_PATH
;
352 #elif (CAA_BITS_PER_LONG == 64)
353 if (!consumerd64_bin
[0]) {
354 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
356 if (!consumerd64_libdir
[0]) {
357 consumerd64_libdir
= INSTALL_LIB_PATH
;
360 #error "Unknown bitness"
364 * runtime env. var. overrides the build default.
366 bin
= getenv("LTTNG_CONSUMERD32_BIN");
368 consumerd32_bin
= bin
;
370 bin
= getenv("LTTNG_CONSUMERD64_BIN");
372 consumerd64_bin
= bin
;
374 libdir
= getenv("LTTNG_CONSUMERD32_LIBDIR");
376 consumerd32_libdir
= libdir
;
378 libdir
= getenv("LTTNG_CONSUMERD64_LIBDIR");
380 consumerd64_libdir
= libdir
;
385 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
387 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
393 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
399 ret
= lttng_poll_add(events
, thread_quit_pipe
[0], LPOLLIN
| LPOLLERR
);
411 * Check if the thread quit pipe was triggered.
413 * Return 1 if it was triggered else 0;
415 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
417 if (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) {
425 * Init thread quit pipe.
427 * Return -1 on error or 0 if all pipes are created.
429 static int init_thread_quit_pipe(void)
433 ret
= pipe(thread_quit_pipe
);
435 PERROR("thread quit pipe");
439 for (i
= 0; i
< 2; i
++) {
440 ret
= fcntl(thread_quit_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
452 * Stop all threads by closing the thread quit pipe.
454 static void stop_threads(void)
458 /* Stopping all threads */
459 DBG("Terminating all threads");
460 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
462 ERR("write error on thread quit pipe");
465 /* Dispatch thread */
466 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
467 futex_nto1_wake(&ust_cmd_queue
.futex
);
471 * Close every consumer sockets.
473 static void close_consumer_sockets(void)
477 if (kconsumer_data
.err_sock
>= 0) {
478 ret
= close(kconsumer_data
.err_sock
);
480 PERROR("kernel consumer err_sock close");
483 if (ustconsumer32_data
.err_sock
>= 0) {
484 ret
= close(ustconsumer32_data
.err_sock
);
486 PERROR("UST consumerd32 err_sock close");
489 if (ustconsumer64_data
.err_sock
>= 0) {
490 ret
= close(ustconsumer64_data
.err_sock
);
492 PERROR("UST consumerd64 err_sock close");
495 if (kconsumer_data
.cmd_sock
>= 0) {
496 ret
= close(kconsumer_data
.cmd_sock
);
498 PERROR("kernel consumer cmd_sock close");
501 if (ustconsumer32_data
.cmd_sock
>= 0) {
502 ret
= close(ustconsumer32_data
.cmd_sock
);
504 PERROR("UST consumerd32 cmd_sock close");
507 if (ustconsumer64_data
.cmd_sock
>= 0) {
508 ret
= close(ustconsumer64_data
.cmd_sock
);
510 PERROR("UST consumerd64 cmd_sock close");
516 * Generate the full lock file path using the rundir.
518 * Return the snprintf() return value thus a negative value is an error.
520 static int generate_lock_file_path(char *path
, size_t len
)
527 /* Build lockfile path from rundir. */
528 ret
= snprintf(path
, len
, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE
, rundir
);
530 PERROR("snprintf lockfile path");
539 static void cleanup(void)
542 struct ltt_session
*sess
, *stmp
;
548 * Close the thread quit pipe. It has already done its job,
549 * since we are now called.
551 utils_close_pipe(thread_quit_pipe
);
554 * If opt_pidfile is undefined, the default file will be wiped when
555 * removing the rundir.
558 ret
= remove(opt_pidfile
);
560 PERROR("remove pidfile %s", opt_pidfile
);
564 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
567 snprintf(path
, PATH_MAX
,
569 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
570 DBG("Removing %s", path
);
573 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
574 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
);
575 DBG("Removing %s", path
);
579 snprintf(path
, PATH_MAX
,
580 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
582 DBG("Removing %s", path
);
585 snprintf(path
, PATH_MAX
,
586 DEFAULT_KCONSUMERD_PATH
,
588 DBG("Removing directory %s", path
);
591 /* ust consumerd 32 */
592 snprintf(path
, PATH_MAX
,
593 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
595 DBG("Removing %s", path
);
598 snprintf(path
, PATH_MAX
,
599 DEFAULT_USTCONSUMERD32_PATH
,
601 DBG("Removing directory %s", path
);
604 /* ust consumerd 64 */
605 snprintf(path
, PATH_MAX
,
606 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
608 DBG("Removing %s", path
);
611 snprintf(path
, PATH_MAX
,
612 DEFAULT_USTCONSUMERD64_PATH
,
614 DBG("Removing directory %s", path
);
617 DBG("Cleaning up all sessions");
619 /* Destroy session list mutex */
620 if (session_list_ptr
!= NULL
) {
621 pthread_mutex_destroy(&session_list_ptr
->lock
);
623 /* Cleanup ALL session */
624 cds_list_for_each_entry_safe(sess
, stmp
,
625 &session_list_ptr
->head
, list
) {
626 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
630 DBG("Closing all UST sockets");
631 ust_app_clean_list();
632 buffer_reg_destroy_registries();
634 if (is_root
&& !opt_no_kernel
) {
635 DBG2("Closing kernel fd");
636 if (kernel_tracer_fd
>= 0) {
637 ret
= close(kernel_tracer_fd
);
642 DBG("Unloading kernel modules");
643 modprobe_remove_lttng_all();
646 close_consumer_sockets();
649 * If the override option is set, the pointer points to a *non* const thus
650 * freeing it even though the variable type is set to const.
652 if (tracing_group_name_override
) {
653 free((void *) tracing_group_name
);
655 if (consumerd32_bin_override
) {
656 free((void *) consumerd32_bin
);
658 if (consumerd64_bin_override
) {
659 free((void *) consumerd64_bin
);
661 if (consumerd32_libdir_override
) {
662 free((void *) consumerd32_libdir
);
664 if (consumerd64_libdir_override
) {
665 free((void *) consumerd64_libdir
);
672 if (opt_load_session_path
) {
673 free(opt_load_session_path
);
677 load_session_destroy_data(load_info
);
682 * Cleanup lock file by deleting it and finaly closing it which will
683 * release the file system lock.
685 if (lockfile_fd
>= 0) {
686 char lockfile_path
[PATH_MAX
];
688 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
690 ret
= remove(lockfile_path
);
692 PERROR("remove lock file");
694 ret
= close(lockfile_fd
);
696 PERROR("close lock file");
702 * We do NOT rmdir rundir because there are other processes
703 * using it, for instance lttng-relayd, which can start in
704 * parallel with this teardown.
710 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
711 "Matthew, BEET driven development works!%c[%dm",
712 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
717 * Send data on a unix socket using the liblttsessiondcomm API.
719 * Return lttcomm error code.
721 static int send_unix_sock(int sock
, void *buf
, size_t len
)
723 /* Check valid length */
728 return lttcomm_send_unix_sock(sock
, buf
, len
);
732 * Free memory of a command context structure.
734 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
736 DBG("Clean command context structure");
738 if ((*cmd_ctx
)->llm
) {
739 free((*cmd_ctx
)->llm
);
741 if ((*cmd_ctx
)->lsm
) {
742 free((*cmd_ctx
)->lsm
);
750 * Notify UST applications using the shm mmap futex.
752 static int notify_ust_apps(int active
)
756 DBG("Notifying applications of session daemon state: %d", active
);
758 /* See shm.c for this call implying mmap, shm and futex calls */
759 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
760 if (wait_shm_mmap
== NULL
) {
764 /* Wake waiting process */
765 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
767 /* Apps notified successfully */
775 * Setup the outgoing data buffer for the response (llm) by allocating the
776 * right amount of memory and copying the original information from the lsm
779 * Return total size of the buffer pointed by buf.
781 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
, size_t size
)
787 cmd_ctx
->llm
= zmalloc(sizeof(struct lttcomm_lttng_msg
) + buf_size
);
788 if (cmd_ctx
->llm
== NULL
) {
794 /* Copy common data */
795 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
796 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
798 cmd_ctx
->llm
->data_size
= size
;
799 cmd_ctx
->lttng_msg_size
= sizeof(struct lttcomm_lttng_msg
) + buf_size
;
808 * Update the kernel poll set of all channel fd available over all tracing
809 * session. Add the wakeup pipe at the end of the set.
811 static int update_kernel_poll(struct lttng_poll_event
*events
)
814 struct ltt_session
*session
;
815 struct ltt_kernel_channel
*channel
;
817 DBG("Updating kernel poll set");
820 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
821 session_lock(session
);
822 if (session
->kernel_session
== NULL
) {
823 session_unlock(session
);
827 cds_list_for_each_entry(channel
,
828 &session
->kernel_session
->channel_list
.head
, list
) {
829 /* Add channel fd to the kernel poll set */
830 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
832 session_unlock(session
);
835 DBG("Channel fd %d added to kernel set", channel
->fd
);
837 session_unlock(session
);
839 session_unlock_list();
844 session_unlock_list();
849 * Find the channel fd from 'fd' over all tracing session. When found, check
850 * for new channel stream and send those stream fds to the kernel consumer.
852 * Useful for CPU hotplug feature.
854 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
857 struct ltt_session
*session
;
858 struct ltt_kernel_session
*ksess
;
859 struct ltt_kernel_channel
*channel
;
861 DBG("Updating kernel streams for channel fd %d", fd
);
864 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
865 session_lock(session
);
866 if (session
->kernel_session
== NULL
) {
867 session_unlock(session
);
870 ksess
= session
->kernel_session
;
872 cds_list_for_each_entry(channel
, &ksess
->channel_list
.head
, list
) {
873 if (channel
->fd
== fd
) {
874 DBG("Channel found, updating kernel streams");
875 ret
= kernel_open_channel_stream(channel
);
879 /* Update the stream global counter */
880 ksess
->stream_count_global
+= ret
;
883 * Have we already sent fds to the consumer? If yes, it means
884 * that tracing is started so it is safe to send our updated
887 if (ksess
->consumer_fds_sent
== 1 && ksess
->consumer
!= NULL
) {
888 struct lttng_ht_iter iter
;
889 struct consumer_socket
*socket
;
892 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
893 &iter
.iter
, socket
, node
.node
) {
894 pthread_mutex_lock(socket
->lock
);
895 ret
= kernel_consumer_send_channel_stream(socket
,
897 session
->output_traces
? 1 : 0);
898 pthread_mutex_unlock(socket
->lock
);
909 session_unlock(session
);
911 session_unlock_list();
915 session_unlock(session
);
916 session_unlock_list();
921 * For each tracing session, update newly registered apps. The session list
922 * lock MUST be acquired before calling this.
924 static void update_ust_app(int app_sock
)
926 struct ltt_session
*sess
, *stmp
;
928 /* Consumer is in an ERROR state. Stop any application update. */
929 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
930 /* Stop the update process since the consumer is dead. */
934 /* For all tracing session(s) */
935 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
937 if (sess
->ust_session
) {
938 ust_app_global_update(sess
->ust_session
, app_sock
);
940 session_unlock(sess
);
945 * This thread manage event coming from the kernel.
947 * Features supported in this thread:
950 static void *thread_manage_kernel(void *data
)
952 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
953 uint32_t revents
, nb_fd
;
955 struct lttng_poll_event events
;
957 DBG("[thread] Thread manage kernel started");
959 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
962 * This first step of the while is to clean this structure which could free
963 * non NULL pointers so initialize it before the loop.
965 lttng_poll_init(&events
);
967 if (testpoint(sessiond_thread_manage_kernel
)) {
968 goto error_testpoint
;
971 health_code_update();
973 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
974 goto error_testpoint
;
978 health_code_update();
980 if (update_poll_flag
== 1) {
981 /* Clean events object. We are about to populate it again. */
982 lttng_poll_clean(&events
);
984 ret
= sessiond_set_thread_pollset(&events
, 2);
986 goto error_poll_create
;
989 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
994 /* This will add the available kernel channel if any. */
995 ret
= update_kernel_poll(&events
);
999 update_poll_flag
= 0;
1002 DBG("Thread kernel polling");
1004 /* Poll infinite value of time */
1006 health_poll_entry();
1007 ret
= lttng_poll_wait(&events
, -1);
1008 DBG("Thread kernel return from poll on %d fds",
1009 LTTNG_POLL_GETNB(&events
));
1013 * Restart interrupted system call.
1015 if (errno
== EINTR
) {
1019 } else if (ret
== 0) {
1020 /* Should not happen since timeout is infinite */
1021 ERR("Return value of poll is 0 with an infinite timeout.\n"
1022 "This should not have happened! Continuing...");
1028 for (i
= 0; i
< nb_fd
; i
++) {
1029 /* Fetch once the poll data */
1030 revents
= LTTNG_POLL_GETEV(&events
, i
);
1031 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1033 health_code_update();
1036 /* No activity for this FD (poll implementation). */
1040 /* Thread quit pipe has been closed. Killing thread. */
1041 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1047 /* Check for data on kernel pipe */
1048 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
1049 (void) lttng_read(kernel_poll_pipe
[0],
1052 * Ret value is useless here, if this pipe gets any actions an
1053 * update is required anyway.
1055 update_poll_flag
= 1;
1059 * New CPU detected by the kernel. Adding kernel stream to
1060 * kernel session and updating the kernel consumer
1062 if (revents
& LPOLLIN
) {
1063 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1069 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1070 * and unregister kernel stream at this point.
1079 lttng_poll_clean(&events
);
1082 utils_close_pipe(kernel_poll_pipe
);
1083 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1086 ERR("Health error occurred in %s", __func__
);
1087 WARN("Kernel thread died unexpectedly. "
1088 "Kernel tracing can continue but CPU hotplug is disabled.");
1090 health_unregister(health_sessiond
);
1091 DBG("Kernel thread dying");
1096 * Signal pthread condition of the consumer data that the thread.
1098 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1100 pthread_mutex_lock(&data
->cond_mutex
);
1103 * The state is set before signaling. It can be any value, it's the waiter
1104 * job to correctly interpret this condition variable associated to the
1105 * consumer pthread_cond.
1107 * A value of 0 means that the corresponding thread of the consumer data
1108 * was not started. 1 indicates that the thread has started and is ready
1109 * for action. A negative value means that there was an error during the
1112 data
->consumer_thread_is_ready
= state
;
1113 (void) pthread_cond_signal(&data
->cond
);
1115 pthread_mutex_unlock(&data
->cond_mutex
);
1119 * This thread manage the consumer error sent back to the session daemon.
1121 static void *thread_manage_consumer(void *data
)
1123 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1124 uint32_t revents
, nb_fd
;
1125 enum lttcomm_return_code code
;
1126 struct lttng_poll_event events
;
1127 struct consumer_data
*consumer_data
= data
;
1129 DBG("[thread] Manage consumer started");
1131 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1133 health_code_update();
1136 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1137 * metadata_sock. Nothing more will be added to this poll set.
1139 ret
= sessiond_set_thread_pollset(&events
, 3);
1145 * The error socket here is already in a listening state which was done
1146 * just before spawning this thread to avoid a race between the consumer
1147 * daemon exec trying to connect and the listen() call.
1149 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1154 health_code_update();
1156 /* Infinite blocking call, waiting for transmission */
1158 health_poll_entry();
1160 if (testpoint(sessiond_thread_manage_consumer
)) {
1164 ret
= lttng_poll_wait(&events
, -1);
1168 * Restart interrupted system call.
1170 if (errno
== EINTR
) {
1178 for (i
= 0; i
< nb_fd
; i
++) {
1179 /* Fetch once the poll data */
1180 revents
= LTTNG_POLL_GETEV(&events
, i
);
1181 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1183 health_code_update();
1186 /* No activity for this FD (poll implementation). */
1190 /* Thread quit pipe has been closed. Killing thread. */
1191 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1197 /* Event on the registration socket */
1198 if (pollfd
== consumer_data
->err_sock
) {
1199 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1200 ERR("consumer err socket poll error");
1206 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1212 * Set the CLOEXEC flag. Return code is useless because either way, the
1215 (void) utils_set_fd_cloexec(sock
);
1217 health_code_update();
1219 DBG2("Receiving code from consumer err_sock");
1221 /* Getting status code from kconsumerd */
1222 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1223 sizeof(enum lttcomm_return_code
));
1228 health_code_update();
1229 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1230 /* Connect both socket, command and metadata. */
1231 consumer_data
->cmd_sock
=
1232 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1233 consumer_data
->metadata_fd
=
1234 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1235 if (consumer_data
->cmd_sock
< 0
1236 || consumer_data
->metadata_fd
< 0) {
1237 PERROR("consumer connect cmd socket");
1238 /* On error, signal condition and quit. */
1239 signal_consumer_condition(consumer_data
, -1);
1242 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1243 /* Create metadata socket lock. */
1244 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1245 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1246 PERROR("zmalloc pthread mutex");
1250 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1252 signal_consumer_condition(consumer_data
, 1);
1253 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1254 DBG("Consumer metadata socket ready (fd: %d)",
1255 consumer_data
->metadata_fd
);
1257 ERR("consumer error when waiting for SOCK_READY : %s",
1258 lttcomm_get_readable_code(-code
));
1262 /* Remove the consumerd error sock since we've established a connexion */
1263 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1268 /* Add new accepted error socket. */
1269 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1274 /* Add metadata socket that is successfully connected. */
1275 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1276 LPOLLIN
| LPOLLRDHUP
);
1281 health_code_update();
1283 /* Infinite blocking call, waiting for transmission */
1286 health_code_update();
1288 /* Exit the thread because the thread quit pipe has been triggered. */
1290 /* Not a health error. */
1295 health_poll_entry();
1296 ret
= lttng_poll_wait(&events
, -1);
1300 * Restart interrupted system call.
1302 if (errno
== EINTR
) {
1310 for (i
= 0; i
< nb_fd
; i
++) {
1311 /* Fetch once the poll data */
1312 revents
= LTTNG_POLL_GETEV(&events
, i
);
1313 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1315 health_code_update();
1318 /* No activity for this FD (poll implementation). */
1323 * Thread quit pipe has been triggered, flag that we should stop
1324 * but continue the current loop to handle potential data from
1327 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1329 if (pollfd
== sock
) {
1330 /* Event on the consumerd socket */
1331 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1332 ERR("consumer err socket second poll error");
1335 health_code_update();
1336 /* Wait for any kconsumerd error */
1337 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1338 sizeof(enum lttcomm_return_code
));
1340 ERR("consumer closed the command socket");
1344 ERR("consumer return code : %s",
1345 lttcomm_get_readable_code(-code
));
1348 } else if (pollfd
== consumer_data
->metadata_fd
) {
1349 /* UST metadata requests */
1350 ret
= ust_consumer_metadata_request(
1351 &consumer_data
->metadata_sock
);
1353 ERR("Handling metadata request");
1357 /* No need for an else branch all FDs are tested prior. */
1359 health_code_update();
1365 * We lock here because we are about to close the sockets and some other
1366 * thread might be using them so get exclusive access which will abort all
1367 * other consumer command by other threads.
1369 pthread_mutex_lock(&consumer_data
->lock
);
1371 /* Immediately set the consumerd state to stopped */
1372 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1373 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1374 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1375 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1376 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1378 /* Code flow error... */
1382 if (consumer_data
->err_sock
>= 0) {
1383 ret
= close(consumer_data
->err_sock
);
1387 consumer_data
->err_sock
= -1;
1389 if (consumer_data
->cmd_sock
>= 0) {
1390 ret
= close(consumer_data
->cmd_sock
);
1394 consumer_data
->cmd_sock
= -1;
1396 if (consumer_data
->metadata_sock
.fd_ptr
&&
1397 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1398 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1410 unlink(consumer_data
->err_unix_sock_path
);
1411 unlink(consumer_data
->cmd_unix_sock_path
);
1412 consumer_data
->pid
= 0;
1413 pthread_mutex_unlock(&consumer_data
->lock
);
1415 /* Cleanup metadata socket mutex. */
1416 if (consumer_data
->metadata_sock
.lock
) {
1417 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1418 free(consumer_data
->metadata_sock
.lock
);
1420 lttng_poll_clean(&events
);
1424 ERR("Health error occurred in %s", __func__
);
1426 health_unregister(health_sessiond
);
1427 DBG("consumer thread cleanup completed");
1433 * This thread manage application communication.
1435 static void *thread_manage_apps(void *data
)
1437 int i
, ret
, pollfd
, err
= -1;
1439 uint32_t revents
, nb_fd
;
1440 struct lttng_poll_event events
;
1442 DBG("[thread] Manage application started");
1444 rcu_register_thread();
1445 rcu_thread_online();
1447 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1449 if (testpoint(sessiond_thread_manage_apps
)) {
1450 goto error_testpoint
;
1453 health_code_update();
1455 ret
= sessiond_set_thread_pollset(&events
, 2);
1457 goto error_poll_create
;
1460 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1465 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1469 health_code_update();
1472 DBG("Apps thread polling");
1474 /* Inifinite blocking call, waiting for transmission */
1476 health_poll_entry();
1477 ret
= lttng_poll_wait(&events
, -1);
1478 DBG("Apps thread return from poll on %d fds",
1479 LTTNG_POLL_GETNB(&events
));
1483 * Restart interrupted system call.
1485 if (errno
== EINTR
) {
1493 for (i
= 0; i
< nb_fd
; i
++) {
1494 /* Fetch once the poll data */
1495 revents
= LTTNG_POLL_GETEV(&events
, i
);
1496 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1498 health_code_update();
1501 /* No activity for this FD (poll implementation). */
1505 /* Thread quit pipe has been closed. Killing thread. */
1506 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1512 /* Inspect the apps cmd pipe */
1513 if (pollfd
== apps_cmd_pipe
[0]) {
1514 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1515 ERR("Apps command pipe error");
1517 } else if (revents
& LPOLLIN
) {
1521 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1522 if (size_ret
< sizeof(sock
)) {
1523 PERROR("read apps cmd pipe");
1527 health_code_update();
1530 * We only monitor the error events of the socket. This
1531 * thread does not handle any incoming data from UST
1534 ret
= lttng_poll_add(&events
, sock
,
1535 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1540 DBG("Apps with sock %d added to poll set", sock
);
1544 * At this point, we know that a registered application made
1545 * the event at poll_wait.
1547 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1548 /* Removing from the poll set */
1549 ret
= lttng_poll_del(&events
, pollfd
);
1554 /* Socket closed on remote end. */
1555 ust_app_unregister(pollfd
);
1559 health_code_update();
1565 lttng_poll_clean(&events
);
1568 utils_close_pipe(apps_cmd_pipe
);
1569 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1572 * We don't clean the UST app hash table here since already registered
1573 * applications can still be controlled so let them be until the session
1574 * daemon dies or the applications stop.
1579 ERR("Health error occurred in %s", __func__
);
1581 health_unregister(health_sessiond
);
1582 DBG("Application communication apps thread cleanup complete");
1583 rcu_thread_offline();
1584 rcu_unregister_thread();
1589 * Send a socket to a thread This is called from the dispatch UST registration
1590 * thread once all sockets are set for the application.
1592 * The sock value can be invalid, we don't really care, the thread will handle
1593 * it and make the necessary cleanup if so.
1595 * On success, return 0 else a negative value being the errno message of the
1598 static int send_socket_to_thread(int fd
, int sock
)
1603 * It's possible that the FD is set as invalid with -1 concurrently just
1604 * before calling this function being a shutdown state of the thread.
1611 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1612 if (ret
< sizeof(sock
)) {
1613 PERROR("write apps pipe %d", fd
);
1620 /* All good. Don't send back the write positive ret value. */
1627 * Sanitize the wait queue of the dispatch registration thread meaning removing
1628 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1629 * notify socket is never received.
1631 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1633 int ret
, nb_fd
= 0, i
;
1634 unsigned int fd_added
= 0;
1635 struct lttng_poll_event events
;
1636 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1640 lttng_poll_init(&events
);
1642 /* Just skip everything for an empty queue. */
1643 if (!wait_queue
->count
) {
1647 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1652 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1653 &wait_queue
->head
, head
) {
1654 assert(wait_node
->app
);
1655 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1656 LPOLLHUP
| LPOLLERR
);
1669 * Poll but don't block so we can quickly identify the faulty events and
1670 * clean them afterwards from the wait queue.
1672 ret
= lttng_poll_wait(&events
, 0);
1678 for (i
= 0; i
< nb_fd
; i
++) {
1679 /* Get faulty FD. */
1680 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1681 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1684 /* No activity for this FD (poll implementation). */
1688 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1689 &wait_queue
->head
, head
) {
1690 if (pollfd
== wait_node
->app
->sock
&&
1691 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1692 cds_list_del(&wait_node
->head
);
1693 wait_queue
->count
--;
1694 ust_app_destroy(wait_node
->app
);
1702 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1706 lttng_poll_clean(&events
);
1710 lttng_poll_clean(&events
);
1712 ERR("Unable to sanitize wait queue");
1717 * Dispatch request from the registration threads to the application
1718 * communication thread.
1720 static void *thread_dispatch_ust_registration(void *data
)
1723 struct cds_wfq_node
*node
;
1724 struct ust_command
*ust_cmd
= NULL
;
1725 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1726 struct ust_reg_wait_queue wait_queue
= {
1730 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1732 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1733 goto error_testpoint
;
1736 health_code_update();
1738 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1740 DBG("[thread] Dispatch UST command started");
1742 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1743 health_code_update();
1745 /* Atomically prepare the queue futex */
1746 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1749 struct ust_app
*app
= NULL
;
1753 * Make sure we don't have node(s) that have hung up before receiving
1754 * the notify socket. This is to clean the list in order to avoid
1755 * memory leaks from notify socket that are never seen.
1757 sanitize_wait_queue(&wait_queue
);
1759 health_code_update();
1760 /* Dequeue command for registration */
1761 node
= cds_wfq_dequeue_blocking(&ust_cmd_queue
.queue
);
1763 DBG("Woken up but nothing in the UST command queue");
1764 /* Continue thread execution */
1768 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1770 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1771 " gid:%d sock:%d name:%s (version %d.%d)",
1772 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1773 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1774 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1775 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1777 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1778 wait_node
= zmalloc(sizeof(*wait_node
));
1780 PERROR("zmalloc wait_node dispatch");
1781 ret
= close(ust_cmd
->sock
);
1783 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1785 lttng_fd_put(LTTNG_FD_APPS
, 1);
1789 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1791 /* Create application object if socket is CMD. */
1792 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1794 if (!wait_node
->app
) {
1795 ret
= close(ust_cmd
->sock
);
1797 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1799 lttng_fd_put(LTTNG_FD_APPS
, 1);
1805 * Add application to the wait queue so we can set the notify
1806 * socket before putting this object in the global ht.
1808 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1813 * We have to continue here since we don't have the notify
1814 * socket and the application MUST be added to the hash table
1815 * only at that moment.
1820 * Look for the application in the local wait queue and set the
1821 * notify socket if found.
1823 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1824 &wait_queue
.head
, head
) {
1825 health_code_update();
1826 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1827 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1828 cds_list_del(&wait_node
->head
);
1830 app
= wait_node
->app
;
1832 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1838 * With no application at this stage the received socket is
1839 * basically useless so close it before we free the cmd data
1840 * structure for good.
1843 ret
= close(ust_cmd
->sock
);
1845 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1847 lttng_fd_put(LTTNG_FD_APPS
, 1);
1854 * @session_lock_list
1856 * Lock the global session list so from the register up to the
1857 * registration done message, no thread can see the application
1858 * and change its state.
1860 session_lock_list();
1864 * Add application to the global hash table. This needs to be
1865 * done before the update to the UST registry can locate the
1870 /* Set app version. This call will print an error if needed. */
1871 (void) ust_app_version(app
);
1873 /* Send notify socket through the notify pipe. */
1874 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1878 session_unlock_list();
1880 * No notify thread, stop the UST tracing. However, this is
1881 * not an internal error of the this thread thus setting
1882 * the health error code to a normal exit.
1889 * Update newly registered application with the tracing
1890 * registry info already enabled information.
1892 update_ust_app(app
->sock
);
1895 * Don't care about return value. Let the manage apps threads
1896 * handle app unregistration upon socket close.
1898 (void) ust_app_register_done(app
->sock
);
1901 * Even if the application socket has been closed, send the app
1902 * to the thread and unregistration will take place at that
1905 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1908 session_unlock_list();
1910 * No apps. thread, stop the UST tracing. However, this is
1911 * not an internal error of the this thread thus setting
1912 * the health error code to a normal exit.
1919 session_unlock_list();
1921 } while (node
!= NULL
);
1923 health_poll_entry();
1924 /* Futex wait on queue. Blocking call on futex() */
1925 futex_nto1_wait(&ust_cmd_queue
.futex
);
1928 /* Normal exit, no error */
1932 /* Clean up wait queue. */
1933 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1934 &wait_queue
.head
, head
) {
1935 cds_list_del(&wait_node
->head
);
1941 DBG("Dispatch thread dying");
1944 ERR("Health error occurred in %s", __func__
);
1946 health_unregister(health_sessiond
);
1951 * This thread manage application registration.
1953 static void *thread_registration_apps(void *data
)
1955 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1956 uint32_t revents
, nb_fd
;
1957 struct lttng_poll_event events
;
1959 * Get allocated in this thread, enqueued to a global queue, dequeued and
1960 * freed in the manage apps thread.
1962 struct ust_command
*ust_cmd
= NULL
;
1964 DBG("[thread] Manage application registration started");
1966 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
1968 if (testpoint(sessiond_thread_registration_apps
)) {
1969 goto error_testpoint
;
1972 ret
= lttcomm_listen_unix_sock(apps_sock
);
1978 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1979 * more will be added to this poll set.
1981 ret
= sessiond_set_thread_pollset(&events
, 2);
1983 goto error_create_poll
;
1986 /* Add the application registration socket */
1987 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1989 goto error_poll_add
;
1992 /* Notify all applications to register */
1993 ret
= notify_ust_apps(1);
1995 ERR("Failed to notify applications or create the wait shared memory.\n"
1996 "Execution continues but there might be problem for already\n"
1997 "running applications that wishes to register.");
2001 DBG("Accepting application registration");
2003 /* Inifinite blocking call, waiting for transmission */
2005 health_poll_entry();
2006 ret
= lttng_poll_wait(&events
, -1);
2010 * Restart interrupted system call.
2012 if (errno
== EINTR
) {
2020 for (i
= 0; i
< nb_fd
; i
++) {
2021 health_code_update();
2023 /* Fetch once the poll data */
2024 revents
= LTTNG_POLL_GETEV(&events
, i
);
2025 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2028 /* No activity for this FD (poll implementation). */
2032 /* Thread quit pipe has been closed. Killing thread. */
2033 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2039 /* Event on the registration socket */
2040 if (pollfd
== apps_sock
) {
2041 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2042 ERR("Register apps socket poll error");
2044 } else if (revents
& LPOLLIN
) {
2045 sock
= lttcomm_accept_unix_sock(apps_sock
);
2051 * Set socket timeout for both receiving and ending.
2052 * app_socket_timeout is in seconds, whereas
2053 * lttcomm_setsockopt_rcv_timeout and
2054 * lttcomm_setsockopt_snd_timeout expect msec as
2057 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2058 app_socket_timeout
* 1000);
2059 (void) lttcomm_setsockopt_snd_timeout(sock
,
2060 app_socket_timeout
* 1000);
2063 * Set the CLOEXEC flag. Return code is useless because
2064 * either way, the show must go on.
2066 (void) utils_set_fd_cloexec(sock
);
2068 /* Create UST registration command for enqueuing */
2069 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2070 if (ust_cmd
== NULL
) {
2071 PERROR("ust command zmalloc");
2076 * Using message-based transmissions to ensure we don't
2077 * have to deal with partially received messages.
2079 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2081 ERR("Exhausted file descriptors allowed for applications.");
2091 health_code_update();
2092 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2095 /* Close socket of the application. */
2100 lttng_fd_put(LTTNG_FD_APPS
, 1);
2104 health_code_update();
2106 ust_cmd
->sock
= sock
;
2109 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2110 " gid:%d sock:%d name:%s (version %d.%d)",
2111 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2112 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2113 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2114 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2117 * Lock free enqueue the registration request. The red pill
2118 * has been taken! This apps will be part of the *system*.
2120 cds_wfq_enqueue(&ust_cmd_queue
.queue
, &ust_cmd
->node
);
2123 * Wake the registration queue futex. Implicit memory
2124 * barrier with the exchange in cds_wfq_enqueue.
2126 futex_nto1_wake(&ust_cmd_queue
.futex
);
2134 /* Notify that the registration thread is gone */
2137 if (apps_sock
>= 0) {
2138 ret
= close(apps_sock
);
2148 lttng_fd_put(LTTNG_FD_APPS
, 1);
2150 unlink(apps_unix_sock_path
);
2153 lttng_poll_clean(&events
);
2157 DBG("UST Registration thread cleanup complete");
2160 ERR("Health error occurred in %s", __func__
);
2162 health_unregister(health_sessiond
);
2168 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2169 * exec or it will fails.
2171 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2174 struct timespec timeout
;
2176 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2177 consumer_data
->consumer_thread_is_ready
= 0;
2179 /* Setup pthread condition */
2180 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2183 PERROR("pthread_condattr_init consumer data");
2188 * Set the monotonic clock in order to make sure we DO NOT jump in time
2189 * between the clock_gettime() call and the timedwait call. See bug #324
2190 * for a more details and how we noticed it.
2192 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2195 PERROR("pthread_condattr_setclock consumer data");
2199 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2202 PERROR("pthread_cond_init consumer data");
2206 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2209 PERROR("pthread_create consumer");
2214 /* We are about to wait on a pthread condition */
2215 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2217 /* Get time for sem_timedwait absolute timeout */
2218 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2220 * Set the timeout for the condition timed wait even if the clock gettime
2221 * call fails since we might loop on that call and we want to avoid to
2222 * increment the timeout too many times.
2224 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2227 * The following loop COULD be skipped in some conditions so this is why we
2228 * set ret to 0 in order to make sure at least one round of the loop is
2234 * Loop until the condition is reached or when a timeout is reached. Note
2235 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2236 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2237 * possible. This loop does not take any chances and works with both of
2240 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2241 if (clock_ret
< 0) {
2242 PERROR("clock_gettime spawn consumer");
2243 /* Infinite wait for the consumerd thread to be ready */
2244 ret
= pthread_cond_wait(&consumer_data
->cond
,
2245 &consumer_data
->cond_mutex
);
2247 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2248 &consumer_data
->cond_mutex
, &timeout
);
2252 /* Release the pthread condition */
2253 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2257 if (ret
== ETIMEDOUT
) {
2261 * Call has timed out so we kill the kconsumerd_thread and return
2264 ERR("Condition timed out. The consumer thread was never ready."
2266 pth_ret
= pthread_cancel(consumer_data
->thread
);
2268 PERROR("pthread_cancel consumer thread");
2271 PERROR("pthread_cond_wait failed consumer thread");
2273 /* Caller is expecting a negative value on failure. */
2278 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2279 if (consumer_data
->pid
== 0) {
2280 ERR("Consumerd did not start");
2281 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2284 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2293 * Join consumer thread
2295 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2299 /* Consumer pid must be a real one. */
2300 if (consumer_data
->pid
> 0) {
2302 ret
= kill(consumer_data
->pid
, SIGTERM
);
2304 ERR("Error killing consumer daemon");
2307 return pthread_join(consumer_data
->thread
, &status
);
2314 * Fork and exec a consumer daemon (consumerd).
2316 * Return pid if successful else -1.
2318 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2322 const char *consumer_to_use
;
2323 const char *verbosity
;
2326 DBG("Spawning consumerd");
2333 if (opt_verbose_consumer
) {
2334 verbosity
= "--verbose";
2335 } else if (lttng_opt_quiet
) {
2336 verbosity
= "--quiet";
2341 switch (consumer_data
->type
) {
2342 case LTTNG_CONSUMER_KERNEL
:
2344 * Find out which consumerd to execute. We will first try the
2345 * 64-bit path, then the sessiond's installation directory, and
2346 * fallback on the 32-bit one,
2348 DBG3("Looking for a kernel consumer at these locations:");
2349 DBG3(" 1) %s", consumerd64_bin
);
2350 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2351 DBG3(" 3) %s", consumerd32_bin
);
2352 if (stat(consumerd64_bin
, &st
) == 0) {
2353 DBG3("Found location #1");
2354 consumer_to_use
= consumerd64_bin
;
2355 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2356 DBG3("Found location #2");
2357 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2358 } else if (stat(consumerd32_bin
, &st
) == 0) {
2359 DBG3("Found location #3");
2360 consumer_to_use
= consumerd32_bin
;
2362 DBG("Could not find any valid consumerd executable");
2366 DBG("Using kernel consumer at: %s", consumer_to_use
);
2367 ret
= execl(consumer_to_use
,
2368 "lttng-consumerd", verbosity
, "-k",
2369 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2370 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2371 "--group", tracing_group_name
,
2374 case LTTNG_CONSUMER64_UST
:
2376 char *tmpnew
= NULL
;
2378 if (consumerd64_libdir
[0] != '\0') {
2382 tmp
= getenv("LD_LIBRARY_PATH");
2386 tmplen
= strlen("LD_LIBRARY_PATH=")
2387 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2388 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2393 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2394 strcat(tmpnew
, consumerd64_libdir
);
2395 if (tmp
[0] != '\0') {
2396 strcat(tmpnew
, ":");
2397 strcat(tmpnew
, tmp
);
2399 ret
= putenv(tmpnew
);
2406 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2407 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2408 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2409 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2410 "--group", tracing_group_name
,
2412 if (consumerd64_libdir
[0] != '\0') {
2417 case LTTNG_CONSUMER32_UST
:
2419 char *tmpnew
= NULL
;
2421 if (consumerd32_libdir
[0] != '\0') {
2425 tmp
= getenv("LD_LIBRARY_PATH");
2429 tmplen
= strlen("LD_LIBRARY_PATH=")
2430 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2431 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2436 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2437 strcat(tmpnew
, consumerd32_libdir
);
2438 if (tmp
[0] != '\0') {
2439 strcat(tmpnew
, ":");
2440 strcat(tmpnew
, tmp
);
2442 ret
= putenv(tmpnew
);
2449 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2450 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2451 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2452 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2453 "--group", tracing_group_name
,
2455 if (consumerd32_libdir
[0] != '\0') {
2461 PERROR("unknown consumer type");
2465 PERROR("Consumer execl()");
2467 /* Reaching this point, we got a failure on our execl(). */
2469 } else if (pid
> 0) {
2472 PERROR("start consumer fork");
2480 * Spawn the consumerd daemon and session daemon thread.
2482 static int start_consumerd(struct consumer_data
*consumer_data
)
2487 * Set the listen() state on the socket since there is a possible race
2488 * between the exec() of the consumer daemon and this call if place in the
2489 * consumer thread. See bug #366 for more details.
2491 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2496 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2497 if (consumer_data
->pid
!= 0) {
2498 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2502 ret
= spawn_consumerd(consumer_data
);
2504 ERR("Spawning consumerd failed");
2505 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2509 /* Setting up the consumer_data pid */
2510 consumer_data
->pid
= ret
;
2511 DBG2("Consumer pid %d", consumer_data
->pid
);
2512 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2514 DBG2("Spawning consumer control thread");
2515 ret
= spawn_consumer_thread(consumer_data
);
2517 ERR("Fatal error spawning consumer control thread");
2525 /* Cleanup already created sockets on error. */
2526 if (consumer_data
->err_sock
>= 0) {
2529 err
= close(consumer_data
->err_sock
);
2531 PERROR("close consumer data error socket");
2538 * Setup necessary data for kernel tracer action.
2540 static int init_kernel_tracer(void)
2544 /* Modprobe lttng kernel modules */
2545 ret
= modprobe_lttng_control();
2550 /* Open debugfs lttng */
2551 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2552 if (kernel_tracer_fd
< 0) {
2553 DBG("Failed to open %s", module_proc_lttng
);
2558 /* Validate kernel version */
2559 ret
= kernel_validate_version(kernel_tracer_fd
);
2564 ret
= modprobe_lttng_data();
2569 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2573 modprobe_remove_lttng_control();
2574 ret
= close(kernel_tracer_fd
);
2578 kernel_tracer_fd
= -1;
2579 return LTTNG_ERR_KERN_VERSION
;
2582 ret
= close(kernel_tracer_fd
);
2588 modprobe_remove_lttng_control();
2591 WARN("No kernel tracer available");
2592 kernel_tracer_fd
= -1;
2594 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2596 return LTTNG_ERR_KERN_NA
;
2602 * Copy consumer output from the tracing session to the domain session. The
2603 * function also applies the right modification on a per domain basis for the
2604 * trace files destination directory.
2606 * Should *NOT* be called with RCU read-side lock held.
2608 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2611 const char *dir_name
;
2612 struct consumer_output
*consumer
;
2615 assert(session
->consumer
);
2618 case LTTNG_DOMAIN_KERNEL
:
2619 DBG3("Copying tracing session consumer output in kernel session");
2621 * XXX: We should audit the session creation and what this function
2622 * does "extra" in order to avoid a destroy since this function is used
2623 * in the domain session creation (kernel and ust) only. Same for UST
2626 if (session
->kernel_session
->consumer
) {
2627 consumer_destroy_output(session
->kernel_session
->consumer
);
2629 session
->kernel_session
->consumer
=
2630 consumer_copy_output(session
->consumer
);
2631 /* Ease our life a bit for the next part */
2632 consumer
= session
->kernel_session
->consumer
;
2633 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2635 case LTTNG_DOMAIN_JUL
:
2636 case LTTNG_DOMAIN_UST
:
2637 DBG3("Copying tracing session consumer output in UST session");
2638 if (session
->ust_session
->consumer
) {
2639 consumer_destroy_output(session
->ust_session
->consumer
);
2641 session
->ust_session
->consumer
=
2642 consumer_copy_output(session
->consumer
);
2643 /* Ease our life a bit for the next part */
2644 consumer
= session
->ust_session
->consumer
;
2645 dir_name
= DEFAULT_UST_TRACE_DIR
;
2648 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2652 /* Append correct directory to subdir */
2653 strncat(consumer
->subdir
, dir_name
,
2654 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2655 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2664 * Create an UST session and add it to the session ust list.
2666 * Should *NOT* be called with RCU read-side lock held.
2668 static int create_ust_session(struct ltt_session
*session
,
2669 struct lttng_domain
*domain
)
2672 struct ltt_ust_session
*lus
= NULL
;
2676 assert(session
->consumer
);
2678 switch (domain
->type
) {
2679 case LTTNG_DOMAIN_JUL
:
2680 case LTTNG_DOMAIN_UST
:
2683 ERR("Unknown UST domain on create session %d", domain
->type
);
2684 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2688 DBG("Creating UST session");
2690 lus
= trace_ust_create_session(session
->id
);
2692 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2696 lus
->uid
= session
->uid
;
2697 lus
->gid
= session
->gid
;
2698 lus
->output_traces
= session
->output_traces
;
2699 lus
->snapshot_mode
= session
->snapshot_mode
;
2700 lus
->live_timer_interval
= session
->live_timer
;
2701 session
->ust_session
= lus
;
2703 /* Copy session output to the newly created UST session */
2704 ret
= copy_session_consumer(domain
->type
, session
);
2705 if (ret
!= LTTNG_OK
) {
2713 session
->ust_session
= NULL
;
2718 * Create a kernel tracer session then create the default channel.
2720 static int create_kernel_session(struct ltt_session
*session
)
2724 DBG("Creating kernel session");
2726 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2728 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2732 /* Code flow safety */
2733 assert(session
->kernel_session
);
2735 /* Copy session output to the newly created Kernel session */
2736 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2737 if (ret
!= LTTNG_OK
) {
2741 /* Create directory(ies) on local filesystem. */
2742 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2743 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2744 ret
= run_as_mkdir_recursive(
2745 session
->kernel_session
->consumer
->dst
.trace_path
,
2746 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2748 if (ret
!= -EEXIST
) {
2749 ERR("Trace directory creation error");
2755 session
->kernel_session
->uid
= session
->uid
;
2756 session
->kernel_session
->gid
= session
->gid
;
2757 session
->kernel_session
->output_traces
= session
->output_traces
;
2758 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2763 trace_kernel_destroy_session(session
->kernel_session
);
2764 session
->kernel_session
= NULL
;
2769 * Count number of session permitted by uid/gid.
2771 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2774 struct ltt_session
*session
;
2776 DBG("Counting number of available session for UID %d GID %d",
2778 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2780 * Only list the sessions the user can control.
2782 if (!session_access_ok(session
, uid
, gid
)) {
2791 * Process the command requested by the lttng client within the command
2792 * context structure. This function make sure that the return structure (llm)
2793 * is set and ready for transmission before returning.
2795 * Return any error encountered or 0 for success.
2797 * "sock" is only used for special-case var. len data.
2799 * Should *NOT* be called with RCU read-side lock held.
2801 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2805 int need_tracing_session
= 1;
2808 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2812 switch (cmd_ctx
->lsm
->cmd_type
) {
2813 case LTTNG_CREATE_SESSION
:
2814 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2815 case LTTNG_CREATE_SESSION_LIVE
:
2816 case LTTNG_DESTROY_SESSION
:
2817 case LTTNG_LIST_SESSIONS
:
2818 case LTTNG_LIST_DOMAINS
:
2819 case LTTNG_START_TRACE
:
2820 case LTTNG_STOP_TRACE
:
2821 case LTTNG_DATA_PENDING
:
2822 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2823 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2824 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2825 case LTTNG_SNAPSHOT_RECORD
:
2826 case LTTNG_SAVE_SESSION
:
2833 if (opt_no_kernel
&& need_domain
2834 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2836 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2838 ret
= LTTNG_ERR_KERN_NA
;
2843 /* Deny register consumer if we already have a spawned consumer. */
2844 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2845 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2846 if (kconsumer_data
.pid
> 0) {
2847 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2848 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2851 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2855 * Check for command that don't needs to allocate a returned payload. We do
2856 * this here so we don't have to make the call for no payload at each
2859 switch(cmd_ctx
->lsm
->cmd_type
) {
2860 case LTTNG_LIST_SESSIONS
:
2861 case LTTNG_LIST_TRACEPOINTS
:
2862 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2863 case LTTNG_LIST_DOMAINS
:
2864 case LTTNG_LIST_CHANNELS
:
2865 case LTTNG_LIST_EVENTS
:
2868 /* Setup lttng message with no payload */
2869 ret
= setup_lttng_msg(cmd_ctx
, 0);
2871 /* This label does not try to unlock the session */
2872 goto init_setup_error
;
2876 /* Commands that DO NOT need a session. */
2877 switch (cmd_ctx
->lsm
->cmd_type
) {
2878 case LTTNG_CREATE_SESSION
:
2879 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2880 case LTTNG_CREATE_SESSION_LIVE
:
2881 case LTTNG_CALIBRATE
:
2882 case LTTNG_LIST_SESSIONS
:
2883 case LTTNG_LIST_TRACEPOINTS
:
2884 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2885 case LTTNG_SAVE_SESSION
:
2886 need_tracing_session
= 0;
2889 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2891 * We keep the session list lock across _all_ commands
2892 * for now, because the per-session lock does not
2893 * handle teardown properly.
2895 session_lock_list();
2896 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2897 if (cmd_ctx
->session
== NULL
) {
2898 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2901 /* Acquire lock for the session */
2902 session_lock(cmd_ctx
->session
);
2908 * Commands that need a valid session but should NOT create one if none
2909 * exists. Instead of creating one and destroying it when the command is
2910 * handled, process that right before so we save some round trip in useless
2913 switch (cmd_ctx
->lsm
->cmd_type
) {
2914 case LTTNG_DISABLE_CHANNEL
:
2915 case LTTNG_DISABLE_EVENT
:
2916 case LTTNG_DISABLE_ALL_EVENT
:
2917 switch (cmd_ctx
->lsm
->domain
.type
) {
2918 case LTTNG_DOMAIN_KERNEL
:
2919 if (!cmd_ctx
->session
->kernel_session
) {
2920 ret
= LTTNG_ERR_NO_CHANNEL
;
2924 case LTTNG_DOMAIN_JUL
:
2925 case LTTNG_DOMAIN_UST
:
2926 if (!cmd_ctx
->session
->ust_session
) {
2927 ret
= LTTNG_ERR_NO_CHANNEL
;
2932 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2944 * Check domain type for specific "pre-action".
2946 switch (cmd_ctx
->lsm
->domain
.type
) {
2947 case LTTNG_DOMAIN_KERNEL
:
2949 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2953 /* Kernel tracer check */
2954 if (kernel_tracer_fd
== -1) {
2955 /* Basically, load kernel tracer modules */
2956 ret
= init_kernel_tracer();
2962 /* Consumer is in an ERROR state. Report back to client */
2963 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2964 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2968 /* Need a session for kernel command */
2969 if (need_tracing_session
) {
2970 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2971 ret
= create_kernel_session(cmd_ctx
->session
);
2973 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2978 /* Start the kernel consumer daemon */
2979 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2980 if (kconsumer_data
.pid
== 0 &&
2981 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2982 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2983 ret
= start_consumerd(&kconsumer_data
);
2985 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2988 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2990 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2994 * The consumer was just spawned so we need to add the socket to
2995 * the consumer output of the session if exist.
2997 ret
= consumer_create_socket(&kconsumer_data
,
2998 cmd_ctx
->session
->kernel_session
->consumer
);
3005 case LTTNG_DOMAIN_JUL
:
3006 case LTTNG_DOMAIN_UST
:
3008 if (!ust_app_supported()) {
3009 ret
= LTTNG_ERR_NO_UST
;
3012 /* Consumer is in an ERROR state. Report back to client */
3013 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3014 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3018 if (need_tracing_session
) {
3019 /* Create UST session if none exist. */
3020 if (cmd_ctx
->session
->ust_session
== NULL
) {
3021 ret
= create_ust_session(cmd_ctx
->session
,
3022 &cmd_ctx
->lsm
->domain
);
3023 if (ret
!= LTTNG_OK
) {
3028 /* Start the UST consumer daemons */
3030 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3031 if (consumerd64_bin
[0] != '\0' &&
3032 ustconsumer64_data
.pid
== 0 &&
3033 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3034 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3035 ret
= start_consumerd(&ustconsumer64_data
);
3037 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3038 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3042 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3043 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3045 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3049 * Setup socket for consumer 64 bit. No need for atomic access
3050 * since it was set above and can ONLY be set in this thread.
3052 ret
= consumer_create_socket(&ustconsumer64_data
,
3053 cmd_ctx
->session
->ust_session
->consumer
);
3059 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3060 if (consumerd32_bin
[0] != '\0' &&
3061 ustconsumer32_data
.pid
== 0 &&
3062 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3063 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3064 ret
= start_consumerd(&ustconsumer32_data
);
3066 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3067 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3071 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3072 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3074 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3078 * Setup socket for consumer 64 bit. No need for atomic access
3079 * since it was set above and can ONLY be set in this thread.
3081 ret
= consumer_create_socket(&ustconsumer32_data
,
3082 cmd_ctx
->session
->ust_session
->consumer
);
3094 /* Validate consumer daemon state when start/stop trace command */
3095 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3096 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3097 switch (cmd_ctx
->lsm
->domain
.type
) {
3098 case LTTNG_DOMAIN_JUL
:
3099 case LTTNG_DOMAIN_UST
:
3100 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3101 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3105 case LTTNG_DOMAIN_KERNEL
:
3106 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3107 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3115 * Check that the UID or GID match that of the tracing session.
3116 * The root user can interact with all sessions.
3118 if (need_tracing_session
) {
3119 if (!session_access_ok(cmd_ctx
->session
,
3120 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3121 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3122 ret
= LTTNG_ERR_EPERM
;
3128 * Send relayd information to consumer as soon as we have a domain and a
3131 if (cmd_ctx
->session
&& need_domain
) {
3133 * Setup relayd if not done yet. If the relayd information was already
3134 * sent to the consumer, this call will gracefully return.
3136 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3137 if (ret
!= LTTNG_OK
) {
3142 /* Process by command type */
3143 switch (cmd_ctx
->lsm
->cmd_type
) {
3144 case LTTNG_ADD_CONTEXT
:
3146 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3147 cmd_ctx
->lsm
->u
.context
.channel_name
,
3148 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3151 case LTTNG_DISABLE_CHANNEL
:
3153 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3154 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3157 case LTTNG_DISABLE_EVENT
:
3159 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3160 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3161 cmd_ctx
->lsm
->u
.disable
.name
);
3164 case LTTNG_DISABLE_ALL_EVENT
:
3166 DBG("Disabling all events");
3168 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3169 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3172 case LTTNG_ENABLE_CHANNEL
:
3174 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3175 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3178 case LTTNG_ENABLE_EVENT
:
3180 struct lttng_event_exclusion
*exclusion
= NULL
;
3181 struct lttng_filter_bytecode
*bytecode
= NULL
;
3182 char *filter_expression
= NULL
;
3184 /* Handle exclusion events and receive it from the client. */
3185 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3186 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3188 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3189 (count
* LTTNG_SYMBOL_NAME_LEN
));
3191 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3195 DBG("Receiving var len exclusion event list from client ...");
3196 exclusion
->count
= count
;
3197 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3198 count
* LTTNG_SYMBOL_NAME_LEN
);
3200 DBG("Nothing recv() from client var len data... continuing");
3203 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3208 /* Get filter expression from client. */
3209 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3210 size_t expression_len
=
3211 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3213 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3214 ret
= LTTNG_ERR_FILTER_INVAL
;
3219 filter_expression
= zmalloc(expression_len
);
3220 if (!filter_expression
) {
3222 ret
= LTTNG_ERR_FILTER_NOMEM
;
3226 /* Receive var. len. data */
3227 DBG("Receiving var len filter's expression from client ...");
3228 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3231 DBG("Nothing recv() from client car len data... continuing");
3233 free(filter_expression
);
3235 ret
= LTTNG_ERR_FILTER_INVAL
;
3240 /* Handle filter and get bytecode from client. */
3241 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3242 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3244 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3245 ret
= LTTNG_ERR_FILTER_INVAL
;
3246 free(filter_expression
);
3251 bytecode
= zmalloc(bytecode_len
);
3253 free(filter_expression
);
3255 ret
= LTTNG_ERR_FILTER_NOMEM
;
3259 /* Receive var. len. data */
3260 DBG("Receiving var len filter's bytecode from client ...");
3261 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3263 DBG("Nothing recv() from client car len data... continuing");
3265 free(filter_expression
);
3268 ret
= LTTNG_ERR_FILTER_INVAL
;
3272 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3273 free(filter_expression
);
3276 ret
= LTTNG_ERR_FILTER_INVAL
;
3281 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3282 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3283 &cmd_ctx
->lsm
->u
.enable
.event
,
3284 filter_expression
, bytecode
, exclusion
,
3285 kernel_poll_pipe
[1]);
3288 case LTTNG_ENABLE_ALL_EVENT
:
3290 DBG("Enabling all events");
3292 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3293 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3294 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, NULL
,
3295 kernel_poll_pipe
[1]);
3298 case LTTNG_LIST_TRACEPOINTS
:
3300 struct lttng_event
*events
;
3303 session_lock_list();
3304 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3305 session_unlock_list();
3306 if (nb_events
< 0) {
3307 /* Return value is a negative lttng_error_code. */
3313 * Setup lttng message with payload size set to the event list size in
3314 * bytes and then copy list into the llm payload.
3316 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3322 /* Copy event list into message payload */
3323 memcpy(cmd_ctx
->llm
->payload
, events
,
3324 sizeof(struct lttng_event
) * nb_events
);
3331 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3333 struct lttng_event_field
*fields
;
3336 session_lock_list();
3337 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3339 session_unlock_list();
3340 if (nb_fields
< 0) {
3341 /* Return value is a negative lttng_error_code. */
3347 * Setup lttng message with payload size set to the event list size in
3348 * bytes and then copy list into the llm payload.
3350 ret
= setup_lttng_msg(cmd_ctx
,
3351 sizeof(struct lttng_event_field
) * nb_fields
);
3357 /* Copy event list into message payload */
3358 memcpy(cmd_ctx
->llm
->payload
, fields
,
3359 sizeof(struct lttng_event_field
) * nb_fields
);
3366 case LTTNG_SET_CONSUMER_URI
:
3369 struct lttng_uri
*uris
;
3371 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3372 len
= nb_uri
* sizeof(struct lttng_uri
);
3375 ret
= LTTNG_ERR_INVALID
;
3379 uris
= zmalloc(len
);
3381 ret
= LTTNG_ERR_FATAL
;
3385 /* Receive variable len data */
3386 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3387 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3389 DBG("No URIs received from client... continuing");
3391 ret
= LTTNG_ERR_SESSION_FAIL
;
3396 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3398 if (ret
!= LTTNG_OK
) {
3405 case LTTNG_START_TRACE
:
3407 ret
= cmd_start_trace(cmd_ctx
->session
);
3410 case LTTNG_STOP_TRACE
:
3412 ret
= cmd_stop_trace(cmd_ctx
->session
);
3415 case LTTNG_CREATE_SESSION
:
3418 struct lttng_uri
*uris
= NULL
;
3420 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3421 len
= nb_uri
* sizeof(struct lttng_uri
);
3424 uris
= zmalloc(len
);
3426 ret
= LTTNG_ERR_FATAL
;
3430 /* Receive variable len data */
3431 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3432 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3434 DBG("No URIs received from client... continuing");
3436 ret
= LTTNG_ERR_SESSION_FAIL
;
3441 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3442 DBG("Creating session with ONE network URI is a bad call");
3443 ret
= LTTNG_ERR_SESSION_FAIL
;
3449 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3450 &cmd_ctx
->creds
, 0);
3456 case LTTNG_DESTROY_SESSION
:
3458 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3460 /* Set session to NULL so we do not unlock it after free. */
3461 cmd_ctx
->session
= NULL
;
3464 case LTTNG_LIST_DOMAINS
:
3467 struct lttng_domain
*domains
;
3469 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3471 /* Return value is a negative lttng_error_code. */
3476 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3482 /* Copy event list into message payload */
3483 memcpy(cmd_ctx
->llm
->payload
, domains
,
3484 nb_dom
* sizeof(struct lttng_domain
));
3491 case LTTNG_LIST_CHANNELS
:
3494 struct lttng_channel
*channels
= NULL
;
3496 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3497 cmd_ctx
->session
, &channels
);
3499 /* Return value is a negative lttng_error_code. */
3504 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3510 /* Copy event list into message payload */
3511 memcpy(cmd_ctx
->llm
->payload
, channels
,
3512 nb_chan
* sizeof(struct lttng_channel
));
3519 case LTTNG_LIST_EVENTS
:
3522 struct lttng_event
*events
= NULL
;
3524 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3525 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3527 /* Return value is a negative lttng_error_code. */
3532 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3538 /* Copy event list into message payload */
3539 memcpy(cmd_ctx
->llm
->payload
, events
,
3540 nb_event
* sizeof(struct lttng_event
));
3547 case LTTNG_LIST_SESSIONS
:
3549 unsigned int nr_sessions
;
3551 session_lock_list();
3552 nr_sessions
= lttng_sessions_count(
3553 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3554 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3556 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3558 session_unlock_list();
3562 /* Filled the session array */
3563 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3564 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3565 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3567 session_unlock_list();
3572 case LTTNG_CALIBRATE
:
3574 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3575 &cmd_ctx
->lsm
->u
.calibrate
);
3578 case LTTNG_REGISTER_CONSUMER
:
3580 struct consumer_data
*cdata
;
3582 switch (cmd_ctx
->lsm
->domain
.type
) {
3583 case LTTNG_DOMAIN_KERNEL
:
3584 cdata
= &kconsumer_data
;
3587 ret
= LTTNG_ERR_UND
;
3591 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3592 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3595 case LTTNG_DATA_PENDING
:
3597 ret
= cmd_data_pending(cmd_ctx
->session
);
3600 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3602 struct lttcomm_lttng_output_id reply
;
3604 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3605 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3606 if (ret
!= LTTNG_OK
) {
3610 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3615 /* Copy output list into message payload */
3616 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3620 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3622 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3623 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3626 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3629 struct lttng_snapshot_output
*outputs
= NULL
;
3631 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3632 if (nb_output
< 0) {
3637 ret
= setup_lttng_msg(cmd_ctx
,
3638 nb_output
* sizeof(struct lttng_snapshot_output
));
3645 /* Copy output list into message payload */
3646 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3647 nb_output
* sizeof(struct lttng_snapshot_output
));
3654 case LTTNG_SNAPSHOT_RECORD
:
3656 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3657 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3658 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3661 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3664 struct lttng_uri
*uris
= NULL
;
3666 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3667 len
= nb_uri
* sizeof(struct lttng_uri
);
3670 uris
= zmalloc(len
);
3672 ret
= LTTNG_ERR_FATAL
;
3676 /* Receive variable len data */
3677 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3678 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3680 DBG("No URIs received from client... continuing");
3682 ret
= LTTNG_ERR_SESSION_FAIL
;
3687 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3688 DBG("Creating session with ONE network URI is a bad call");
3689 ret
= LTTNG_ERR_SESSION_FAIL
;
3695 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3696 nb_uri
, &cmd_ctx
->creds
);
3700 case LTTNG_CREATE_SESSION_LIVE
:
3703 struct lttng_uri
*uris
= NULL
;
3705 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3706 len
= nb_uri
* sizeof(struct lttng_uri
);
3709 uris
= zmalloc(len
);
3711 ret
= LTTNG_ERR_FATAL
;
3715 /* Receive variable len data */
3716 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3717 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3719 DBG("No URIs received from client... continuing");
3721 ret
= LTTNG_ERR_SESSION_FAIL
;
3726 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3727 DBG("Creating session with ONE network URI is a bad call");
3728 ret
= LTTNG_ERR_SESSION_FAIL
;
3734 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3735 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3739 case LTTNG_SAVE_SESSION
:
3741 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
3746 ret
= LTTNG_ERR_UND
;
3751 if (cmd_ctx
->llm
== NULL
) {
3752 DBG("Missing llm structure. Allocating one.");
3753 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3757 /* Set return code */
3758 cmd_ctx
->llm
->ret_code
= ret
;
3760 if (cmd_ctx
->session
) {
3761 session_unlock(cmd_ctx
->session
);
3763 if (need_tracing_session
) {
3764 session_unlock_list();
3771 * Thread managing health check socket.
3773 static void *thread_manage_health(void *data
)
3775 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3776 uint32_t revents
, nb_fd
;
3777 struct lttng_poll_event events
;
3778 struct health_comm_msg msg
;
3779 struct health_comm_reply reply
;
3781 DBG("[thread] Manage health check started");
3783 rcu_register_thread();
3785 /* We might hit an error path before this is created. */
3786 lttng_poll_init(&events
);
3788 /* Create unix socket */
3789 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3791 ERR("Unable to create health check Unix socket");
3797 /* lttng health client socket path permissions */
3798 ret
= chown(health_unix_sock_path
, 0,
3799 utils_get_group_id(tracing_group_name
));
3801 ERR("Unable to set group on %s", health_unix_sock_path
);
3807 ret
= chmod(health_unix_sock_path
,
3808 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3810 ERR("Unable to set permissions on %s", health_unix_sock_path
);
3818 * Set the CLOEXEC flag. Return code is useless because either way, the
3821 (void) utils_set_fd_cloexec(sock
);
3823 ret
= lttcomm_listen_unix_sock(sock
);
3829 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3830 * more will be added to this poll set.
3832 ret
= sessiond_set_thread_pollset(&events
, 2);
3837 /* Add the application registration socket */
3838 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3843 sessiond_notify_ready();
3846 DBG("Health check ready");
3848 /* Inifinite blocking call, waiting for transmission */
3850 ret
= lttng_poll_wait(&events
, -1);
3853 * Restart interrupted system call.
3855 if (errno
== EINTR
) {
3863 for (i
= 0; i
< nb_fd
; i
++) {
3864 /* Fetch once the poll data */
3865 revents
= LTTNG_POLL_GETEV(&events
, i
);
3866 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3869 /* No activity for this FD (poll implementation). */
3873 /* Thread quit pipe has been closed. Killing thread. */
3874 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3880 /* Event on the registration socket */
3881 if (pollfd
== sock
) {
3882 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3883 ERR("Health socket poll error");
3889 new_sock
= lttcomm_accept_unix_sock(sock
);
3895 * Set the CLOEXEC flag. Return code is useless because either way, the
3898 (void) utils_set_fd_cloexec(new_sock
);
3900 DBG("Receiving data from client for health...");
3901 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3903 DBG("Nothing recv() from client... continuing");
3904 ret
= close(new_sock
);
3912 rcu_thread_online();
3914 memset(&reply
, 0, sizeof(reply
));
3915 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
3917 * health_check_state returns 0 if health is
3920 if (!health_check_state(health_sessiond
, i
)) {
3921 reply
.ret_code
|= 1ULL << i
;
3925 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
3927 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3929 ERR("Failed to send health data back to client");
3932 /* End of transmission */
3933 ret
= close(new_sock
);
3943 ERR("Health error occurred in %s", __func__
);
3945 DBG("Health check thread dying");
3946 unlink(health_unix_sock_path
);
3954 lttng_poll_clean(&events
);
3956 rcu_unregister_thread();
3961 * This thread manage all clients request using the unix client socket for
3964 static void *thread_manage_clients(void *data
)
3966 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3968 uint32_t revents
, nb_fd
;
3969 struct command_ctx
*cmd_ctx
= NULL
;
3970 struct lttng_poll_event events
;
3972 DBG("[thread] Manage client started");
3974 rcu_register_thread();
3976 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
3978 health_code_update();
3980 ret
= lttcomm_listen_unix_sock(client_sock
);
3986 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3987 * more will be added to this poll set.
3989 ret
= sessiond_set_thread_pollset(&events
, 2);
3991 goto error_create_poll
;
3994 /* Add the application registration socket */
3995 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4000 sessiond_notify_ready();
4001 ret
= sem_post(&load_info
->message_thread_ready
);
4003 PERROR("sem_post message_thread_ready");
4007 /* This testpoint is after we signal readiness to the parent. */
4008 if (testpoint(sessiond_thread_manage_clients
)) {
4012 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4016 health_code_update();
4019 DBG("Accepting client command ...");
4021 /* Inifinite blocking call, waiting for transmission */
4023 health_poll_entry();
4024 ret
= lttng_poll_wait(&events
, -1);
4028 * Restart interrupted system call.
4030 if (errno
== EINTR
) {
4038 for (i
= 0; i
< nb_fd
; i
++) {
4039 /* Fetch once the poll data */
4040 revents
= LTTNG_POLL_GETEV(&events
, i
);
4041 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4043 health_code_update();
4046 /* No activity for this FD (poll implementation). */
4050 /* Thread quit pipe has been closed. Killing thread. */
4051 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4057 /* Event on the registration socket */
4058 if (pollfd
== client_sock
) {
4059 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4060 ERR("Client socket poll error");
4066 DBG("Wait for client response");
4068 health_code_update();
4070 sock
= lttcomm_accept_unix_sock(client_sock
);
4076 * Set the CLOEXEC flag. Return code is useless because either way, the
4079 (void) utils_set_fd_cloexec(sock
);
4081 /* Set socket option for credentials retrieval */
4082 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4087 /* Allocate context command to process the client request */
4088 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4089 if (cmd_ctx
== NULL
) {
4090 PERROR("zmalloc cmd_ctx");
4094 /* Allocate data buffer for reception */
4095 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4096 if (cmd_ctx
->lsm
== NULL
) {
4097 PERROR("zmalloc cmd_ctx->lsm");
4101 cmd_ctx
->llm
= NULL
;
4102 cmd_ctx
->session
= NULL
;
4104 health_code_update();
4107 * Data is received from the lttng client. The struct
4108 * lttcomm_session_msg (lsm) contains the command and data request of
4111 DBG("Receiving data from client ...");
4112 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4113 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4115 DBG("Nothing recv() from client... continuing");
4121 clean_command_ctx(&cmd_ctx
);
4125 health_code_update();
4127 // TODO: Validate cmd_ctx including sanity check for
4128 // security purpose.
4130 rcu_thread_online();
4132 * This function dispatch the work to the kernel or userspace tracer
4133 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4134 * informations for the client. The command context struct contains
4135 * everything this function may needs.
4137 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4138 rcu_thread_offline();
4146 * TODO: Inform client somehow of the fatal error. At
4147 * this point, ret < 0 means that a zmalloc failed
4148 * (ENOMEM). Error detected but still accept
4149 * command, unless a socket error has been
4152 clean_command_ctx(&cmd_ctx
);
4156 health_code_update();
4158 DBG("Sending response (size: %d, retcode: %s)",
4159 cmd_ctx
->lttng_msg_size
,
4160 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
4161 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4163 ERR("Failed to send data back to client");
4166 /* End of transmission */
4173 clean_command_ctx(&cmd_ctx
);
4175 health_code_update();
4187 lttng_poll_clean(&events
);
4188 clean_command_ctx(&cmd_ctx
);
4192 unlink(client_unix_sock_path
);
4193 if (client_sock
>= 0) {
4194 ret
= close(client_sock
);
4202 ERR("Health error occurred in %s", __func__
);
4205 health_unregister(health_sessiond
);
4207 DBG("Client thread dying");
4209 rcu_unregister_thread();
4215 * usage function on stderr
4217 static void usage(void)
4219 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4220 fprintf(stderr
, " -h, --help Display this usage.\n");
4221 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4222 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4223 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4224 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4225 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4226 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4227 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4228 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4229 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4230 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4231 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4232 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4233 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4234 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4235 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4236 fprintf(stderr
, " -V, --version Show version number.\n");
4237 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4238 fprintf(stderr
, " -q, --quiet No output at all.\n");
4239 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4240 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4241 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4242 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4243 fprintf(stderr
, " --jul-tcp-port JUL application registration TCP port\n");
4244 fprintf(stderr
, " -f --config PATH Load daemon configuration file\n");
4245 fprintf(stderr
, " -l --load PATH Load session configuration\n");
4246 fprintf(stderr
, " --kmod-probes Specify kernel module probes to load\n");
4250 * Take an option from the getopt output and set it in the right variable to be
4253 * Return 0 on success else a negative value.
4255 static int set_option(int opt
, const char *arg
, const char *optname
)
4259 if (arg
&& arg
[0] == '\0') {
4261 * This only happens if the value is read from daemon config
4262 * file. This means the option requires an argument and the
4263 * configuration file contains a line such as:
4272 fprintf(stderr
, "option %s", optname
);
4274 fprintf(stderr
, " with arg %s\n", arg
);
4278 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4281 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4291 * If the override option is set, the pointer points to a
4292 * *non* const thus freeing it even though the variable type is
4295 if (tracing_group_name_override
) {
4296 free((void *) tracing_group_name
);
4298 tracing_group_name
= strdup(arg
);
4299 if (!tracing_group_name
) {
4303 tracing_group_name_override
= 1;
4309 fprintf(stdout
, "%s\n", VERSION
);
4315 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4318 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4321 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4324 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4327 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4330 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4336 lttng_opt_quiet
= 1;
4339 /* Verbose level can increase using multiple -v */
4341 /* Value obtained from config file */
4342 lttng_opt_verbose
= config_parse_value(arg
);
4344 /* -v used on command line */
4345 lttng_opt_verbose
++;
4347 /* Clamp value to [0, 3] */
4348 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4349 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4353 opt_verbose_consumer
= config_parse_value(arg
);
4355 opt_verbose_consumer
+= 1;
4359 if (consumerd32_bin_override
) {
4360 free((void *) consumerd32_bin
);
4362 consumerd32_bin
= strdup(arg
);
4363 if (!consumerd32_bin
) {
4367 consumerd32_bin_override
= 1;
4370 if (consumerd32_libdir_override
) {
4371 free((void *) consumerd32_libdir
);
4373 consumerd32_libdir
= strdup(arg
);
4374 if (!consumerd32_libdir
) {
4378 consumerd32_libdir_override
= 1;
4381 if (consumerd64_bin_override
) {
4382 free((void *) consumerd64_bin
);
4384 consumerd64_bin
= strdup(arg
);
4385 if (!consumerd64_bin
) {
4389 consumerd64_bin_override
= 1;
4392 if (consumerd64_libdir_override
) {
4393 free((void *) consumerd64_libdir
);
4395 consumerd64_libdir
= strdup(arg
);
4396 if (!consumerd64_libdir
) {
4400 consumerd64_libdir_override
= 1;
4404 opt_pidfile
= strdup(arg
);
4410 case 'J': /* JUL TCP port. */
4415 v
= strtoul(arg
, NULL
, 0);
4416 if (errno
!= 0 || !isdigit(arg
[0])) {
4417 ERR("Wrong value in --jul-tcp-port parameter: %s", arg
);
4420 if (v
== 0 || v
>= 65535) {
4421 ERR("Port overflow in --jul-tcp-port parameter: %s", arg
);
4424 jul_tcp_port
= (uint32_t) v
;
4425 DBG3("JUL TCP port set to non default: %u", jul_tcp_port
);
4429 free(opt_load_session_path
);
4430 opt_load_session_path
= strdup(arg
);
4431 if (!opt_load_session_path
) {
4436 case 'P': /* probe modules list */
4437 free(kmod_probes_list
);
4438 kmod_probes_list
= strdup(arg
);
4439 if (!kmod_probes_list
) {
4445 /* This is handled in set_options() thus silent break. */
4448 /* Unknown option or other error.
4449 * Error is printed by getopt, just return */
4454 if (ret
== -EINVAL
) {
4455 const char *opt_name
= "unknown";
4458 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
4460 if (opt
== long_options
[i
].val
) {
4461 opt_name
= long_options
[i
].name
;
4466 WARN("Invalid argument provided for option \"%s\", using default value.",
4474 * config_entry_handler_cb used to handle options read from a config file.
4475 * See config_entry_handler_cb comment in common/config/config.h for the
4476 * return value conventions.
4478 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4482 if (!entry
|| !entry
->name
|| !entry
->value
) {
4487 /* Check if the option is to be ignored */
4488 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4489 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4494 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4497 /* Ignore if not fully matched. */
4498 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4503 * If the option takes no argument on the command line, we have to
4504 * check if the value is "true". We support non-zero numeric values,
4507 if (!long_options
[i
].has_arg
) {
4508 ret
= config_parse_value(entry
->value
);
4511 WARN("Invalid configuration value \"%s\" for option %s",
4512 entry
->value
, entry
->name
);
4514 /* False, skip boolean config option. */
4519 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4523 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4530 * daemon configuration loading and argument parsing
4532 static int set_options(int argc
, char **argv
)
4534 int ret
= 0, c
= 0, option_index
= 0;
4535 int orig_optopt
= optopt
, orig_optind
= optind
;
4537 const char *config_path
= NULL
;
4539 optstring
= utils_generate_optstring(long_options
,
4540 sizeof(long_options
) / sizeof(struct option
));
4546 /* Check for the --config option */
4547 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4548 &option_index
)) != -1) {
4552 } else if (c
!= 'f') {
4553 /* if not equal to --config option. */
4557 config_path
= utils_expand_path(optarg
);
4559 ERR("Failed to resolve path: %s", optarg
);
4563 ret
= config_get_section_entries(config_path
, config_section_name
,
4564 config_entry_handler
, NULL
);
4567 ERR("Invalid configuration option at line %i", ret
);
4573 /* Reset getopt's global state */
4574 optopt
= orig_optopt
;
4575 optind
= orig_optind
;
4577 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
4582 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
4594 * Creates the two needed socket by the daemon.
4595 * apps_sock - The communication socket for all UST apps.
4596 * client_sock - The communication of the cli tool (lttng).
4598 static int init_daemon_socket(void)
4603 old_umask
= umask(0);
4605 /* Create client tool unix socket */
4606 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4607 if (client_sock
< 0) {
4608 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4613 /* Set the cloexec flag */
4614 ret
= utils_set_fd_cloexec(client_sock
);
4616 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4617 "Continuing but note that the consumer daemon will have a "
4618 "reference to this socket on exec()", client_sock
);
4621 /* File permission MUST be 660 */
4622 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4624 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4629 /* Create the application unix socket */
4630 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4631 if (apps_sock
< 0) {
4632 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4637 /* Set the cloexec flag */
4638 ret
= utils_set_fd_cloexec(apps_sock
);
4640 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4641 "Continuing but note that the consumer daemon will have a "
4642 "reference to this socket on exec()", apps_sock
);
4645 /* File permission MUST be 666 */
4646 ret
= chmod(apps_unix_sock_path
,
4647 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4649 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4654 DBG3("Session daemon client socket %d and application socket %d created",
4655 client_sock
, apps_sock
);
4663 * Check if the global socket is available, and if a daemon is answering at the
4664 * other side. If yes, error is returned.
4666 static int check_existing_daemon(void)
4668 /* Is there anybody out there ? */
4669 if (lttng_session_daemon_alive()) {
4677 * Set the tracing group gid onto the client socket.
4679 * Race window between mkdir and chown is OK because we are going from more
4680 * permissive (root.root) to less permissive (root.tracing).
4682 static int set_permissions(char *rundir
)
4687 gid
= utils_get_group_id(tracing_group_name
);
4689 /* Set lttng run dir */
4690 ret
= chown(rundir
, 0, gid
);
4692 ERR("Unable to set group on %s", rundir
);
4697 * Ensure all applications and tracing group can search the run
4698 * dir. Allow everyone to read the directory, since it does not
4699 * buy us anything to hide its content.
4701 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
4703 ERR("Unable to set permissions on %s", rundir
);
4707 /* lttng client socket path */
4708 ret
= chown(client_unix_sock_path
, 0, gid
);
4710 ERR("Unable to set group on %s", client_unix_sock_path
);
4714 /* kconsumer error socket path */
4715 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
4717 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4721 /* 64-bit ustconsumer error socket path */
4722 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
4724 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4728 /* 32-bit ustconsumer compat32 error socket path */
4729 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
4731 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4735 DBG("All permissions are set");
4741 * Create the lttng run directory needed for all global sockets and pipe.
4743 static int create_lttng_rundir(const char *rundir
)
4747 DBG3("Creating LTTng run directory: %s", rundir
);
4749 ret
= mkdir(rundir
, S_IRWXU
);
4751 if (errno
!= EEXIST
) {
4752 ERR("Unable to create %s", rundir
);
4764 * Setup sockets and directory needed by the kconsumerd communication with the
4767 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4771 char path
[PATH_MAX
];
4773 switch (consumer_data
->type
) {
4774 case LTTNG_CONSUMER_KERNEL
:
4775 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4777 case LTTNG_CONSUMER64_UST
:
4778 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4780 case LTTNG_CONSUMER32_UST
:
4781 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4784 ERR("Consumer type unknown");
4789 DBG2("Creating consumer directory: %s", path
);
4791 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
4793 if (errno
!= EEXIST
) {
4795 ERR("Failed to create %s", path
);
4801 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
4803 ERR("Unable to set group on %s", path
);
4809 /* Create the kconsumerd error unix socket */
4810 consumer_data
->err_sock
=
4811 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4812 if (consumer_data
->err_sock
< 0) {
4813 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4819 * Set the CLOEXEC flag. Return code is useless because either way, the
4822 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4824 PERROR("utils_set_fd_cloexec");
4825 /* continue anyway */
4828 /* File permission MUST be 660 */
4829 ret
= chmod(consumer_data
->err_unix_sock_path
,
4830 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4832 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4842 * Signal handler for the daemon
4844 * Simply stop all worker threads, leaving main() return gracefully after
4845 * joining all threads and calling cleanup().
4847 static void sighandler(int sig
)
4851 DBG("SIGPIPE caught");
4854 DBG("SIGINT caught");
4858 DBG("SIGTERM caught");
4862 CMM_STORE_SHARED(recv_child_signal
, 1);
4870 * Setup signal handler for :
4871 * SIGINT, SIGTERM, SIGPIPE
4873 static int set_signal_handler(void)
4876 struct sigaction sa
;
4879 if ((ret
= sigemptyset(&sigset
)) < 0) {
4880 PERROR("sigemptyset");
4884 sa
.sa_handler
= sighandler
;
4885 sa
.sa_mask
= sigset
;
4887 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4888 PERROR("sigaction");
4892 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4893 PERROR("sigaction");
4897 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4898 PERROR("sigaction");
4902 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
4903 PERROR("sigaction");
4907 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4913 * Set open files limit to unlimited. This daemon can open a large number of
4914 * file descriptors in order to consumer multiple kernel traces.
4916 static void set_ulimit(void)
4921 /* The kernel does not allowed an infinite limit for open files */
4922 lim
.rlim_cur
= 65535;
4923 lim
.rlim_max
= 65535;
4925 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4927 PERROR("failed to set open files limit");
4932 * Write pidfile using the rundir and opt_pidfile.
4934 static void write_pidfile(void)
4937 char pidfile_path
[PATH_MAX
];
4942 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4944 /* Build pidfile path from rundir and opt_pidfile. */
4945 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4946 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4948 PERROR("snprintf pidfile path");
4954 * Create pid file in rundir. Return value is of no importance. The
4955 * execution will continue even though we are not able to write the file.
4957 (void) utils_create_pid_file(getpid(), pidfile_path
);
4964 * Create lockfile using the rundir and return its fd.
4966 static int create_lockfile(void)
4969 char lockfile_path
[PATH_MAX
];
4971 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
4976 ret
= utils_create_lock_file(lockfile_path
);
4982 * Write JUL TCP port using the rundir.
4984 static void write_julport(void)
4987 char path
[PATH_MAX
];
4991 ret
= snprintf(path
, sizeof(path
), "%s/"
4992 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
, rundir
);
4994 PERROR("snprintf julport path");
4999 * Create TCP JUL port file in rundir. Return value is of no importance.
5000 * The execution will continue even though we are not able to write the
5003 (void) utils_create_pid_file(jul_tcp_port
, path
);
5010 * Start the load session thread and dettach from it so the main thread can
5011 * continue. This does not return a value since whatever the outcome, the main
5012 * thread will continue.
5014 static void start_load_session_thread(void)
5018 /* Create session loading thread. */
5019 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
5022 PERROR("pthread_create load_session_thread");
5026 ret
= pthread_detach(load_session_thread
);
5028 PERROR("pthread_detach load_session_thread");
5031 /* Everything went well so don't cleanup anything. */
5034 /* The cleanup() function will destroy the load_info data. */
5041 int main(int argc
, char **argv
)
5045 const char *home_path
, *env_app_timeout
;
5047 init_kernel_workarounds();
5049 rcu_register_thread();
5051 if ((ret
= set_signal_handler()) < 0) {
5055 setup_consumerd_path();
5057 page_size
= sysconf(_SC_PAGESIZE
);
5058 if (page_size
< 0) {
5059 PERROR("sysconf _SC_PAGESIZE");
5060 page_size
= LONG_MAX
;
5061 WARN("Fallback page size to %ld", page_size
);
5064 /* Parse arguments and load the daemon configuration file */
5066 if ((ret
= set_options(argc
, argv
)) < 0) {
5071 if (opt_daemon
|| opt_background
) {
5074 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5081 * We are in the child. Make sure all other file descriptors are
5082 * closed, in case we are called with more opened file descriptors than
5083 * the standard ones.
5085 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5090 /* Create thread quit pipe */
5091 if ((ret
= init_thread_quit_pipe()) < 0) {
5095 /* Check if daemon is UID = 0 */
5096 is_root
= !getuid();
5099 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5105 /* Create global run dir with root access */
5106 ret
= create_lttng_rundir(rundir
);
5111 if (strlen(apps_unix_sock_path
) == 0) {
5112 snprintf(apps_unix_sock_path
, PATH_MAX
,
5113 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5116 if (strlen(client_unix_sock_path
) == 0) {
5117 snprintf(client_unix_sock_path
, PATH_MAX
,
5118 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5121 /* Set global SHM for ust */
5122 if (strlen(wait_shm_path
) == 0) {
5123 snprintf(wait_shm_path
, PATH_MAX
,
5124 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5127 if (strlen(health_unix_sock_path
) == 0) {
5128 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
5129 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5132 /* Setup kernel consumerd path */
5133 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5134 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5135 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5136 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5138 DBG2("Kernel consumer err path: %s",
5139 kconsumer_data
.err_unix_sock_path
);
5140 DBG2("Kernel consumer cmd path: %s",
5141 kconsumer_data
.cmd_unix_sock_path
);
5143 home_path
= utils_get_home_dir();
5144 if (home_path
== NULL
) {
5145 /* TODO: Add --socket PATH option */
5146 ERR("Can't get HOME directory for sockets creation.");
5152 * Create rundir from home path. This will create something like
5155 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5161 ret
= create_lttng_rundir(rundir
);
5166 if (strlen(apps_unix_sock_path
) == 0) {
5167 snprintf(apps_unix_sock_path
, PATH_MAX
,
5168 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
5171 /* Set the cli tool unix socket path */
5172 if (strlen(client_unix_sock_path
) == 0) {
5173 snprintf(client_unix_sock_path
, PATH_MAX
,
5174 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
5177 /* Set global SHM for ust */
5178 if (strlen(wait_shm_path
) == 0) {
5179 snprintf(wait_shm_path
, PATH_MAX
,
5180 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
5183 /* Set health check Unix path */
5184 if (strlen(health_unix_sock_path
) == 0) {
5185 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
5186 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
5190 lockfile_fd
= create_lockfile();
5191 if (lockfile_fd
< 0) {
5195 /* Set consumer initial state */
5196 kernel_consumerd_state
= CONSUMER_STOPPED
;
5197 ust_consumerd_state
= CONSUMER_STOPPED
;
5199 DBG("Client socket path %s", client_unix_sock_path
);
5200 DBG("Application socket path %s", apps_unix_sock_path
);
5201 DBG("Application wait path %s", wait_shm_path
);
5202 DBG("LTTng run directory path: %s", rundir
);
5204 /* 32 bits consumerd path setup */
5205 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5206 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5207 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5208 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5210 DBG2("UST consumer 32 bits err path: %s",
5211 ustconsumer32_data
.err_unix_sock_path
);
5212 DBG2("UST consumer 32 bits cmd path: %s",
5213 ustconsumer32_data
.cmd_unix_sock_path
);
5215 /* 64 bits consumerd path setup */
5216 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5217 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5218 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5219 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5221 DBG2("UST consumer 64 bits err path: %s",
5222 ustconsumer64_data
.err_unix_sock_path
);
5223 DBG2("UST consumer 64 bits cmd path: %s",
5224 ustconsumer64_data
.cmd_unix_sock_path
);
5227 * See if daemon already exist.
5229 if ((ret
= check_existing_daemon()) < 0) {
5230 ERR("Already running daemon.\n");
5232 * We do not goto exit because we must not cleanup()
5233 * because a daemon is already running.
5239 * Init UST app hash table. Alloc hash table before this point since
5240 * cleanup() can get called after that point.
5244 /* Initialize JUL domain subsystem. */
5245 if ((ret
= jul_init()) < 0) {
5246 /* ENOMEM at this point. */
5250 /* After this point, we can safely call cleanup() with "goto exit" */
5253 * These actions must be executed as root. We do that *after* setting up
5254 * the sockets path because we MUST make the check for another daemon using
5255 * those paths *before* trying to set the kernel consumer sockets and init
5259 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
5264 /* Setup kernel tracer */
5265 if (!opt_no_kernel
) {
5266 init_kernel_tracer();
5269 /* Set ulimit for open files */
5272 /* init lttng_fd tracking must be done after set_ulimit. */
5275 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
5280 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
5285 /* Setup the needed unix socket */
5286 if ((ret
= init_daemon_socket()) < 0) {
5290 /* Set credentials to socket */
5291 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
5295 /* Get parent pid if -S, --sig-parent is specified. */
5296 if (opt_sig_parent
) {
5300 /* Setup the kernel pipe for waking up the kernel thread */
5301 if (is_root
&& !opt_no_kernel
) {
5302 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
5307 /* Setup the thread ht_cleanup communication pipe. */
5308 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
5312 /* Setup the thread apps communication pipe. */
5313 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
5317 /* Setup the thread apps notify communication pipe. */
5318 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
5322 /* Initialize global buffer per UID and PID registry. */
5323 buffer_reg_init_uid_registry();
5324 buffer_reg_init_pid_registry();
5326 /* Init UST command queue. */
5327 cds_wfq_init(&ust_cmd_queue
.queue
);
5330 * Get session list pointer. This pointer MUST NOT be free(). This list is
5331 * statically declared in session.c
5333 session_list_ptr
= session_get_list();
5335 /* Set up max poll set size */
5336 lttng_poll_set_max_size();
5340 /* Check for the application socket timeout env variable. */
5341 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5342 if (env_app_timeout
) {
5343 app_socket_timeout
= atoi(env_app_timeout
);
5345 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5351 /* Initialize communication library */
5353 /* This is to get the TCP timeout value. */
5354 lttcomm_inet_init();
5356 if (load_session_init_data(&load_info
) < 0) {
5359 load_info
->path
= opt_load_session_path
;
5362 * Initialize the health check subsystem. This call should set the
5363 * appropriate time values.
5365 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5366 if (!health_sessiond
) {
5367 PERROR("health_app_create error");
5368 goto exit_health_sessiond_cleanup
;
5371 /* Create thread to clean up RCU hash tables */
5372 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5373 thread_ht_cleanup
, (void *) NULL
);
5375 PERROR("pthread_create ht_cleanup");
5376 goto exit_ht_cleanup
;
5379 /* Create health-check thread */
5380 ret
= pthread_create(&health_thread
, NULL
,
5381 thread_manage_health
, (void *) NULL
);
5383 PERROR("pthread_create health");
5387 /* Create thread to manage the client socket */
5388 ret
= pthread_create(&client_thread
, NULL
,
5389 thread_manage_clients
, (void *) NULL
);
5391 PERROR("pthread_create clients");
5395 /* Create thread to dispatch registration */
5396 ret
= pthread_create(&dispatch_thread
, NULL
,
5397 thread_dispatch_ust_registration
, (void *) NULL
);
5399 PERROR("pthread_create dispatch");
5403 /* Create thread to manage application registration. */
5404 ret
= pthread_create(®_apps_thread
, NULL
,
5405 thread_registration_apps
, (void *) NULL
);
5407 PERROR("pthread_create registration");
5411 /* Create thread to manage application socket */
5412 ret
= pthread_create(&apps_thread
, NULL
,
5413 thread_manage_apps
, (void *) NULL
);
5415 PERROR("pthread_create apps");
5419 /* Create thread to manage application notify socket */
5420 ret
= pthread_create(&apps_notify_thread
, NULL
,
5421 ust_thread_manage_notify
, (void *) NULL
);
5423 PERROR("pthread_create notify");
5424 goto exit_apps_notify
;
5427 /* Create JUL registration thread. */
5428 ret
= pthread_create(&jul_reg_thread
, NULL
,
5429 jul_thread_manage_registration
, (void *) NULL
);
5431 PERROR("pthread_create JUL");
5435 /* Don't start this thread if kernel tracing is not requested nor root */
5436 if (is_root
&& !opt_no_kernel
) {
5437 /* Create kernel thread to manage kernel event */
5438 ret
= pthread_create(&kernel_thread
, NULL
,
5439 thread_manage_kernel
, (void *) NULL
);
5441 PERROR("pthread_create kernel");
5446 /* Load possible session(s). */
5447 start_load_session_thread();
5449 if (is_root
&& !opt_no_kernel
) {
5450 ret
= pthread_join(kernel_thread
, &status
);
5452 PERROR("pthread_join");
5453 goto error
; /* join error, exit without cleanup */
5458 ret
= pthread_join(jul_reg_thread
, &status
);
5460 PERROR("pthread_join JUL");
5461 goto error
; /* join error, exit without cleanup */
5465 ret
= pthread_join(apps_notify_thread
, &status
);
5467 PERROR("pthread_join apps notify");
5468 goto error
; /* join error, exit without cleanup */
5472 ret
= pthread_join(apps_thread
, &status
);
5474 PERROR("pthread_join apps");
5475 goto error
; /* join error, exit without cleanup */
5480 ret
= pthread_join(reg_apps_thread
, &status
);
5482 PERROR("pthread_join");
5483 goto error
; /* join error, exit without cleanup */
5487 ret
= pthread_join(dispatch_thread
, &status
);
5489 PERROR("pthread_join");
5490 goto error
; /* join error, exit without cleanup */
5494 ret
= pthread_join(client_thread
, &status
);
5496 PERROR("pthread_join");
5497 goto error
; /* join error, exit without cleanup */
5500 ret
= join_consumer_thread(&kconsumer_data
);
5502 PERROR("join_consumer");
5503 goto error
; /* join error, exit without cleanup */
5506 ret
= join_consumer_thread(&ustconsumer32_data
);
5508 PERROR("join_consumer ust32");
5509 goto error
; /* join error, exit without cleanup */
5512 ret
= join_consumer_thread(&ustconsumer64_data
);
5514 PERROR("join_consumer ust64");
5515 goto error
; /* join error, exit without cleanup */
5519 ret
= pthread_join(health_thread
, &status
);
5521 PERROR("pthread_join health thread");
5522 goto error
; /* join error, exit without cleanup */
5526 ret
= pthread_join(ht_cleanup_thread
, &status
);
5528 PERROR("pthread_join ht cleanup thread");
5529 goto error
; /* join error, exit without cleanup */
5532 health_app_destroy(health_sessiond
);
5533 exit_health_sessiond_cleanup
:
5536 * cleanup() is called when no other thread is running.
5538 rcu_thread_online();
5540 rcu_thread_offline();
5541 rcu_unregister_thread();