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/compat/getenv.h>
45 #include <common/defaults.h>
46 #include <common/kernel-consumer/kernel-consumer.h>
47 #include <common/futex.h>
48 #include <common/relayd/relayd.h>
49 #include <common/utils.h>
50 #include <common/daemonize.h>
51 #include <common/config/session-config.h>
53 #include "lttng-sessiond.h"
54 #include "buffer-registry.h"
61 #include "kernel-consumer.h"
65 #include "ust-consumer.h"
68 #include "health-sessiond.h"
69 #include "testpoint.h"
70 #include "ust-thread.h"
71 #include "agent-thread.h"
73 #include "load-session-thread.h"
77 #define CONSUMERD_FILE "lttng-consumerd"
80 static const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
81 static int tracing_group_name_override
;
82 static char *opt_pidfile
;
83 static int opt_sig_parent
;
84 static int opt_verbose_consumer
;
85 static int opt_daemon
, opt_background
;
86 static int opt_no_kernel
;
87 static char *opt_load_session_path
;
88 static pid_t ppid
; /* Parent PID for --sig-parent option */
89 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
91 static int lockfile_fd
= -1;
93 /* Set to 1 when a SIGUSR1 signal is received. */
94 static int recv_child_signal
;
97 * Consumer daemon specific control data. Every value not initialized here is
98 * set to 0 by the static definition.
100 static struct consumer_data kconsumer_data
= {
101 .type
= LTTNG_CONSUMER_KERNEL
,
102 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
103 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
106 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
107 .lock
= PTHREAD_MUTEX_INITIALIZER
,
108 .cond
= PTHREAD_COND_INITIALIZER
,
109 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
111 static struct consumer_data ustconsumer64_data
= {
112 .type
= LTTNG_CONSUMER64_UST
,
113 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
114 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
117 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
118 .lock
= PTHREAD_MUTEX_INITIALIZER
,
119 .cond
= PTHREAD_COND_INITIALIZER
,
120 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
122 static struct consumer_data ustconsumer32_data
= {
123 .type
= LTTNG_CONSUMER32_UST
,
124 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
125 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
128 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
129 .lock
= PTHREAD_MUTEX_INITIALIZER
,
130 .cond
= PTHREAD_COND_INITIALIZER
,
131 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
134 /* Command line options */
135 static const struct option long_options
[] = {
136 { "client-sock", required_argument
, 0, 'c' },
137 { "apps-sock", required_argument
, 0, 'a' },
138 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
139 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
140 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
141 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
142 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
143 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
144 { "consumerd32-path", required_argument
, 0, '\0' },
145 { "consumerd32-libdir", required_argument
, 0, '\0' },
146 { "consumerd64-path", required_argument
, 0, '\0' },
147 { "consumerd64-libdir", required_argument
, 0, '\0' },
148 { "daemonize", no_argument
, 0, 'd' },
149 { "background", no_argument
, 0, 'b' },
150 { "sig-parent", no_argument
, 0, 'S' },
151 { "help", no_argument
, 0, 'h' },
152 { "group", required_argument
, 0, 'g' },
153 { "version", no_argument
, 0, 'V' },
154 { "quiet", no_argument
, 0, 'q' },
155 { "verbose", no_argument
, 0, 'v' },
156 { "verbose-consumer", no_argument
, 0, '\0' },
157 { "no-kernel", no_argument
, 0, '\0' },
158 { "pidfile", required_argument
, 0, 'p' },
159 { "agent-tcp-port", required_argument
, 0, '\0' },
160 { "config", required_argument
, 0, 'f' },
161 { "load", required_argument
, 0, 'l' },
162 { "kmod-probes", required_argument
, 0, '\0' },
163 { "extra-kmod-probes", required_argument
, 0, '\0' },
167 /* Command line options to ignore from configuration file */
168 static const char *config_ignore_options
[] = { "help", "version", "config" };
170 /* Shared between threads */
171 static int dispatch_thread_exit
;
173 /* Global application Unix socket path */
174 static char apps_unix_sock_path
[PATH_MAX
];
175 /* Global client Unix socket path */
176 static char client_unix_sock_path
[PATH_MAX
];
177 /* global wait shm path for UST */
178 static char wait_shm_path
[PATH_MAX
];
179 /* Global health check unix path */
180 static char health_unix_sock_path
[PATH_MAX
];
182 /* Sockets and FDs */
183 static int client_sock
= -1;
184 static int apps_sock
= -1;
185 int kernel_tracer_fd
= -1;
186 static int kernel_poll_pipe
[2] = { -1, -1 };
189 * Quit pipe for all threads. This permits a single cancellation point
190 * for all threads when receiving an event on the pipe.
192 static int thread_quit_pipe
[2] = { -1, -1 };
193 static int ht_cleanup_quit_pipe
[2] = { -1, -1 };
196 * This pipe is used to inform the thread managing application communication
197 * that a command is queued and ready to be processed.
199 static int apps_cmd_pipe
[2] = { -1, -1 };
201 int apps_cmd_notify_pipe
[2] = { -1, -1 };
203 /* Pthread, Mutexes and Semaphores */
204 static pthread_t apps_thread
;
205 static pthread_t apps_notify_thread
;
206 static pthread_t reg_apps_thread
;
207 static pthread_t client_thread
;
208 static pthread_t kernel_thread
;
209 static pthread_t dispatch_thread
;
210 static pthread_t health_thread
;
211 static pthread_t ht_cleanup_thread
;
212 static pthread_t agent_reg_thread
;
213 static pthread_t load_session_thread
;
216 * UST registration command queue. This queue is tied with a futex and uses a N
217 * wakers / 1 waiter implemented and detailed in futex.c/.h
219 * The thread_registration_apps and thread_dispatch_ust_registration uses this
220 * queue along with the wait/wake scheme. The thread_manage_apps receives down
221 * the line new application socket and monitors it for any I/O error or clean
222 * close that triggers an unregistration of the application.
224 static struct ust_cmd_queue ust_cmd_queue
;
227 * Pointer initialized before thread creation.
229 * This points to the tracing session list containing the session count and a
230 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
231 * MUST NOT be taken if you call a public function in session.c.
233 * The lock is nested inside the structure: session_list_ptr->lock. Please use
234 * session_lock_list and session_unlock_list for lock acquisition.
236 static struct ltt_session_list
*session_list_ptr
;
238 int ust_consumerd64_fd
= -1;
239 int ust_consumerd32_fd
= -1;
241 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
242 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
243 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
244 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
245 static int consumerd32_bin_override
;
246 static int consumerd64_bin_override
;
247 static int consumerd32_libdir_override
;
248 static int consumerd64_libdir_override
;
250 static const char *module_proc_lttng
= "/proc/lttng";
253 * Consumer daemon state which is changed when spawning it, killing it or in
254 * case of a fatal error.
256 enum consumerd_state
{
257 CONSUMER_STARTED
= 1,
258 CONSUMER_STOPPED
= 2,
263 * This consumer daemon state is used to validate if a client command will be
264 * able to reach the consumer. If not, the client is informed. For instance,
265 * doing a "lttng start" when the consumer state is set to ERROR will return an
266 * error to the client.
268 * The following example shows a possible race condition of this scheme:
270 * consumer thread error happens
272 * client cmd checks state -> still OK
273 * consumer thread exit, sets error
274 * client cmd try to talk to consumer
277 * However, since the consumer is a different daemon, we have no way of making
278 * sure the command will reach it safely even with this state flag. This is why
279 * we consider that up to the state validation during command processing, the
280 * command is safe. After that, we can not guarantee the correctness of the
281 * client request vis-a-vis the consumer.
283 static enum consumerd_state ust_consumerd_state
;
284 static enum consumerd_state kernel_consumerd_state
;
287 * Socket timeout for receiving and sending in seconds.
289 static int app_socket_timeout
;
291 /* Set in main() with the current page size. */
294 /* Application health monitoring */
295 struct health_app
*health_sessiond
;
297 /* Agent TCP port for registration. Used by the agent thread. */
298 unsigned int agent_tcp_port
= DEFAULT_AGENT_TCP_PORT
;
300 /* Am I root or not. */
301 int is_root
; /* Set to 1 if the daemon is running as root */
303 const char * const config_section_name
= "sessiond";
305 /* Load session thread information to operate. */
306 struct load_session_thread_data
*load_info
;
308 /* Global hash tables */
309 struct lttng_ht
*agent_apps_ht_by_sock
= NULL
;
312 * Whether sessiond is ready for commands/health check requests.
313 * NR_LTTNG_SESSIOND_READY must match the number of calls to
314 * sessiond_notify_ready().
316 #define NR_LTTNG_SESSIOND_READY 3
317 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
319 /* Notify parents that we are ready for cmd and health check */
321 void sessiond_notify_ready(void)
323 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
325 * Notify parent pid that we are ready to accept command
326 * for client side. This ppid is the one from the
327 * external process that spawned us.
329 if (opt_sig_parent
) {
334 * Notify the parent of the fork() process that we are
337 if (opt_daemon
|| opt_background
) {
338 kill(child_ppid
, SIGUSR1
);
344 void setup_consumerd_path(void)
346 const char *bin
, *libdir
;
349 * Allow INSTALL_BIN_PATH to be used as a target path for the
350 * native architecture size consumer if CONFIG_CONSUMER*_PATH
351 * has not been defined.
353 #if (CAA_BITS_PER_LONG == 32)
354 if (!consumerd32_bin
[0]) {
355 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
357 if (!consumerd32_libdir
[0]) {
358 consumerd32_libdir
= INSTALL_LIB_PATH
;
360 #elif (CAA_BITS_PER_LONG == 64)
361 if (!consumerd64_bin
[0]) {
362 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
364 if (!consumerd64_libdir
[0]) {
365 consumerd64_libdir
= INSTALL_LIB_PATH
;
368 #error "Unknown bitness"
372 * runtime env. var. overrides the build default.
374 bin
= lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
376 consumerd32_bin
= bin
;
378 bin
= lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
380 consumerd64_bin
= bin
;
382 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
384 consumerd32_libdir
= libdir
;
386 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
388 consumerd64_libdir
= libdir
;
393 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
400 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
406 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
418 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
420 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
422 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
426 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
428 int sessiond_set_ht_cleanup_thread_pollset(struct lttng_poll_event
*events
,
431 return __sessiond_set_thread_pollset(events
, size
,
432 ht_cleanup_quit_pipe
);
436 int __sessiond_check_thread_quit_pipe(int fd
, uint32_t events
, int a_pipe
)
438 if (fd
== a_pipe
&& (events
& LPOLLIN
)) {
445 * Check if the thread quit pipe was triggered.
447 * Return 1 if it was triggered else 0;
449 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
451 return __sessiond_check_thread_quit_pipe(fd
, events
,
452 thread_quit_pipe
[0]);
456 * Check if the ht_cleanup thread quit pipe was triggered.
458 * Return 1 if it was triggered else 0;
460 int sessiond_check_ht_cleanup_quit(int fd
, uint32_t events
)
462 return __sessiond_check_thread_quit_pipe(fd
, events
,
463 ht_cleanup_quit_pipe
[0]);
467 * Init thread quit pipe.
469 * Return -1 on error or 0 if all pipes are created.
471 static int __init_thread_quit_pipe(int *a_pipe
)
477 PERROR("thread quit pipe");
481 for (i
= 0; i
< 2; i
++) {
482 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
493 static int init_thread_quit_pipe(void)
495 return __init_thread_quit_pipe(thread_quit_pipe
);
498 static int init_ht_cleanup_quit_pipe(void)
500 return __init_thread_quit_pipe(ht_cleanup_quit_pipe
);
504 * Stop all threads by closing the thread quit pipe.
506 static void stop_threads(void)
510 /* Stopping all threads */
511 DBG("Terminating all threads");
512 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
514 ERR("write error on thread quit pipe");
517 /* Dispatch thread */
518 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
519 futex_nto1_wake(&ust_cmd_queue
.futex
);
523 * Close every consumer sockets.
525 static void close_consumer_sockets(void)
529 if (kconsumer_data
.err_sock
>= 0) {
530 ret
= close(kconsumer_data
.err_sock
);
532 PERROR("kernel consumer err_sock close");
535 if (ustconsumer32_data
.err_sock
>= 0) {
536 ret
= close(ustconsumer32_data
.err_sock
);
538 PERROR("UST consumerd32 err_sock close");
541 if (ustconsumer64_data
.err_sock
>= 0) {
542 ret
= close(ustconsumer64_data
.err_sock
);
544 PERROR("UST consumerd64 err_sock close");
547 if (kconsumer_data
.cmd_sock
>= 0) {
548 ret
= close(kconsumer_data
.cmd_sock
);
550 PERROR("kernel consumer cmd_sock close");
553 if (ustconsumer32_data
.cmd_sock
>= 0) {
554 ret
= close(ustconsumer32_data
.cmd_sock
);
556 PERROR("UST consumerd32 cmd_sock close");
559 if (ustconsumer64_data
.cmd_sock
>= 0) {
560 ret
= close(ustconsumer64_data
.cmd_sock
);
562 PERROR("UST consumerd64 cmd_sock close");
568 * Generate the full lock file path using the rundir.
570 * Return the snprintf() return value thus a negative value is an error.
572 static int generate_lock_file_path(char *path
, size_t len
)
579 /* Build lockfile path from rundir. */
580 ret
= snprintf(path
, len
, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE
, rundir
);
582 PERROR("snprintf lockfile path");
589 * Wait on consumer process termination.
591 * Need to be called with the consumer data lock held or from a context
592 * ensuring no concurrent access to data (e.g: cleanup).
594 static void wait_consumer(struct consumer_data
*consumer_data
)
599 if (consumer_data
->pid
<= 0) {
603 DBG("Waiting for complete teardown of consumerd (PID: %d)",
605 ret
= waitpid(consumer_data
->pid
, &status
, 0);
607 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
609 if (!WIFEXITED(status
)) {
610 ERR("consumerd termination with error: %d",
613 consumer_data
->pid
= 0;
617 * Cleanup the session daemon's data structures.
619 static void sessiond_cleanup(void)
622 struct ltt_session
*sess
, *stmp
;
625 DBG("Cleanup sessiond");
628 * Close the thread quit pipe. It has already done its job,
629 * since we are now called.
631 utils_close_pipe(thread_quit_pipe
);
634 * If opt_pidfile is undefined, the default file will be wiped when
635 * removing the rundir.
638 ret
= remove(opt_pidfile
);
640 PERROR("remove pidfile %s", opt_pidfile
);
644 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
647 snprintf(path
, PATH_MAX
,
649 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
650 DBG("Removing %s", path
);
653 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
654 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
);
655 DBG("Removing %s", path
);
659 snprintf(path
, PATH_MAX
,
660 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
662 DBG("Removing %s", path
);
665 snprintf(path
, PATH_MAX
,
666 DEFAULT_KCONSUMERD_PATH
,
668 DBG("Removing directory %s", path
);
671 /* ust consumerd 32 */
672 snprintf(path
, PATH_MAX
,
673 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
675 DBG("Removing %s", path
);
678 snprintf(path
, PATH_MAX
,
679 DEFAULT_USTCONSUMERD32_PATH
,
681 DBG("Removing directory %s", path
);
684 /* ust consumerd 64 */
685 snprintf(path
, PATH_MAX
,
686 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
688 DBG("Removing %s", path
);
691 snprintf(path
, PATH_MAX
,
692 DEFAULT_USTCONSUMERD64_PATH
,
694 DBG("Removing directory %s", path
);
697 DBG("Cleaning up all sessions");
699 /* Destroy session list mutex */
700 if (session_list_ptr
!= NULL
) {
701 pthread_mutex_destroy(&session_list_ptr
->lock
);
703 /* Cleanup ALL session */
704 cds_list_for_each_entry_safe(sess
, stmp
,
705 &session_list_ptr
->head
, list
) {
706 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
710 wait_consumer(&kconsumer_data
);
711 wait_consumer(&ustconsumer64_data
);
712 wait_consumer(&ustconsumer32_data
);
714 DBG("Cleaning up all agent apps");
715 agent_app_ht_clean();
717 DBG("Closing all UST sockets");
718 ust_app_clean_list();
719 buffer_reg_destroy_registries();
721 if (is_root
&& !opt_no_kernel
) {
722 DBG2("Closing kernel fd");
723 if (kernel_tracer_fd
>= 0) {
724 ret
= close(kernel_tracer_fd
);
729 DBG("Unloading kernel modules");
730 modprobe_remove_lttng_all();
734 close_consumer_sockets();
737 load_session_destroy_data(load_info
);
742 * Cleanup lock file by deleting it and finaly closing it which will
743 * release the file system lock.
745 if (lockfile_fd
>= 0) {
746 char lockfile_path
[PATH_MAX
];
748 ret
= generate_lock_file_path(lockfile_path
,
749 sizeof(lockfile_path
));
751 ret
= remove(lockfile_path
);
753 PERROR("remove lock file");
755 ret
= close(lockfile_fd
);
757 PERROR("close lock file");
763 * We do NOT rmdir rundir because there are other processes
764 * using it, for instance lttng-relayd, which can start in
765 * parallel with this teardown.
772 * Cleanup the daemon's option data structures.
774 static void sessiond_cleanup_options(void)
776 DBG("Cleaning up options");
779 * If the override option is set, the pointer points to a *non* const
780 * thus freeing it even though the variable type is set to const.
782 if (tracing_group_name_override
) {
783 free((void *) tracing_group_name
);
785 if (consumerd32_bin_override
) {
786 free((void *) consumerd32_bin
);
788 if (consumerd64_bin_override
) {
789 free((void *) consumerd64_bin
);
791 if (consumerd32_libdir_override
) {
792 free((void *) consumerd32_libdir
);
794 if (consumerd64_libdir_override
) {
795 free((void *) consumerd64_libdir
);
799 free(opt_load_session_path
);
800 free(kmod_probes_list
);
801 free(kmod_extra_probes_list
);
803 run_as_destroy_worker();
806 DBG("%c[%d;%dm*** assert failed :-) *** ==> %c[%dm%c[%d;%dm"
807 "Matthew, BEET driven development works!%c[%dm",
808 27, 1, 31, 27, 0, 27, 1, 33, 27, 0);
813 * Send data on a unix socket using the liblttsessiondcomm API.
815 * Return lttcomm error code.
817 static int send_unix_sock(int sock
, void *buf
, size_t len
)
819 /* Check valid length */
824 return lttcomm_send_unix_sock(sock
, buf
, len
);
828 * Free memory of a command context structure.
830 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
832 DBG("Clean command context structure");
834 if ((*cmd_ctx
)->llm
) {
835 free((*cmd_ctx
)->llm
);
837 if ((*cmd_ctx
)->lsm
) {
838 free((*cmd_ctx
)->lsm
);
846 * Notify UST applications using the shm mmap futex.
848 static int notify_ust_apps(int active
)
852 DBG("Notifying applications of session daemon state: %d", active
);
854 /* See shm.c for this call implying mmap, shm and futex calls */
855 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
856 if (wait_shm_mmap
== NULL
) {
860 /* Wake waiting process */
861 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
863 /* Apps notified successfully */
871 * Setup the outgoing data buffer for the response (llm) by allocating the
872 * right amount of memory and copying the original information from the lsm
875 * Return 0 on success, negative value on error.
877 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
,
878 const void *payload_buf
, size_t payload_len
,
879 const void *cmd_header_buf
, size_t cmd_header_len
)
882 const size_t header_len
= sizeof(struct lttcomm_lttng_msg
);
883 const size_t cmd_header_offset
= header_len
;
884 const size_t payload_offset
= cmd_header_offset
+ cmd_header_len
;
885 const size_t total_msg_size
= header_len
+ cmd_header_len
+ payload_len
;
887 cmd_ctx
->llm
= zmalloc(total_msg_size
);
889 if (cmd_ctx
->llm
== NULL
) {
895 /* Copy common data */
896 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
897 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
898 cmd_ctx
->llm
->cmd_header_size
= cmd_header_len
;
899 cmd_ctx
->llm
->data_size
= payload_len
;
900 cmd_ctx
->lttng_msg_size
= total_msg_size
;
902 /* Copy command header */
903 if (cmd_header_len
) {
904 memcpy(((uint8_t *) cmd_ctx
->llm
) + cmd_header_offset
, cmd_header_buf
,
910 memcpy(((uint8_t *) cmd_ctx
->llm
) + payload_offset
, payload_buf
,
919 * Version of setup_lttng_msg() without command header.
921 static int setup_lttng_msg_no_cmd_header(struct command_ctx
*cmd_ctx
,
922 void *payload_buf
, size_t payload_len
)
924 return setup_lttng_msg(cmd_ctx
, payload_buf
, payload_len
, NULL
, 0);
927 * Update the kernel poll set of all channel fd available over all tracing
928 * session. Add the wakeup pipe at the end of the set.
930 static int update_kernel_poll(struct lttng_poll_event
*events
)
933 struct ltt_session
*session
;
934 struct ltt_kernel_channel
*channel
;
936 DBG("Updating kernel poll set");
939 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
940 session_lock(session
);
941 if (session
->kernel_session
== NULL
) {
942 session_unlock(session
);
946 cds_list_for_each_entry(channel
,
947 &session
->kernel_session
->channel_list
.head
, list
) {
948 /* Add channel fd to the kernel poll set */
949 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
951 session_unlock(session
);
954 DBG("Channel fd %d added to kernel set", channel
->fd
);
956 session_unlock(session
);
958 session_unlock_list();
963 session_unlock_list();
968 * Find the channel fd from 'fd' over all tracing session. When found, check
969 * for new channel stream and send those stream fds to the kernel consumer.
971 * Useful for CPU hotplug feature.
973 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
976 struct ltt_session
*session
;
977 struct ltt_kernel_session
*ksess
;
978 struct ltt_kernel_channel
*channel
;
980 DBG("Updating kernel streams for channel fd %d", fd
);
983 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
984 session_lock(session
);
985 if (session
->kernel_session
== NULL
) {
986 session_unlock(session
);
989 ksess
= session
->kernel_session
;
991 cds_list_for_each_entry(channel
,
992 &ksess
->channel_list
.head
, list
) {
993 struct lttng_ht_iter iter
;
994 struct consumer_socket
*socket
;
996 if (channel
->fd
!= fd
) {
999 DBG("Channel found, updating kernel streams");
1000 ret
= kernel_open_channel_stream(channel
);
1004 /* Update the stream global counter */
1005 ksess
->stream_count_global
+= ret
;
1008 * Have we already sent fds to the consumer? If yes, it
1009 * means that tracing is started so it is safe to send
1010 * our updated stream fds.
1012 if (ksess
->consumer_fds_sent
!= 1
1013 || ksess
->consumer
== NULL
) {
1019 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
1020 &iter
.iter
, socket
, node
.node
) {
1021 pthread_mutex_lock(socket
->lock
);
1022 ret
= kernel_consumer_send_channel_stream(socket
,
1024 session
->output_traces
? 1 : 0);
1025 pthread_mutex_unlock(socket
->lock
);
1033 session_unlock(session
);
1035 session_unlock_list();
1039 session_unlock(session
);
1040 session_unlock_list();
1045 * For each tracing session, update newly registered apps. The session list
1046 * lock MUST be acquired before calling this.
1048 static void update_ust_app(int app_sock
)
1050 struct ltt_session
*sess
, *stmp
;
1052 /* Consumer is in an ERROR state. Stop any application update. */
1053 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
1054 /* Stop the update process since the consumer is dead. */
1058 /* For all tracing session(s) */
1059 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
1060 struct ust_app
*app
;
1063 if (!sess
->ust_session
) {
1064 goto unlock_session
;
1068 assert(app_sock
>= 0);
1069 app
= ust_app_find_by_sock(app_sock
);
1072 * Application can be unregistered before so
1073 * this is possible hence simply stopping the
1076 DBG3("UST app update failed to find app sock %d",
1080 ust_app_global_update(sess
->ust_session
, app
);
1084 session_unlock(sess
);
1089 * This thread manage event coming from the kernel.
1091 * Features supported in this thread:
1094 static void *thread_manage_kernel(void *data
)
1096 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1097 uint32_t revents
, nb_fd
;
1099 struct lttng_poll_event events
;
1101 DBG("[thread] Thread manage kernel started");
1103 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1106 * This first step of the while is to clean this structure which could free
1107 * non NULL pointers so initialize it before the loop.
1109 lttng_poll_init(&events
);
1111 if (testpoint(sessiond_thread_manage_kernel
)) {
1112 goto error_testpoint
;
1115 health_code_update();
1117 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1118 goto error_testpoint
;
1122 health_code_update();
1124 if (update_poll_flag
== 1) {
1125 /* Clean events object. We are about to populate it again. */
1126 lttng_poll_clean(&events
);
1128 ret
= sessiond_set_thread_pollset(&events
, 2);
1130 goto error_poll_create
;
1133 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1138 /* This will add the available kernel channel if any. */
1139 ret
= update_kernel_poll(&events
);
1143 update_poll_flag
= 0;
1146 DBG("Thread kernel polling");
1148 /* Poll infinite value of time */
1150 health_poll_entry();
1151 ret
= lttng_poll_wait(&events
, -1);
1152 DBG("Thread kernel return from poll on %d fds",
1153 LTTNG_POLL_GETNB(&events
));
1157 * Restart interrupted system call.
1159 if (errno
== EINTR
) {
1163 } else if (ret
== 0) {
1164 /* Should not happen since timeout is infinite */
1165 ERR("Return value of poll is 0 with an infinite timeout.\n"
1166 "This should not have happened! Continuing...");
1172 for (i
= 0; i
< nb_fd
; i
++) {
1173 /* Fetch once the poll data */
1174 revents
= LTTNG_POLL_GETEV(&events
, i
);
1175 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1177 health_code_update();
1180 /* No activity for this FD (poll implementation). */
1184 /* Thread quit pipe has been closed. Killing thread. */
1185 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1191 /* Check for data on kernel pipe */
1192 if (revents
& LPOLLIN
) {
1193 if (pollfd
== kernel_poll_pipe
[0]) {
1194 (void) lttng_read(kernel_poll_pipe
[0],
1197 * Ret value is useless here, if this pipe gets any actions an
1198 * update is required anyway.
1200 update_poll_flag
= 1;
1204 * New CPU detected by the kernel. Adding kernel stream to
1205 * kernel session and updating the kernel consumer
1207 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1213 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1214 update_poll_flag
= 1;
1217 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1225 lttng_poll_clean(&events
);
1228 utils_close_pipe(kernel_poll_pipe
);
1229 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1232 ERR("Health error occurred in %s", __func__
);
1233 WARN("Kernel thread died unexpectedly. "
1234 "Kernel tracing can continue but CPU hotplug is disabled.");
1236 health_unregister(health_sessiond
);
1237 DBG("Kernel thread dying");
1242 * Signal pthread condition of the consumer data that the thread.
1244 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1246 pthread_mutex_lock(&data
->cond_mutex
);
1249 * The state is set before signaling. It can be any value, it's the waiter
1250 * job to correctly interpret this condition variable associated to the
1251 * consumer pthread_cond.
1253 * A value of 0 means that the corresponding thread of the consumer data
1254 * was not started. 1 indicates that the thread has started and is ready
1255 * for action. A negative value means that there was an error during the
1258 data
->consumer_thread_is_ready
= state
;
1259 (void) pthread_cond_signal(&data
->cond
);
1261 pthread_mutex_unlock(&data
->cond_mutex
);
1265 * This thread manage the consumer error sent back to the session daemon.
1267 static void *thread_manage_consumer(void *data
)
1269 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1270 uint32_t revents
, nb_fd
;
1271 enum lttcomm_return_code code
;
1272 struct lttng_poll_event events
;
1273 struct consumer_data
*consumer_data
= data
;
1275 DBG("[thread] Manage consumer started");
1277 rcu_register_thread();
1278 rcu_thread_online();
1280 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1282 health_code_update();
1285 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1286 * metadata_sock. Nothing more will be added to this poll set.
1288 ret
= sessiond_set_thread_pollset(&events
, 3);
1294 * The error socket here is already in a listening state which was done
1295 * just before spawning this thread to avoid a race between the consumer
1296 * daemon exec trying to connect and the listen() call.
1298 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1303 health_code_update();
1305 /* Infinite blocking call, waiting for transmission */
1307 health_poll_entry();
1309 if (testpoint(sessiond_thread_manage_consumer
)) {
1313 ret
= lttng_poll_wait(&events
, -1);
1317 * Restart interrupted system call.
1319 if (errno
== EINTR
) {
1327 for (i
= 0; i
< nb_fd
; i
++) {
1328 /* Fetch once the poll data */
1329 revents
= LTTNG_POLL_GETEV(&events
, i
);
1330 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1332 health_code_update();
1335 /* No activity for this FD (poll implementation). */
1339 /* Thread quit pipe has been closed. Killing thread. */
1340 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1346 /* Event on the registration socket */
1347 if (pollfd
== consumer_data
->err_sock
) {
1348 if (revents
& LPOLLIN
) {
1350 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1351 ERR("consumer err socket poll error");
1354 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1360 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1366 * Set the CLOEXEC flag. Return code is useless because either way, the
1369 (void) utils_set_fd_cloexec(sock
);
1371 health_code_update();
1373 DBG2("Receiving code from consumer err_sock");
1375 /* Getting status code from kconsumerd */
1376 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1377 sizeof(enum lttcomm_return_code
));
1382 health_code_update();
1383 if (code
== LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1384 /* Connect both socket, command and metadata. */
1385 consumer_data
->cmd_sock
=
1386 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1387 consumer_data
->metadata_fd
=
1388 lttcomm_connect_unix_sock(consumer_data
->cmd_unix_sock_path
);
1389 if (consumer_data
->cmd_sock
< 0
1390 || consumer_data
->metadata_fd
< 0) {
1391 PERROR("consumer connect cmd socket");
1392 /* On error, signal condition and quit. */
1393 signal_consumer_condition(consumer_data
, -1);
1396 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1397 /* Create metadata socket lock. */
1398 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1399 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1400 PERROR("zmalloc pthread mutex");
1404 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1406 signal_consumer_condition(consumer_data
, 1);
1407 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1408 DBG("Consumer metadata socket ready (fd: %d)",
1409 consumer_data
->metadata_fd
);
1411 ERR("consumer error when waiting for SOCK_READY : %s",
1412 lttcomm_get_readable_code(-code
));
1416 /* Remove the consumerd error sock since we've established a connexion */
1417 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1422 /* Add new accepted error socket. */
1423 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1428 /* Add metadata socket that is successfully connected. */
1429 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1430 LPOLLIN
| LPOLLRDHUP
);
1435 health_code_update();
1437 /* Infinite blocking call, waiting for transmission */
1440 health_code_update();
1442 /* Exit the thread because the thread quit pipe has been triggered. */
1444 /* Not a health error. */
1449 health_poll_entry();
1450 ret
= lttng_poll_wait(&events
, -1);
1454 * Restart interrupted system call.
1456 if (errno
== EINTR
) {
1464 for (i
= 0; i
< nb_fd
; i
++) {
1465 /* Fetch once the poll data */
1466 revents
= LTTNG_POLL_GETEV(&events
, i
);
1467 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1469 health_code_update();
1472 /* No activity for this FD (poll implementation). */
1477 * Thread quit pipe has been triggered, flag that we should stop
1478 * but continue the current loop to handle potential data from
1481 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1483 if (pollfd
== sock
) {
1484 /* Event on the consumerd socket */
1485 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1486 && !(revents
& LPOLLIN
)) {
1487 ERR("consumer err socket second poll error");
1490 health_code_update();
1491 /* Wait for any kconsumerd error */
1492 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1493 sizeof(enum lttcomm_return_code
));
1495 ERR("consumer closed the command socket");
1499 ERR("consumer return code : %s",
1500 lttcomm_get_readable_code(-code
));
1503 } else if (pollfd
== consumer_data
->metadata_fd
) {
1504 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1505 && !(revents
& LPOLLIN
)) {
1506 ERR("consumer err metadata socket second poll error");
1509 /* UST metadata requests */
1510 ret
= ust_consumer_metadata_request(
1511 &consumer_data
->metadata_sock
);
1513 ERR("Handling metadata request");
1517 /* No need for an else branch all FDs are tested prior. */
1519 health_code_update();
1525 * We lock here because we are about to close the sockets and some other
1526 * thread might be using them so get exclusive access which will abort all
1527 * other consumer command by other threads.
1529 pthread_mutex_lock(&consumer_data
->lock
);
1531 /* Immediately set the consumerd state to stopped */
1532 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1533 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1534 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1535 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1536 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1538 /* Code flow error... */
1542 if (consumer_data
->err_sock
>= 0) {
1543 ret
= close(consumer_data
->err_sock
);
1547 consumer_data
->err_sock
= -1;
1549 if (consumer_data
->cmd_sock
>= 0) {
1550 ret
= close(consumer_data
->cmd_sock
);
1554 consumer_data
->cmd_sock
= -1;
1556 if (consumer_data
->metadata_sock
.fd_ptr
&&
1557 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1558 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1570 unlink(consumer_data
->err_unix_sock_path
);
1571 unlink(consumer_data
->cmd_unix_sock_path
);
1572 pthread_mutex_unlock(&consumer_data
->lock
);
1574 /* Cleanup metadata socket mutex. */
1575 if (consumer_data
->metadata_sock
.lock
) {
1576 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1577 free(consumer_data
->metadata_sock
.lock
);
1579 lttng_poll_clean(&events
);
1583 ERR("Health error occurred in %s", __func__
);
1585 health_unregister(health_sessiond
);
1586 DBG("consumer thread cleanup completed");
1588 rcu_thread_offline();
1589 rcu_unregister_thread();
1595 * This thread manage application communication.
1597 static void *thread_manage_apps(void *data
)
1599 int i
, ret
, pollfd
, err
= -1;
1601 uint32_t revents
, nb_fd
;
1602 struct lttng_poll_event events
;
1604 DBG("[thread] Manage application started");
1606 rcu_register_thread();
1607 rcu_thread_online();
1609 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1611 if (testpoint(sessiond_thread_manage_apps
)) {
1612 goto error_testpoint
;
1615 health_code_update();
1617 ret
= sessiond_set_thread_pollset(&events
, 2);
1619 goto error_poll_create
;
1622 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1627 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1631 health_code_update();
1634 DBG("Apps thread polling");
1636 /* Inifinite blocking call, waiting for transmission */
1638 health_poll_entry();
1639 ret
= lttng_poll_wait(&events
, -1);
1640 DBG("Apps thread return from poll on %d fds",
1641 LTTNG_POLL_GETNB(&events
));
1645 * Restart interrupted system call.
1647 if (errno
== EINTR
) {
1655 for (i
= 0; i
< nb_fd
; i
++) {
1656 /* Fetch once the poll data */
1657 revents
= LTTNG_POLL_GETEV(&events
, i
);
1658 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1660 health_code_update();
1663 /* No activity for this FD (poll implementation). */
1667 /* Thread quit pipe has been closed. Killing thread. */
1668 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1674 /* Inspect the apps cmd pipe */
1675 if (pollfd
== apps_cmd_pipe
[0]) {
1676 if (revents
& LPOLLIN
) {
1680 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1681 if (size_ret
< sizeof(sock
)) {
1682 PERROR("read apps cmd pipe");
1686 health_code_update();
1689 * Since this is a command socket (write then read),
1690 * we only monitor the error events of the socket.
1692 ret
= lttng_poll_add(&events
, sock
,
1693 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1698 DBG("Apps with sock %d added to poll set", sock
);
1699 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1700 ERR("Apps command pipe error");
1703 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1708 * At this point, we know that a registered application made
1709 * the event at poll_wait.
1711 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1712 /* Removing from the poll set */
1713 ret
= lttng_poll_del(&events
, pollfd
);
1718 /* Socket closed on remote end. */
1719 ust_app_unregister(pollfd
);
1721 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1726 health_code_update();
1732 lttng_poll_clean(&events
);
1735 utils_close_pipe(apps_cmd_pipe
);
1736 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1739 * We don't clean the UST app hash table here since already registered
1740 * applications can still be controlled so let them be until the session
1741 * daemon dies or the applications stop.
1746 ERR("Health error occurred in %s", __func__
);
1748 health_unregister(health_sessiond
);
1749 DBG("Application communication apps thread cleanup complete");
1750 rcu_thread_offline();
1751 rcu_unregister_thread();
1756 * Send a socket to a thread This is called from the dispatch UST registration
1757 * thread once all sockets are set for the application.
1759 * The sock value can be invalid, we don't really care, the thread will handle
1760 * it and make the necessary cleanup if so.
1762 * On success, return 0 else a negative value being the errno message of the
1765 static int send_socket_to_thread(int fd
, int sock
)
1770 * It's possible that the FD is set as invalid with -1 concurrently just
1771 * before calling this function being a shutdown state of the thread.
1778 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1779 if (ret
< sizeof(sock
)) {
1780 PERROR("write apps pipe %d", fd
);
1787 /* All good. Don't send back the write positive ret value. */
1794 * Sanitize the wait queue of the dispatch registration thread meaning removing
1795 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1796 * notify socket is never received.
1798 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1800 int ret
, nb_fd
= 0, i
;
1801 unsigned int fd_added
= 0;
1802 struct lttng_poll_event events
;
1803 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1807 lttng_poll_init(&events
);
1809 /* Just skip everything for an empty queue. */
1810 if (!wait_queue
->count
) {
1814 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1819 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1820 &wait_queue
->head
, head
) {
1821 assert(wait_node
->app
);
1822 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1823 LPOLLHUP
| LPOLLERR
);
1836 * Poll but don't block so we can quickly identify the faulty events and
1837 * clean them afterwards from the wait queue.
1839 ret
= lttng_poll_wait(&events
, 0);
1845 for (i
= 0; i
< nb_fd
; i
++) {
1846 /* Get faulty FD. */
1847 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1848 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1851 /* No activity for this FD (poll implementation). */
1855 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1856 &wait_queue
->head
, head
) {
1857 if (pollfd
== wait_node
->app
->sock
&&
1858 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1859 cds_list_del(&wait_node
->head
);
1860 wait_queue
->count
--;
1861 ust_app_destroy(wait_node
->app
);
1864 * Silence warning of use-after-free in
1865 * cds_list_for_each_entry_safe which uses
1866 * __typeof__(*wait_node).
1871 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1878 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1882 lttng_poll_clean(&events
);
1886 lttng_poll_clean(&events
);
1888 ERR("Unable to sanitize wait queue");
1893 * Dispatch request from the registration threads to the application
1894 * communication thread.
1896 static void *thread_dispatch_ust_registration(void *data
)
1899 struct cds_wfcq_node
*node
;
1900 struct ust_command
*ust_cmd
= NULL
;
1901 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1902 struct ust_reg_wait_queue wait_queue
= {
1906 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1908 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1909 goto error_testpoint
;
1912 health_code_update();
1914 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1916 DBG("[thread] Dispatch UST command started");
1918 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1919 health_code_update();
1921 /* Atomically prepare the queue futex */
1922 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1925 struct ust_app
*app
= NULL
;
1929 * Make sure we don't have node(s) that have hung up before receiving
1930 * the notify socket. This is to clean the list in order to avoid
1931 * memory leaks from notify socket that are never seen.
1933 sanitize_wait_queue(&wait_queue
);
1935 health_code_update();
1936 /* Dequeue command for registration */
1937 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1939 DBG("Woken up but nothing in the UST command queue");
1940 /* Continue thread execution */
1944 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1946 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1947 " gid:%d sock:%d name:%s (version %d.%d)",
1948 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1949 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1950 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1951 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1953 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1954 wait_node
= zmalloc(sizeof(*wait_node
));
1956 PERROR("zmalloc wait_node dispatch");
1957 ret
= close(ust_cmd
->sock
);
1959 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1961 lttng_fd_put(LTTNG_FD_APPS
, 1);
1965 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1967 /* Create application object if socket is CMD. */
1968 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1970 if (!wait_node
->app
) {
1971 ret
= close(ust_cmd
->sock
);
1973 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1975 lttng_fd_put(LTTNG_FD_APPS
, 1);
1981 * Add application to the wait queue so we can set the notify
1982 * socket before putting this object in the global ht.
1984 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1989 * We have to continue here since we don't have the notify
1990 * socket and the application MUST be added to the hash table
1991 * only at that moment.
1996 * Look for the application in the local wait queue and set the
1997 * notify socket if found.
1999 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2000 &wait_queue
.head
, head
) {
2001 health_code_update();
2002 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
2003 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
2004 cds_list_del(&wait_node
->head
);
2006 app
= wait_node
->app
;
2008 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
2014 * With no application at this stage the received socket is
2015 * basically useless so close it before we free the cmd data
2016 * structure for good.
2019 ret
= close(ust_cmd
->sock
);
2021 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
2023 lttng_fd_put(LTTNG_FD_APPS
, 1);
2030 * @session_lock_list
2032 * Lock the global session list so from the register up to the
2033 * registration done message, no thread can see the application
2034 * and change its state.
2036 session_lock_list();
2040 * Add application to the global hash table. This needs to be
2041 * done before the update to the UST registry can locate the
2046 /* Set app version. This call will print an error if needed. */
2047 (void) ust_app_version(app
);
2049 /* Send notify socket through the notify pipe. */
2050 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
2054 session_unlock_list();
2056 * No notify thread, stop the UST tracing. However, this is
2057 * not an internal error of the this thread thus setting
2058 * the health error code to a normal exit.
2065 * Update newly registered application with the tracing
2066 * registry info already enabled information.
2068 update_ust_app(app
->sock
);
2071 * Don't care about return value. Let the manage apps threads
2072 * handle app unregistration upon socket close.
2074 (void) ust_app_register_done(app
);
2077 * Even if the application socket has been closed, send the app
2078 * to the thread and unregistration will take place at that
2081 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2084 session_unlock_list();
2086 * No apps. thread, stop the UST tracing. However, this is
2087 * not an internal error of the this thread thus setting
2088 * the health error code to a normal exit.
2095 session_unlock_list();
2097 } while (node
!= NULL
);
2099 health_poll_entry();
2100 /* Futex wait on queue. Blocking call on futex() */
2101 futex_nto1_wait(&ust_cmd_queue
.futex
);
2104 /* Normal exit, no error */
2108 /* Clean up wait queue. */
2109 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2110 &wait_queue
.head
, head
) {
2111 cds_list_del(&wait_node
->head
);
2116 /* Empty command queue. */
2118 /* Dequeue command for registration */
2119 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2123 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2124 ret
= close(ust_cmd
->sock
);
2126 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2128 lttng_fd_put(LTTNG_FD_APPS
, 1);
2133 DBG("Dispatch thread dying");
2136 ERR("Health error occurred in %s", __func__
);
2138 health_unregister(health_sessiond
);
2143 * This thread manage application registration.
2145 static void *thread_registration_apps(void *data
)
2147 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2148 uint32_t revents
, nb_fd
;
2149 struct lttng_poll_event events
;
2151 * Get allocated in this thread, enqueued to a global queue, dequeued and
2152 * freed in the manage apps thread.
2154 struct ust_command
*ust_cmd
= NULL
;
2156 DBG("[thread] Manage application registration started");
2158 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2160 if (testpoint(sessiond_thread_registration_apps
)) {
2161 goto error_testpoint
;
2164 ret
= lttcomm_listen_unix_sock(apps_sock
);
2170 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2171 * more will be added to this poll set.
2173 ret
= sessiond_set_thread_pollset(&events
, 2);
2175 goto error_create_poll
;
2178 /* Add the application registration socket */
2179 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2181 goto error_poll_add
;
2184 /* Notify all applications to register */
2185 ret
= notify_ust_apps(1);
2187 ERR("Failed to notify applications or create the wait shared memory.\n"
2188 "Execution continues but there might be problem for already\n"
2189 "running applications that wishes to register.");
2193 DBG("Accepting application registration");
2195 /* Inifinite blocking call, waiting for transmission */
2197 health_poll_entry();
2198 ret
= lttng_poll_wait(&events
, -1);
2202 * Restart interrupted system call.
2204 if (errno
== EINTR
) {
2212 for (i
= 0; i
< nb_fd
; i
++) {
2213 health_code_update();
2215 /* Fetch once the poll data */
2216 revents
= LTTNG_POLL_GETEV(&events
, i
);
2217 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2220 /* No activity for this FD (poll implementation). */
2224 /* Thread quit pipe has been closed. Killing thread. */
2225 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2231 /* Event on the registration socket */
2232 if (pollfd
== apps_sock
) {
2233 if (revents
& LPOLLIN
) {
2234 sock
= lttcomm_accept_unix_sock(apps_sock
);
2240 * Set socket timeout for both receiving and ending.
2241 * app_socket_timeout is in seconds, whereas
2242 * lttcomm_setsockopt_rcv_timeout and
2243 * lttcomm_setsockopt_snd_timeout expect msec as
2246 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2247 app_socket_timeout
* 1000);
2248 (void) lttcomm_setsockopt_snd_timeout(sock
,
2249 app_socket_timeout
* 1000);
2252 * Set the CLOEXEC flag. Return code is useless because
2253 * either way, the show must go on.
2255 (void) utils_set_fd_cloexec(sock
);
2257 /* Create UST registration command for enqueuing */
2258 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2259 if (ust_cmd
== NULL
) {
2260 PERROR("ust command zmalloc");
2269 * Using message-based transmissions to ensure we don't
2270 * have to deal with partially received messages.
2272 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2274 ERR("Exhausted file descriptors allowed for applications.");
2284 health_code_update();
2285 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2288 /* Close socket of the application. */
2293 lttng_fd_put(LTTNG_FD_APPS
, 1);
2297 health_code_update();
2299 ust_cmd
->sock
= sock
;
2302 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2303 " gid:%d sock:%d name:%s (version %d.%d)",
2304 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2305 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2306 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2307 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2310 * Lock free enqueue the registration request. The red pill
2311 * has been taken! This apps will be part of the *system*.
2313 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2316 * Wake the registration queue futex. Implicit memory
2317 * barrier with the exchange in cds_wfcq_enqueue.
2319 futex_nto1_wake(&ust_cmd_queue
.futex
);
2320 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2321 ERR("Register apps socket poll error");
2324 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2333 /* Notify that the registration thread is gone */
2336 if (apps_sock
>= 0) {
2337 ret
= close(apps_sock
);
2347 lttng_fd_put(LTTNG_FD_APPS
, 1);
2349 unlink(apps_unix_sock_path
);
2352 lttng_poll_clean(&events
);
2356 DBG("UST Registration thread cleanup complete");
2359 ERR("Health error occurred in %s", __func__
);
2361 health_unregister(health_sessiond
);
2367 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2368 * exec or it will fails.
2370 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2373 struct timespec timeout
;
2375 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2376 consumer_data
->consumer_thread_is_ready
= 0;
2378 /* Setup pthread condition */
2379 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2382 PERROR("pthread_condattr_init consumer data");
2387 * Set the monotonic clock in order to make sure we DO NOT jump in time
2388 * between the clock_gettime() call and the timedwait call. See bug #324
2389 * for a more details and how we noticed it.
2391 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2394 PERROR("pthread_condattr_setclock consumer data");
2398 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2401 PERROR("pthread_cond_init consumer data");
2405 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2409 PERROR("pthread_create consumer");
2414 /* We are about to wait on a pthread condition */
2415 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2417 /* Get time for sem_timedwait absolute timeout */
2418 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2420 * Set the timeout for the condition timed wait even if the clock gettime
2421 * call fails since we might loop on that call and we want to avoid to
2422 * increment the timeout too many times.
2424 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2427 * The following loop COULD be skipped in some conditions so this is why we
2428 * set ret to 0 in order to make sure at least one round of the loop is
2434 * Loop until the condition is reached or when a timeout is reached. Note
2435 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2436 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2437 * possible. This loop does not take any chances and works with both of
2440 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2441 if (clock_ret
< 0) {
2442 PERROR("clock_gettime spawn consumer");
2443 /* Infinite wait for the consumerd thread to be ready */
2444 ret
= pthread_cond_wait(&consumer_data
->cond
,
2445 &consumer_data
->cond_mutex
);
2447 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2448 &consumer_data
->cond_mutex
, &timeout
);
2452 /* Release the pthread condition */
2453 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2457 if (ret
== ETIMEDOUT
) {
2461 * Call has timed out so we kill the kconsumerd_thread and return
2464 ERR("Condition timed out. The consumer thread was never ready."
2466 pth_ret
= pthread_cancel(consumer_data
->thread
);
2468 PERROR("pthread_cancel consumer thread");
2471 PERROR("pthread_cond_wait failed consumer thread");
2473 /* Caller is expecting a negative value on failure. */
2478 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2479 if (consumer_data
->pid
== 0) {
2480 ERR("Consumerd did not start");
2481 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2484 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2493 * Join consumer thread
2495 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2499 /* Consumer pid must be a real one. */
2500 if (consumer_data
->pid
> 0) {
2502 ret
= kill(consumer_data
->pid
, SIGTERM
);
2504 PERROR("Error killing consumer daemon");
2507 return pthread_join(consumer_data
->thread
, &status
);
2514 * Fork and exec a consumer daemon (consumerd).
2516 * Return pid if successful else -1.
2518 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2522 const char *consumer_to_use
;
2523 const char *verbosity
;
2526 DBG("Spawning consumerd");
2533 if (opt_verbose_consumer
) {
2534 verbosity
= "--verbose";
2535 } else if (lttng_opt_quiet
) {
2536 verbosity
= "--quiet";
2541 switch (consumer_data
->type
) {
2542 case LTTNG_CONSUMER_KERNEL
:
2544 * Find out which consumerd to execute. We will first try the
2545 * 64-bit path, then the sessiond's installation directory, and
2546 * fallback on the 32-bit one,
2548 DBG3("Looking for a kernel consumer at these locations:");
2549 DBG3(" 1) %s", consumerd64_bin
);
2550 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2551 DBG3(" 3) %s", consumerd32_bin
);
2552 if (stat(consumerd64_bin
, &st
) == 0) {
2553 DBG3("Found location #1");
2554 consumer_to_use
= consumerd64_bin
;
2555 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2556 DBG3("Found location #2");
2557 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2558 } else if (stat(consumerd32_bin
, &st
) == 0) {
2559 DBG3("Found location #3");
2560 consumer_to_use
= consumerd32_bin
;
2562 DBG("Could not find any valid consumerd executable");
2566 DBG("Using kernel consumer at: %s", consumer_to_use
);
2567 ret
= execl(consumer_to_use
,
2568 "lttng-consumerd", verbosity
, "-k",
2569 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2570 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2571 "--group", tracing_group_name
,
2574 case LTTNG_CONSUMER64_UST
:
2576 char *tmpnew
= NULL
;
2578 if (consumerd64_libdir
[0] != '\0') {
2582 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2586 tmplen
= strlen("LD_LIBRARY_PATH=")
2587 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2588 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2593 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2594 strcat(tmpnew
, consumerd64_libdir
);
2595 if (tmp
[0] != '\0') {
2596 strcat(tmpnew
, ":");
2597 strcat(tmpnew
, tmp
);
2599 ret
= putenv(tmpnew
);
2606 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2607 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2608 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2609 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2610 "--group", tracing_group_name
,
2612 if (consumerd64_libdir
[0] != '\0') {
2617 case LTTNG_CONSUMER32_UST
:
2619 char *tmpnew
= NULL
;
2621 if (consumerd32_libdir
[0] != '\0') {
2625 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2629 tmplen
= strlen("LD_LIBRARY_PATH=")
2630 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2631 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2636 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2637 strcat(tmpnew
, consumerd32_libdir
);
2638 if (tmp
[0] != '\0') {
2639 strcat(tmpnew
, ":");
2640 strcat(tmpnew
, tmp
);
2642 ret
= putenv(tmpnew
);
2649 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2650 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2651 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2652 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2653 "--group", tracing_group_name
,
2655 if (consumerd32_libdir
[0] != '\0') {
2661 PERROR("unknown consumer type");
2665 PERROR("Consumer execl()");
2667 /* Reaching this point, we got a failure on our execl(). */
2669 } else if (pid
> 0) {
2672 PERROR("start consumer fork");
2680 * Spawn the consumerd daemon and session daemon thread.
2682 static int start_consumerd(struct consumer_data
*consumer_data
)
2687 * Set the listen() state on the socket since there is a possible race
2688 * between the exec() of the consumer daemon and this call if place in the
2689 * consumer thread. See bug #366 for more details.
2691 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2696 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2697 if (consumer_data
->pid
!= 0) {
2698 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2702 ret
= spawn_consumerd(consumer_data
);
2704 ERR("Spawning consumerd failed");
2705 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2709 /* Setting up the consumer_data pid */
2710 consumer_data
->pid
= ret
;
2711 DBG2("Consumer pid %d", consumer_data
->pid
);
2712 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2714 DBG2("Spawning consumer control thread");
2715 ret
= spawn_consumer_thread(consumer_data
);
2717 ERR("Fatal error spawning consumer control thread");
2725 /* Cleanup already created sockets on error. */
2726 if (consumer_data
->err_sock
>= 0) {
2729 err
= close(consumer_data
->err_sock
);
2731 PERROR("close consumer data error socket");
2738 * Setup necessary data for kernel tracer action.
2740 static int init_kernel_tracer(void)
2744 /* Modprobe lttng kernel modules */
2745 ret
= modprobe_lttng_control();
2750 /* Open debugfs lttng */
2751 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2752 if (kernel_tracer_fd
< 0) {
2753 DBG("Failed to open %s", module_proc_lttng
);
2758 /* Validate kernel version */
2759 ret
= kernel_validate_version(kernel_tracer_fd
);
2764 ret
= modprobe_lttng_data();
2769 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2773 modprobe_remove_lttng_control();
2774 ret
= close(kernel_tracer_fd
);
2778 kernel_tracer_fd
= -1;
2779 return LTTNG_ERR_KERN_VERSION
;
2782 ret
= close(kernel_tracer_fd
);
2788 modprobe_remove_lttng_control();
2791 WARN("No kernel tracer available");
2792 kernel_tracer_fd
= -1;
2794 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2796 return LTTNG_ERR_KERN_NA
;
2802 * Copy consumer output from the tracing session to the domain session. The
2803 * function also applies the right modification on a per domain basis for the
2804 * trace files destination directory.
2806 * Should *NOT* be called with RCU read-side lock held.
2808 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2811 const char *dir_name
;
2812 struct consumer_output
*consumer
;
2815 assert(session
->consumer
);
2818 case LTTNG_DOMAIN_KERNEL
:
2819 DBG3("Copying tracing session consumer output in kernel session");
2821 * XXX: We should audit the session creation and what this function
2822 * does "extra" in order to avoid a destroy since this function is used
2823 * in the domain session creation (kernel and ust) only. Same for UST
2826 if (session
->kernel_session
->consumer
) {
2827 consumer_output_put(session
->kernel_session
->consumer
);
2829 session
->kernel_session
->consumer
=
2830 consumer_copy_output(session
->consumer
);
2831 /* Ease our life a bit for the next part */
2832 consumer
= session
->kernel_session
->consumer
;
2833 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2835 case LTTNG_DOMAIN_JUL
:
2836 case LTTNG_DOMAIN_LOG4J
:
2837 case LTTNG_DOMAIN_PYTHON
:
2838 case LTTNG_DOMAIN_UST
:
2839 DBG3("Copying tracing session consumer output in UST session");
2840 if (session
->ust_session
->consumer
) {
2841 consumer_output_put(session
->ust_session
->consumer
);
2843 session
->ust_session
->consumer
=
2844 consumer_copy_output(session
->consumer
);
2845 /* Ease our life a bit for the next part */
2846 consumer
= session
->ust_session
->consumer
;
2847 dir_name
= DEFAULT_UST_TRACE_DIR
;
2850 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2854 /* Append correct directory to subdir */
2855 strncat(consumer
->subdir
, dir_name
,
2856 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2857 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2866 * Create an UST session and add it to the session ust list.
2868 * Should *NOT* be called with RCU read-side lock held.
2870 static int create_ust_session(struct ltt_session
*session
,
2871 struct lttng_domain
*domain
)
2874 struct ltt_ust_session
*lus
= NULL
;
2878 assert(session
->consumer
);
2880 switch (domain
->type
) {
2881 case LTTNG_DOMAIN_JUL
:
2882 case LTTNG_DOMAIN_LOG4J
:
2883 case LTTNG_DOMAIN_PYTHON
:
2884 case LTTNG_DOMAIN_UST
:
2887 ERR("Unknown UST domain on create session %d", domain
->type
);
2888 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2892 DBG("Creating UST session");
2894 lus
= trace_ust_create_session(session
->id
);
2896 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2900 lus
->uid
= session
->uid
;
2901 lus
->gid
= session
->gid
;
2902 lus
->output_traces
= session
->output_traces
;
2903 lus
->snapshot_mode
= session
->snapshot_mode
;
2904 lus
->live_timer_interval
= session
->live_timer
;
2905 session
->ust_session
= lus
;
2906 if (session
->shm_path
[0]) {
2907 strncpy(lus
->root_shm_path
, session
->shm_path
,
2908 sizeof(lus
->root_shm_path
));
2909 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2910 strncpy(lus
->shm_path
, session
->shm_path
,
2911 sizeof(lus
->shm_path
));
2912 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2913 strncat(lus
->shm_path
, "/ust",
2914 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2916 /* Copy session output to the newly created UST session */
2917 ret
= copy_session_consumer(domain
->type
, session
);
2918 if (ret
!= LTTNG_OK
) {
2926 session
->ust_session
= NULL
;
2931 * Create a kernel tracer session then create the default channel.
2933 static int create_kernel_session(struct ltt_session
*session
)
2937 DBG("Creating kernel session");
2939 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2941 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2945 /* Code flow safety */
2946 assert(session
->kernel_session
);
2948 /* Copy session output to the newly created Kernel session */
2949 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2950 if (ret
!= LTTNG_OK
) {
2954 /* Create directory(ies) on local filesystem. */
2955 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2956 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2957 ret
= run_as_mkdir_recursive(
2958 session
->kernel_session
->consumer
->dst
.trace_path
,
2959 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2961 if (errno
!= EEXIST
) {
2962 ERR("Trace directory creation error");
2968 session
->kernel_session
->uid
= session
->uid
;
2969 session
->kernel_session
->gid
= session
->gid
;
2970 session
->kernel_session
->output_traces
= session
->output_traces
;
2971 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2976 trace_kernel_destroy_session(session
->kernel_session
);
2977 session
->kernel_session
= NULL
;
2982 * Count number of session permitted by uid/gid.
2984 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2987 struct ltt_session
*session
;
2989 DBG("Counting number of available session for UID %d GID %d",
2991 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2993 * Only list the sessions the user can control.
2995 if (!session_access_ok(session
, uid
, gid
)) {
3004 * Process the command requested by the lttng client within the command
3005 * context structure. This function make sure that the return structure (llm)
3006 * is set and ready for transmission before returning.
3008 * Return any error encountered or 0 for success.
3010 * "sock" is only used for special-case var. len data.
3012 * Should *NOT* be called with RCU read-side lock held.
3014 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
3018 int need_tracing_session
= 1;
3021 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
3023 assert(!rcu_read_ongoing());
3027 switch (cmd_ctx
->lsm
->cmd_type
) {
3028 case LTTNG_CREATE_SESSION
:
3029 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3030 case LTTNG_CREATE_SESSION_LIVE
:
3031 case LTTNG_DESTROY_SESSION
:
3032 case LTTNG_LIST_SESSIONS
:
3033 case LTTNG_LIST_DOMAINS
:
3034 case LTTNG_START_TRACE
:
3035 case LTTNG_STOP_TRACE
:
3036 case LTTNG_DATA_PENDING
:
3037 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3038 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3039 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3040 case LTTNG_SNAPSHOT_RECORD
:
3041 case LTTNG_SAVE_SESSION
:
3042 case LTTNG_SET_SESSION_SHM_PATH
:
3043 case LTTNG_METADATA_REGENERATE
:
3050 if (opt_no_kernel
&& need_domain
3051 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3053 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3055 ret
= LTTNG_ERR_KERN_NA
;
3060 /* Deny register consumer if we already have a spawned consumer. */
3061 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3062 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3063 if (kconsumer_data
.pid
> 0) {
3064 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3065 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3068 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3072 * Check for command that don't needs to allocate a returned payload. We do
3073 * this here so we don't have to make the call for no payload at each
3076 switch(cmd_ctx
->lsm
->cmd_type
) {
3077 case LTTNG_LIST_SESSIONS
:
3078 case LTTNG_LIST_TRACEPOINTS
:
3079 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3080 case LTTNG_LIST_DOMAINS
:
3081 case LTTNG_LIST_CHANNELS
:
3082 case LTTNG_LIST_EVENTS
:
3083 case LTTNG_LIST_SYSCALLS
:
3084 case LTTNG_LIST_TRACKER_PIDS
:
3085 case LTTNG_DATA_PENDING
:
3088 /* Setup lttng message with no payload */
3089 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0);
3091 /* This label does not try to unlock the session */
3092 goto init_setup_error
;
3096 /* Commands that DO NOT need a session. */
3097 switch (cmd_ctx
->lsm
->cmd_type
) {
3098 case LTTNG_CREATE_SESSION
:
3099 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3100 case LTTNG_CREATE_SESSION_LIVE
:
3101 case LTTNG_CALIBRATE
:
3102 case LTTNG_LIST_SESSIONS
:
3103 case LTTNG_LIST_TRACEPOINTS
:
3104 case LTTNG_LIST_SYSCALLS
:
3105 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3106 case LTTNG_SAVE_SESSION
:
3107 need_tracing_session
= 0;
3110 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3112 * We keep the session list lock across _all_ commands
3113 * for now, because the per-session lock does not
3114 * handle teardown properly.
3116 session_lock_list();
3117 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3118 if (cmd_ctx
->session
== NULL
) {
3119 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3122 /* Acquire lock for the session */
3123 session_lock(cmd_ctx
->session
);
3129 * Commands that need a valid session but should NOT create one if none
3130 * exists. Instead of creating one and destroying it when the command is
3131 * handled, process that right before so we save some round trip in useless
3134 switch (cmd_ctx
->lsm
->cmd_type
) {
3135 case LTTNG_DISABLE_CHANNEL
:
3136 case LTTNG_DISABLE_EVENT
:
3137 switch (cmd_ctx
->lsm
->domain
.type
) {
3138 case LTTNG_DOMAIN_KERNEL
:
3139 if (!cmd_ctx
->session
->kernel_session
) {
3140 ret
= LTTNG_ERR_NO_CHANNEL
;
3144 case LTTNG_DOMAIN_JUL
:
3145 case LTTNG_DOMAIN_LOG4J
:
3146 case LTTNG_DOMAIN_PYTHON
:
3147 case LTTNG_DOMAIN_UST
:
3148 if (!cmd_ctx
->session
->ust_session
) {
3149 ret
= LTTNG_ERR_NO_CHANNEL
;
3154 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3166 * Check domain type for specific "pre-action".
3168 switch (cmd_ctx
->lsm
->domain
.type
) {
3169 case LTTNG_DOMAIN_KERNEL
:
3171 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3175 /* Kernel tracer check */
3176 if (kernel_tracer_fd
== -1) {
3177 /* Basically, load kernel tracer modules */
3178 ret
= init_kernel_tracer();
3184 /* Consumer is in an ERROR state. Report back to client */
3185 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3186 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3190 /* Need a session for kernel command */
3191 if (need_tracing_session
) {
3192 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3193 ret
= create_kernel_session(cmd_ctx
->session
);
3195 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3200 /* Start the kernel consumer daemon */
3201 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3202 if (kconsumer_data
.pid
== 0 &&
3203 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3204 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3205 ret
= start_consumerd(&kconsumer_data
);
3207 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3210 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3212 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3216 * The consumer was just spawned so we need to add the socket to
3217 * the consumer output of the session if exist.
3219 ret
= consumer_create_socket(&kconsumer_data
,
3220 cmd_ctx
->session
->kernel_session
->consumer
);
3227 case LTTNG_DOMAIN_JUL
:
3228 case LTTNG_DOMAIN_LOG4J
:
3229 case LTTNG_DOMAIN_PYTHON
:
3230 case LTTNG_DOMAIN_UST
:
3232 if (!ust_app_supported()) {
3233 ret
= LTTNG_ERR_NO_UST
;
3236 /* Consumer is in an ERROR state. Report back to client */
3237 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3238 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3242 if (need_tracing_session
) {
3243 /* Create UST session if none exist. */
3244 if (cmd_ctx
->session
->ust_session
== NULL
) {
3245 ret
= create_ust_session(cmd_ctx
->session
,
3246 &cmd_ctx
->lsm
->domain
);
3247 if (ret
!= LTTNG_OK
) {
3252 /* Start the UST consumer daemons */
3254 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3255 if (consumerd64_bin
[0] != '\0' &&
3256 ustconsumer64_data
.pid
== 0 &&
3257 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3258 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3259 ret
= start_consumerd(&ustconsumer64_data
);
3261 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3262 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3266 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3267 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3269 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3273 * Setup socket for consumer 64 bit. No need for atomic access
3274 * since it was set above and can ONLY be set in this thread.
3276 ret
= consumer_create_socket(&ustconsumer64_data
,
3277 cmd_ctx
->session
->ust_session
->consumer
);
3283 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3284 if (consumerd32_bin
[0] != '\0' &&
3285 ustconsumer32_data
.pid
== 0 &&
3286 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3287 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3288 ret
= start_consumerd(&ustconsumer32_data
);
3290 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3291 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3295 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3296 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3298 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3302 * Setup socket for consumer 64 bit. No need for atomic access
3303 * since it was set above and can ONLY be set in this thread.
3305 ret
= consumer_create_socket(&ustconsumer32_data
,
3306 cmd_ctx
->session
->ust_session
->consumer
);
3318 /* Validate consumer daemon state when start/stop trace command */
3319 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3320 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3321 switch (cmd_ctx
->lsm
->domain
.type
) {
3322 case LTTNG_DOMAIN_NONE
:
3324 case LTTNG_DOMAIN_JUL
:
3325 case LTTNG_DOMAIN_LOG4J
:
3326 case LTTNG_DOMAIN_PYTHON
:
3327 case LTTNG_DOMAIN_UST
:
3328 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3329 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3333 case LTTNG_DOMAIN_KERNEL
:
3334 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3335 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3340 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3346 * Check that the UID or GID match that of the tracing session.
3347 * The root user can interact with all sessions.
3349 if (need_tracing_session
) {
3350 if (!session_access_ok(cmd_ctx
->session
,
3351 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3352 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3353 ret
= LTTNG_ERR_EPERM
;
3359 * Send relayd information to consumer as soon as we have a domain and a
3362 if (cmd_ctx
->session
&& need_domain
) {
3364 * Setup relayd if not done yet. If the relayd information was already
3365 * sent to the consumer, this call will gracefully return.
3367 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3368 if (ret
!= LTTNG_OK
) {
3373 /* Process by command type */
3374 switch (cmd_ctx
->lsm
->cmd_type
) {
3375 case LTTNG_ADD_CONTEXT
:
3378 * An LTTNG_ADD_CONTEXT command might have a supplementary
3379 * payload if the context being added is an application context.
3381 if (cmd_ctx
->lsm
->u
.context
.ctx
.ctx
==
3382 LTTNG_EVENT_CONTEXT_APP_CONTEXT
) {
3383 char *provider_name
= NULL
, *context_name
= NULL
;
3384 size_t provider_name_len
=
3385 cmd_ctx
->lsm
->u
.context
.provider_name_len
;
3386 size_t context_name_len
=
3387 cmd_ctx
->lsm
->u
.context
.context_name_len
;
3389 if (provider_name_len
== 0 || context_name_len
== 0) {
3391 * Application provider and context names MUST
3394 ret
= -LTTNG_ERR_INVALID
;
3398 provider_name
= zmalloc(provider_name_len
+ 1);
3399 if (!provider_name
) {
3400 ret
= -LTTNG_ERR_NOMEM
;
3403 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
=
3406 context_name
= zmalloc(context_name_len
+ 1);
3407 if (!context_name
) {
3408 ret
= -LTTNG_ERR_NOMEM
;
3409 goto error_add_context
;
3411 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
=
3414 ret
= lttcomm_recv_unix_sock(sock
, provider_name
,
3417 goto error_add_context
;
3420 ret
= lttcomm_recv_unix_sock(sock
, context_name
,
3423 goto error_add_context
;
3428 * cmd_add_context assumes ownership of the provider and context
3431 ret
= cmd_add_context(cmd_ctx
->session
,
3432 cmd_ctx
->lsm
->domain
.type
,
3433 cmd_ctx
->lsm
->u
.context
.channel_name
,
3434 &cmd_ctx
->lsm
->u
.context
.ctx
,
3435 kernel_poll_pipe
[1]);
3437 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
= NULL
;
3438 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
= NULL
;
3440 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
);
3441 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
);
3447 case LTTNG_DISABLE_CHANNEL
:
3449 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3450 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3453 case LTTNG_DISABLE_EVENT
:
3457 * FIXME: handle filter; for now we just receive the filter's
3458 * bytecode along with the filter expression which are sent by
3459 * liblttng-ctl and discard them.
3461 * This fixes an issue where the client may block while sending
3462 * the filter payload and encounter an error because the session
3463 * daemon closes the socket without ever handling this data.
3465 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3466 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3469 char data
[LTTNG_FILTER_MAX_LEN
];
3471 DBG("Discarding disable event command payload of size %zu", count
);
3473 ret
= lttcomm_recv_unix_sock(sock
, data
,
3474 count
> sizeof(data
) ? sizeof(data
) : count
);
3479 count
-= (size_t) ret
;
3482 /* FIXME: passing packed structure to non-packed pointer */
3483 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3484 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3485 &cmd_ctx
->lsm
->u
.disable
.event
);
3488 case LTTNG_ENABLE_CHANNEL
:
3490 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3491 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3494 case LTTNG_TRACK_PID
:
3496 ret
= cmd_track_pid(cmd_ctx
->session
,
3497 cmd_ctx
->lsm
->domain
.type
,
3498 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3501 case LTTNG_UNTRACK_PID
:
3503 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3504 cmd_ctx
->lsm
->domain
.type
,
3505 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3508 case LTTNG_ENABLE_EVENT
:
3510 struct lttng_event_exclusion
*exclusion
= NULL
;
3511 struct lttng_filter_bytecode
*bytecode
= NULL
;
3512 char *filter_expression
= NULL
;
3514 /* Handle exclusion events and receive it from the client. */
3515 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3516 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3518 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3519 (count
* LTTNG_SYMBOL_NAME_LEN
));
3521 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3525 DBG("Receiving var len exclusion event list from client ...");
3526 exclusion
->count
= count
;
3527 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3528 count
* LTTNG_SYMBOL_NAME_LEN
);
3530 DBG("Nothing recv() from client var len data... continuing");
3533 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3538 /* Get filter expression from client. */
3539 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3540 size_t expression_len
=
3541 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3543 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3544 ret
= LTTNG_ERR_FILTER_INVAL
;
3549 filter_expression
= zmalloc(expression_len
);
3550 if (!filter_expression
) {
3552 ret
= LTTNG_ERR_FILTER_NOMEM
;
3556 /* Receive var. len. data */
3557 DBG("Receiving var len filter's expression from client ...");
3558 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3561 DBG("Nothing recv() from client car len data... continuing");
3563 free(filter_expression
);
3565 ret
= LTTNG_ERR_FILTER_INVAL
;
3570 /* Handle filter and get bytecode from client. */
3571 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3572 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3574 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3575 ret
= LTTNG_ERR_FILTER_INVAL
;
3576 free(filter_expression
);
3581 bytecode
= zmalloc(bytecode_len
);
3583 free(filter_expression
);
3585 ret
= LTTNG_ERR_FILTER_NOMEM
;
3589 /* Receive var. len. data */
3590 DBG("Receiving var len filter's bytecode from client ...");
3591 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3593 DBG("Nothing recv() from client car len data... continuing");
3595 free(filter_expression
);
3598 ret
= LTTNG_ERR_FILTER_INVAL
;
3602 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3603 free(filter_expression
);
3606 ret
= LTTNG_ERR_FILTER_INVAL
;
3611 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3612 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3613 &cmd_ctx
->lsm
->u
.enable
.event
,
3614 filter_expression
, bytecode
, exclusion
,
3615 kernel_poll_pipe
[1]);
3618 case LTTNG_LIST_TRACEPOINTS
:
3620 struct lttng_event
*events
;
3623 session_lock_list();
3624 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3625 session_unlock_list();
3626 if (nb_events
< 0) {
3627 /* Return value is a negative lttng_error_code. */
3633 * Setup lttng message with payload size set to the event list size in
3634 * bytes and then copy list into the llm payload.
3636 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3637 sizeof(struct lttng_event
) * nb_events
);
3647 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3649 struct lttng_event_field
*fields
;
3652 session_lock_list();
3653 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3655 session_unlock_list();
3656 if (nb_fields
< 0) {
3657 /* Return value is a negative lttng_error_code. */
3663 * Setup lttng message with payload size set to the event list size in
3664 * bytes and then copy list into the llm payload.
3666 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, fields
,
3667 sizeof(struct lttng_event_field
) * nb_fields
);
3677 case LTTNG_LIST_SYSCALLS
:
3679 struct lttng_event
*events
;
3682 nb_events
= cmd_list_syscalls(&events
);
3683 if (nb_events
< 0) {
3684 /* Return value is a negative lttng_error_code. */
3690 * Setup lttng message with payload size set to the event list size in
3691 * bytes and then copy list into the llm payload.
3693 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3694 sizeof(struct lttng_event
) * nb_events
);
3704 case LTTNG_LIST_TRACKER_PIDS
:
3706 int32_t *pids
= NULL
;
3709 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3710 cmd_ctx
->lsm
->domain
.type
, &pids
);
3712 /* Return value is a negative lttng_error_code. */
3718 * Setup lttng message with payload size set to the event list size in
3719 * bytes and then copy list into the llm payload.
3721 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, pids
,
3722 sizeof(int32_t) * nr_pids
);
3732 case LTTNG_SET_CONSUMER_URI
:
3735 struct lttng_uri
*uris
;
3737 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3738 len
= nb_uri
* sizeof(struct lttng_uri
);
3741 ret
= LTTNG_ERR_INVALID
;
3745 uris
= zmalloc(len
);
3747 ret
= LTTNG_ERR_FATAL
;
3751 /* Receive variable len data */
3752 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3753 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3755 DBG("No URIs received from client... continuing");
3757 ret
= LTTNG_ERR_SESSION_FAIL
;
3762 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3764 if (ret
!= LTTNG_OK
) {
3771 case LTTNG_START_TRACE
:
3773 ret
= cmd_start_trace(cmd_ctx
->session
);
3776 case LTTNG_STOP_TRACE
:
3778 ret
= cmd_stop_trace(cmd_ctx
->session
);
3781 case LTTNG_CREATE_SESSION
:
3784 struct lttng_uri
*uris
= NULL
;
3786 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3787 len
= nb_uri
* sizeof(struct lttng_uri
);
3790 uris
= zmalloc(len
);
3792 ret
= LTTNG_ERR_FATAL
;
3796 /* Receive variable len data */
3797 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3798 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3800 DBG("No URIs received from client... continuing");
3802 ret
= LTTNG_ERR_SESSION_FAIL
;
3807 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3808 DBG("Creating session with ONE network URI is a bad call");
3809 ret
= LTTNG_ERR_SESSION_FAIL
;
3815 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3816 &cmd_ctx
->creds
, 0);
3822 case LTTNG_DESTROY_SESSION
:
3824 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3826 /* Set session to NULL so we do not unlock it after free. */
3827 cmd_ctx
->session
= NULL
;
3830 case LTTNG_LIST_DOMAINS
:
3833 struct lttng_domain
*domains
= NULL
;
3835 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3837 /* Return value is a negative lttng_error_code. */
3842 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, domains
,
3843 nb_dom
* sizeof(struct lttng_domain
));
3853 case LTTNG_LIST_CHANNELS
:
3855 ssize_t payload_size
;
3856 struct lttng_channel
*channels
= NULL
;
3858 payload_size
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3859 cmd_ctx
->session
, &channels
);
3860 if (payload_size
< 0) {
3861 /* Return value is a negative lttng_error_code. */
3862 ret
= -payload_size
;
3866 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, channels
,
3877 case LTTNG_LIST_EVENTS
:
3880 struct lttng_event
*events
= NULL
;
3881 struct lttcomm_event_command_header cmd_header
;
3884 memset(&cmd_header
, 0, sizeof(cmd_header
));
3885 /* Extended infos are included at the end of events */
3886 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
,
3887 cmd_ctx
->session
, cmd_ctx
->lsm
->u
.list
.channel_name
,
3888 &events
, &total_size
);
3891 /* Return value is a negative lttng_error_code. */
3896 cmd_header
.nb_events
= nb_event
;
3897 ret
= setup_lttng_msg(cmd_ctx
, events
, total_size
,
3898 &cmd_header
, sizeof(cmd_header
));
3908 case LTTNG_LIST_SESSIONS
:
3910 unsigned int nr_sessions
;
3911 void *sessions_payload
;
3914 session_lock_list();
3915 nr_sessions
= lttng_sessions_count(
3916 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3917 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3918 payload_len
= sizeof(struct lttng_session
) * nr_sessions
;
3919 sessions_payload
= zmalloc(payload_len
);
3921 if (!sessions_payload
) {
3922 session_unlock_list();
3927 cmd_list_lttng_sessions(sessions_payload
,
3928 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3929 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3930 session_unlock_list();
3932 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, sessions_payload
,
3934 free(sessions_payload
);
3943 case LTTNG_CALIBRATE
:
3945 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3946 &cmd_ctx
->lsm
->u
.calibrate
);
3949 case LTTNG_REGISTER_CONSUMER
:
3951 struct consumer_data
*cdata
;
3953 switch (cmd_ctx
->lsm
->domain
.type
) {
3954 case LTTNG_DOMAIN_KERNEL
:
3955 cdata
= &kconsumer_data
;
3958 ret
= LTTNG_ERR_UND
;
3962 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3963 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3966 case LTTNG_DATA_PENDING
:
3969 uint8_t pending_ret_byte
;
3971 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
3976 * This function may returns 0 or 1 to indicate whether or not
3977 * there is data pending. In case of error, it should return an
3978 * LTTNG_ERR code. However, some code paths may still return
3979 * a nondescript error code, which we handle by returning an
3982 if (pending_ret
== 0 || pending_ret
== 1) {
3984 * ret will be set to LTTNG_OK at the end of
3987 } else if (pending_ret
< 0) {
3988 ret
= LTTNG_ERR_UNK
;
3995 pending_ret_byte
= (uint8_t) pending_ret
;
3997 /* 1 byte to return whether or not data is pending */
3998 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
,
3999 &pending_ret_byte
, 1);
4008 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
4010 struct lttcomm_lttng_output_id reply
;
4012 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
4013 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
4014 if (ret
!= LTTNG_OK
) {
4018 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &reply
,
4024 /* Copy output list into message payload */
4028 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
4030 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
4031 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
4034 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
4037 struct lttng_snapshot_output
*outputs
= NULL
;
4039 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
4040 if (nb_output
< 0) {
4045 assert((nb_output
> 0 && outputs
) || nb_output
== 0);
4046 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, outputs
,
4047 nb_output
* sizeof(struct lttng_snapshot_output
));
4057 case LTTNG_SNAPSHOT_RECORD
:
4059 ret
= cmd_snapshot_record(cmd_ctx
->session
,
4060 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
4061 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
4064 case LTTNG_CREATE_SESSION_SNAPSHOT
:
4067 struct lttng_uri
*uris
= NULL
;
4069 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4070 len
= nb_uri
* sizeof(struct lttng_uri
);
4073 uris
= zmalloc(len
);
4075 ret
= LTTNG_ERR_FATAL
;
4079 /* Receive variable len data */
4080 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4081 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4083 DBG("No URIs received from client... continuing");
4085 ret
= LTTNG_ERR_SESSION_FAIL
;
4090 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4091 DBG("Creating session with ONE network URI is a bad call");
4092 ret
= LTTNG_ERR_SESSION_FAIL
;
4098 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
4099 nb_uri
, &cmd_ctx
->creds
);
4103 case LTTNG_CREATE_SESSION_LIVE
:
4106 struct lttng_uri
*uris
= NULL
;
4108 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4109 len
= nb_uri
* sizeof(struct lttng_uri
);
4112 uris
= zmalloc(len
);
4114 ret
= LTTNG_ERR_FATAL
;
4118 /* Receive variable len data */
4119 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4120 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4122 DBG("No URIs received from client... continuing");
4124 ret
= LTTNG_ERR_SESSION_FAIL
;
4129 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4130 DBG("Creating session with ONE network URI is a bad call");
4131 ret
= LTTNG_ERR_SESSION_FAIL
;
4137 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4138 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4142 case LTTNG_SAVE_SESSION
:
4144 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4148 case LTTNG_SET_SESSION_SHM_PATH
:
4150 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4151 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4154 case LTTNG_METADATA_REGENERATE
:
4156 ret
= cmd_metadata_regenerate(cmd_ctx
->session
);
4160 ret
= LTTNG_ERR_UND
;
4165 if (cmd_ctx
->llm
== NULL
) {
4166 DBG("Missing llm structure. Allocating one.");
4167 if (setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0) < 0) {
4171 /* Set return code */
4172 cmd_ctx
->llm
->ret_code
= ret
;
4174 if (cmd_ctx
->session
) {
4175 session_unlock(cmd_ctx
->session
);
4177 if (need_tracing_session
) {
4178 session_unlock_list();
4181 assert(!rcu_read_ongoing());
4186 * Thread managing health check socket.
4188 static void *thread_manage_health(void *data
)
4190 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4191 uint32_t revents
, nb_fd
;
4192 struct lttng_poll_event events
;
4193 struct health_comm_msg msg
;
4194 struct health_comm_reply reply
;
4196 DBG("[thread] Manage health check started");
4198 rcu_register_thread();
4200 /* We might hit an error path before this is created. */
4201 lttng_poll_init(&events
);
4203 /* Create unix socket */
4204 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4206 ERR("Unable to create health check Unix socket");
4212 /* lttng health client socket path permissions */
4213 ret
= chown(health_unix_sock_path
, 0,
4214 utils_get_group_id(tracing_group_name
));
4216 ERR("Unable to set group on %s", health_unix_sock_path
);
4222 ret
= chmod(health_unix_sock_path
,
4223 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4225 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4233 * Set the CLOEXEC flag. Return code is useless because either way, the
4236 (void) utils_set_fd_cloexec(sock
);
4238 ret
= lttcomm_listen_unix_sock(sock
);
4244 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4245 * more will be added to this poll set.
4247 ret
= sessiond_set_thread_pollset(&events
, 2);
4252 /* Add the application registration socket */
4253 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4258 sessiond_notify_ready();
4261 DBG("Health check ready");
4263 /* Inifinite blocking call, waiting for transmission */
4265 ret
= lttng_poll_wait(&events
, -1);
4268 * Restart interrupted system call.
4270 if (errno
== EINTR
) {
4278 for (i
= 0; i
< nb_fd
; i
++) {
4279 /* Fetch once the poll data */
4280 revents
= LTTNG_POLL_GETEV(&events
, i
);
4281 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4284 /* No activity for this FD (poll implementation). */
4288 /* Thread quit pipe has been closed. Killing thread. */
4289 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4295 /* Event on the registration socket */
4296 if (pollfd
== sock
) {
4297 if (revents
& LPOLLIN
) {
4299 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4300 ERR("Health socket poll error");
4303 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4309 new_sock
= lttcomm_accept_unix_sock(sock
);
4315 * Set the CLOEXEC flag. Return code is useless because either way, the
4318 (void) utils_set_fd_cloexec(new_sock
);
4320 DBG("Receiving data from client for health...");
4321 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4323 DBG("Nothing recv() from client... continuing");
4324 ret
= close(new_sock
);
4332 rcu_thread_online();
4334 memset(&reply
, 0, sizeof(reply
));
4335 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4337 * health_check_state returns 0 if health is
4340 if (!health_check_state(health_sessiond
, i
)) {
4341 reply
.ret_code
|= 1ULL << i
;
4345 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4347 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4349 ERR("Failed to send health data back to client");
4352 /* End of transmission */
4353 ret
= close(new_sock
);
4363 ERR("Health error occurred in %s", __func__
);
4365 DBG("Health check thread dying");
4366 unlink(health_unix_sock_path
);
4374 lttng_poll_clean(&events
);
4376 rcu_unregister_thread();
4381 * This thread manage all clients request using the unix client socket for
4384 static void *thread_manage_clients(void *data
)
4386 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4388 uint32_t revents
, nb_fd
;
4389 struct command_ctx
*cmd_ctx
= NULL
;
4390 struct lttng_poll_event events
;
4392 DBG("[thread] Manage client started");
4394 rcu_register_thread();
4396 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4398 health_code_update();
4400 ret
= lttcomm_listen_unix_sock(client_sock
);
4406 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4407 * more will be added to this poll set.
4409 ret
= sessiond_set_thread_pollset(&events
, 2);
4411 goto error_create_poll
;
4414 /* Add the application registration socket */
4415 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4420 sessiond_notify_ready();
4421 ret
= sem_post(&load_info
->message_thread_ready
);
4423 PERROR("sem_post message_thread_ready");
4427 /* This testpoint is after we signal readiness to the parent. */
4428 if (testpoint(sessiond_thread_manage_clients
)) {
4432 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4436 health_code_update();
4439 DBG("Accepting client command ...");
4441 /* Inifinite blocking call, waiting for transmission */
4443 health_poll_entry();
4444 ret
= lttng_poll_wait(&events
, -1);
4448 * Restart interrupted system call.
4450 if (errno
== EINTR
) {
4458 for (i
= 0; i
< nb_fd
; i
++) {
4459 /* Fetch once the poll data */
4460 revents
= LTTNG_POLL_GETEV(&events
, i
);
4461 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4463 health_code_update();
4466 /* No activity for this FD (poll implementation). */
4470 /* Thread quit pipe has been closed. Killing thread. */
4471 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4477 /* Event on the registration socket */
4478 if (pollfd
== client_sock
) {
4479 if (revents
& LPOLLIN
) {
4481 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4482 ERR("Client socket poll error");
4485 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4491 DBG("Wait for client response");
4493 health_code_update();
4495 sock
= lttcomm_accept_unix_sock(client_sock
);
4501 * Set the CLOEXEC flag. Return code is useless because either way, the
4504 (void) utils_set_fd_cloexec(sock
);
4506 /* Set socket option for credentials retrieval */
4507 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4512 /* Allocate context command to process the client request */
4513 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4514 if (cmd_ctx
== NULL
) {
4515 PERROR("zmalloc cmd_ctx");
4519 /* Allocate data buffer for reception */
4520 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4521 if (cmd_ctx
->lsm
== NULL
) {
4522 PERROR("zmalloc cmd_ctx->lsm");
4526 cmd_ctx
->llm
= NULL
;
4527 cmd_ctx
->session
= NULL
;
4529 health_code_update();
4532 * Data is received from the lttng client. The struct
4533 * lttcomm_session_msg (lsm) contains the command and data request of
4536 DBG("Receiving data from client ...");
4537 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4538 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4540 DBG("Nothing recv() from client... continuing");
4546 clean_command_ctx(&cmd_ctx
);
4550 health_code_update();
4552 // TODO: Validate cmd_ctx including sanity check for
4553 // security purpose.
4555 rcu_thread_online();
4557 * This function dispatch the work to the kernel or userspace tracer
4558 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4559 * informations for the client. The command context struct contains
4560 * everything this function may needs.
4562 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4563 rcu_thread_offline();
4571 * TODO: Inform client somehow of the fatal error. At
4572 * this point, ret < 0 means that a zmalloc failed
4573 * (ENOMEM). Error detected but still accept
4574 * command, unless a socket error has been
4577 clean_command_ctx(&cmd_ctx
);
4581 health_code_update();
4583 DBG("Sending response (size: %d, retcode: %s (%d))",
4584 cmd_ctx
->lttng_msg_size
,
4585 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4586 cmd_ctx
->llm
->ret_code
);
4587 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4589 ERR("Failed to send data back to client");
4592 /* End of transmission */
4599 clean_command_ctx(&cmd_ctx
);
4601 health_code_update();
4613 lttng_poll_clean(&events
);
4614 clean_command_ctx(&cmd_ctx
);
4618 unlink(client_unix_sock_path
);
4619 if (client_sock
>= 0) {
4620 ret
= close(client_sock
);
4628 ERR("Health error occurred in %s", __func__
);
4631 health_unregister(health_sessiond
);
4633 DBG("Client thread dying");
4635 rcu_unregister_thread();
4638 * Since we are creating the consumer threads, we own them, so we need
4639 * to join them before our thread exits.
4641 ret
= join_consumer_thread(&kconsumer_data
);
4644 PERROR("join_consumer");
4647 ret
= join_consumer_thread(&ustconsumer32_data
);
4650 PERROR("join_consumer ust32");
4653 ret
= join_consumer_thread(&ustconsumer64_data
);
4656 PERROR("join_consumer ust64");
4661 static int string_match(const char *str1
, const char *str2
)
4663 return (str1
&& str2
) && !strcmp(str1
, str2
);
4667 * Take an option from the getopt output and set it in the right variable to be
4670 * Return 0 on success else a negative value.
4672 static int set_option(int opt
, const char *arg
, const char *optname
)
4676 if (string_match(optname
, "client-sock") || opt
== 'c') {
4677 if (!arg
|| *arg
== '\0') {
4681 if (lttng_is_setuid_setgid()) {
4682 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4683 "-c, --client-sock");
4685 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4687 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4688 if (!arg
|| *arg
== '\0') {
4692 if (lttng_is_setuid_setgid()) {
4693 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4696 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4698 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4700 } else if (string_match(optname
, "background") || opt
== 'b') {
4702 } else if (string_match(optname
, "group") || opt
== 'g') {
4703 if (!arg
|| *arg
== '\0') {
4707 if (lttng_is_setuid_setgid()) {
4708 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4712 * If the override option is set, the pointer points to a
4713 * *non* const thus freeing it even though the variable type is
4716 if (tracing_group_name_override
) {
4717 free((void *) tracing_group_name
);
4719 tracing_group_name
= strdup(arg
);
4720 if (!tracing_group_name
) {
4724 tracing_group_name_override
= 1;
4726 } else if (string_match(optname
, "help") || opt
== 'h') {
4727 ret
= utils_show_man_page(8, "lttng-sessiond");
4729 ERR("Cannot view man page lttng-sessiond(8)");
4732 exit(ret
? EXIT_FAILURE
: EXIT_SUCCESS
);
4733 } else if (string_match(optname
, "version") || opt
== 'V') {
4734 fprintf(stdout
, "%s\n", VERSION
);
4736 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4738 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4739 if (!arg
|| *arg
== '\0') {
4743 if (lttng_is_setuid_setgid()) {
4744 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4745 "--kconsumerd-err-sock");
4747 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4749 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4750 if (!arg
|| *arg
== '\0') {
4754 if (lttng_is_setuid_setgid()) {
4755 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4756 "--kconsumerd-cmd-sock");
4758 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4760 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4761 if (!arg
|| *arg
== '\0') {
4765 if (lttng_is_setuid_setgid()) {
4766 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4767 "--ustconsumerd64-err-sock");
4769 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4771 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4772 if (!arg
|| *arg
== '\0') {
4776 if (lttng_is_setuid_setgid()) {
4777 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4778 "--ustconsumerd64-cmd-sock");
4780 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4782 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4783 if (!arg
|| *arg
== '\0') {
4787 if (lttng_is_setuid_setgid()) {
4788 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4789 "--ustconsumerd32-err-sock");
4791 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4793 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4794 if (!arg
|| *arg
== '\0') {
4798 if (lttng_is_setuid_setgid()) {
4799 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4800 "--ustconsumerd32-cmd-sock");
4802 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4804 } else if (string_match(optname
, "no-kernel")) {
4806 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4807 lttng_opt_quiet
= 1;
4808 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4809 /* Verbose level can increase using multiple -v */
4811 /* Value obtained from config file */
4812 lttng_opt_verbose
= config_parse_value(arg
);
4814 /* -v used on command line */
4815 lttng_opt_verbose
++;
4817 /* Clamp value to [0, 3] */
4818 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4819 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4820 } else if (string_match(optname
, "verbose-consumer")) {
4822 opt_verbose_consumer
= config_parse_value(arg
);
4824 opt_verbose_consumer
++;
4826 } else if (string_match(optname
, "consumerd32-path")) {
4827 if (!arg
|| *arg
== '\0') {
4831 if (lttng_is_setuid_setgid()) {
4832 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4833 "--consumerd32-path");
4835 if (consumerd32_bin_override
) {
4836 free((void *) consumerd32_bin
);
4838 consumerd32_bin
= strdup(arg
);
4839 if (!consumerd32_bin
) {
4843 consumerd32_bin_override
= 1;
4845 } else if (string_match(optname
, "consumerd32-libdir")) {
4846 if (!arg
|| *arg
== '\0') {
4850 if (lttng_is_setuid_setgid()) {
4851 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4852 "--consumerd32-libdir");
4854 if (consumerd32_libdir_override
) {
4855 free((void *) consumerd32_libdir
);
4857 consumerd32_libdir
= strdup(arg
);
4858 if (!consumerd32_libdir
) {
4862 consumerd32_libdir_override
= 1;
4864 } else if (string_match(optname
, "consumerd64-path")) {
4865 if (!arg
|| *arg
== '\0') {
4869 if (lttng_is_setuid_setgid()) {
4870 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4871 "--consumerd64-path");
4873 if (consumerd64_bin_override
) {
4874 free((void *) consumerd64_bin
);
4876 consumerd64_bin
= strdup(arg
);
4877 if (!consumerd64_bin
) {
4881 consumerd64_bin_override
= 1;
4883 } else if (string_match(optname
, "consumerd64-libdir")) {
4884 if (!arg
|| *arg
== '\0') {
4888 if (lttng_is_setuid_setgid()) {
4889 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4890 "--consumerd64-libdir");
4892 if (consumerd64_libdir_override
) {
4893 free((void *) consumerd64_libdir
);
4895 consumerd64_libdir
= strdup(arg
);
4896 if (!consumerd64_libdir
) {
4900 consumerd64_libdir_override
= 1;
4902 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4903 if (!arg
|| *arg
== '\0') {
4907 if (lttng_is_setuid_setgid()) {
4908 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4912 opt_pidfile
= strdup(arg
);
4918 } else if (string_match(optname
, "agent-tcp-port")) {
4919 if (!arg
|| *arg
== '\0') {
4923 if (lttng_is_setuid_setgid()) {
4924 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4925 "--agent-tcp-port");
4934 v
= strtoul(arg
, NULL
, 0);
4935 if (errno
!= 0 || !isdigit(arg
[0])) {
4936 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4939 if (v
== 0 || v
>= 65535) {
4940 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4943 agent_tcp_port
= (uint32_t) v
;
4944 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4946 } else if (string_match(optname
, "load") || opt
== 'l') {
4947 if (!arg
|| *arg
== '\0') {
4951 if (lttng_is_setuid_setgid()) {
4952 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4955 free(opt_load_session_path
);
4956 opt_load_session_path
= strdup(arg
);
4957 if (!opt_load_session_path
) {
4962 } else if (string_match(optname
, "kmod-probes")) {
4963 if (!arg
|| *arg
== '\0') {
4967 if (lttng_is_setuid_setgid()) {
4968 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4971 free(kmod_probes_list
);
4972 kmod_probes_list
= strdup(arg
);
4973 if (!kmod_probes_list
) {
4978 } else if (string_match(optname
, "extra-kmod-probes")) {
4979 if (!arg
|| *arg
== '\0') {
4983 if (lttng_is_setuid_setgid()) {
4984 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4985 "--extra-kmod-probes");
4987 free(kmod_extra_probes_list
);
4988 kmod_extra_probes_list
= strdup(arg
);
4989 if (!kmod_extra_probes_list
) {
4994 } else if (string_match(optname
, "config") || opt
== 'f') {
4995 /* This is handled in set_options() thus silent skip. */
4998 /* Unknown option or other error.
4999 * Error is printed by getopt, just return */
5004 if (ret
== -EINVAL
) {
5005 const char *opt_name
= "unknown";
5008 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
5010 if (opt
== long_options
[i
].val
) {
5011 opt_name
= long_options
[i
].name
;
5016 WARN("Invalid argument provided for option \"%s\", using default value.",
5024 * config_entry_handler_cb used to handle options read from a config file.
5025 * See config_entry_handler_cb comment in common/config/session-config.h for the
5026 * return value conventions.
5028 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
5032 if (!entry
|| !entry
->name
|| !entry
->value
) {
5037 /* Check if the option is to be ignored */
5038 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
5039 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
5044 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
5047 /* Ignore if not fully matched. */
5048 if (strcmp(entry
->name
, long_options
[i
].name
)) {
5053 * If the option takes no argument on the command line, we have to
5054 * check if the value is "true". We support non-zero numeric values,
5057 if (!long_options
[i
].has_arg
) {
5058 ret
= config_parse_value(entry
->value
);
5061 WARN("Invalid configuration value \"%s\" for option %s",
5062 entry
->value
, entry
->name
);
5064 /* False, skip boolean config option. */
5069 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
5073 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
5080 * daemon configuration loading and argument parsing
5082 static int set_options(int argc
, char **argv
)
5084 int ret
= 0, c
= 0, option_index
= 0;
5085 int orig_optopt
= optopt
, orig_optind
= optind
;
5087 const char *config_path
= NULL
;
5089 optstring
= utils_generate_optstring(long_options
,
5090 sizeof(long_options
) / sizeof(struct option
));
5096 /* Check for the --config option */
5097 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5098 &option_index
)) != -1) {
5102 } else if (c
!= 'f') {
5103 /* if not equal to --config option. */
5107 if (lttng_is_setuid_setgid()) {
5108 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5111 config_path
= utils_expand_path(optarg
);
5113 ERR("Failed to resolve path: %s", optarg
);
5118 ret
= config_get_section_entries(config_path
, config_section_name
,
5119 config_entry_handler
, NULL
);
5122 ERR("Invalid configuration option at line %i", ret
);
5128 /* Reset getopt's global state */
5129 optopt
= orig_optopt
;
5130 optind
= orig_optind
;
5134 * getopt_long() will not set option_index if it encounters a
5137 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5144 * Pass NULL as the long option name if popt left the index
5147 ret
= set_option(c
, optarg
,
5148 option_index
< 0 ? NULL
:
5149 long_options
[option_index
].name
);
5161 * Creates the two needed socket by the daemon.
5162 * apps_sock - The communication socket for all UST apps.
5163 * client_sock - The communication of the cli tool (lttng).
5165 static int init_daemon_socket(void)
5170 old_umask
= umask(0);
5172 /* Create client tool unix socket */
5173 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
5174 if (client_sock
< 0) {
5175 ERR("Create unix sock failed: %s", client_unix_sock_path
);
5180 /* Set the cloexec flag */
5181 ret
= utils_set_fd_cloexec(client_sock
);
5183 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5184 "Continuing but note that the consumer daemon will have a "
5185 "reference to this socket on exec()", client_sock
);
5188 /* File permission MUST be 660 */
5189 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5191 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5196 /* Create the application unix socket */
5197 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5198 if (apps_sock
< 0) {
5199 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5204 /* Set the cloexec flag */
5205 ret
= utils_set_fd_cloexec(apps_sock
);
5207 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5208 "Continuing but note that the consumer daemon will have a "
5209 "reference to this socket on exec()", apps_sock
);
5212 /* File permission MUST be 666 */
5213 ret
= chmod(apps_unix_sock_path
,
5214 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5216 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5221 DBG3("Session daemon client socket %d and application socket %d created",
5222 client_sock
, apps_sock
);
5230 * Check if the global socket is available, and if a daemon is answering at the
5231 * other side. If yes, error is returned.
5233 static int check_existing_daemon(void)
5235 /* Is there anybody out there ? */
5236 if (lttng_session_daemon_alive()) {
5244 * Set the tracing group gid onto the client socket.
5246 * Race window between mkdir and chown is OK because we are going from more
5247 * permissive (root.root) to less permissive (root.tracing).
5249 static int set_permissions(char *rundir
)
5254 gid
= utils_get_group_id(tracing_group_name
);
5256 /* Set lttng run dir */
5257 ret
= chown(rundir
, 0, gid
);
5259 ERR("Unable to set group on %s", rundir
);
5264 * Ensure all applications and tracing group can search the run
5265 * dir. Allow everyone to read the directory, since it does not
5266 * buy us anything to hide its content.
5268 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5270 ERR("Unable to set permissions on %s", rundir
);
5274 /* lttng client socket path */
5275 ret
= chown(client_unix_sock_path
, 0, gid
);
5277 ERR("Unable to set group on %s", client_unix_sock_path
);
5281 /* kconsumer error socket path */
5282 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5284 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5288 /* 64-bit ustconsumer error socket path */
5289 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5291 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5295 /* 32-bit ustconsumer compat32 error socket path */
5296 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5298 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5302 DBG("All permissions are set");
5308 * Create the lttng run directory needed for all global sockets and pipe.
5310 static int create_lttng_rundir(const char *rundir
)
5314 DBG3("Creating LTTng run directory: %s", rundir
);
5316 ret
= mkdir(rundir
, S_IRWXU
);
5318 if (errno
!= EEXIST
) {
5319 ERR("Unable to create %s", rundir
);
5331 * Setup sockets and directory needed by the kconsumerd communication with the
5334 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5338 char path
[PATH_MAX
];
5340 switch (consumer_data
->type
) {
5341 case LTTNG_CONSUMER_KERNEL
:
5342 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5344 case LTTNG_CONSUMER64_UST
:
5345 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5347 case LTTNG_CONSUMER32_UST
:
5348 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5351 ERR("Consumer type unknown");
5356 DBG2("Creating consumer directory: %s", path
);
5358 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5360 if (errno
!= EEXIST
) {
5362 ERR("Failed to create %s", path
);
5368 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5370 ERR("Unable to set group on %s", path
);
5376 /* Create the kconsumerd error unix socket */
5377 consumer_data
->err_sock
=
5378 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5379 if (consumer_data
->err_sock
< 0) {
5380 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5386 * Set the CLOEXEC flag. Return code is useless because either way, the
5389 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5391 PERROR("utils_set_fd_cloexec");
5392 /* continue anyway */
5395 /* File permission MUST be 660 */
5396 ret
= chmod(consumer_data
->err_unix_sock_path
,
5397 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5399 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5409 * Signal handler for the daemon
5411 * Simply stop all worker threads, leaving main() return gracefully after
5412 * joining all threads and calling cleanup().
5414 static void sighandler(int sig
)
5418 DBG("SIGPIPE caught");
5421 DBG("SIGINT caught");
5425 DBG("SIGTERM caught");
5429 CMM_STORE_SHARED(recv_child_signal
, 1);
5437 * Setup signal handler for :
5438 * SIGINT, SIGTERM, SIGPIPE
5440 static int set_signal_handler(void)
5443 struct sigaction sa
;
5446 if ((ret
= sigemptyset(&sigset
)) < 0) {
5447 PERROR("sigemptyset");
5451 sa
.sa_handler
= sighandler
;
5452 sa
.sa_mask
= sigset
;
5454 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5455 PERROR("sigaction");
5459 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5460 PERROR("sigaction");
5464 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5465 PERROR("sigaction");
5469 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5470 PERROR("sigaction");
5474 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5480 * Set open files limit to unlimited. This daemon can open a large number of
5481 * file descriptors in order to consumer multiple kernel traces.
5483 static void set_ulimit(void)
5488 /* The kernel does not allowed an infinite limit for open files */
5489 lim
.rlim_cur
= 65535;
5490 lim
.rlim_max
= 65535;
5492 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5494 PERROR("failed to set open files limit");
5499 * Write pidfile using the rundir and opt_pidfile.
5501 static int write_pidfile(void)
5504 char pidfile_path
[PATH_MAX
];
5509 if (lttng_strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
))) {
5514 /* Build pidfile path from rundir and opt_pidfile. */
5515 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5516 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5518 PERROR("snprintf pidfile path");
5524 * Create pid file in rundir.
5526 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5532 * Create lockfile using the rundir and return its fd.
5534 static int create_lockfile(void)
5537 char lockfile_path
[PATH_MAX
];
5539 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5544 ret
= utils_create_lock_file(lockfile_path
);
5550 * Write agent TCP port using the rundir.
5552 static int write_agent_port(void)
5555 char path
[PATH_MAX
];
5559 ret
= snprintf(path
, sizeof(path
), "%s/"
5560 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5562 PERROR("snprintf agent port path");
5567 * Create TCP agent port file in rundir.
5569 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5578 int main(int argc
, char **argv
)
5580 int ret
= 0, retval
= 0;
5582 const char *home_path
, *env_app_timeout
;
5584 init_kernel_workarounds();
5586 rcu_register_thread();
5588 if (set_signal_handler()) {
5590 goto exit_set_signal_handler
;
5593 setup_consumerd_path();
5595 page_size
= sysconf(_SC_PAGESIZE
);
5596 if (page_size
< 0) {
5597 PERROR("sysconf _SC_PAGESIZE");
5598 page_size
= LONG_MAX
;
5599 WARN("Fallback page size to %ld", page_size
);
5603 * Parse arguments and load the daemon configuration file.
5605 * We have an exit_options exit path to free memory reserved by
5606 * set_options. This is needed because the rest of sessiond_cleanup()
5607 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5608 * depends on set_options.
5611 if (set_options(argc
, argv
)) {
5617 if (opt_daemon
|| opt_background
) {
5620 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5628 * We are in the child. Make sure all other file descriptors are
5629 * closed, in case we are called with more opened file
5630 * descriptors than the standard ones.
5632 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5637 if (run_as_create_worker(argv
[0]) < 0) {
5638 goto exit_create_run_as_worker_cleanup
;
5642 * Starting from here, we can create threads. This needs to be after
5643 * lttng_daemonize due to RCU.
5647 * Initialize the health check subsystem. This call should set the
5648 * appropriate time values.
5650 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5651 if (!health_sessiond
) {
5652 PERROR("health_app_create error");
5654 goto exit_health_sessiond_cleanup
;
5657 if (init_ht_cleanup_quit_pipe()) {
5659 goto exit_ht_cleanup_quit_pipe
;
5662 /* Setup the thread ht_cleanup communication pipe. */
5663 if (utils_create_pipe_cloexec(ht_cleanup_pipe
)) {
5665 goto exit_ht_cleanup_pipe
;
5668 /* Set up max poll set size */
5669 if (lttng_poll_set_max_size()) {
5671 goto exit_set_max_size
;
5674 /* Create thread to clean up RCU hash tables */
5675 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5676 thread_ht_cleanup
, (void *) NULL
);
5679 PERROR("pthread_create ht_cleanup");
5681 goto exit_ht_cleanup
;
5684 /* Create thread quit pipe */
5685 if (init_thread_quit_pipe()) {
5687 goto exit_init_data
;
5690 /* Check if daemon is UID = 0 */
5691 is_root
= !getuid();
5694 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5697 goto exit_init_data
;
5700 /* Create global run dir with root access */
5701 if (create_lttng_rundir(rundir
)) {
5703 goto exit_init_data
;
5706 if (strlen(apps_unix_sock_path
) == 0) {
5707 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5708 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5711 goto exit_init_data
;
5715 if (strlen(client_unix_sock_path
) == 0) {
5716 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5717 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5720 goto exit_init_data
;
5724 /* Set global SHM for ust */
5725 if (strlen(wait_shm_path
) == 0) {
5726 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5727 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5730 goto exit_init_data
;
5734 if (strlen(health_unix_sock_path
) == 0) {
5735 ret
= snprintf(health_unix_sock_path
,
5736 sizeof(health_unix_sock_path
),
5737 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5740 goto exit_init_data
;
5744 /* Setup kernel consumerd path */
5745 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5746 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5749 goto exit_init_data
;
5751 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5752 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5755 goto exit_init_data
;
5758 DBG2("Kernel consumer err path: %s",
5759 kconsumer_data
.err_unix_sock_path
);
5760 DBG2("Kernel consumer cmd path: %s",
5761 kconsumer_data
.cmd_unix_sock_path
);
5763 home_path
= utils_get_home_dir();
5764 if (home_path
== NULL
) {
5765 /* TODO: Add --socket PATH option */
5766 ERR("Can't get HOME directory for sockets creation.");
5768 goto exit_init_data
;
5772 * Create rundir from home path. This will create something like
5775 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5778 goto exit_init_data
;
5781 if (create_lttng_rundir(rundir
)) {
5783 goto exit_init_data
;
5786 if (strlen(apps_unix_sock_path
) == 0) {
5787 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5788 DEFAULT_HOME_APPS_UNIX_SOCK
,
5792 goto exit_init_data
;
5796 /* Set the cli tool unix socket path */
5797 if (strlen(client_unix_sock_path
) == 0) {
5798 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5799 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5803 goto exit_init_data
;
5807 /* Set global SHM for ust */
5808 if (strlen(wait_shm_path
) == 0) {
5809 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5810 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5814 goto exit_init_data
;
5818 /* Set health check Unix path */
5819 if (strlen(health_unix_sock_path
) == 0) {
5820 ret
= snprintf(health_unix_sock_path
,
5821 sizeof(health_unix_sock_path
),
5822 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5826 goto exit_init_data
;
5831 lockfile_fd
= create_lockfile();
5832 if (lockfile_fd
< 0) {
5834 goto exit_init_data
;
5837 /* Set consumer initial state */
5838 kernel_consumerd_state
= CONSUMER_STOPPED
;
5839 ust_consumerd_state
= CONSUMER_STOPPED
;
5841 DBG("Client socket path %s", client_unix_sock_path
);
5842 DBG("Application socket path %s", apps_unix_sock_path
);
5843 DBG("Application wait path %s", wait_shm_path
);
5844 DBG("LTTng run directory path: %s", rundir
);
5846 /* 32 bits consumerd path setup */
5847 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5848 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5850 PERROR("snprintf 32-bit consumer error socket path");
5852 goto exit_init_data
;
5854 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5855 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5857 PERROR("snprintf 32-bit consumer command socket path");
5859 goto exit_init_data
;
5862 DBG2("UST consumer 32 bits err path: %s",
5863 ustconsumer32_data
.err_unix_sock_path
);
5864 DBG2("UST consumer 32 bits cmd path: %s",
5865 ustconsumer32_data
.cmd_unix_sock_path
);
5867 /* 64 bits consumerd path setup */
5868 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5869 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5871 PERROR("snprintf 64-bit consumer error socket path");
5873 goto exit_init_data
;
5875 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5876 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5878 PERROR("snprintf 64-bit consumer command socket path");
5880 goto exit_init_data
;
5883 DBG2("UST consumer 64 bits err path: %s",
5884 ustconsumer64_data
.err_unix_sock_path
);
5885 DBG2("UST consumer 64 bits cmd path: %s",
5886 ustconsumer64_data
.cmd_unix_sock_path
);
5889 * See if daemon already exist.
5891 if (check_existing_daemon()) {
5892 ERR("Already running daemon.\n");
5894 * We do not goto exit because we must not cleanup()
5895 * because a daemon is already running.
5898 goto exit_init_data
;
5902 * Init UST app hash table. Alloc hash table before this point since
5903 * cleanup() can get called after that point.
5905 if (ust_app_ht_alloc()) {
5906 ERR("Failed to allocate UST app hash table");
5908 goto exit_init_data
;
5912 * Initialize agent app hash table. We allocate the hash table here
5913 * since cleanup() can get called after this point.
5915 if (agent_app_ht_alloc()) {
5916 ERR("Failed to allocate Agent app hash table");
5918 goto exit_init_data
;
5922 * These actions must be executed as root. We do that *after* setting up
5923 * the sockets path because we MUST make the check for another daemon using
5924 * those paths *before* trying to set the kernel consumer sockets and init
5928 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5930 goto exit_init_data
;
5933 /* Setup kernel tracer */
5934 if (!opt_no_kernel
) {
5935 init_kernel_tracer();
5936 if (kernel_tracer_fd
>= 0) {
5937 ret
= syscall_init_table();
5939 ERR("Unable to populate syscall table. "
5940 "Syscall tracing won't work "
5941 "for this session daemon.");
5946 /* Set ulimit for open files */
5949 /* init lttng_fd tracking must be done after set_ulimit. */
5952 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5954 goto exit_init_data
;
5957 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5959 goto exit_init_data
;
5962 /* Setup the needed unix socket */
5963 if (init_daemon_socket()) {
5965 goto exit_init_data
;
5968 /* Set credentials to socket */
5969 if (is_root
&& set_permissions(rundir
)) {
5971 goto exit_init_data
;
5974 /* Get parent pid if -S, --sig-parent is specified. */
5975 if (opt_sig_parent
) {
5979 /* Setup the kernel pipe for waking up the kernel thread */
5980 if (is_root
&& !opt_no_kernel
) {
5981 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5983 goto exit_init_data
;
5987 /* Setup the thread apps communication pipe. */
5988 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5990 goto exit_init_data
;
5993 /* Setup the thread apps notify communication pipe. */
5994 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
5996 goto exit_init_data
;
5999 /* Initialize global buffer per UID and PID registry. */
6000 buffer_reg_init_uid_registry();
6001 buffer_reg_init_pid_registry();
6003 /* Init UST command queue. */
6004 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
6007 * Get session list pointer. This pointer MUST NOT be free'd. This list
6008 * is statically declared in session.c
6010 session_list_ptr
= session_get_list();
6014 /* Check for the application socket timeout env variable. */
6015 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
6016 if (env_app_timeout
) {
6017 app_socket_timeout
= atoi(env_app_timeout
);
6019 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
6022 ret
= write_pidfile();
6024 ERR("Error in write_pidfile");
6026 goto exit_init_data
;
6028 ret
= write_agent_port();
6030 ERR("Error in write_agent_port");
6032 goto exit_init_data
;
6035 /* Initialize communication library */
6037 /* Initialize TCP timeout values */
6038 lttcomm_inet_init();
6040 if (load_session_init_data(&load_info
) < 0) {
6042 goto exit_init_data
;
6044 load_info
->path
= opt_load_session_path
;
6046 /* Create health-check thread */
6047 ret
= pthread_create(&health_thread
, NULL
,
6048 thread_manage_health
, (void *) NULL
);
6051 PERROR("pthread_create health");
6056 /* Create thread to manage the client socket */
6057 ret
= pthread_create(&client_thread
, NULL
,
6058 thread_manage_clients
, (void *) NULL
);
6061 PERROR("pthread_create clients");
6066 /* Create thread to dispatch registration */
6067 ret
= pthread_create(&dispatch_thread
, NULL
,
6068 thread_dispatch_ust_registration
, (void *) NULL
);
6071 PERROR("pthread_create dispatch");
6076 /* Create thread to manage application registration. */
6077 ret
= pthread_create(®_apps_thread
, NULL
,
6078 thread_registration_apps
, (void *) NULL
);
6081 PERROR("pthread_create registration");
6086 /* Create thread to manage application socket */
6087 ret
= pthread_create(&apps_thread
, NULL
,
6088 thread_manage_apps
, (void *) NULL
);
6091 PERROR("pthread_create apps");
6096 /* Create thread to manage application notify socket */
6097 ret
= pthread_create(&apps_notify_thread
, NULL
,
6098 ust_thread_manage_notify
, (void *) NULL
);
6101 PERROR("pthread_create notify");
6103 goto exit_apps_notify
;
6106 /* Create agent registration thread. */
6107 ret
= pthread_create(&agent_reg_thread
, NULL
,
6108 agent_thread_manage_registration
, (void *) NULL
);
6111 PERROR("pthread_create agent");
6113 goto exit_agent_reg
;
6116 /* Don't start this thread if kernel tracing is not requested nor root */
6117 if (is_root
&& !opt_no_kernel
) {
6118 /* Create kernel thread to manage kernel event */
6119 ret
= pthread_create(&kernel_thread
, NULL
,
6120 thread_manage_kernel
, (void *) NULL
);
6123 PERROR("pthread_create kernel");
6129 /* Create session loading thread. */
6130 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
6134 PERROR("pthread_create load_session_thread");
6136 goto exit_load_session
;
6140 * This is where we start awaiting program completion (e.g. through
6141 * signal that asks threads to teardown).
6144 ret
= pthread_join(load_session_thread
, &status
);
6147 PERROR("pthread_join load_session_thread");
6152 if (is_root
&& !opt_no_kernel
) {
6153 ret
= pthread_join(kernel_thread
, &status
);
6156 PERROR("pthread_join");
6162 ret
= pthread_join(agent_reg_thread
, &status
);
6165 PERROR("pthread_join agent");
6170 ret
= pthread_join(apps_notify_thread
, &status
);
6173 PERROR("pthread_join apps notify");
6178 ret
= pthread_join(apps_thread
, &status
);
6181 PERROR("pthread_join apps");
6186 ret
= pthread_join(reg_apps_thread
, &status
);
6189 PERROR("pthread_join");
6195 * Join dispatch thread after joining reg_apps_thread to ensure
6196 * we don't leak applications in the queue.
6198 ret
= pthread_join(dispatch_thread
, &status
);
6201 PERROR("pthread_join");
6206 ret
= pthread_join(client_thread
, &status
);
6209 PERROR("pthread_join");
6214 ret
= pthread_join(health_thread
, &status
);
6217 PERROR("pthread_join health thread");
6224 * sessiond_cleanup() is called when no other thread is running, except
6225 * the ht_cleanup thread, which is needed to destroy the hash tables.
6227 rcu_thread_online();
6229 rcu_thread_offline();
6230 rcu_unregister_thread();
6233 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6234 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6235 * the queue is empty before shutting down the clean-up thread.
6239 ret
= notify_thread_pipe(ht_cleanup_quit_pipe
[1]);
6241 ERR("write error on ht_cleanup quit pipe");
6245 ret
= pthread_join(ht_cleanup_thread
, &status
);
6248 PERROR("pthread_join ht cleanup thread");
6254 utils_close_pipe(ht_cleanup_pipe
);
6255 exit_ht_cleanup_pipe
:
6258 * Close the ht_cleanup quit pipe.
6260 utils_close_pipe(ht_cleanup_quit_pipe
);
6261 exit_ht_cleanup_quit_pipe
:
6263 health_app_destroy(health_sessiond
);
6264 exit_health_sessiond_cleanup
:
6265 exit_create_run_as_worker_cleanup
:
6268 sessiond_cleanup_options();
6270 exit_set_signal_handler
: