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 on %d fds", LTTNG_POLL_GETNB(&events
));
1004 /* Poll infinite value of time */
1006 health_poll_entry();
1007 ret
= lttng_poll_wait(&events
, -1);
1011 * Restart interrupted system call.
1013 if (errno
== EINTR
) {
1017 } else if (ret
== 0) {
1018 /* Should not happen since timeout is infinite */
1019 ERR("Return value of poll is 0 with an infinite timeout.\n"
1020 "This should not have happened! Continuing...");
1026 for (i
= 0; i
< nb_fd
; i
++) {
1027 /* Fetch once the poll data */
1028 revents
= LTTNG_POLL_GETEV(&events
, i
);
1029 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1031 health_code_update();
1033 /* Thread quit pipe has been closed. Killing thread. */
1034 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1040 /* Check for data on kernel pipe */
1041 if (pollfd
== kernel_poll_pipe
[0] && (revents
& LPOLLIN
)) {
1042 (void) lttng_read(kernel_poll_pipe
[0],
1045 * Ret value is useless here, if this pipe gets any actions an
1046 * update is required anyway.
1048 update_poll_flag
= 1;
1052 * New CPU detected by the kernel. Adding kernel stream to
1053 * kernel session and updating the kernel consumer
1055 if (revents
& LPOLLIN
) {
1056 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1062 * TODO: We might want to handle the LPOLLERR | LPOLLHUP
1063 * and unregister kernel stream at this point.
1072 lttng_poll_clean(&events
);
1075 utils_close_pipe(kernel_poll_pipe
);
1076 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1079 ERR("Health error occurred in %s", __func__
);
1080 WARN("Kernel thread died unexpectedly. "
1081 "Kernel tracing can continue but CPU hotplug is disabled.");
1083 health_unregister(health_sessiond
);
1084 DBG("Kernel thread dying");
1089 * Signal pthread condition of the consumer data that the thread.
1091 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1093 pthread_mutex_lock(&data
->cond_mutex
);
1096 * The state is set before signaling. It can be any value, it's the waiter
1097 * job to correctly interpret this condition variable associated to the
1098 * consumer pthread_cond.
1100 * A value of 0 means that the corresponding thread of the consumer data
1101 * was not started. 1 indicates that the thread has started and is ready
1102 * for action. A negative value means that there was an error during the
1105 data
->consumer_thread_is_ready
= state
;
1106 (void) pthread_cond_signal(&data
->cond
);
1108 pthread_mutex_unlock(&data
->cond_mutex
);
1112 * This thread manage the consumer error sent back to the session daemon.
1114 static void *thread_manage_consumer(void *data
)
1116 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1117 uint32_t revents
, nb_fd
;
1118 enum lttcomm_return_code code
;
1119 struct lttng_poll_event events
;
1120 struct consumer_data
*consumer_data
= data
;
1122 DBG("[thread] Manage consumer started");
1124 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1126 health_code_update();
1129 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1130 * metadata_sock. Nothing more will be added to this poll set.
1132 ret
= sessiond_set_thread_pollset(&events
, 3);
1138 * The error socket here is already in a listening state which was done
1139 * just before spawning this thread to avoid a race between the consumer
1140 * daemon exec trying to connect and the listen() call.
1142 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1147 health_code_update();
1149 /* Infinite blocking call, waiting for transmission */
1151 health_poll_entry();
1153 if (testpoint(sessiond_thread_manage_consumer
)) {
1157 ret
= lttng_poll_wait(&events
, -1);
1161 * Restart interrupted system call.
1163 if (errno
== EINTR
) {
1171 for (i
= 0; i
< nb_fd
; i
++) {
1172 /* Fetch once the poll data */
1173 revents
= LTTNG_POLL_GETEV(&events
, i
);
1174 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1176 health_code_update();
1178 /* Thread quit pipe has been closed. Killing thread. */
1179 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1185 /* Event on the registration socket */
1186 if (pollfd
== consumer_data
->err_sock
) {
1187 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1188 ERR("consumer err socket poll error");
1194 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1200 * Set the CLOEXEC flag. Return code is useless because either way, the
1203 (void) utils_set_fd_cloexec(sock
);
1205 health_code_update();
1207 DBG2("Receiving code from consumer err_sock");
1209 /* Getting status code from kconsumerd */
1210 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1211 sizeof(enum lttcomm_return_code
));
1216 health_code_update();
1217 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1218 /* Connect both socket, command and metadata. */
1219 consumer_data
->cmd_sock
=
1220 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1221 consumer_data
->metadata_fd
=
1222 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1223 if (consumer_data
->cmd_sock
< 0
1224 || consumer_data
->metadata_fd
< 0) {
1225 PERROR("consumer connect cmd socket");
1226 /* On error, signal condition and quit. */
1227 signal_consumer_condition(consumer_data
, -1);
1230 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1231 /* Create metadata socket lock. */
1232 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1233 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1234 PERROR("zmalloc pthread mutex");
1238 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1240 signal_consumer_condition(consumer_data
, 1);
1241 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1242 DBG("Consumer metadata socket ready (fd: %d)",
1243 consumer_data
->metadata_fd
);
1245 ERR("consumer error when waiting for SOCK_READY : %s",
1246 lttcomm_get_readable_code(-code
));
1250 /* Remove the consumerd error sock since we've established a connexion */
1251 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1256 /* Add new accepted error socket. */
1257 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1262 /* Add metadata socket that is successfully connected. */
1263 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1264 LPOLLIN
| LPOLLRDHUP
);
1269 health_code_update();
1271 /* Infinite blocking call, waiting for transmission */
1274 health_poll_entry();
1275 ret
= lttng_poll_wait(&events
, -1);
1279 * Restart interrupted system call.
1281 if (errno
== EINTR
) {
1289 for (i
= 0; i
< nb_fd
; i
++) {
1290 /* Fetch once the poll data */
1291 revents
= LTTNG_POLL_GETEV(&events
, i
);
1292 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1294 health_code_update();
1296 /* Thread quit pipe has been closed. Killing thread. */
1297 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1303 if (pollfd
== sock
) {
1304 /* Event on the consumerd socket */
1305 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1306 ERR("consumer err socket second poll error");
1309 health_code_update();
1310 /* Wait for any kconsumerd error */
1311 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1312 sizeof(enum lttcomm_return_code
));
1314 ERR("consumer closed the command socket");
1318 ERR("consumer return code : %s",
1319 lttcomm_get_readable_code(-code
));
1322 } else if (pollfd
== consumer_data
->metadata_fd
) {
1323 /* UST metadata requests */
1324 ret
= ust_consumer_metadata_request(
1325 &consumer_data
->metadata_sock
);
1327 ERR("Handling metadata request");
1332 ERR("Unknown pollfd");
1336 health_code_update();
1342 * We lock here because we are about to close the sockets and some other
1343 * thread might be using them so get exclusive access which will abort all
1344 * other consumer command by other threads.
1346 pthread_mutex_lock(&consumer_data
->lock
);
1348 /* Immediately set the consumerd state to stopped */
1349 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1350 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1351 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1352 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1353 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1355 /* Code flow error... */
1359 if (consumer_data
->err_sock
>= 0) {
1360 ret
= close(consumer_data
->err_sock
);
1364 consumer_data
->err_sock
= -1;
1366 if (consumer_data
->cmd_sock
>= 0) {
1367 ret
= close(consumer_data
->cmd_sock
);
1371 consumer_data
->cmd_sock
= -1;
1373 if (consumer_data
->metadata_sock
.fd_ptr
&&
1374 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1375 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1387 unlink(consumer_data
->err_unix_sock_path
);
1388 unlink(consumer_data
->cmd_unix_sock_path
);
1389 consumer_data
->pid
= 0;
1390 pthread_mutex_unlock(&consumer_data
->lock
);
1392 /* Cleanup metadata socket mutex. */
1393 if (consumer_data
->metadata_sock
.lock
) {
1394 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1395 free(consumer_data
->metadata_sock
.lock
);
1397 lttng_poll_clean(&events
);
1401 ERR("Health error occurred in %s", __func__
);
1403 health_unregister(health_sessiond
);
1404 DBG("consumer thread cleanup completed");
1410 * This thread manage application communication.
1412 static void *thread_manage_apps(void *data
)
1414 int i
, ret
, pollfd
, err
= -1;
1416 uint32_t revents
, nb_fd
;
1417 struct lttng_poll_event events
;
1419 DBG("[thread] Manage application started");
1421 rcu_register_thread();
1422 rcu_thread_online();
1424 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1426 if (testpoint(sessiond_thread_manage_apps
)) {
1427 goto error_testpoint
;
1430 health_code_update();
1432 ret
= sessiond_set_thread_pollset(&events
, 2);
1434 goto error_poll_create
;
1437 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1442 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1446 health_code_update();
1449 DBG("Apps thread polling on %d fds", LTTNG_POLL_GETNB(&events
));
1451 /* Inifinite blocking call, waiting for transmission */
1453 health_poll_entry();
1454 ret
= lttng_poll_wait(&events
, -1);
1458 * Restart interrupted system call.
1460 if (errno
== EINTR
) {
1468 for (i
= 0; i
< nb_fd
; i
++) {
1469 /* Fetch once the poll data */
1470 revents
= LTTNG_POLL_GETEV(&events
, i
);
1471 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1473 health_code_update();
1475 /* Thread quit pipe has been closed. Killing thread. */
1476 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1482 /* Inspect the apps cmd pipe */
1483 if (pollfd
== apps_cmd_pipe
[0]) {
1484 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1485 ERR("Apps command pipe error");
1487 } else if (revents
& LPOLLIN
) {
1491 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1492 if (size_ret
< sizeof(sock
)) {
1493 PERROR("read apps cmd pipe");
1497 health_code_update();
1500 * We only monitor the error events of the socket. This
1501 * thread does not handle any incoming data from UST
1504 ret
= lttng_poll_add(&events
, sock
,
1505 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1510 DBG("Apps with sock %d added to poll set", sock
);
1514 * At this point, we know that a registered application made
1515 * the event at poll_wait.
1517 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1518 /* Removing from the poll set */
1519 ret
= lttng_poll_del(&events
, pollfd
);
1524 /* Socket closed on remote end. */
1525 ust_app_unregister(pollfd
);
1529 health_code_update();
1535 lttng_poll_clean(&events
);
1538 utils_close_pipe(apps_cmd_pipe
);
1539 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1542 * We don't clean the UST app hash table here since already registered
1543 * applications can still be controlled so let them be until the session
1544 * daemon dies or the applications stop.
1549 ERR("Health error occurred in %s", __func__
);
1551 health_unregister(health_sessiond
);
1552 DBG("Application communication apps thread cleanup complete");
1553 rcu_thread_offline();
1554 rcu_unregister_thread();
1559 * Send a socket to a thread This is called from the dispatch UST registration
1560 * thread once all sockets are set for the application.
1562 * The sock value can be invalid, we don't really care, the thread will handle
1563 * it and make the necessary cleanup if so.
1565 * On success, return 0 else a negative value being the errno message of the
1568 static int send_socket_to_thread(int fd
, int sock
)
1573 * It's possible that the FD is set as invalid with -1 concurrently just
1574 * before calling this function being a shutdown state of the thread.
1581 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1582 if (ret
< sizeof(sock
)) {
1583 PERROR("write apps pipe %d", fd
);
1590 /* All good. Don't send back the write positive ret value. */
1597 * Sanitize the wait queue of the dispatch registration thread meaning removing
1598 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1599 * notify socket is never received.
1601 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1603 int ret
, nb_fd
= 0, i
;
1604 unsigned int fd_added
= 0;
1605 struct lttng_poll_event events
;
1606 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1610 lttng_poll_init(&events
);
1612 /* Just skip everything for an empty queue. */
1613 if (!wait_queue
->count
) {
1617 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1622 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1623 &wait_queue
->head
, head
) {
1624 assert(wait_node
->app
);
1625 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1626 LPOLLHUP
| LPOLLERR
);
1639 * Poll but don't block so we can quickly identify the faulty events and
1640 * clean them afterwards from the wait queue.
1642 ret
= lttng_poll_wait(&events
, 0);
1648 for (i
= 0; i
< nb_fd
; i
++) {
1649 /* Get faulty FD. */
1650 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1651 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1653 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1654 &wait_queue
->head
, head
) {
1655 if (pollfd
== wait_node
->app
->sock
&&
1656 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1657 cds_list_del(&wait_node
->head
);
1658 wait_queue
->count
--;
1659 ust_app_destroy(wait_node
->app
);
1667 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1671 lttng_poll_clean(&events
);
1675 lttng_poll_clean(&events
);
1677 ERR("Unable to sanitize wait queue");
1682 * Dispatch request from the registration threads to the application
1683 * communication thread.
1685 static void *thread_dispatch_ust_registration(void *data
)
1688 struct cds_wfcq_node
*node
;
1689 struct ust_command
*ust_cmd
= NULL
;
1690 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1691 struct ust_reg_wait_queue wait_queue
= {
1695 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1697 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1698 goto error_testpoint
;
1701 health_code_update();
1703 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1705 DBG("[thread] Dispatch UST command started");
1707 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1708 health_code_update();
1710 /* Atomically prepare the queue futex */
1711 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1714 struct ust_app
*app
= NULL
;
1718 * Make sure we don't have node(s) that have hung up before receiving
1719 * the notify socket. This is to clean the list in order to avoid
1720 * memory leaks from notify socket that are never seen.
1722 sanitize_wait_queue(&wait_queue
);
1724 health_code_update();
1725 /* Dequeue command for registration */
1726 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1728 DBG("Woken up but nothing in the UST command queue");
1729 /* Continue thread execution */
1733 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1735 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1736 " gid:%d sock:%d name:%s (version %d.%d)",
1737 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1738 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1739 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1740 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1742 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1743 wait_node
= zmalloc(sizeof(*wait_node
));
1745 PERROR("zmalloc wait_node dispatch");
1746 ret
= close(ust_cmd
->sock
);
1748 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1750 lttng_fd_put(LTTNG_FD_APPS
, 1);
1754 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1756 /* Create application object if socket is CMD. */
1757 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1759 if (!wait_node
->app
) {
1760 ret
= close(ust_cmd
->sock
);
1762 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1764 lttng_fd_put(LTTNG_FD_APPS
, 1);
1770 * Add application to the wait queue so we can set the notify
1771 * socket before putting this object in the global ht.
1773 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1778 * We have to continue here since we don't have the notify
1779 * socket and the application MUST be added to the hash table
1780 * only at that moment.
1785 * Look for the application in the local wait queue and set the
1786 * notify socket if found.
1788 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1789 &wait_queue
.head
, head
) {
1790 health_code_update();
1791 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1792 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1793 cds_list_del(&wait_node
->head
);
1795 app
= wait_node
->app
;
1797 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1803 * With no application at this stage the received socket is
1804 * basically useless so close it before we free the cmd data
1805 * structure for good.
1808 ret
= close(ust_cmd
->sock
);
1810 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1812 lttng_fd_put(LTTNG_FD_APPS
, 1);
1819 * @session_lock_list
1821 * Lock the global session list so from the register up to the
1822 * registration done message, no thread can see the application
1823 * and change its state.
1825 session_lock_list();
1829 * Add application to the global hash table. This needs to be
1830 * done before the update to the UST registry can locate the
1835 /* Set app version. This call will print an error if needed. */
1836 (void) ust_app_version(app
);
1838 /* Send notify socket through the notify pipe. */
1839 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1843 session_unlock_list();
1845 * No notify thread, stop the UST tracing. However, this is
1846 * not an internal error of the this thread thus setting
1847 * the health error code to a normal exit.
1854 * Update newly registered application with the tracing
1855 * registry info already enabled information.
1857 update_ust_app(app
->sock
);
1860 * Don't care about return value. Let the manage apps threads
1861 * handle app unregistration upon socket close.
1863 (void) ust_app_register_done(app
->sock
);
1866 * Even if the application socket has been closed, send the app
1867 * to the thread and unregistration will take place at that
1870 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1873 session_unlock_list();
1875 * No apps. thread, stop the UST tracing. However, this is
1876 * not an internal error of the this thread thus setting
1877 * the health error code to a normal exit.
1884 session_unlock_list();
1886 } while (node
!= NULL
);
1888 health_poll_entry();
1889 /* Futex wait on queue. Blocking call on futex() */
1890 futex_nto1_wait(&ust_cmd_queue
.futex
);
1893 /* Normal exit, no error */
1897 /* Clean up wait queue. */
1898 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1899 &wait_queue
.head
, head
) {
1900 cds_list_del(&wait_node
->head
);
1906 DBG("Dispatch thread dying");
1909 ERR("Health error occurred in %s", __func__
);
1911 health_unregister(health_sessiond
);
1916 * This thread manage application registration.
1918 static void *thread_registration_apps(void *data
)
1920 int sock
= -1, i
, ret
, pollfd
, err
= -1;
1921 uint32_t revents
, nb_fd
;
1922 struct lttng_poll_event events
;
1924 * Get allocated in this thread, enqueued to a global queue, dequeued and
1925 * freed in the manage apps thread.
1927 struct ust_command
*ust_cmd
= NULL
;
1929 DBG("[thread] Manage application registration started");
1931 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
1933 if (testpoint(sessiond_thread_registration_apps
)) {
1934 goto error_testpoint
;
1937 ret
= lttcomm_listen_unix_sock(apps_sock
);
1943 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
1944 * more will be added to this poll set.
1946 ret
= sessiond_set_thread_pollset(&events
, 2);
1948 goto error_create_poll
;
1951 /* Add the application registration socket */
1952 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
1954 goto error_poll_add
;
1957 /* Notify all applications to register */
1958 ret
= notify_ust_apps(1);
1960 ERR("Failed to notify applications or create the wait shared memory.\n"
1961 "Execution continues but there might be problem for already\n"
1962 "running applications that wishes to register.");
1966 DBG("Accepting application registration");
1968 /* Inifinite blocking call, waiting for transmission */
1970 health_poll_entry();
1971 ret
= lttng_poll_wait(&events
, -1);
1975 * Restart interrupted system call.
1977 if (errno
== EINTR
) {
1985 for (i
= 0; i
< nb_fd
; i
++) {
1986 health_code_update();
1988 /* Fetch once the poll data */
1989 revents
= LTTNG_POLL_GETEV(&events
, i
);
1990 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1992 /* Thread quit pipe has been closed. Killing thread. */
1993 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1999 /* Event on the registration socket */
2000 if (pollfd
== apps_sock
) {
2001 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2002 ERR("Register apps socket poll error");
2004 } else if (revents
& LPOLLIN
) {
2005 sock
= lttcomm_accept_unix_sock(apps_sock
);
2011 * Set socket timeout for both receiving and ending.
2012 * app_socket_timeout is in seconds, whereas
2013 * lttcomm_setsockopt_rcv_timeout and
2014 * lttcomm_setsockopt_snd_timeout expect msec as
2017 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2018 app_socket_timeout
* 1000);
2019 (void) lttcomm_setsockopt_snd_timeout(sock
,
2020 app_socket_timeout
* 1000);
2023 * Set the CLOEXEC flag. Return code is useless because
2024 * either way, the show must go on.
2026 (void) utils_set_fd_cloexec(sock
);
2028 /* Create UST registration command for enqueuing */
2029 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2030 if (ust_cmd
== NULL
) {
2031 PERROR("ust command zmalloc");
2036 * Using message-based transmissions to ensure we don't
2037 * have to deal with partially received messages.
2039 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2041 ERR("Exhausted file descriptors allowed for applications.");
2051 health_code_update();
2052 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2055 /* Close socket of the application. */
2060 lttng_fd_put(LTTNG_FD_APPS
, 1);
2064 health_code_update();
2066 ust_cmd
->sock
= sock
;
2069 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2070 " gid:%d sock:%d name:%s (version %d.%d)",
2071 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2072 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2073 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2074 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2077 * Lock free enqueue the registration request. The red pill
2078 * has been taken! This apps will be part of the *system*.
2080 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2083 * Wake the registration queue futex. Implicit memory
2084 * barrier with the exchange in cds_wfcq_enqueue.
2086 futex_nto1_wake(&ust_cmd_queue
.futex
);
2094 /* Notify that the registration thread is gone */
2097 if (apps_sock
>= 0) {
2098 ret
= close(apps_sock
);
2108 lttng_fd_put(LTTNG_FD_APPS
, 1);
2110 unlink(apps_unix_sock_path
);
2113 lttng_poll_clean(&events
);
2117 DBG("UST Registration thread cleanup complete");
2120 ERR("Health error occurred in %s", __func__
);
2122 health_unregister(health_sessiond
);
2128 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2129 * exec or it will fails.
2131 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2134 struct timespec timeout
;
2136 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2137 consumer_data
->consumer_thread_is_ready
= 0;
2139 /* Setup pthread condition */
2140 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2143 PERROR("pthread_condattr_init consumer data");
2148 * Set the monotonic clock in order to make sure we DO NOT jump in time
2149 * between the clock_gettime() call and the timedwait call. See bug #324
2150 * for a more details and how we noticed it.
2152 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2155 PERROR("pthread_condattr_setclock consumer data");
2159 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2162 PERROR("pthread_cond_init consumer data");
2166 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2169 PERROR("pthread_create consumer");
2174 /* We are about to wait on a pthread condition */
2175 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2177 /* Get time for sem_timedwait absolute timeout */
2178 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2180 * Set the timeout for the condition timed wait even if the clock gettime
2181 * call fails since we might loop on that call and we want to avoid to
2182 * increment the timeout too many times.
2184 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2187 * The following loop COULD be skipped in some conditions so this is why we
2188 * set ret to 0 in order to make sure at least one round of the loop is
2194 * Loop until the condition is reached or when a timeout is reached. Note
2195 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2196 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2197 * possible. This loop does not take any chances and works with both of
2200 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2201 if (clock_ret
< 0) {
2202 PERROR("clock_gettime spawn consumer");
2203 /* Infinite wait for the consumerd thread to be ready */
2204 ret
= pthread_cond_wait(&consumer_data
->cond
,
2205 &consumer_data
->cond_mutex
);
2207 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2208 &consumer_data
->cond_mutex
, &timeout
);
2212 /* Release the pthread condition */
2213 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2217 if (ret
== ETIMEDOUT
) {
2221 * Call has timed out so we kill the kconsumerd_thread and return
2224 ERR("Condition timed out. The consumer thread was never ready."
2226 pth_ret
= pthread_cancel(consumer_data
->thread
);
2228 PERROR("pthread_cancel consumer thread");
2231 PERROR("pthread_cond_wait failed consumer thread");
2233 /* Caller is expecting a negative value on failure. */
2238 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2239 if (consumer_data
->pid
== 0) {
2240 ERR("Consumerd did not start");
2241 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2244 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2253 * Join consumer thread
2255 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2259 /* Consumer pid must be a real one. */
2260 if (consumer_data
->pid
> 0) {
2262 ret
= kill(consumer_data
->pid
, SIGTERM
);
2264 ERR("Error killing consumer daemon");
2267 return pthread_join(consumer_data
->thread
, &status
);
2274 * Fork and exec a consumer daemon (consumerd).
2276 * Return pid if successful else -1.
2278 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2282 const char *consumer_to_use
;
2283 const char *verbosity
;
2286 DBG("Spawning consumerd");
2293 if (opt_verbose_consumer
) {
2294 verbosity
= "--verbose";
2295 } else if (lttng_opt_quiet
) {
2296 verbosity
= "--quiet";
2301 switch (consumer_data
->type
) {
2302 case LTTNG_CONSUMER_KERNEL
:
2304 * Find out which consumerd to execute. We will first try the
2305 * 64-bit path, then the sessiond's installation directory, and
2306 * fallback on the 32-bit one,
2308 DBG3("Looking for a kernel consumer at these locations:");
2309 DBG3(" 1) %s", consumerd64_bin
);
2310 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2311 DBG3(" 3) %s", consumerd32_bin
);
2312 if (stat(consumerd64_bin
, &st
) == 0) {
2313 DBG3("Found location #1");
2314 consumer_to_use
= consumerd64_bin
;
2315 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2316 DBG3("Found location #2");
2317 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2318 } else if (stat(consumerd32_bin
, &st
) == 0) {
2319 DBG3("Found location #3");
2320 consumer_to_use
= consumerd32_bin
;
2322 DBG("Could not find any valid consumerd executable");
2326 DBG("Using kernel consumer at: %s", consumer_to_use
);
2327 ret
= execl(consumer_to_use
,
2328 "lttng-consumerd", verbosity
, "-k",
2329 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2330 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2331 "--group", tracing_group_name
,
2334 case LTTNG_CONSUMER64_UST
:
2336 char *tmpnew
= NULL
;
2338 if (consumerd64_libdir
[0] != '\0') {
2342 tmp
= getenv("LD_LIBRARY_PATH");
2346 tmplen
= strlen("LD_LIBRARY_PATH=")
2347 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2348 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2353 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2354 strcat(tmpnew
, consumerd64_libdir
);
2355 if (tmp
[0] != '\0') {
2356 strcat(tmpnew
, ":");
2357 strcat(tmpnew
, tmp
);
2359 ret
= putenv(tmpnew
);
2366 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2367 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2368 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2369 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2370 "--group", tracing_group_name
,
2372 if (consumerd64_libdir
[0] != '\0') {
2377 case LTTNG_CONSUMER32_UST
:
2379 char *tmpnew
= NULL
;
2381 if (consumerd32_libdir
[0] != '\0') {
2385 tmp
= getenv("LD_LIBRARY_PATH");
2389 tmplen
= strlen("LD_LIBRARY_PATH=")
2390 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2391 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2396 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2397 strcat(tmpnew
, consumerd32_libdir
);
2398 if (tmp
[0] != '\0') {
2399 strcat(tmpnew
, ":");
2400 strcat(tmpnew
, tmp
);
2402 ret
= putenv(tmpnew
);
2409 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2410 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2411 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2412 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2413 "--group", tracing_group_name
,
2415 if (consumerd32_libdir
[0] != '\0') {
2421 PERROR("unknown consumer type");
2425 PERROR("Consumer execl()");
2427 /* Reaching this point, we got a failure on our execl(). */
2429 } else if (pid
> 0) {
2432 PERROR("start consumer fork");
2440 * Spawn the consumerd daemon and session daemon thread.
2442 static int start_consumerd(struct consumer_data
*consumer_data
)
2447 * Set the listen() state on the socket since there is a possible race
2448 * between the exec() of the consumer daemon and this call if place in the
2449 * consumer thread. See bug #366 for more details.
2451 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2456 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2457 if (consumer_data
->pid
!= 0) {
2458 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2462 ret
= spawn_consumerd(consumer_data
);
2464 ERR("Spawning consumerd failed");
2465 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2469 /* Setting up the consumer_data pid */
2470 consumer_data
->pid
= ret
;
2471 DBG2("Consumer pid %d", consumer_data
->pid
);
2472 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2474 DBG2("Spawning consumer control thread");
2475 ret
= spawn_consumer_thread(consumer_data
);
2477 ERR("Fatal error spawning consumer control thread");
2485 /* Cleanup already created sockets on error. */
2486 if (consumer_data
->err_sock
>= 0) {
2489 err
= close(consumer_data
->err_sock
);
2491 PERROR("close consumer data error socket");
2498 * Setup necessary data for kernel tracer action.
2500 static int init_kernel_tracer(void)
2504 /* Modprobe lttng kernel modules */
2505 ret
= modprobe_lttng_control();
2510 /* Open debugfs lttng */
2511 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2512 if (kernel_tracer_fd
< 0) {
2513 DBG("Failed to open %s", module_proc_lttng
);
2518 /* Validate kernel version */
2519 ret
= kernel_validate_version(kernel_tracer_fd
);
2524 ret
= modprobe_lttng_data();
2529 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2533 modprobe_remove_lttng_control();
2534 ret
= close(kernel_tracer_fd
);
2538 kernel_tracer_fd
= -1;
2539 return LTTNG_ERR_KERN_VERSION
;
2542 ret
= close(kernel_tracer_fd
);
2548 modprobe_remove_lttng_control();
2551 WARN("No kernel tracer available");
2552 kernel_tracer_fd
= -1;
2554 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2556 return LTTNG_ERR_KERN_NA
;
2562 * Copy consumer output from the tracing session to the domain session. The
2563 * function also applies the right modification on a per domain basis for the
2564 * trace files destination directory.
2566 * Should *NOT* be called with RCU read-side lock held.
2568 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2571 const char *dir_name
;
2572 struct consumer_output
*consumer
;
2575 assert(session
->consumer
);
2578 case LTTNG_DOMAIN_KERNEL
:
2579 DBG3("Copying tracing session consumer output in kernel session");
2581 * XXX: We should audit the session creation and what this function
2582 * does "extra" in order to avoid a destroy since this function is used
2583 * in the domain session creation (kernel and ust) only. Same for UST
2586 if (session
->kernel_session
->consumer
) {
2587 consumer_destroy_output(session
->kernel_session
->consumer
);
2589 session
->kernel_session
->consumer
=
2590 consumer_copy_output(session
->consumer
);
2591 /* Ease our life a bit for the next part */
2592 consumer
= session
->kernel_session
->consumer
;
2593 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2595 case LTTNG_DOMAIN_JUL
:
2596 case LTTNG_DOMAIN_UST
:
2597 DBG3("Copying tracing session consumer output in UST session");
2598 if (session
->ust_session
->consumer
) {
2599 consumer_destroy_output(session
->ust_session
->consumer
);
2601 session
->ust_session
->consumer
=
2602 consumer_copy_output(session
->consumer
);
2603 /* Ease our life a bit for the next part */
2604 consumer
= session
->ust_session
->consumer
;
2605 dir_name
= DEFAULT_UST_TRACE_DIR
;
2608 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2612 /* Append correct directory to subdir */
2613 strncat(consumer
->subdir
, dir_name
,
2614 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2615 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2624 * Create an UST session and add it to the session ust list.
2626 * Should *NOT* be called with RCU read-side lock held.
2628 static int create_ust_session(struct ltt_session
*session
,
2629 struct lttng_domain
*domain
)
2632 struct ltt_ust_session
*lus
= NULL
;
2636 assert(session
->consumer
);
2638 switch (domain
->type
) {
2639 case LTTNG_DOMAIN_JUL
:
2640 case LTTNG_DOMAIN_UST
:
2643 ERR("Unknown UST domain on create session %d", domain
->type
);
2644 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2648 DBG("Creating UST session");
2650 lus
= trace_ust_create_session(session
->id
);
2652 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2656 lus
->uid
= session
->uid
;
2657 lus
->gid
= session
->gid
;
2658 lus
->output_traces
= session
->output_traces
;
2659 lus
->snapshot_mode
= session
->snapshot_mode
;
2660 lus
->live_timer_interval
= session
->live_timer
;
2661 session
->ust_session
= lus
;
2663 /* Copy session output to the newly created UST session */
2664 ret
= copy_session_consumer(domain
->type
, session
);
2665 if (ret
!= LTTNG_OK
) {
2673 session
->ust_session
= NULL
;
2678 * Create a kernel tracer session then create the default channel.
2680 static int create_kernel_session(struct ltt_session
*session
)
2684 DBG("Creating kernel session");
2686 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2688 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2692 /* Code flow safety */
2693 assert(session
->kernel_session
);
2695 /* Copy session output to the newly created Kernel session */
2696 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2697 if (ret
!= LTTNG_OK
) {
2701 /* Create directory(ies) on local filesystem. */
2702 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2703 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2704 ret
= run_as_mkdir_recursive(
2705 session
->kernel_session
->consumer
->dst
.trace_path
,
2706 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2708 if (ret
!= -EEXIST
) {
2709 ERR("Trace directory creation error");
2715 session
->kernel_session
->uid
= session
->uid
;
2716 session
->kernel_session
->gid
= session
->gid
;
2717 session
->kernel_session
->output_traces
= session
->output_traces
;
2718 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2723 trace_kernel_destroy_session(session
->kernel_session
);
2724 session
->kernel_session
= NULL
;
2729 * Count number of session permitted by uid/gid.
2731 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2734 struct ltt_session
*session
;
2736 DBG("Counting number of available session for UID %d GID %d",
2738 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2740 * Only list the sessions the user can control.
2742 if (!session_access_ok(session
, uid
, gid
)) {
2751 * Process the command requested by the lttng client within the command
2752 * context structure. This function make sure that the return structure (llm)
2753 * is set and ready for transmission before returning.
2755 * Return any error encountered or 0 for success.
2757 * "sock" is only used for special-case var. len data.
2759 * Should *NOT* be called with RCU read-side lock held.
2761 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2765 int need_tracing_session
= 1;
2768 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2772 switch (cmd_ctx
->lsm
->cmd_type
) {
2773 case LTTNG_CREATE_SESSION
:
2774 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2775 case LTTNG_CREATE_SESSION_LIVE
:
2776 case LTTNG_DESTROY_SESSION
:
2777 case LTTNG_LIST_SESSIONS
:
2778 case LTTNG_LIST_DOMAINS
:
2779 case LTTNG_START_TRACE
:
2780 case LTTNG_STOP_TRACE
:
2781 case LTTNG_DATA_PENDING
:
2782 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2783 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2784 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2785 case LTTNG_SNAPSHOT_RECORD
:
2786 case LTTNG_SAVE_SESSION
:
2793 if (opt_no_kernel
&& need_domain
2794 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2796 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2798 ret
= LTTNG_ERR_KERN_NA
;
2803 /* Deny register consumer if we already have a spawned consumer. */
2804 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2805 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2806 if (kconsumer_data
.pid
> 0) {
2807 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2808 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2811 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2815 * Check for command that don't needs to allocate a returned payload. We do
2816 * this here so we don't have to make the call for no payload at each
2819 switch(cmd_ctx
->lsm
->cmd_type
) {
2820 case LTTNG_LIST_SESSIONS
:
2821 case LTTNG_LIST_TRACEPOINTS
:
2822 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2823 case LTTNG_LIST_DOMAINS
:
2824 case LTTNG_LIST_CHANNELS
:
2825 case LTTNG_LIST_EVENTS
:
2828 /* Setup lttng message with no payload */
2829 ret
= setup_lttng_msg(cmd_ctx
, 0);
2831 /* This label does not try to unlock the session */
2832 goto init_setup_error
;
2836 /* Commands that DO NOT need a session. */
2837 switch (cmd_ctx
->lsm
->cmd_type
) {
2838 case LTTNG_CREATE_SESSION
:
2839 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2840 case LTTNG_CREATE_SESSION_LIVE
:
2841 case LTTNG_CALIBRATE
:
2842 case LTTNG_LIST_SESSIONS
:
2843 case LTTNG_LIST_TRACEPOINTS
:
2844 case LTTNG_LIST_TRACEPOINT_FIELDS
:
2845 case LTTNG_SAVE_SESSION
:
2846 need_tracing_session
= 0;
2849 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
2851 * We keep the session list lock across _all_ commands
2852 * for now, because the per-session lock does not
2853 * handle teardown properly.
2855 session_lock_list();
2856 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
2857 if (cmd_ctx
->session
== NULL
) {
2858 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
2861 /* Acquire lock for the session */
2862 session_lock(cmd_ctx
->session
);
2868 * Commands that need a valid session but should NOT create one if none
2869 * exists. Instead of creating one and destroying it when the command is
2870 * handled, process that right before so we save some round trip in useless
2873 switch (cmd_ctx
->lsm
->cmd_type
) {
2874 case LTTNG_DISABLE_CHANNEL
:
2875 case LTTNG_DISABLE_EVENT
:
2876 case LTTNG_DISABLE_ALL_EVENT
:
2877 switch (cmd_ctx
->lsm
->domain
.type
) {
2878 case LTTNG_DOMAIN_KERNEL
:
2879 if (!cmd_ctx
->session
->kernel_session
) {
2880 ret
= LTTNG_ERR_NO_CHANNEL
;
2884 case LTTNG_DOMAIN_JUL
:
2885 case LTTNG_DOMAIN_UST
:
2886 if (!cmd_ctx
->session
->ust_session
) {
2887 ret
= LTTNG_ERR_NO_CHANNEL
;
2892 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2904 * Check domain type for specific "pre-action".
2906 switch (cmd_ctx
->lsm
->domain
.type
) {
2907 case LTTNG_DOMAIN_KERNEL
:
2909 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2913 /* Kernel tracer check */
2914 if (kernel_tracer_fd
== -1) {
2915 /* Basically, load kernel tracer modules */
2916 ret
= init_kernel_tracer();
2922 /* Consumer is in an ERROR state. Report back to client */
2923 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
2924 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
2928 /* Need a session for kernel command */
2929 if (need_tracing_session
) {
2930 if (cmd_ctx
->session
->kernel_session
== NULL
) {
2931 ret
= create_kernel_session(cmd_ctx
->session
);
2933 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2938 /* Start the kernel consumer daemon */
2939 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2940 if (kconsumer_data
.pid
== 0 &&
2941 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2942 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2943 ret
= start_consumerd(&kconsumer_data
);
2945 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2948 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
2950 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2954 * The consumer was just spawned so we need to add the socket to
2955 * the consumer output of the session if exist.
2957 ret
= consumer_create_socket(&kconsumer_data
,
2958 cmd_ctx
->session
->kernel_session
->consumer
);
2965 case LTTNG_DOMAIN_JUL
:
2966 case LTTNG_DOMAIN_UST
:
2968 if (!ust_app_supported()) {
2969 ret
= LTTNG_ERR_NO_UST
;
2972 /* Consumer is in an ERROR state. Report back to client */
2973 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
2974 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
2978 if (need_tracing_session
) {
2979 /* Create UST session if none exist. */
2980 if (cmd_ctx
->session
->ust_session
== NULL
) {
2981 ret
= create_ust_session(cmd_ctx
->session
,
2982 &cmd_ctx
->lsm
->domain
);
2983 if (ret
!= LTTNG_OK
) {
2988 /* Start the UST consumer daemons */
2990 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
2991 if (consumerd64_bin
[0] != '\0' &&
2992 ustconsumer64_data
.pid
== 0 &&
2993 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
2994 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
2995 ret
= start_consumerd(&ustconsumer64_data
);
2997 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
2998 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3002 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3003 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3005 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3009 * Setup socket for consumer 64 bit. No need for atomic access
3010 * since it was set above and can ONLY be set in this thread.
3012 ret
= consumer_create_socket(&ustconsumer64_data
,
3013 cmd_ctx
->session
->ust_session
->consumer
);
3019 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3020 if (consumerd32_bin
[0] != '\0' &&
3021 ustconsumer32_data
.pid
== 0 &&
3022 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3023 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3024 ret
= start_consumerd(&ustconsumer32_data
);
3026 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3027 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3031 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3032 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3034 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3038 * Setup socket for consumer 64 bit. No need for atomic access
3039 * since it was set above and can ONLY be set in this thread.
3041 ret
= consumer_create_socket(&ustconsumer32_data
,
3042 cmd_ctx
->session
->ust_session
->consumer
);
3054 /* Validate consumer daemon state when start/stop trace command */
3055 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3056 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3057 switch (cmd_ctx
->lsm
->domain
.type
) {
3058 case LTTNG_DOMAIN_JUL
:
3059 case LTTNG_DOMAIN_UST
:
3060 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3061 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3065 case LTTNG_DOMAIN_KERNEL
:
3066 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3067 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3075 * Check that the UID or GID match that of the tracing session.
3076 * The root user can interact with all sessions.
3078 if (need_tracing_session
) {
3079 if (!session_access_ok(cmd_ctx
->session
,
3080 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3081 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3082 ret
= LTTNG_ERR_EPERM
;
3088 * Send relayd information to consumer as soon as we have a domain and a
3091 if (cmd_ctx
->session
&& need_domain
) {
3093 * Setup relayd if not done yet. If the relayd information was already
3094 * sent to the consumer, this call will gracefully return.
3096 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3097 if (ret
!= LTTNG_OK
) {
3102 /* Process by command type */
3103 switch (cmd_ctx
->lsm
->cmd_type
) {
3104 case LTTNG_ADD_CONTEXT
:
3106 ret
= cmd_add_context(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3107 cmd_ctx
->lsm
->u
.context
.channel_name
,
3108 &cmd_ctx
->lsm
->u
.context
.ctx
, kernel_poll_pipe
[1]);
3111 case LTTNG_DISABLE_CHANNEL
:
3113 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3114 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3117 case LTTNG_DISABLE_EVENT
:
3119 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3120 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3121 cmd_ctx
->lsm
->u
.disable
.name
);
3124 case LTTNG_DISABLE_ALL_EVENT
:
3126 DBG("Disabling all events");
3128 ret
= cmd_disable_event_all(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3129 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3132 case LTTNG_ENABLE_CHANNEL
:
3134 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3135 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3138 case LTTNG_ENABLE_EVENT
:
3140 struct lttng_event_exclusion
*exclusion
= NULL
;
3141 struct lttng_filter_bytecode
*bytecode
= NULL
;
3142 char *filter_expression
= NULL
;
3144 /* Handle exclusion events and receive it from the client. */
3145 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3146 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3148 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3149 (count
* LTTNG_SYMBOL_NAME_LEN
));
3151 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3155 DBG("Receiving var len exclusion event list from client ...");
3156 exclusion
->count
= count
;
3157 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3158 count
* LTTNG_SYMBOL_NAME_LEN
);
3160 DBG("Nothing recv() from client var len data... continuing");
3163 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3168 /* Get filter expression from client. */
3169 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3170 size_t expression_len
=
3171 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3173 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3174 ret
= LTTNG_ERR_FILTER_INVAL
;
3179 filter_expression
= zmalloc(expression_len
);
3180 if (!filter_expression
) {
3182 ret
= LTTNG_ERR_FILTER_NOMEM
;
3186 /* Receive var. len. data */
3187 DBG("Receiving var len filter's expression from client ...");
3188 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3191 DBG("Nothing recv() from client car len data... continuing");
3193 free(filter_expression
);
3195 ret
= LTTNG_ERR_FILTER_INVAL
;
3200 /* Handle filter and get bytecode from client. */
3201 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3202 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3204 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3205 ret
= LTTNG_ERR_FILTER_INVAL
;
3210 bytecode
= zmalloc(bytecode_len
);
3213 ret
= LTTNG_ERR_FILTER_NOMEM
;
3217 /* Receive var. len. data */
3218 DBG("Receiving var len filter's bytecode from client ...");
3219 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3221 DBG("Nothing recv() from client car len data... continuing");
3225 ret
= LTTNG_ERR_FILTER_INVAL
;
3229 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3232 ret
= LTTNG_ERR_FILTER_INVAL
;
3237 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3238 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3239 &cmd_ctx
->lsm
->u
.enable
.event
,
3240 filter_expression
, bytecode
, exclusion
,
3241 kernel_poll_pipe
[1]);
3244 case LTTNG_ENABLE_ALL_EVENT
:
3246 DBG("Enabling all events");
3248 ret
= cmd_enable_event_all(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3249 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3250 cmd_ctx
->lsm
->u
.enable
.event
.type
, NULL
, NULL
,
3251 kernel_poll_pipe
[1]);
3254 case LTTNG_LIST_TRACEPOINTS
:
3256 struct lttng_event
*events
;
3259 session_lock_list();
3260 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3261 session_unlock_list();
3262 if (nb_events
< 0) {
3263 /* Return value is a negative lttng_error_code. */
3269 * Setup lttng message with payload size set to the event list size in
3270 * bytes and then copy list into the llm payload.
3272 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_event
) * nb_events
);
3278 /* Copy event list into message payload */
3279 memcpy(cmd_ctx
->llm
->payload
, events
,
3280 sizeof(struct lttng_event
) * nb_events
);
3287 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3289 struct lttng_event_field
*fields
;
3292 session_lock_list();
3293 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3295 session_unlock_list();
3296 if (nb_fields
< 0) {
3297 /* Return value is a negative lttng_error_code. */
3303 * Setup lttng message with payload size set to the event list size in
3304 * bytes and then copy list into the llm payload.
3306 ret
= setup_lttng_msg(cmd_ctx
,
3307 sizeof(struct lttng_event_field
) * nb_fields
);
3313 /* Copy event list into message payload */
3314 memcpy(cmd_ctx
->llm
->payload
, fields
,
3315 sizeof(struct lttng_event_field
) * nb_fields
);
3322 case LTTNG_SET_CONSUMER_URI
:
3325 struct lttng_uri
*uris
;
3327 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3328 len
= nb_uri
* sizeof(struct lttng_uri
);
3331 ret
= LTTNG_ERR_INVALID
;
3335 uris
= zmalloc(len
);
3337 ret
= LTTNG_ERR_FATAL
;
3341 /* Receive variable len data */
3342 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3343 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3345 DBG("No URIs received from client... continuing");
3347 ret
= LTTNG_ERR_SESSION_FAIL
;
3352 ret
= cmd_set_consumer_uri(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3354 if (ret
!= LTTNG_OK
) {
3360 * XXX: 0 means that this URI should be applied on the session. Should
3361 * be a DOMAIN enuam.
3363 if (cmd_ctx
->lsm
->domain
.type
== 0) {
3364 /* Add the URI for the UST session if a consumer is present. */
3365 if (cmd_ctx
->session
->ust_session
&&
3366 cmd_ctx
->session
->ust_session
->consumer
) {
3367 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_UST
, cmd_ctx
->session
,
3369 } else if (cmd_ctx
->session
->kernel_session
&&
3370 cmd_ctx
->session
->kernel_session
->consumer
) {
3371 ret
= cmd_set_consumer_uri(LTTNG_DOMAIN_KERNEL
,
3372 cmd_ctx
->session
, nb_uri
, uris
);
3380 case LTTNG_START_TRACE
:
3382 ret
= cmd_start_trace(cmd_ctx
->session
);
3385 case LTTNG_STOP_TRACE
:
3387 ret
= cmd_stop_trace(cmd_ctx
->session
);
3390 case LTTNG_CREATE_SESSION
:
3393 struct lttng_uri
*uris
= NULL
;
3395 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3396 len
= nb_uri
* sizeof(struct lttng_uri
);
3399 uris
= zmalloc(len
);
3401 ret
= LTTNG_ERR_FATAL
;
3405 /* Receive variable len data */
3406 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3407 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3409 DBG("No URIs received from client... continuing");
3411 ret
= LTTNG_ERR_SESSION_FAIL
;
3416 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3417 DBG("Creating session with ONE network URI is a bad call");
3418 ret
= LTTNG_ERR_SESSION_FAIL
;
3424 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3425 &cmd_ctx
->creds
, 0);
3431 case LTTNG_DESTROY_SESSION
:
3433 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3435 /* Set session to NULL so we do not unlock it after free. */
3436 cmd_ctx
->session
= NULL
;
3439 case LTTNG_LIST_DOMAINS
:
3442 struct lttng_domain
*domains
;
3444 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3446 /* Return value is a negative lttng_error_code. */
3451 ret
= setup_lttng_msg(cmd_ctx
, nb_dom
* sizeof(struct lttng_domain
));
3457 /* Copy event list into message payload */
3458 memcpy(cmd_ctx
->llm
->payload
, domains
,
3459 nb_dom
* sizeof(struct lttng_domain
));
3466 case LTTNG_LIST_CHANNELS
:
3469 struct lttng_channel
*channels
;
3471 nb_chan
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3472 cmd_ctx
->session
, &channels
);
3474 /* Return value is a negative lttng_error_code. */
3479 ret
= setup_lttng_msg(cmd_ctx
, nb_chan
* sizeof(struct lttng_channel
));
3485 /* Copy event list into message payload */
3486 memcpy(cmd_ctx
->llm
->payload
, channels
,
3487 nb_chan
* sizeof(struct lttng_channel
));
3494 case LTTNG_LIST_EVENTS
:
3497 struct lttng_event
*events
= NULL
;
3499 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
, cmd_ctx
->session
,
3500 cmd_ctx
->lsm
->u
.list
.channel_name
, &events
);
3502 /* Return value is a negative lttng_error_code. */
3507 ret
= setup_lttng_msg(cmd_ctx
, nb_event
* sizeof(struct lttng_event
));
3513 /* Copy event list into message payload */
3514 memcpy(cmd_ctx
->llm
->payload
, events
,
3515 nb_event
* sizeof(struct lttng_event
));
3522 case LTTNG_LIST_SESSIONS
:
3524 unsigned int nr_sessions
;
3526 session_lock_list();
3527 nr_sessions
= lttng_sessions_count(
3528 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3529 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3531 ret
= setup_lttng_msg(cmd_ctx
, sizeof(struct lttng_session
) * nr_sessions
);
3533 session_unlock_list();
3537 /* Filled the session array */
3538 cmd_list_lttng_sessions((struct lttng_session
*)(cmd_ctx
->llm
->payload
),
3539 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3540 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3542 session_unlock_list();
3547 case LTTNG_CALIBRATE
:
3549 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3550 &cmd_ctx
->lsm
->u
.calibrate
);
3553 case LTTNG_REGISTER_CONSUMER
:
3555 struct consumer_data
*cdata
;
3557 switch (cmd_ctx
->lsm
->domain
.type
) {
3558 case LTTNG_DOMAIN_KERNEL
:
3559 cdata
= &kconsumer_data
;
3562 ret
= LTTNG_ERR_UND
;
3566 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3567 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3570 case LTTNG_DATA_PENDING
:
3572 ret
= cmd_data_pending(cmd_ctx
->session
);
3575 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3577 struct lttcomm_lttng_output_id reply
;
3579 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3580 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3581 if (ret
!= LTTNG_OK
) {
3585 ret
= setup_lttng_msg(cmd_ctx
, sizeof(reply
));
3590 /* Copy output list into message payload */
3591 memcpy(cmd_ctx
->llm
->payload
, &reply
, sizeof(reply
));
3595 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3597 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3598 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3601 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3604 struct lttng_snapshot_output
*outputs
= NULL
;
3606 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3607 if (nb_output
< 0) {
3612 ret
= setup_lttng_msg(cmd_ctx
,
3613 nb_output
* sizeof(struct lttng_snapshot_output
));
3620 /* Copy output list into message payload */
3621 memcpy(cmd_ctx
->llm
->payload
, outputs
,
3622 nb_output
* sizeof(struct lttng_snapshot_output
));
3629 case LTTNG_SNAPSHOT_RECORD
:
3631 ret
= cmd_snapshot_record(cmd_ctx
->session
,
3632 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
3633 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
3636 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3639 struct lttng_uri
*uris
= NULL
;
3641 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3642 len
= nb_uri
* sizeof(struct lttng_uri
);
3645 uris
= zmalloc(len
);
3647 ret
= LTTNG_ERR_FATAL
;
3651 /* Receive variable len data */
3652 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3653 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3655 DBG("No URIs received from client... continuing");
3657 ret
= LTTNG_ERR_SESSION_FAIL
;
3662 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3663 DBG("Creating session with ONE network URI is a bad call");
3664 ret
= LTTNG_ERR_SESSION_FAIL
;
3670 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
3671 nb_uri
, &cmd_ctx
->creds
);
3675 case LTTNG_CREATE_SESSION_LIVE
:
3678 struct lttng_uri
*uris
= NULL
;
3680 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3681 len
= nb_uri
* sizeof(struct lttng_uri
);
3684 uris
= zmalloc(len
);
3686 ret
= LTTNG_ERR_FATAL
;
3690 /* Receive variable len data */
3691 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3692 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3694 DBG("No URIs received from client... continuing");
3696 ret
= LTTNG_ERR_SESSION_FAIL
;
3701 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3702 DBG("Creating session with ONE network URI is a bad call");
3703 ret
= LTTNG_ERR_SESSION_FAIL
;
3709 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
3710 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
3714 case LTTNG_SAVE_SESSION
:
3716 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
3721 ret
= LTTNG_ERR_UND
;
3726 if (cmd_ctx
->llm
== NULL
) {
3727 DBG("Missing llm structure. Allocating one.");
3728 if (setup_lttng_msg(cmd_ctx
, 0) < 0) {
3732 /* Set return code */
3733 cmd_ctx
->llm
->ret_code
= ret
;
3735 if (cmd_ctx
->session
) {
3736 session_unlock(cmd_ctx
->session
);
3738 if (need_tracing_session
) {
3739 session_unlock_list();
3746 * Thread managing health check socket.
3748 static void *thread_manage_health(void *data
)
3750 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
3751 uint32_t revents
, nb_fd
;
3752 struct lttng_poll_event events
;
3753 struct health_comm_msg msg
;
3754 struct health_comm_reply reply
;
3756 DBG("[thread] Manage health check started");
3758 rcu_register_thread();
3760 /* We might hit an error path before this is created. */
3761 lttng_poll_init(&events
);
3763 /* Create unix socket */
3764 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
3766 ERR("Unable to create health check Unix socket");
3772 /* lttng health client socket path permissions */
3773 ret
= chown(health_unix_sock_path
, 0,
3774 utils_get_group_id(tracing_group_name
));
3776 ERR("Unable to set group on %s", health_unix_sock_path
);
3782 ret
= chmod(health_unix_sock_path
,
3783 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
3785 ERR("Unable to set permissions on %s", health_unix_sock_path
);
3793 * Set the CLOEXEC flag. Return code is useless because either way, the
3796 (void) utils_set_fd_cloexec(sock
);
3798 ret
= lttcomm_listen_unix_sock(sock
);
3804 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3805 * more will be added to this poll set.
3807 ret
= sessiond_set_thread_pollset(&events
, 2);
3812 /* Add the application registration socket */
3813 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
3818 sessiond_notify_ready();
3821 DBG("Health check ready");
3823 /* Inifinite blocking call, waiting for transmission */
3825 ret
= lttng_poll_wait(&events
, -1);
3828 * Restart interrupted system call.
3830 if (errno
== EINTR
) {
3838 for (i
= 0; i
< nb_fd
; i
++) {
3839 /* Fetch once the poll data */
3840 revents
= LTTNG_POLL_GETEV(&events
, i
);
3841 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
3843 /* Thread quit pipe has been closed. Killing thread. */
3844 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
3850 /* Event on the registration socket */
3851 if (pollfd
== sock
) {
3852 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
3853 ERR("Health socket poll error");
3859 new_sock
= lttcomm_accept_unix_sock(sock
);
3865 * Set the CLOEXEC flag. Return code is useless because either way, the
3868 (void) utils_set_fd_cloexec(new_sock
);
3870 DBG("Receiving data from client for health...");
3871 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
3873 DBG("Nothing recv() from client... continuing");
3874 ret
= close(new_sock
);
3882 rcu_thread_online();
3884 memset(&reply
, 0, sizeof(reply
));
3885 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
3887 * health_check_state returns 0 if health is
3890 if (!health_check_state(health_sessiond
, i
)) {
3891 reply
.ret_code
|= 1ULL << i
;
3895 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
3897 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
3899 ERR("Failed to send health data back to client");
3902 /* End of transmission */
3903 ret
= close(new_sock
);
3913 ERR("Health error occurred in %s", __func__
);
3915 DBG("Health check thread dying");
3916 unlink(health_unix_sock_path
);
3924 lttng_poll_clean(&events
);
3926 rcu_unregister_thread();
3931 * This thread manage all clients request using the unix client socket for
3934 static void *thread_manage_clients(void *data
)
3936 int sock
= -1, ret
, i
, pollfd
, err
= -1;
3938 uint32_t revents
, nb_fd
;
3939 struct command_ctx
*cmd_ctx
= NULL
;
3940 struct lttng_poll_event events
;
3942 DBG("[thread] Manage client started");
3944 rcu_register_thread();
3946 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
3948 health_code_update();
3950 ret
= lttcomm_listen_unix_sock(client_sock
);
3956 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
3957 * more will be added to this poll set.
3959 ret
= sessiond_set_thread_pollset(&events
, 2);
3961 goto error_create_poll
;
3964 /* Add the application registration socket */
3965 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
3970 sessiond_notify_ready();
3971 ret
= sem_post(&load_info
->message_thread_ready
);
3973 PERROR("sem_post message_thread_ready");
3977 /* This testpoint is after we signal readiness to the parent. */
3978 if (testpoint(sessiond_thread_manage_clients
)) {
3982 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
3986 health_code_update();
3989 DBG("Accepting client command ...");
3991 /* Inifinite blocking call, waiting for transmission */
3993 health_poll_entry();
3994 ret
= lttng_poll_wait(&events
, -1);
3998 * Restart interrupted system call.
4000 if (errno
== EINTR
) {
4008 for (i
= 0; i
< nb_fd
; i
++) {
4009 /* Fetch once the poll data */
4010 revents
= LTTNG_POLL_GETEV(&events
, i
);
4011 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4013 health_code_update();
4015 /* Thread quit pipe has been closed. Killing thread. */
4016 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4022 /* Event on the registration socket */
4023 if (pollfd
== client_sock
) {
4024 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4025 ERR("Client socket poll error");
4031 DBG("Wait for client response");
4033 health_code_update();
4035 sock
= lttcomm_accept_unix_sock(client_sock
);
4041 * Set the CLOEXEC flag. Return code is useless because either way, the
4044 (void) utils_set_fd_cloexec(sock
);
4046 /* Set socket option for credentials retrieval */
4047 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4052 /* Allocate context command to process the client request */
4053 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4054 if (cmd_ctx
== NULL
) {
4055 PERROR("zmalloc cmd_ctx");
4059 /* Allocate data buffer for reception */
4060 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4061 if (cmd_ctx
->lsm
== NULL
) {
4062 PERROR("zmalloc cmd_ctx->lsm");
4066 cmd_ctx
->llm
= NULL
;
4067 cmd_ctx
->session
= NULL
;
4069 health_code_update();
4072 * Data is received from the lttng client. The struct
4073 * lttcomm_session_msg (lsm) contains the command and data request of
4076 DBG("Receiving data from client ...");
4077 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4078 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4080 DBG("Nothing recv() from client... continuing");
4086 clean_command_ctx(&cmd_ctx
);
4090 health_code_update();
4092 // TODO: Validate cmd_ctx including sanity check for
4093 // security purpose.
4095 rcu_thread_online();
4097 * This function dispatch the work to the kernel or userspace tracer
4098 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4099 * informations for the client. The command context struct contains
4100 * everything this function may needs.
4102 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4103 rcu_thread_offline();
4111 * TODO: Inform client somehow of the fatal error. At
4112 * this point, ret < 0 means that a zmalloc failed
4113 * (ENOMEM). Error detected but still accept
4114 * command, unless a socket error has been
4117 clean_command_ctx(&cmd_ctx
);
4121 health_code_update();
4123 DBG("Sending response (size: %d, retcode: %s)",
4124 cmd_ctx
->lttng_msg_size
,
4125 lttng_strerror(-cmd_ctx
->llm
->ret_code
));
4126 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4128 ERR("Failed to send data back to client");
4131 /* End of transmission */
4138 clean_command_ctx(&cmd_ctx
);
4140 health_code_update();
4152 lttng_poll_clean(&events
);
4153 clean_command_ctx(&cmd_ctx
);
4157 unlink(client_unix_sock_path
);
4158 if (client_sock
>= 0) {
4159 ret
= close(client_sock
);
4167 ERR("Health error occurred in %s", __func__
);
4170 health_unregister(health_sessiond
);
4172 DBG("Client thread dying");
4174 rcu_unregister_thread();
4180 * usage function on stderr
4182 static void usage(void)
4184 fprintf(stderr
, "Usage: %s OPTIONS\n\nOptions:\n", progname
);
4185 fprintf(stderr
, " -h, --help Display this usage.\n");
4186 fprintf(stderr
, " -c, --client-sock PATH Specify path for the client unix socket\n");
4187 fprintf(stderr
, " -a, --apps-sock PATH Specify path for apps unix socket\n");
4188 fprintf(stderr
, " --kconsumerd-err-sock PATH Specify path for the kernel consumer error socket\n");
4189 fprintf(stderr
, " --kconsumerd-cmd-sock PATH Specify path for the kernel consumer command socket\n");
4190 fprintf(stderr
, " --ustconsumerd32-err-sock PATH Specify path for the 32-bit UST consumer error socket\n");
4191 fprintf(stderr
, " --ustconsumerd64-err-sock PATH Specify path for the 64-bit UST consumer error socket\n");
4192 fprintf(stderr
, " --ustconsumerd32-cmd-sock PATH Specify path for the 32-bit UST consumer command socket\n");
4193 fprintf(stderr
, " --ustconsumerd64-cmd-sock PATH Specify path for the 64-bit UST consumer command socket\n");
4194 fprintf(stderr
, " --consumerd32-path PATH Specify path for the 32-bit UST consumer daemon binary\n");
4195 fprintf(stderr
, " --consumerd32-libdir PATH Specify path for the 32-bit UST consumer daemon libraries\n");
4196 fprintf(stderr
, " --consumerd64-path PATH Specify path for the 64-bit UST consumer daemon binary\n");
4197 fprintf(stderr
, " --consumerd64-libdir PATH Specify path for the 64-bit UST consumer daemon libraries\n");
4198 fprintf(stderr
, " -d, --daemonize Start as a daemon.\n");
4199 fprintf(stderr
, " -b, --background Start as a daemon, keeping console open.\n");
4200 fprintf(stderr
, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
4201 fprintf(stderr
, " -V, --version Show version number.\n");
4202 fprintf(stderr
, " -S, --sig-parent Send SIGUSR1 to parent pid to notify readiness.\n");
4203 fprintf(stderr
, " -q, --quiet No output at all.\n");
4204 fprintf(stderr
, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
4205 fprintf(stderr
, " -p, --pidfile FILE Write a pid to FILE name overriding the default value.\n");
4206 fprintf(stderr
, " --verbose-consumer Verbose mode for consumer. Activate DBG() macro.\n");
4207 fprintf(stderr
, " --no-kernel Disable kernel tracer\n");
4208 fprintf(stderr
, " --jul-tcp-port JUL application registration TCP port\n");
4209 fprintf(stderr
, " -f --config Load daemon configuration file\n");
4210 fprintf(stderr
, " -l --load PATH Load session configuration\n");
4211 fprintf(stderr
, " --kmod-probes Specify kernel module probes to load\n");
4215 * Take an option from the getopt output and set it in the right variable to be
4218 * Return 0 on success else a negative value.
4220 static int set_option(int opt
, const char *arg
, const char *optname
)
4226 fprintf(stderr
, "option %s", optname
);
4228 fprintf(stderr
, " with arg %s\n", arg
);
4232 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4235 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4245 * If the override option is set, the pointer points to a
4246 * *non* const thus freeing it even though the variable type is
4249 if (tracing_group_name_override
) {
4250 free((void *) tracing_group_name
);
4252 tracing_group_name
= strdup(arg
);
4253 if (!tracing_group_name
) {
4257 tracing_group_name_override
= 1;
4263 fprintf(stdout
, "%s\n", VERSION
);
4269 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4272 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4275 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4278 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4281 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4284 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4290 lttng_opt_quiet
= 1;
4293 /* Verbose level can increase using multiple -v */
4295 /* Value obtained from config file */
4296 lttng_opt_verbose
= config_parse_value(arg
);
4298 /* -v used on command line */
4299 lttng_opt_verbose
++;
4301 /* Clamp value to [0, 3] */
4302 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4303 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4307 opt_verbose_consumer
= config_parse_value(arg
);
4309 opt_verbose_consumer
+= 1;
4313 if (consumerd32_bin_override
) {
4314 free((void *) consumerd32_bin
);
4316 consumerd32_bin
= strdup(arg
);
4317 if (!consumerd32_bin
) {
4321 consumerd32_bin_override
= 1;
4324 if (consumerd32_libdir_override
) {
4325 free((void *) consumerd32_libdir
);
4327 consumerd32_libdir
= strdup(arg
);
4328 if (!consumerd32_libdir
) {
4332 consumerd32_libdir_override
= 1;
4335 if (consumerd64_bin_override
) {
4336 free((void *) consumerd64_bin
);
4338 consumerd64_bin
= strdup(arg
);
4339 if (!consumerd64_bin
) {
4343 consumerd64_bin_override
= 1;
4346 if (consumerd64_libdir_override
) {
4347 free((void *) consumerd64_libdir
);
4349 consumerd64_libdir
= strdup(arg
);
4350 if (!consumerd64_libdir
) {
4354 consumerd64_libdir_override
= 1;
4358 opt_pidfile
= strdup(arg
);
4364 case 'J': /* JUL TCP port. */
4369 v
= strtoul(arg
, NULL
, 0);
4370 if (errno
!= 0 || !isdigit(arg
[0])) {
4371 ERR("Wrong value in --jul-tcp-port parameter: %s", arg
);
4374 if (v
== 0 || v
>= 65535) {
4375 ERR("Port overflow in --jul-tcp-port parameter: %s", arg
);
4378 jul_tcp_port
= (uint32_t) v
;
4379 DBG3("JUL TCP port set to non default: %u", jul_tcp_port
);
4383 free(opt_load_session_path
);
4384 opt_load_session_path
= strdup(arg
);
4385 if (!opt_load_session_path
) {
4390 case 'P': /* probe modules list */
4391 free(kmod_probes_list
);
4392 kmod_probes_list
= strdup(arg
);
4393 if (!kmod_probes_list
) {
4399 /* This is handled in set_options() thus silent break. */
4402 /* Unknown option or other error.
4403 * Error is printed by getopt, just return */
4411 * config_entry_handler_cb used to handle options read from a config file.
4412 * See config_entry_handler_cb comment in common/config/config.h for the
4413 * return value conventions.
4415 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
4419 if (!entry
|| !entry
->name
|| !entry
->value
) {
4424 /* Check if the option is to be ignored */
4425 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
4426 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
4431 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
4434 /* Ignore if not fully matched. */
4435 if (strcmp(entry
->name
, long_options
[i
].name
)) {
4440 * If the option takes no argument on the command line, we have to
4441 * check if the value is "true". We support non-zero numeric values,
4444 if (!long_options
[i
].has_arg
) {
4445 ret
= config_parse_value(entry
->value
);
4448 WARN("Invalid configuration value \"%s\" for option %s",
4449 entry
->value
, entry
->name
);
4451 /* False, skip boolean config option. */
4456 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
4460 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
4467 * daemon configuration loading and argument parsing
4469 static int set_options(int argc
, char **argv
)
4471 int ret
= 0, c
= 0, option_index
= 0;
4472 int orig_optopt
= optopt
, orig_optind
= optind
;
4474 const char *config_path
= NULL
;
4476 optstring
= utils_generate_optstring(long_options
,
4477 sizeof(long_options
) / sizeof(struct option
));
4483 /* Check for the --config option */
4484 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
4485 &option_index
)) != -1) {
4489 } else if (c
!= 'f') {
4490 /* if not equal to --config option. */
4494 config_path
= utils_expand_path(optarg
);
4496 ERR("Failed to resolve path: %s", optarg
);
4500 ret
= config_get_section_entries(config_path
, config_section_name
,
4501 config_entry_handler
, NULL
);
4504 ERR("Invalid configuration option at line %i", ret
);
4510 /* Reset getopt's global state */
4511 optopt
= orig_optopt
;
4512 optind
= orig_optind
;
4514 c
= getopt_long(argc
, argv
, optstring
, long_options
, &option_index
);
4519 ret
= set_option(c
, optarg
, long_options
[option_index
].name
);
4531 * Creates the two needed socket by the daemon.
4532 * apps_sock - The communication socket for all UST apps.
4533 * client_sock - The communication of the cli tool (lttng).
4535 static int init_daemon_socket(void)
4540 old_umask
= umask(0);
4542 /* Create client tool unix socket */
4543 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
4544 if (client_sock
< 0) {
4545 ERR("Create unix sock failed: %s", client_unix_sock_path
);
4550 /* Set the cloexec flag */
4551 ret
= utils_set_fd_cloexec(client_sock
);
4553 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
4554 "Continuing but note that the consumer daemon will have a "
4555 "reference to this socket on exec()", client_sock
);
4558 /* File permission MUST be 660 */
4559 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4561 ERR("Set file permissions failed: %s", client_unix_sock_path
);
4566 /* Create the application unix socket */
4567 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
4568 if (apps_sock
< 0) {
4569 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
4574 /* Set the cloexec flag */
4575 ret
= utils_set_fd_cloexec(apps_sock
);
4577 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
4578 "Continuing but note that the consumer daemon will have a "
4579 "reference to this socket on exec()", apps_sock
);
4582 /* File permission MUST be 666 */
4583 ret
= chmod(apps_unix_sock_path
,
4584 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
4586 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
4591 DBG3("Session daemon client socket %d and application socket %d created",
4592 client_sock
, apps_sock
);
4600 * Check if the global socket is available, and if a daemon is answering at the
4601 * other side. If yes, error is returned.
4603 static int check_existing_daemon(void)
4605 /* Is there anybody out there ? */
4606 if (lttng_session_daemon_alive()) {
4614 * Set the tracing group gid onto the client socket.
4616 * Race window between mkdir and chown is OK because we are going from more
4617 * permissive (root.root) to less permissive (root.tracing).
4619 static int set_permissions(char *rundir
)
4624 gid
= utils_get_group_id(tracing_group_name
);
4626 /* Set lttng run dir */
4627 ret
= chown(rundir
, 0, gid
);
4629 ERR("Unable to set group on %s", rundir
);
4634 * Ensure all applications and tracing group can search the run
4635 * dir. Allow everyone to read the directory, since it does not
4636 * buy us anything to hide its content.
4638 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
4640 ERR("Unable to set permissions on %s", rundir
);
4644 /* lttng client socket path */
4645 ret
= chown(client_unix_sock_path
, 0, gid
);
4647 ERR("Unable to set group on %s", client_unix_sock_path
);
4651 /* kconsumer error socket path */
4652 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
4654 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
4658 /* 64-bit ustconsumer error socket path */
4659 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
4661 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
4665 /* 32-bit ustconsumer compat32 error socket path */
4666 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
4668 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
4672 DBG("All permissions are set");
4678 * Create the lttng run directory needed for all global sockets and pipe.
4680 static int create_lttng_rundir(const char *rundir
)
4684 DBG3("Creating LTTng run directory: %s", rundir
);
4686 ret
= mkdir(rundir
, S_IRWXU
);
4688 if (errno
!= EEXIST
) {
4689 ERR("Unable to create %s", rundir
);
4701 * Setup sockets and directory needed by the kconsumerd communication with the
4704 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
4708 char path
[PATH_MAX
];
4710 switch (consumer_data
->type
) {
4711 case LTTNG_CONSUMER_KERNEL
:
4712 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
4714 case LTTNG_CONSUMER64_UST
:
4715 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
4717 case LTTNG_CONSUMER32_UST
:
4718 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
4721 ERR("Consumer type unknown");
4726 DBG2("Creating consumer directory: %s", path
);
4728 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
4730 if (errno
!= EEXIST
) {
4732 ERR("Failed to create %s", path
);
4738 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
4740 ERR("Unable to set group on %s", path
);
4746 /* Create the kconsumerd error unix socket */
4747 consumer_data
->err_sock
=
4748 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
4749 if (consumer_data
->err_sock
< 0) {
4750 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
4756 * Set the CLOEXEC flag. Return code is useless because either way, the
4759 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
4761 PERROR("utils_set_fd_cloexec");
4762 /* continue anyway */
4765 /* File permission MUST be 660 */
4766 ret
= chmod(consumer_data
->err_unix_sock_path
,
4767 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4769 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
4779 * Signal handler for the daemon
4781 * Simply stop all worker threads, leaving main() return gracefully after
4782 * joining all threads and calling cleanup().
4784 static void sighandler(int sig
)
4788 DBG("SIGPIPE caught");
4791 DBG("SIGINT caught");
4795 DBG("SIGTERM caught");
4799 CMM_STORE_SHARED(recv_child_signal
, 1);
4807 * Setup signal handler for :
4808 * SIGINT, SIGTERM, SIGPIPE
4810 static int set_signal_handler(void)
4813 struct sigaction sa
;
4816 if ((ret
= sigemptyset(&sigset
)) < 0) {
4817 PERROR("sigemptyset");
4821 sa
.sa_handler
= sighandler
;
4822 sa
.sa_mask
= sigset
;
4824 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
4825 PERROR("sigaction");
4829 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
4830 PERROR("sigaction");
4834 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
4835 PERROR("sigaction");
4839 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
4840 PERROR("sigaction");
4844 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
4850 * Set open files limit to unlimited. This daemon can open a large number of
4851 * file descriptors in order to consumer multiple kernel traces.
4853 static void set_ulimit(void)
4858 /* The kernel does not allowed an infinite limit for open files */
4859 lim
.rlim_cur
= 65535;
4860 lim
.rlim_max
= 65535;
4862 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
4864 PERROR("failed to set open files limit");
4869 * Write pidfile using the rundir and opt_pidfile.
4871 static void write_pidfile(void)
4874 char pidfile_path
[PATH_MAX
];
4879 strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
));
4881 /* Build pidfile path from rundir and opt_pidfile. */
4882 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
4883 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
4885 PERROR("snprintf pidfile path");
4891 * Create pid file in rundir. Return value is of no importance. The
4892 * execution will continue even though we are not able to write the file.
4894 (void) utils_create_pid_file(getpid(), pidfile_path
);
4901 * Create lockfile using the rundir and return its fd.
4903 static int create_lockfile(void)
4906 char lockfile_path
[PATH_MAX
];
4908 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
4913 ret
= utils_create_lock_file(lockfile_path
);
4919 * Write JUL TCP port using the rundir.
4921 static void write_julport(void)
4924 char path
[PATH_MAX
];
4928 ret
= snprintf(path
, sizeof(path
), "%s/"
4929 DEFAULT_LTTNG_SESSIOND_JULPORT_FILE
, rundir
);
4931 PERROR("snprintf julport path");
4936 * Create TCP JUL port file in rundir. Return value is of no importance.
4937 * The execution will continue even though we are not able to write the
4940 (void) utils_create_pid_file(jul_tcp_port
, path
);
4947 * Start the load session thread and dettach from it so the main thread can
4948 * continue. This does not return a value since whatever the outcome, the main
4949 * thread will continue.
4951 static void start_load_session_thread(void)
4955 /* Create session loading thread. */
4956 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
4959 PERROR("pthread_create load_session_thread");
4963 ret
= pthread_detach(load_session_thread
);
4965 PERROR("pthread_detach load_session_thread");
4968 /* Everything went well so don't cleanup anything. */
4971 /* The cleanup() function will destroy the load_info data. */
4978 int main(int argc
, char **argv
)
4982 const char *home_path
, *env_app_timeout
;
4984 init_kernel_workarounds();
4986 rcu_register_thread();
4988 if ((ret
= set_signal_handler()) < 0) {
4992 setup_consumerd_path();
4994 page_size
= sysconf(_SC_PAGESIZE
);
4995 if (page_size
< 0) {
4996 PERROR("sysconf _SC_PAGESIZE");
4997 page_size
= LONG_MAX
;
4998 WARN("Fallback page size to %ld", page_size
);
5001 /* Parse arguments and load the daemon configuration file */
5003 if ((ret
= set_options(argc
, argv
)) < 0) {
5008 if (opt_daemon
|| opt_background
) {
5011 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5018 * We are in the child. Make sure all other file descriptors are
5019 * closed, in case we are called with more opened file descriptors than
5020 * the standard ones.
5022 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5027 /* Create thread quit pipe */
5028 if ((ret
= init_thread_quit_pipe()) < 0) {
5032 /* Check if daemon is UID = 0 */
5033 is_root
= !getuid();
5036 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5038 /* Create global run dir with root access */
5039 ret
= create_lttng_rundir(rundir
);
5044 if (strlen(apps_unix_sock_path
) == 0) {
5045 snprintf(apps_unix_sock_path
, PATH_MAX
,
5046 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5049 if (strlen(client_unix_sock_path
) == 0) {
5050 snprintf(client_unix_sock_path
, PATH_MAX
,
5051 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5054 /* Set global SHM for ust */
5055 if (strlen(wait_shm_path
) == 0) {
5056 snprintf(wait_shm_path
, PATH_MAX
,
5057 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5060 if (strlen(health_unix_sock_path
) == 0) {
5061 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
5062 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5065 /* Setup kernel consumerd path */
5066 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5067 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5068 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5069 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5071 DBG2("Kernel consumer err path: %s",
5072 kconsumer_data
.err_unix_sock_path
);
5073 DBG2("Kernel consumer cmd path: %s",
5074 kconsumer_data
.cmd_unix_sock_path
);
5076 home_path
= utils_get_home_dir();
5077 if (home_path
== NULL
) {
5078 /* TODO: Add --socket PATH option */
5079 ERR("Can't get HOME directory for sockets creation.");
5085 * Create rundir from home path. This will create something like
5088 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5094 ret
= create_lttng_rundir(rundir
);
5099 if (strlen(apps_unix_sock_path
) == 0) {
5100 snprintf(apps_unix_sock_path
, PATH_MAX
,
5101 DEFAULT_HOME_APPS_UNIX_SOCK
, home_path
);
5104 /* Set the cli tool unix socket path */
5105 if (strlen(client_unix_sock_path
) == 0) {
5106 snprintf(client_unix_sock_path
, PATH_MAX
,
5107 DEFAULT_HOME_CLIENT_UNIX_SOCK
, home_path
);
5110 /* Set global SHM for ust */
5111 if (strlen(wait_shm_path
) == 0) {
5112 snprintf(wait_shm_path
, PATH_MAX
,
5113 DEFAULT_HOME_APPS_WAIT_SHM_PATH
, getuid());
5116 /* Set health check Unix path */
5117 if (strlen(health_unix_sock_path
) == 0) {
5118 snprintf(health_unix_sock_path
, sizeof(health_unix_sock_path
),
5119 DEFAULT_HOME_HEALTH_UNIX_SOCK
, home_path
);
5123 lockfile_fd
= create_lockfile();
5124 if (lockfile_fd
< 0) {
5128 /* Set consumer initial state */
5129 kernel_consumerd_state
= CONSUMER_STOPPED
;
5130 ust_consumerd_state
= CONSUMER_STOPPED
;
5132 DBG("Client socket path %s", client_unix_sock_path
);
5133 DBG("Application socket path %s", apps_unix_sock_path
);
5134 DBG("Application wait path %s", wait_shm_path
);
5135 DBG("LTTng run directory path: %s", rundir
);
5137 /* 32 bits consumerd path setup */
5138 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5139 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5140 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5141 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5143 DBG2("UST consumer 32 bits err path: %s",
5144 ustconsumer32_data
.err_unix_sock_path
);
5145 DBG2("UST consumer 32 bits cmd path: %s",
5146 ustconsumer32_data
.cmd_unix_sock_path
);
5148 /* 64 bits consumerd path setup */
5149 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5150 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5151 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5152 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5154 DBG2("UST consumer 64 bits err path: %s",
5155 ustconsumer64_data
.err_unix_sock_path
);
5156 DBG2("UST consumer 64 bits cmd path: %s",
5157 ustconsumer64_data
.cmd_unix_sock_path
);
5160 * See if daemon already exist.
5162 if ((ret
= check_existing_daemon()) < 0) {
5163 ERR("Already running daemon.\n");
5165 * We do not goto exit because we must not cleanup()
5166 * because a daemon is already running.
5172 * Init UST app hash table. Alloc hash table before this point since
5173 * cleanup() can get called after that point.
5177 /* Initialize JUL domain subsystem. */
5178 if ((ret
= jul_init()) < 0) {
5179 /* ENOMEM at this point. */
5183 /* After this point, we can safely call cleanup() with "goto exit" */
5186 * These actions must be executed as root. We do that *after* setting up
5187 * the sockets path because we MUST make the check for another daemon using
5188 * those paths *before* trying to set the kernel consumer sockets and init
5192 ret
= set_consumer_sockets(&kconsumer_data
, rundir
);
5197 /* Setup kernel tracer */
5198 if (!opt_no_kernel
) {
5199 init_kernel_tracer();
5202 /* Set ulimit for open files */
5205 /* init lttng_fd tracking must be done after set_ulimit. */
5208 ret
= set_consumer_sockets(&ustconsumer64_data
, rundir
);
5213 ret
= set_consumer_sockets(&ustconsumer32_data
, rundir
);
5218 /* Setup the needed unix socket */
5219 if ((ret
= init_daemon_socket()) < 0) {
5223 /* Set credentials to socket */
5224 if (is_root
&& ((ret
= set_permissions(rundir
)) < 0)) {
5228 /* Get parent pid if -S, --sig-parent is specified. */
5229 if (opt_sig_parent
) {
5233 /* Setup the kernel pipe for waking up the kernel thread */
5234 if (is_root
&& !opt_no_kernel
) {
5235 if ((ret
= utils_create_pipe_cloexec(kernel_poll_pipe
)) < 0) {
5240 /* Setup the thread ht_cleanup communication pipe. */
5241 if (utils_create_pipe_cloexec(ht_cleanup_pipe
) < 0) {
5245 /* Setup the thread apps communication pipe. */
5246 if ((ret
= utils_create_pipe_cloexec(apps_cmd_pipe
)) < 0) {
5250 /* Setup the thread apps notify communication pipe. */
5251 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
) < 0) {
5255 /* Initialize global buffer per UID and PID registry. */
5256 buffer_reg_init_uid_registry();
5257 buffer_reg_init_pid_registry();
5259 /* Init UST command queue. */
5260 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
5263 * Get session list pointer. This pointer MUST NOT be free(). This list is
5264 * statically declared in session.c
5266 session_list_ptr
= session_get_list();
5268 /* Set up max poll set size */
5269 lttng_poll_set_max_size();
5273 /* Check for the application socket timeout env variable. */
5274 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
5275 if (env_app_timeout
) {
5276 app_socket_timeout
= atoi(env_app_timeout
);
5278 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
5284 /* Initialize communication library */
5286 /* This is to get the TCP timeout value. */
5287 lttcomm_inet_init();
5289 if (load_session_init_data(&load_info
) < 0) {
5292 load_info
->path
= opt_load_session_path
;
5295 * Initialize the health check subsystem. This call should set the
5296 * appropriate time values.
5298 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5299 if (!health_sessiond
) {
5300 PERROR("health_app_create error");
5301 goto exit_health_sessiond_cleanup
;
5304 /* Create thread to clean up RCU hash tables */
5305 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5306 thread_ht_cleanup
, (void *) NULL
);
5308 PERROR("pthread_create ht_cleanup");
5309 goto exit_ht_cleanup
;
5312 /* Create health-check thread */
5313 ret
= pthread_create(&health_thread
, NULL
,
5314 thread_manage_health
, (void *) NULL
);
5316 PERROR("pthread_create health");
5320 /* Create thread to manage the client socket */
5321 ret
= pthread_create(&client_thread
, NULL
,
5322 thread_manage_clients
, (void *) NULL
);
5324 PERROR("pthread_create clients");
5328 /* Create thread to dispatch registration */
5329 ret
= pthread_create(&dispatch_thread
, NULL
,
5330 thread_dispatch_ust_registration
, (void *) NULL
);
5332 PERROR("pthread_create dispatch");
5336 /* Create thread to manage application registration. */
5337 ret
= pthread_create(®_apps_thread
, NULL
,
5338 thread_registration_apps
, (void *) NULL
);
5340 PERROR("pthread_create registration");
5344 /* Create thread to manage application socket */
5345 ret
= pthread_create(&apps_thread
, NULL
,
5346 thread_manage_apps
, (void *) NULL
);
5348 PERROR("pthread_create apps");
5352 /* Create thread to manage application notify socket */
5353 ret
= pthread_create(&apps_notify_thread
, NULL
,
5354 ust_thread_manage_notify
, (void *) NULL
);
5356 PERROR("pthread_create notify");
5357 goto exit_apps_notify
;
5360 /* Create JUL registration thread. */
5361 ret
= pthread_create(&jul_reg_thread
, NULL
,
5362 jul_thread_manage_registration
, (void *) NULL
);
5364 PERROR("pthread_create JUL");
5368 /* Don't start this thread if kernel tracing is not requested nor root */
5369 if (is_root
&& !opt_no_kernel
) {
5370 /* Create kernel thread to manage kernel event */
5371 ret
= pthread_create(&kernel_thread
, NULL
,
5372 thread_manage_kernel
, (void *) NULL
);
5374 PERROR("pthread_create kernel");
5379 /* Load possible session(s). */
5380 start_load_session_thread();
5382 if (is_root
&& !opt_no_kernel
) {
5383 ret
= pthread_join(kernel_thread
, &status
);
5385 PERROR("pthread_join");
5386 goto error
; /* join error, exit without cleanup */
5391 ret
= pthread_join(jul_reg_thread
, &status
);
5393 PERROR("pthread_join JUL");
5394 goto error
; /* join error, exit without cleanup */
5398 ret
= pthread_join(apps_notify_thread
, &status
);
5400 PERROR("pthread_join apps notify");
5401 goto error
; /* join error, exit without cleanup */
5405 ret
= pthread_join(apps_thread
, &status
);
5407 PERROR("pthread_join apps");
5408 goto error
; /* join error, exit without cleanup */
5413 ret
= pthread_join(reg_apps_thread
, &status
);
5415 PERROR("pthread_join");
5416 goto error
; /* join error, exit without cleanup */
5420 ret
= pthread_join(dispatch_thread
, &status
);
5422 PERROR("pthread_join");
5423 goto error
; /* join error, exit without cleanup */
5427 ret
= pthread_join(client_thread
, &status
);
5429 PERROR("pthread_join");
5430 goto error
; /* join error, exit without cleanup */
5433 ret
= join_consumer_thread(&kconsumer_data
);
5435 PERROR("join_consumer");
5436 goto error
; /* join error, exit without cleanup */
5439 ret
= join_consumer_thread(&ustconsumer32_data
);
5441 PERROR("join_consumer ust32");
5442 goto error
; /* join error, exit without cleanup */
5445 ret
= join_consumer_thread(&ustconsumer64_data
);
5447 PERROR("join_consumer ust64");
5448 goto error
; /* join error, exit without cleanup */
5452 ret
= pthread_join(health_thread
, &status
);
5454 PERROR("pthread_join health thread");
5455 goto error
; /* join error, exit without cleanup */
5459 ret
= pthread_join(ht_cleanup_thread
, &status
);
5461 PERROR("pthread_join ht cleanup thread");
5462 goto error
; /* join error, exit without cleanup */
5465 health_app_destroy(health_sessiond
);
5466 exit_health_sessiond_cleanup
:
5469 * cleanup() is called when no other thread is running.
5471 rcu_thread_online();
5473 rcu_thread_offline();
5474 rcu_unregister_thread();