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 rcu_register_thread();
1908 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1910 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1911 goto error_testpoint
;
1914 health_code_update();
1916 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1918 DBG("[thread] Dispatch UST command started");
1920 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1921 health_code_update();
1923 /* Atomically prepare the queue futex */
1924 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1927 struct ust_app
*app
= NULL
;
1931 * Make sure we don't have node(s) that have hung up before receiving
1932 * the notify socket. This is to clean the list in order to avoid
1933 * memory leaks from notify socket that are never seen.
1935 sanitize_wait_queue(&wait_queue
);
1937 health_code_update();
1938 /* Dequeue command for registration */
1939 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1941 DBG("Woken up but nothing in the UST command queue");
1942 /* Continue thread execution */
1946 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1948 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1949 " gid:%d sock:%d name:%s (version %d.%d)",
1950 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1951 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1952 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1953 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1955 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1956 wait_node
= zmalloc(sizeof(*wait_node
));
1958 PERROR("zmalloc wait_node dispatch");
1959 ret
= close(ust_cmd
->sock
);
1961 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1963 lttng_fd_put(LTTNG_FD_APPS
, 1);
1967 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1969 /* Create application object if socket is CMD. */
1970 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1972 if (!wait_node
->app
) {
1973 ret
= close(ust_cmd
->sock
);
1975 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1977 lttng_fd_put(LTTNG_FD_APPS
, 1);
1983 * Add application to the wait queue so we can set the notify
1984 * socket before putting this object in the global ht.
1986 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1991 * We have to continue here since we don't have the notify
1992 * socket and the application MUST be added to the hash table
1993 * only at that moment.
1998 * Look for the application in the local wait queue and set the
1999 * notify socket if found.
2001 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2002 &wait_queue
.head
, head
) {
2003 health_code_update();
2004 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
2005 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
2006 cds_list_del(&wait_node
->head
);
2008 app
= wait_node
->app
;
2010 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
2016 * With no application at this stage the received socket is
2017 * basically useless so close it before we free the cmd data
2018 * structure for good.
2021 ret
= close(ust_cmd
->sock
);
2023 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
2025 lttng_fd_put(LTTNG_FD_APPS
, 1);
2032 * @session_lock_list
2034 * Lock the global session list so from the register up to the
2035 * registration done message, no thread can see the application
2036 * and change its state.
2038 session_lock_list();
2042 * Add application to the global hash table. This needs to be
2043 * done before the update to the UST registry can locate the
2048 /* Set app version. This call will print an error if needed. */
2049 (void) ust_app_version(app
);
2051 /* Send notify socket through the notify pipe. */
2052 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
2056 session_unlock_list();
2058 * No notify thread, stop the UST tracing. However, this is
2059 * not an internal error of the this thread thus setting
2060 * the health error code to a normal exit.
2067 * Update newly registered application with the tracing
2068 * registry info already enabled information.
2070 update_ust_app(app
->sock
);
2073 * Don't care about return value. Let the manage apps threads
2074 * handle app unregistration upon socket close.
2076 (void) ust_app_register_done(app
);
2079 * Even if the application socket has been closed, send the app
2080 * to the thread and unregistration will take place at that
2083 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2086 session_unlock_list();
2088 * No apps. thread, stop the UST tracing. However, this is
2089 * not an internal error of the this thread thus setting
2090 * the health error code to a normal exit.
2097 session_unlock_list();
2099 } while (node
!= NULL
);
2101 health_poll_entry();
2102 /* Futex wait on queue. Blocking call on futex() */
2103 futex_nto1_wait(&ust_cmd_queue
.futex
);
2106 /* Normal exit, no error */
2110 /* Clean up wait queue. */
2111 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2112 &wait_queue
.head
, head
) {
2113 cds_list_del(&wait_node
->head
);
2118 /* Empty command queue. */
2120 /* Dequeue command for registration */
2121 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2125 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2126 ret
= close(ust_cmd
->sock
);
2128 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2130 lttng_fd_put(LTTNG_FD_APPS
, 1);
2135 DBG("Dispatch thread dying");
2138 ERR("Health error occurred in %s", __func__
);
2140 health_unregister(health_sessiond
);
2141 rcu_unregister_thread();
2146 * This thread manage application registration.
2148 static void *thread_registration_apps(void *data
)
2150 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2151 uint32_t revents
, nb_fd
;
2152 struct lttng_poll_event events
;
2154 * Get allocated in this thread, enqueued to a global queue, dequeued and
2155 * freed in the manage apps thread.
2157 struct ust_command
*ust_cmd
= NULL
;
2159 DBG("[thread] Manage application registration started");
2161 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2163 if (testpoint(sessiond_thread_registration_apps
)) {
2164 goto error_testpoint
;
2167 ret
= lttcomm_listen_unix_sock(apps_sock
);
2173 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2174 * more will be added to this poll set.
2176 ret
= sessiond_set_thread_pollset(&events
, 2);
2178 goto error_create_poll
;
2181 /* Add the application registration socket */
2182 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2184 goto error_poll_add
;
2187 /* Notify all applications to register */
2188 ret
= notify_ust_apps(1);
2190 ERR("Failed to notify applications or create the wait shared memory.\n"
2191 "Execution continues but there might be problem for already\n"
2192 "running applications that wishes to register.");
2196 DBG("Accepting application registration");
2198 /* Inifinite blocking call, waiting for transmission */
2200 health_poll_entry();
2201 ret
= lttng_poll_wait(&events
, -1);
2205 * Restart interrupted system call.
2207 if (errno
== EINTR
) {
2215 for (i
= 0; i
< nb_fd
; i
++) {
2216 health_code_update();
2218 /* Fetch once the poll data */
2219 revents
= LTTNG_POLL_GETEV(&events
, i
);
2220 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2223 /* No activity for this FD (poll implementation). */
2227 /* Thread quit pipe has been closed. Killing thread. */
2228 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2234 /* Event on the registration socket */
2235 if (pollfd
== apps_sock
) {
2236 if (revents
& LPOLLIN
) {
2237 sock
= lttcomm_accept_unix_sock(apps_sock
);
2243 * Set socket timeout for both receiving and ending.
2244 * app_socket_timeout is in seconds, whereas
2245 * lttcomm_setsockopt_rcv_timeout and
2246 * lttcomm_setsockopt_snd_timeout expect msec as
2249 if (app_socket_timeout
>= 0) {
2250 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2251 app_socket_timeout
* 1000);
2252 (void) lttcomm_setsockopt_snd_timeout(sock
,
2253 app_socket_timeout
* 1000);
2257 * Set the CLOEXEC flag. Return code is useless because
2258 * either way, the show must go on.
2260 (void) utils_set_fd_cloexec(sock
);
2262 /* Create UST registration command for enqueuing */
2263 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2264 if (ust_cmd
== NULL
) {
2265 PERROR("ust command zmalloc");
2274 * Using message-based transmissions to ensure we don't
2275 * have to deal with partially received messages.
2277 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2279 ERR("Exhausted file descriptors allowed for applications.");
2289 health_code_update();
2290 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2293 /* Close socket of the application. */
2298 lttng_fd_put(LTTNG_FD_APPS
, 1);
2302 health_code_update();
2304 ust_cmd
->sock
= sock
;
2307 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2308 " gid:%d sock:%d name:%s (version %d.%d)",
2309 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2310 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2311 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2312 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2315 * Lock free enqueue the registration request. The red pill
2316 * has been taken! This apps will be part of the *system*.
2318 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2321 * Wake the registration queue futex. Implicit memory
2322 * barrier with the exchange in cds_wfcq_enqueue.
2324 futex_nto1_wake(&ust_cmd_queue
.futex
);
2325 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2326 ERR("Register apps socket poll error");
2329 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2338 /* Notify that the registration thread is gone */
2341 if (apps_sock
>= 0) {
2342 ret
= close(apps_sock
);
2352 lttng_fd_put(LTTNG_FD_APPS
, 1);
2354 unlink(apps_unix_sock_path
);
2357 lttng_poll_clean(&events
);
2361 DBG("UST Registration thread cleanup complete");
2364 ERR("Health error occurred in %s", __func__
);
2366 health_unregister(health_sessiond
);
2372 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2373 * exec or it will fails.
2375 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2378 struct timespec timeout
;
2380 /* Make sure we set the readiness flag to 0 because we are NOT ready */
2381 consumer_data
->consumer_thread_is_ready
= 0;
2383 /* Setup pthread condition */
2384 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2387 PERROR("pthread_condattr_init consumer data");
2392 * Set the monotonic clock in order to make sure we DO NOT jump in time
2393 * between the clock_gettime() call and the timedwait call. See bug #324
2394 * for a more details and how we noticed it.
2396 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2399 PERROR("pthread_condattr_setclock consumer data");
2403 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2406 PERROR("pthread_cond_init consumer data");
2410 ret
= pthread_create(&consumer_data
->thread
, NULL
, thread_manage_consumer
,
2414 PERROR("pthread_create consumer");
2419 /* We are about to wait on a pthread condition */
2420 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2422 /* Get time for sem_timedwait absolute timeout */
2423 clock_ret
= clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2425 * Set the timeout for the condition timed wait even if the clock gettime
2426 * call fails since we might loop on that call and we want to avoid to
2427 * increment the timeout too many times.
2429 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2432 * The following loop COULD be skipped in some conditions so this is why we
2433 * set ret to 0 in order to make sure at least one round of the loop is
2439 * Loop until the condition is reached or when a timeout is reached. Note
2440 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2441 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2442 * possible. This loop does not take any chances and works with both of
2445 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2446 if (clock_ret
< 0) {
2447 PERROR("clock_gettime spawn consumer");
2448 /* Infinite wait for the consumerd thread to be ready */
2449 ret
= pthread_cond_wait(&consumer_data
->cond
,
2450 &consumer_data
->cond_mutex
);
2452 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2453 &consumer_data
->cond_mutex
, &timeout
);
2457 /* Release the pthread condition */
2458 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2462 if (ret
== ETIMEDOUT
) {
2466 * Call has timed out so we kill the kconsumerd_thread and return
2469 ERR("Condition timed out. The consumer thread was never ready."
2471 pth_ret
= pthread_cancel(consumer_data
->thread
);
2473 PERROR("pthread_cancel consumer thread");
2476 PERROR("pthread_cond_wait failed consumer thread");
2478 /* Caller is expecting a negative value on failure. */
2483 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2484 if (consumer_data
->pid
== 0) {
2485 ERR("Consumerd did not start");
2486 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2489 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2498 * Join consumer thread
2500 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2504 /* Consumer pid must be a real one. */
2505 if (consumer_data
->pid
> 0) {
2507 ret
= kill(consumer_data
->pid
, SIGTERM
);
2509 PERROR("Error killing consumer daemon");
2512 return pthread_join(consumer_data
->thread
, &status
);
2519 * Fork and exec a consumer daemon (consumerd).
2521 * Return pid if successful else -1.
2523 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2527 const char *consumer_to_use
;
2528 const char *verbosity
;
2531 DBG("Spawning consumerd");
2538 if (opt_verbose_consumer
) {
2539 verbosity
= "--verbose";
2540 } else if (lttng_opt_quiet
) {
2541 verbosity
= "--quiet";
2546 switch (consumer_data
->type
) {
2547 case LTTNG_CONSUMER_KERNEL
:
2549 * Find out which consumerd to execute. We will first try the
2550 * 64-bit path, then the sessiond's installation directory, and
2551 * fallback on the 32-bit one,
2553 DBG3("Looking for a kernel consumer at these locations:");
2554 DBG3(" 1) %s", consumerd64_bin
);
2555 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2556 DBG3(" 3) %s", consumerd32_bin
);
2557 if (stat(consumerd64_bin
, &st
) == 0) {
2558 DBG3("Found location #1");
2559 consumer_to_use
= consumerd64_bin
;
2560 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2561 DBG3("Found location #2");
2562 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2563 } else if (stat(consumerd32_bin
, &st
) == 0) {
2564 DBG3("Found location #3");
2565 consumer_to_use
= consumerd32_bin
;
2567 DBG("Could not find any valid consumerd executable");
2571 DBG("Using kernel consumer at: %s", consumer_to_use
);
2572 ret
= execl(consumer_to_use
,
2573 "lttng-consumerd", verbosity
, "-k",
2574 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2575 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2576 "--group", tracing_group_name
,
2579 case LTTNG_CONSUMER64_UST
:
2581 char *tmpnew
= NULL
;
2583 if (consumerd64_libdir
[0] != '\0') {
2587 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2591 tmplen
= strlen("LD_LIBRARY_PATH=")
2592 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2593 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2598 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2599 strcat(tmpnew
, consumerd64_libdir
);
2600 if (tmp
[0] != '\0') {
2601 strcat(tmpnew
, ":");
2602 strcat(tmpnew
, tmp
);
2604 ret
= putenv(tmpnew
);
2611 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2612 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2613 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2614 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2615 "--group", tracing_group_name
,
2617 if (consumerd64_libdir
[0] != '\0') {
2622 case LTTNG_CONSUMER32_UST
:
2624 char *tmpnew
= NULL
;
2626 if (consumerd32_libdir
[0] != '\0') {
2630 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2634 tmplen
= strlen("LD_LIBRARY_PATH=")
2635 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2636 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2641 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2642 strcat(tmpnew
, consumerd32_libdir
);
2643 if (tmp
[0] != '\0') {
2644 strcat(tmpnew
, ":");
2645 strcat(tmpnew
, tmp
);
2647 ret
= putenv(tmpnew
);
2654 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2655 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2656 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2657 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2658 "--group", tracing_group_name
,
2660 if (consumerd32_libdir
[0] != '\0') {
2666 PERROR("unknown consumer type");
2670 PERROR("Consumer execl()");
2672 /* Reaching this point, we got a failure on our execl(). */
2674 } else if (pid
> 0) {
2677 PERROR("start consumer fork");
2685 * Spawn the consumerd daemon and session daemon thread.
2687 static int start_consumerd(struct consumer_data
*consumer_data
)
2692 * Set the listen() state on the socket since there is a possible race
2693 * between the exec() of the consumer daemon and this call if place in the
2694 * consumer thread. See bug #366 for more details.
2696 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2701 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2702 if (consumer_data
->pid
!= 0) {
2703 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2707 ret
= spawn_consumerd(consumer_data
);
2709 ERR("Spawning consumerd failed");
2710 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2714 /* Setting up the consumer_data pid */
2715 consumer_data
->pid
= ret
;
2716 DBG2("Consumer pid %d", consumer_data
->pid
);
2717 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2719 DBG2("Spawning consumer control thread");
2720 ret
= spawn_consumer_thread(consumer_data
);
2722 ERR("Fatal error spawning consumer control thread");
2730 /* Cleanup already created sockets on error. */
2731 if (consumer_data
->err_sock
>= 0) {
2734 err
= close(consumer_data
->err_sock
);
2736 PERROR("close consumer data error socket");
2743 * Setup necessary data for kernel tracer action.
2745 static int init_kernel_tracer(void)
2749 /* Modprobe lttng kernel modules */
2750 ret
= modprobe_lttng_control();
2755 /* Open debugfs lttng */
2756 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2757 if (kernel_tracer_fd
< 0) {
2758 DBG("Failed to open %s", module_proc_lttng
);
2763 /* Validate kernel version */
2764 ret
= kernel_validate_version(kernel_tracer_fd
);
2769 ret
= modprobe_lttng_data();
2774 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2778 modprobe_remove_lttng_control();
2779 ret
= close(kernel_tracer_fd
);
2783 kernel_tracer_fd
= -1;
2784 return LTTNG_ERR_KERN_VERSION
;
2787 ret
= close(kernel_tracer_fd
);
2793 modprobe_remove_lttng_control();
2796 WARN("No kernel tracer available");
2797 kernel_tracer_fd
= -1;
2799 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2801 return LTTNG_ERR_KERN_NA
;
2807 * Copy consumer output from the tracing session to the domain session. The
2808 * function also applies the right modification on a per domain basis for the
2809 * trace files destination directory.
2811 * Should *NOT* be called with RCU read-side lock held.
2813 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2816 const char *dir_name
;
2817 struct consumer_output
*consumer
;
2820 assert(session
->consumer
);
2823 case LTTNG_DOMAIN_KERNEL
:
2824 DBG3("Copying tracing session consumer output in kernel session");
2826 * XXX: We should audit the session creation and what this function
2827 * does "extra" in order to avoid a destroy since this function is used
2828 * in the domain session creation (kernel and ust) only. Same for UST
2831 if (session
->kernel_session
->consumer
) {
2832 consumer_output_put(session
->kernel_session
->consumer
);
2834 session
->kernel_session
->consumer
=
2835 consumer_copy_output(session
->consumer
);
2836 /* Ease our life a bit for the next part */
2837 consumer
= session
->kernel_session
->consumer
;
2838 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2840 case LTTNG_DOMAIN_JUL
:
2841 case LTTNG_DOMAIN_LOG4J
:
2842 case LTTNG_DOMAIN_PYTHON
:
2843 case LTTNG_DOMAIN_UST
:
2844 DBG3("Copying tracing session consumer output in UST session");
2845 if (session
->ust_session
->consumer
) {
2846 consumer_output_put(session
->ust_session
->consumer
);
2848 session
->ust_session
->consumer
=
2849 consumer_copy_output(session
->consumer
);
2850 /* Ease our life a bit for the next part */
2851 consumer
= session
->ust_session
->consumer
;
2852 dir_name
= DEFAULT_UST_TRACE_DIR
;
2855 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2859 /* Append correct directory to subdir */
2860 strncat(consumer
->subdir
, dir_name
,
2861 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2862 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2871 * Create an UST session and add it to the session ust list.
2873 * Should *NOT* be called with RCU read-side lock held.
2875 static int create_ust_session(struct ltt_session
*session
,
2876 struct lttng_domain
*domain
)
2879 struct ltt_ust_session
*lus
= NULL
;
2883 assert(session
->consumer
);
2885 switch (domain
->type
) {
2886 case LTTNG_DOMAIN_JUL
:
2887 case LTTNG_DOMAIN_LOG4J
:
2888 case LTTNG_DOMAIN_PYTHON
:
2889 case LTTNG_DOMAIN_UST
:
2892 ERR("Unknown UST domain on create session %d", domain
->type
);
2893 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2897 DBG("Creating UST session");
2899 lus
= trace_ust_create_session(session
->id
);
2901 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2905 lus
->uid
= session
->uid
;
2906 lus
->gid
= session
->gid
;
2907 lus
->output_traces
= session
->output_traces
;
2908 lus
->snapshot_mode
= session
->snapshot_mode
;
2909 lus
->live_timer_interval
= session
->live_timer
;
2910 session
->ust_session
= lus
;
2911 if (session
->shm_path
[0]) {
2912 strncpy(lus
->root_shm_path
, session
->shm_path
,
2913 sizeof(lus
->root_shm_path
));
2914 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2915 strncpy(lus
->shm_path
, session
->shm_path
,
2916 sizeof(lus
->shm_path
));
2917 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2918 strncat(lus
->shm_path
, "/ust",
2919 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2921 /* Copy session output to the newly created UST session */
2922 ret
= copy_session_consumer(domain
->type
, session
);
2923 if (ret
!= LTTNG_OK
) {
2931 session
->ust_session
= NULL
;
2936 * Create a kernel tracer session then create the default channel.
2938 static int create_kernel_session(struct ltt_session
*session
)
2942 DBG("Creating kernel session");
2944 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2946 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2950 /* Code flow safety */
2951 assert(session
->kernel_session
);
2953 /* Copy session output to the newly created Kernel session */
2954 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2955 if (ret
!= LTTNG_OK
) {
2959 /* Create directory(ies) on local filesystem. */
2960 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2961 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2962 ret
= run_as_mkdir_recursive(
2963 session
->kernel_session
->consumer
->dst
.trace_path
,
2964 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2966 if (errno
!= EEXIST
) {
2967 ERR("Trace directory creation error");
2973 session
->kernel_session
->uid
= session
->uid
;
2974 session
->kernel_session
->gid
= session
->gid
;
2975 session
->kernel_session
->output_traces
= session
->output_traces
;
2976 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2981 trace_kernel_destroy_session(session
->kernel_session
);
2982 session
->kernel_session
= NULL
;
2987 * Count number of session permitted by uid/gid.
2989 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2992 struct ltt_session
*session
;
2994 DBG("Counting number of available session for UID %d GID %d",
2996 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2998 * Only list the sessions the user can control.
3000 if (!session_access_ok(session
, uid
, gid
)) {
3009 * Process the command requested by the lttng client within the command
3010 * context structure. This function make sure that the return structure (llm)
3011 * is set and ready for transmission before returning.
3013 * Return any error encountered or 0 for success.
3015 * "sock" is only used for special-case var. len data.
3017 * Should *NOT* be called with RCU read-side lock held.
3019 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
3023 int need_tracing_session
= 1;
3026 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
3028 assert(!rcu_read_ongoing());
3032 switch (cmd_ctx
->lsm
->cmd_type
) {
3033 case LTTNG_CREATE_SESSION
:
3034 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3035 case LTTNG_CREATE_SESSION_LIVE
:
3036 case LTTNG_DESTROY_SESSION
:
3037 case LTTNG_LIST_SESSIONS
:
3038 case LTTNG_LIST_DOMAINS
:
3039 case LTTNG_START_TRACE
:
3040 case LTTNG_STOP_TRACE
:
3041 case LTTNG_DATA_PENDING
:
3042 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3043 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3044 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3045 case LTTNG_SNAPSHOT_RECORD
:
3046 case LTTNG_SAVE_SESSION
:
3047 case LTTNG_SET_SESSION_SHM_PATH
:
3048 case LTTNG_METADATA_REGENERATE
:
3055 if (opt_no_kernel
&& need_domain
3056 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3058 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3060 ret
= LTTNG_ERR_KERN_NA
;
3065 /* Deny register consumer if we already have a spawned consumer. */
3066 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3067 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3068 if (kconsumer_data
.pid
> 0) {
3069 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3070 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3073 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3077 * Check for command that don't needs to allocate a returned payload. We do
3078 * this here so we don't have to make the call for no payload at each
3081 switch(cmd_ctx
->lsm
->cmd_type
) {
3082 case LTTNG_LIST_SESSIONS
:
3083 case LTTNG_LIST_TRACEPOINTS
:
3084 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3085 case LTTNG_LIST_DOMAINS
:
3086 case LTTNG_LIST_CHANNELS
:
3087 case LTTNG_LIST_EVENTS
:
3088 case LTTNG_LIST_SYSCALLS
:
3089 case LTTNG_LIST_TRACKER_PIDS
:
3090 case LTTNG_DATA_PENDING
:
3093 /* Setup lttng message with no payload */
3094 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0);
3096 /* This label does not try to unlock the session */
3097 goto init_setup_error
;
3101 /* Commands that DO NOT need a session. */
3102 switch (cmd_ctx
->lsm
->cmd_type
) {
3103 case LTTNG_CREATE_SESSION
:
3104 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3105 case LTTNG_CREATE_SESSION_LIVE
:
3106 case LTTNG_CALIBRATE
:
3107 case LTTNG_LIST_SESSIONS
:
3108 case LTTNG_LIST_TRACEPOINTS
:
3109 case LTTNG_LIST_SYSCALLS
:
3110 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3111 case LTTNG_SAVE_SESSION
:
3112 need_tracing_session
= 0;
3115 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3117 * We keep the session list lock across _all_ commands
3118 * for now, because the per-session lock does not
3119 * handle teardown properly.
3121 session_lock_list();
3122 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3123 if (cmd_ctx
->session
== NULL
) {
3124 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3127 /* Acquire lock for the session */
3128 session_lock(cmd_ctx
->session
);
3134 * Commands that need a valid session but should NOT create one if none
3135 * exists. Instead of creating one and destroying it when the command is
3136 * handled, process that right before so we save some round trip in useless
3139 switch (cmd_ctx
->lsm
->cmd_type
) {
3140 case LTTNG_DISABLE_CHANNEL
:
3141 case LTTNG_DISABLE_EVENT
:
3142 switch (cmd_ctx
->lsm
->domain
.type
) {
3143 case LTTNG_DOMAIN_KERNEL
:
3144 if (!cmd_ctx
->session
->kernel_session
) {
3145 ret
= LTTNG_ERR_NO_CHANNEL
;
3149 case LTTNG_DOMAIN_JUL
:
3150 case LTTNG_DOMAIN_LOG4J
:
3151 case LTTNG_DOMAIN_PYTHON
:
3152 case LTTNG_DOMAIN_UST
:
3153 if (!cmd_ctx
->session
->ust_session
) {
3154 ret
= LTTNG_ERR_NO_CHANNEL
;
3159 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3171 * Check domain type for specific "pre-action".
3173 switch (cmd_ctx
->lsm
->domain
.type
) {
3174 case LTTNG_DOMAIN_KERNEL
:
3176 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3180 /* Kernel tracer check */
3181 if (kernel_tracer_fd
== -1) {
3182 /* Basically, load kernel tracer modules */
3183 ret
= init_kernel_tracer();
3189 /* Consumer is in an ERROR state. Report back to client */
3190 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3191 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3195 /* Need a session for kernel command */
3196 if (need_tracing_session
) {
3197 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3198 ret
= create_kernel_session(cmd_ctx
->session
);
3200 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3205 /* Start the kernel consumer daemon */
3206 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3207 if (kconsumer_data
.pid
== 0 &&
3208 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3209 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3210 ret
= start_consumerd(&kconsumer_data
);
3212 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3215 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3217 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3221 * The consumer was just spawned so we need to add the socket to
3222 * the consumer output of the session if exist.
3224 ret
= consumer_create_socket(&kconsumer_data
,
3225 cmd_ctx
->session
->kernel_session
->consumer
);
3232 case LTTNG_DOMAIN_JUL
:
3233 case LTTNG_DOMAIN_LOG4J
:
3234 case LTTNG_DOMAIN_PYTHON
:
3235 case LTTNG_DOMAIN_UST
:
3237 if (!ust_app_supported()) {
3238 ret
= LTTNG_ERR_NO_UST
;
3241 /* Consumer is in an ERROR state. Report back to client */
3242 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3243 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3247 if (need_tracing_session
) {
3248 /* Create UST session if none exist. */
3249 if (cmd_ctx
->session
->ust_session
== NULL
) {
3250 ret
= create_ust_session(cmd_ctx
->session
,
3251 &cmd_ctx
->lsm
->domain
);
3252 if (ret
!= LTTNG_OK
) {
3257 /* Start the UST consumer daemons */
3259 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3260 if (consumerd64_bin
[0] != '\0' &&
3261 ustconsumer64_data
.pid
== 0 &&
3262 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3263 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3264 ret
= start_consumerd(&ustconsumer64_data
);
3266 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3267 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3271 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3272 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3274 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3278 * Setup socket for consumer 64 bit. No need for atomic access
3279 * since it was set above and can ONLY be set in this thread.
3281 ret
= consumer_create_socket(&ustconsumer64_data
,
3282 cmd_ctx
->session
->ust_session
->consumer
);
3288 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3289 if (consumerd32_bin
[0] != '\0' &&
3290 ustconsumer32_data
.pid
== 0 &&
3291 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3292 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3293 ret
= start_consumerd(&ustconsumer32_data
);
3295 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3296 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3300 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3301 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3303 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3307 * Setup socket for consumer 64 bit. No need for atomic access
3308 * since it was set above and can ONLY be set in this thread.
3310 ret
= consumer_create_socket(&ustconsumer32_data
,
3311 cmd_ctx
->session
->ust_session
->consumer
);
3323 /* Validate consumer daemon state when start/stop trace command */
3324 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3325 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3326 switch (cmd_ctx
->lsm
->domain
.type
) {
3327 case LTTNG_DOMAIN_NONE
:
3329 case LTTNG_DOMAIN_JUL
:
3330 case LTTNG_DOMAIN_LOG4J
:
3331 case LTTNG_DOMAIN_PYTHON
:
3332 case LTTNG_DOMAIN_UST
:
3333 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3334 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3338 case LTTNG_DOMAIN_KERNEL
:
3339 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3340 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3345 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3351 * Check that the UID or GID match that of the tracing session.
3352 * The root user can interact with all sessions.
3354 if (need_tracing_session
) {
3355 if (!session_access_ok(cmd_ctx
->session
,
3356 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3357 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3358 ret
= LTTNG_ERR_EPERM
;
3364 * Send relayd information to consumer as soon as we have a domain and a
3367 if (cmd_ctx
->session
&& need_domain
) {
3369 * Setup relayd if not done yet. If the relayd information was already
3370 * sent to the consumer, this call will gracefully return.
3372 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3373 if (ret
!= LTTNG_OK
) {
3378 /* Process by command type */
3379 switch (cmd_ctx
->lsm
->cmd_type
) {
3380 case LTTNG_ADD_CONTEXT
:
3383 * An LTTNG_ADD_CONTEXT command might have a supplementary
3384 * payload if the context being added is an application context.
3386 if (cmd_ctx
->lsm
->u
.context
.ctx
.ctx
==
3387 LTTNG_EVENT_CONTEXT_APP_CONTEXT
) {
3388 char *provider_name
= NULL
, *context_name
= NULL
;
3389 size_t provider_name_len
=
3390 cmd_ctx
->lsm
->u
.context
.provider_name_len
;
3391 size_t context_name_len
=
3392 cmd_ctx
->lsm
->u
.context
.context_name_len
;
3394 if (provider_name_len
== 0 || context_name_len
== 0) {
3396 * Application provider and context names MUST
3399 ret
= -LTTNG_ERR_INVALID
;
3403 provider_name
= zmalloc(provider_name_len
+ 1);
3404 if (!provider_name
) {
3405 ret
= -LTTNG_ERR_NOMEM
;
3408 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
=
3411 context_name
= zmalloc(context_name_len
+ 1);
3412 if (!context_name
) {
3413 ret
= -LTTNG_ERR_NOMEM
;
3414 goto error_add_context
;
3416 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
=
3419 ret
= lttcomm_recv_unix_sock(sock
, provider_name
,
3422 goto error_add_context
;
3425 ret
= lttcomm_recv_unix_sock(sock
, context_name
,
3428 goto error_add_context
;
3433 * cmd_add_context assumes ownership of the provider and context
3436 ret
= cmd_add_context(cmd_ctx
->session
,
3437 cmd_ctx
->lsm
->domain
.type
,
3438 cmd_ctx
->lsm
->u
.context
.channel_name
,
3439 &cmd_ctx
->lsm
->u
.context
.ctx
,
3440 kernel_poll_pipe
[1]);
3442 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
= NULL
;
3443 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
= NULL
;
3445 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
);
3446 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
);
3452 case LTTNG_DISABLE_CHANNEL
:
3454 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3455 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3458 case LTTNG_DISABLE_EVENT
:
3462 * FIXME: handle filter; for now we just receive the filter's
3463 * bytecode along with the filter expression which are sent by
3464 * liblttng-ctl and discard them.
3466 * This fixes an issue where the client may block while sending
3467 * the filter payload and encounter an error because the session
3468 * daemon closes the socket without ever handling this data.
3470 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3471 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3474 char data
[LTTNG_FILTER_MAX_LEN
];
3476 DBG("Discarding disable event command payload of size %zu", count
);
3478 ret
= lttcomm_recv_unix_sock(sock
, data
,
3479 count
> sizeof(data
) ? sizeof(data
) : count
);
3484 count
-= (size_t) ret
;
3487 /* FIXME: passing packed structure to non-packed pointer */
3488 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3489 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3490 &cmd_ctx
->lsm
->u
.disable
.event
);
3493 case LTTNG_ENABLE_CHANNEL
:
3495 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3496 &cmd_ctx
->lsm
->u
.channel
.chan
, kernel_poll_pipe
[1]);
3499 case LTTNG_TRACK_PID
:
3501 ret
= cmd_track_pid(cmd_ctx
->session
,
3502 cmd_ctx
->lsm
->domain
.type
,
3503 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3506 case LTTNG_UNTRACK_PID
:
3508 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3509 cmd_ctx
->lsm
->domain
.type
,
3510 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3513 case LTTNG_ENABLE_EVENT
:
3515 struct lttng_event_exclusion
*exclusion
= NULL
;
3516 struct lttng_filter_bytecode
*bytecode
= NULL
;
3517 char *filter_expression
= NULL
;
3519 /* Handle exclusion events and receive it from the client. */
3520 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3521 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3523 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3524 (count
* LTTNG_SYMBOL_NAME_LEN
));
3526 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3530 DBG("Receiving var len exclusion event list from client ...");
3531 exclusion
->count
= count
;
3532 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3533 count
* LTTNG_SYMBOL_NAME_LEN
);
3535 DBG("Nothing recv() from client var len data... continuing");
3538 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3543 /* Get filter expression from client. */
3544 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3545 size_t expression_len
=
3546 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3548 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3549 ret
= LTTNG_ERR_FILTER_INVAL
;
3554 filter_expression
= zmalloc(expression_len
);
3555 if (!filter_expression
) {
3557 ret
= LTTNG_ERR_FILTER_NOMEM
;
3561 /* Receive var. len. data */
3562 DBG("Receiving var len filter's expression from client ...");
3563 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3566 DBG("Nothing recv() from client car len data... continuing");
3568 free(filter_expression
);
3570 ret
= LTTNG_ERR_FILTER_INVAL
;
3575 /* Handle filter and get bytecode from client. */
3576 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3577 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3579 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3580 ret
= LTTNG_ERR_FILTER_INVAL
;
3581 free(filter_expression
);
3586 bytecode
= zmalloc(bytecode_len
);
3588 free(filter_expression
);
3590 ret
= LTTNG_ERR_FILTER_NOMEM
;
3594 /* Receive var. len. data */
3595 DBG("Receiving var len filter's bytecode from client ...");
3596 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3598 DBG("Nothing recv() from client car len data... continuing");
3600 free(filter_expression
);
3603 ret
= LTTNG_ERR_FILTER_INVAL
;
3607 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3608 free(filter_expression
);
3611 ret
= LTTNG_ERR_FILTER_INVAL
;
3616 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3617 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3618 &cmd_ctx
->lsm
->u
.enable
.event
,
3619 filter_expression
, bytecode
, exclusion
,
3620 kernel_poll_pipe
[1]);
3623 case LTTNG_LIST_TRACEPOINTS
:
3625 struct lttng_event
*events
;
3628 session_lock_list();
3629 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3630 session_unlock_list();
3631 if (nb_events
< 0) {
3632 /* Return value is a negative lttng_error_code. */
3638 * Setup lttng message with payload size set to the event list size in
3639 * bytes and then copy list into the llm payload.
3641 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3642 sizeof(struct lttng_event
) * nb_events
);
3652 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3654 struct lttng_event_field
*fields
;
3657 session_lock_list();
3658 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3660 session_unlock_list();
3661 if (nb_fields
< 0) {
3662 /* Return value is a negative lttng_error_code. */
3668 * Setup lttng message with payload size set to the event list size in
3669 * bytes and then copy list into the llm payload.
3671 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, fields
,
3672 sizeof(struct lttng_event_field
) * nb_fields
);
3682 case LTTNG_LIST_SYSCALLS
:
3684 struct lttng_event
*events
;
3687 nb_events
= cmd_list_syscalls(&events
);
3688 if (nb_events
< 0) {
3689 /* Return value is a negative lttng_error_code. */
3695 * Setup lttng message with payload size set to the event list size in
3696 * bytes and then copy list into the llm payload.
3698 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3699 sizeof(struct lttng_event
) * nb_events
);
3709 case LTTNG_LIST_TRACKER_PIDS
:
3711 int32_t *pids
= NULL
;
3714 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3715 cmd_ctx
->lsm
->domain
.type
, &pids
);
3717 /* Return value is a negative lttng_error_code. */
3723 * Setup lttng message with payload size set to the event list size in
3724 * bytes and then copy list into the llm payload.
3726 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, pids
,
3727 sizeof(int32_t) * nr_pids
);
3737 case LTTNG_SET_CONSUMER_URI
:
3740 struct lttng_uri
*uris
;
3742 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3743 len
= nb_uri
* sizeof(struct lttng_uri
);
3746 ret
= LTTNG_ERR_INVALID
;
3750 uris
= zmalloc(len
);
3752 ret
= LTTNG_ERR_FATAL
;
3756 /* Receive variable len data */
3757 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3758 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3760 DBG("No URIs received from client... continuing");
3762 ret
= LTTNG_ERR_SESSION_FAIL
;
3767 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3769 if (ret
!= LTTNG_OK
) {
3776 case LTTNG_START_TRACE
:
3778 ret
= cmd_start_trace(cmd_ctx
->session
);
3781 case LTTNG_STOP_TRACE
:
3783 ret
= cmd_stop_trace(cmd_ctx
->session
);
3786 case LTTNG_CREATE_SESSION
:
3789 struct lttng_uri
*uris
= NULL
;
3791 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3792 len
= nb_uri
* sizeof(struct lttng_uri
);
3795 uris
= zmalloc(len
);
3797 ret
= LTTNG_ERR_FATAL
;
3801 /* Receive variable len data */
3802 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3803 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3805 DBG("No URIs received from client... continuing");
3807 ret
= LTTNG_ERR_SESSION_FAIL
;
3812 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3813 DBG("Creating session with ONE network URI is a bad call");
3814 ret
= LTTNG_ERR_SESSION_FAIL
;
3820 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3821 &cmd_ctx
->creds
, 0);
3827 case LTTNG_DESTROY_SESSION
:
3829 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3831 /* Set session to NULL so we do not unlock it after free. */
3832 cmd_ctx
->session
= NULL
;
3835 case LTTNG_LIST_DOMAINS
:
3838 struct lttng_domain
*domains
= NULL
;
3840 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3842 /* Return value is a negative lttng_error_code. */
3847 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, domains
,
3848 nb_dom
* sizeof(struct lttng_domain
));
3858 case LTTNG_LIST_CHANNELS
:
3860 ssize_t payload_size
;
3861 struct lttng_channel
*channels
= NULL
;
3863 payload_size
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3864 cmd_ctx
->session
, &channels
);
3865 if (payload_size
< 0) {
3866 /* Return value is a negative lttng_error_code. */
3867 ret
= -payload_size
;
3871 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, channels
,
3882 case LTTNG_LIST_EVENTS
:
3885 struct lttng_event
*events
= NULL
;
3886 struct lttcomm_event_command_header cmd_header
;
3889 memset(&cmd_header
, 0, sizeof(cmd_header
));
3890 /* Extended infos are included at the end of events */
3891 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
,
3892 cmd_ctx
->session
, cmd_ctx
->lsm
->u
.list
.channel_name
,
3893 &events
, &total_size
);
3896 /* Return value is a negative lttng_error_code. */
3901 cmd_header
.nb_events
= nb_event
;
3902 ret
= setup_lttng_msg(cmd_ctx
, events
, total_size
,
3903 &cmd_header
, sizeof(cmd_header
));
3913 case LTTNG_LIST_SESSIONS
:
3915 unsigned int nr_sessions
;
3916 void *sessions_payload
;
3919 session_lock_list();
3920 nr_sessions
= lttng_sessions_count(
3921 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3922 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3923 payload_len
= sizeof(struct lttng_session
) * nr_sessions
;
3924 sessions_payload
= zmalloc(payload_len
);
3926 if (!sessions_payload
) {
3927 session_unlock_list();
3932 cmd_list_lttng_sessions(sessions_payload
,
3933 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3934 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3935 session_unlock_list();
3937 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, sessions_payload
,
3939 free(sessions_payload
);
3948 case LTTNG_CALIBRATE
:
3950 ret
= cmd_calibrate(cmd_ctx
->lsm
->domain
.type
,
3951 &cmd_ctx
->lsm
->u
.calibrate
);
3954 case LTTNG_REGISTER_CONSUMER
:
3956 struct consumer_data
*cdata
;
3958 switch (cmd_ctx
->lsm
->domain
.type
) {
3959 case LTTNG_DOMAIN_KERNEL
:
3960 cdata
= &kconsumer_data
;
3963 ret
= LTTNG_ERR_UND
;
3967 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3968 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3971 case LTTNG_DATA_PENDING
:
3974 uint8_t pending_ret_byte
;
3976 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
3981 * This function may returns 0 or 1 to indicate whether or not
3982 * there is data pending. In case of error, it should return an
3983 * LTTNG_ERR code. However, some code paths may still return
3984 * a nondescript error code, which we handle by returning an
3987 if (pending_ret
== 0 || pending_ret
== 1) {
3989 * ret will be set to LTTNG_OK at the end of
3992 } else if (pending_ret
< 0) {
3993 ret
= LTTNG_ERR_UNK
;
4000 pending_ret_byte
= (uint8_t) pending_ret
;
4002 /* 1 byte to return whether or not data is pending */
4003 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
,
4004 &pending_ret_byte
, 1);
4013 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
4015 struct lttcomm_lttng_output_id reply
;
4017 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
4018 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
4019 if (ret
!= LTTNG_OK
) {
4023 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &reply
,
4029 /* Copy output list into message payload */
4033 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
4035 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
4036 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
4039 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
4042 struct lttng_snapshot_output
*outputs
= NULL
;
4044 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
4045 if (nb_output
< 0) {
4050 assert((nb_output
> 0 && outputs
) || nb_output
== 0);
4051 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, outputs
,
4052 nb_output
* sizeof(struct lttng_snapshot_output
));
4062 case LTTNG_SNAPSHOT_RECORD
:
4064 ret
= cmd_snapshot_record(cmd_ctx
->session
,
4065 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
4066 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
4069 case LTTNG_CREATE_SESSION_SNAPSHOT
:
4072 struct lttng_uri
*uris
= NULL
;
4074 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4075 len
= nb_uri
* sizeof(struct lttng_uri
);
4078 uris
= zmalloc(len
);
4080 ret
= LTTNG_ERR_FATAL
;
4084 /* Receive variable len data */
4085 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4086 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4088 DBG("No URIs received from client... continuing");
4090 ret
= LTTNG_ERR_SESSION_FAIL
;
4095 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4096 DBG("Creating session with ONE network URI is a bad call");
4097 ret
= LTTNG_ERR_SESSION_FAIL
;
4103 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
4104 nb_uri
, &cmd_ctx
->creds
);
4108 case LTTNG_CREATE_SESSION_LIVE
:
4111 struct lttng_uri
*uris
= NULL
;
4113 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4114 len
= nb_uri
* sizeof(struct lttng_uri
);
4117 uris
= zmalloc(len
);
4119 ret
= LTTNG_ERR_FATAL
;
4123 /* Receive variable len data */
4124 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4125 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4127 DBG("No URIs received from client... continuing");
4129 ret
= LTTNG_ERR_SESSION_FAIL
;
4134 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4135 DBG("Creating session with ONE network URI is a bad call");
4136 ret
= LTTNG_ERR_SESSION_FAIL
;
4142 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4143 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4147 case LTTNG_SAVE_SESSION
:
4149 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4153 case LTTNG_SET_SESSION_SHM_PATH
:
4155 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4156 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4159 case LTTNG_METADATA_REGENERATE
:
4161 ret
= cmd_metadata_regenerate(cmd_ctx
->session
);
4165 ret
= LTTNG_ERR_UND
;
4170 if (cmd_ctx
->llm
== NULL
) {
4171 DBG("Missing llm structure. Allocating one.");
4172 if (setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0) < 0) {
4176 /* Set return code */
4177 cmd_ctx
->llm
->ret_code
= ret
;
4179 if (cmd_ctx
->session
) {
4180 session_unlock(cmd_ctx
->session
);
4182 if (need_tracing_session
) {
4183 session_unlock_list();
4186 assert(!rcu_read_ongoing());
4191 * Thread managing health check socket.
4193 static void *thread_manage_health(void *data
)
4195 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4196 uint32_t revents
, nb_fd
;
4197 struct lttng_poll_event events
;
4198 struct health_comm_msg msg
;
4199 struct health_comm_reply reply
;
4201 DBG("[thread] Manage health check started");
4203 rcu_register_thread();
4205 /* We might hit an error path before this is created. */
4206 lttng_poll_init(&events
);
4208 /* Create unix socket */
4209 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4211 ERR("Unable to create health check Unix socket");
4217 /* lttng health client socket path permissions */
4218 ret
= chown(health_unix_sock_path
, 0,
4219 utils_get_group_id(tracing_group_name
));
4221 ERR("Unable to set group on %s", health_unix_sock_path
);
4227 ret
= chmod(health_unix_sock_path
,
4228 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4230 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4238 * Set the CLOEXEC flag. Return code is useless because either way, the
4241 (void) utils_set_fd_cloexec(sock
);
4243 ret
= lttcomm_listen_unix_sock(sock
);
4249 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4250 * more will be added to this poll set.
4252 ret
= sessiond_set_thread_pollset(&events
, 2);
4257 /* Add the application registration socket */
4258 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4263 sessiond_notify_ready();
4266 DBG("Health check ready");
4268 /* Inifinite blocking call, waiting for transmission */
4270 ret
= lttng_poll_wait(&events
, -1);
4273 * Restart interrupted system call.
4275 if (errno
== EINTR
) {
4283 for (i
= 0; i
< nb_fd
; i
++) {
4284 /* Fetch once the poll data */
4285 revents
= LTTNG_POLL_GETEV(&events
, i
);
4286 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4289 /* No activity for this FD (poll implementation). */
4293 /* Thread quit pipe has been closed. Killing thread. */
4294 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4300 /* Event on the registration socket */
4301 if (pollfd
== sock
) {
4302 if (revents
& LPOLLIN
) {
4304 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4305 ERR("Health socket poll error");
4308 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4314 new_sock
= lttcomm_accept_unix_sock(sock
);
4320 * Set the CLOEXEC flag. Return code is useless because either way, the
4323 (void) utils_set_fd_cloexec(new_sock
);
4325 DBG("Receiving data from client for health...");
4326 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4328 DBG("Nothing recv() from client... continuing");
4329 ret
= close(new_sock
);
4337 rcu_thread_online();
4339 memset(&reply
, 0, sizeof(reply
));
4340 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4342 * health_check_state returns 0 if health is
4345 if (!health_check_state(health_sessiond
, i
)) {
4346 reply
.ret_code
|= 1ULL << i
;
4350 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4352 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4354 ERR("Failed to send health data back to client");
4357 /* End of transmission */
4358 ret
= close(new_sock
);
4368 ERR("Health error occurred in %s", __func__
);
4370 DBG("Health check thread dying");
4371 unlink(health_unix_sock_path
);
4379 lttng_poll_clean(&events
);
4381 rcu_unregister_thread();
4386 * This thread manage all clients request using the unix client socket for
4389 static void *thread_manage_clients(void *data
)
4391 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4393 uint32_t revents
, nb_fd
;
4394 struct command_ctx
*cmd_ctx
= NULL
;
4395 struct lttng_poll_event events
;
4397 DBG("[thread] Manage client started");
4399 rcu_register_thread();
4401 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4403 health_code_update();
4405 ret
= lttcomm_listen_unix_sock(client_sock
);
4411 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4412 * more will be added to this poll set.
4414 ret
= sessiond_set_thread_pollset(&events
, 2);
4416 goto error_create_poll
;
4419 /* Add the application registration socket */
4420 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4425 sessiond_notify_ready();
4426 ret
= sem_post(&load_info
->message_thread_ready
);
4428 PERROR("sem_post message_thread_ready");
4432 /* This testpoint is after we signal readiness to the parent. */
4433 if (testpoint(sessiond_thread_manage_clients
)) {
4437 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4441 health_code_update();
4444 DBG("Accepting client command ...");
4446 /* Inifinite blocking call, waiting for transmission */
4448 health_poll_entry();
4449 ret
= lttng_poll_wait(&events
, -1);
4453 * Restart interrupted system call.
4455 if (errno
== EINTR
) {
4463 for (i
= 0; i
< nb_fd
; i
++) {
4464 /* Fetch once the poll data */
4465 revents
= LTTNG_POLL_GETEV(&events
, i
);
4466 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4468 health_code_update();
4471 /* No activity for this FD (poll implementation). */
4475 /* Thread quit pipe has been closed. Killing thread. */
4476 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4482 /* Event on the registration socket */
4483 if (pollfd
== client_sock
) {
4484 if (revents
& LPOLLIN
) {
4486 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4487 ERR("Client socket poll error");
4490 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4496 DBG("Wait for client response");
4498 health_code_update();
4500 sock
= lttcomm_accept_unix_sock(client_sock
);
4506 * Set the CLOEXEC flag. Return code is useless because either way, the
4509 (void) utils_set_fd_cloexec(sock
);
4511 /* Set socket option for credentials retrieval */
4512 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4517 /* Allocate context command to process the client request */
4518 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4519 if (cmd_ctx
== NULL
) {
4520 PERROR("zmalloc cmd_ctx");
4524 /* Allocate data buffer for reception */
4525 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4526 if (cmd_ctx
->lsm
== NULL
) {
4527 PERROR("zmalloc cmd_ctx->lsm");
4531 cmd_ctx
->llm
= NULL
;
4532 cmd_ctx
->session
= NULL
;
4534 health_code_update();
4537 * Data is received from the lttng client. The struct
4538 * lttcomm_session_msg (lsm) contains the command and data request of
4541 DBG("Receiving data from client ...");
4542 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4543 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4545 DBG("Nothing recv() from client... continuing");
4551 clean_command_ctx(&cmd_ctx
);
4555 health_code_update();
4557 // TODO: Validate cmd_ctx including sanity check for
4558 // security purpose.
4560 rcu_thread_online();
4562 * This function dispatch the work to the kernel or userspace tracer
4563 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4564 * informations for the client. The command context struct contains
4565 * everything this function may needs.
4567 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4568 rcu_thread_offline();
4576 * TODO: Inform client somehow of the fatal error. At
4577 * this point, ret < 0 means that a zmalloc failed
4578 * (ENOMEM). Error detected but still accept
4579 * command, unless a socket error has been
4582 clean_command_ctx(&cmd_ctx
);
4586 health_code_update();
4588 DBG("Sending response (size: %d, retcode: %s (%d))",
4589 cmd_ctx
->lttng_msg_size
,
4590 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4591 cmd_ctx
->llm
->ret_code
);
4592 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4594 ERR("Failed to send data back to client");
4597 /* End of transmission */
4604 clean_command_ctx(&cmd_ctx
);
4606 health_code_update();
4618 lttng_poll_clean(&events
);
4619 clean_command_ctx(&cmd_ctx
);
4623 unlink(client_unix_sock_path
);
4624 if (client_sock
>= 0) {
4625 ret
= close(client_sock
);
4633 ERR("Health error occurred in %s", __func__
);
4636 health_unregister(health_sessiond
);
4638 DBG("Client thread dying");
4640 rcu_unregister_thread();
4643 * Since we are creating the consumer threads, we own them, so we need
4644 * to join them before our thread exits.
4646 ret
= join_consumer_thread(&kconsumer_data
);
4649 PERROR("join_consumer");
4652 ret
= join_consumer_thread(&ustconsumer32_data
);
4655 PERROR("join_consumer ust32");
4658 ret
= join_consumer_thread(&ustconsumer64_data
);
4661 PERROR("join_consumer ust64");
4666 static int string_match(const char *str1
, const char *str2
)
4668 return (str1
&& str2
) && !strcmp(str1
, str2
);
4672 * Take an option from the getopt output and set it in the right variable to be
4675 * Return 0 on success else a negative value.
4677 static int set_option(int opt
, const char *arg
, const char *optname
)
4681 if (string_match(optname
, "client-sock") || opt
== 'c') {
4682 if (!arg
|| *arg
== '\0') {
4686 if (lttng_is_setuid_setgid()) {
4687 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4688 "-c, --client-sock");
4690 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4692 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4693 if (!arg
|| *arg
== '\0') {
4697 if (lttng_is_setuid_setgid()) {
4698 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4701 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4703 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4705 } else if (string_match(optname
, "background") || opt
== 'b') {
4707 } else if (string_match(optname
, "group") || opt
== 'g') {
4708 if (!arg
|| *arg
== '\0') {
4712 if (lttng_is_setuid_setgid()) {
4713 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4717 * If the override option is set, the pointer points to a
4718 * *non* const thus freeing it even though the variable type is
4721 if (tracing_group_name_override
) {
4722 free((void *) tracing_group_name
);
4724 tracing_group_name
= strdup(arg
);
4725 if (!tracing_group_name
) {
4729 tracing_group_name_override
= 1;
4731 } else if (string_match(optname
, "help") || opt
== 'h') {
4732 ret
= utils_show_man_page(8, "lttng-sessiond");
4734 ERR("Cannot view man page lttng-sessiond(8)");
4737 exit(ret
? EXIT_FAILURE
: EXIT_SUCCESS
);
4738 } else if (string_match(optname
, "version") || opt
== 'V') {
4739 fprintf(stdout
, "%s\n", VERSION
);
4741 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4743 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4744 if (!arg
|| *arg
== '\0') {
4748 if (lttng_is_setuid_setgid()) {
4749 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4750 "--kconsumerd-err-sock");
4752 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4754 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4755 if (!arg
|| *arg
== '\0') {
4759 if (lttng_is_setuid_setgid()) {
4760 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4761 "--kconsumerd-cmd-sock");
4763 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4765 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4766 if (!arg
|| *arg
== '\0') {
4770 if (lttng_is_setuid_setgid()) {
4771 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4772 "--ustconsumerd64-err-sock");
4774 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4776 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4777 if (!arg
|| *arg
== '\0') {
4781 if (lttng_is_setuid_setgid()) {
4782 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4783 "--ustconsumerd64-cmd-sock");
4785 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4787 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4788 if (!arg
|| *arg
== '\0') {
4792 if (lttng_is_setuid_setgid()) {
4793 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4794 "--ustconsumerd32-err-sock");
4796 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4798 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4799 if (!arg
|| *arg
== '\0') {
4803 if (lttng_is_setuid_setgid()) {
4804 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4805 "--ustconsumerd32-cmd-sock");
4807 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4809 } else if (string_match(optname
, "no-kernel")) {
4811 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4812 lttng_opt_quiet
= 1;
4813 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4814 /* Verbose level can increase using multiple -v */
4816 /* Value obtained from config file */
4817 lttng_opt_verbose
= config_parse_value(arg
);
4819 /* -v used on command line */
4820 lttng_opt_verbose
++;
4822 /* Clamp value to [0, 3] */
4823 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4824 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4825 } else if (string_match(optname
, "verbose-consumer")) {
4827 opt_verbose_consumer
= config_parse_value(arg
);
4829 opt_verbose_consumer
++;
4831 } else if (string_match(optname
, "consumerd32-path")) {
4832 if (!arg
|| *arg
== '\0') {
4836 if (lttng_is_setuid_setgid()) {
4837 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4838 "--consumerd32-path");
4840 if (consumerd32_bin_override
) {
4841 free((void *) consumerd32_bin
);
4843 consumerd32_bin
= strdup(arg
);
4844 if (!consumerd32_bin
) {
4848 consumerd32_bin_override
= 1;
4850 } else if (string_match(optname
, "consumerd32-libdir")) {
4851 if (!arg
|| *arg
== '\0') {
4855 if (lttng_is_setuid_setgid()) {
4856 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4857 "--consumerd32-libdir");
4859 if (consumerd32_libdir_override
) {
4860 free((void *) consumerd32_libdir
);
4862 consumerd32_libdir
= strdup(arg
);
4863 if (!consumerd32_libdir
) {
4867 consumerd32_libdir_override
= 1;
4869 } else if (string_match(optname
, "consumerd64-path")) {
4870 if (!arg
|| *arg
== '\0') {
4874 if (lttng_is_setuid_setgid()) {
4875 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4876 "--consumerd64-path");
4878 if (consumerd64_bin_override
) {
4879 free((void *) consumerd64_bin
);
4881 consumerd64_bin
= strdup(arg
);
4882 if (!consumerd64_bin
) {
4886 consumerd64_bin_override
= 1;
4888 } else if (string_match(optname
, "consumerd64-libdir")) {
4889 if (!arg
|| *arg
== '\0') {
4893 if (lttng_is_setuid_setgid()) {
4894 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4895 "--consumerd64-libdir");
4897 if (consumerd64_libdir_override
) {
4898 free((void *) consumerd64_libdir
);
4900 consumerd64_libdir
= strdup(arg
);
4901 if (!consumerd64_libdir
) {
4905 consumerd64_libdir_override
= 1;
4907 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4908 if (!arg
|| *arg
== '\0') {
4912 if (lttng_is_setuid_setgid()) {
4913 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4917 opt_pidfile
= strdup(arg
);
4923 } else if (string_match(optname
, "agent-tcp-port")) {
4924 if (!arg
|| *arg
== '\0') {
4928 if (lttng_is_setuid_setgid()) {
4929 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4930 "--agent-tcp-port");
4939 v
= strtoul(arg
, NULL
, 0);
4940 if (errno
!= 0 || !isdigit(arg
[0])) {
4941 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4944 if (v
== 0 || v
>= 65535) {
4945 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4948 agent_tcp_port
= (uint32_t) v
;
4949 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4951 } else if (string_match(optname
, "load") || opt
== 'l') {
4952 if (!arg
|| *arg
== '\0') {
4956 if (lttng_is_setuid_setgid()) {
4957 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4960 free(opt_load_session_path
);
4961 opt_load_session_path
= strdup(arg
);
4962 if (!opt_load_session_path
) {
4967 } else if (string_match(optname
, "kmod-probes")) {
4968 if (!arg
|| *arg
== '\0') {
4972 if (lttng_is_setuid_setgid()) {
4973 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4976 free(kmod_probes_list
);
4977 kmod_probes_list
= strdup(arg
);
4978 if (!kmod_probes_list
) {
4983 } else if (string_match(optname
, "extra-kmod-probes")) {
4984 if (!arg
|| *arg
== '\0') {
4988 if (lttng_is_setuid_setgid()) {
4989 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4990 "--extra-kmod-probes");
4992 free(kmod_extra_probes_list
);
4993 kmod_extra_probes_list
= strdup(arg
);
4994 if (!kmod_extra_probes_list
) {
4999 } else if (string_match(optname
, "config") || opt
== 'f') {
5000 /* This is handled in set_options() thus silent skip. */
5003 /* Unknown option or other error.
5004 * Error is printed by getopt, just return */
5009 if (ret
== -EINVAL
) {
5010 const char *opt_name
= "unknown";
5013 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
5015 if (opt
== long_options
[i
].val
) {
5016 opt_name
= long_options
[i
].name
;
5021 WARN("Invalid argument provided for option \"%s\", using default value.",
5029 * config_entry_handler_cb used to handle options read from a config file.
5030 * See config_entry_handler_cb comment in common/config/session-config.h for the
5031 * return value conventions.
5033 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
5037 if (!entry
|| !entry
->name
|| !entry
->value
) {
5042 /* Check if the option is to be ignored */
5043 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
5044 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
5049 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
5052 /* Ignore if not fully matched. */
5053 if (strcmp(entry
->name
, long_options
[i
].name
)) {
5058 * If the option takes no argument on the command line, we have to
5059 * check if the value is "true". We support non-zero numeric values,
5062 if (!long_options
[i
].has_arg
) {
5063 ret
= config_parse_value(entry
->value
);
5066 WARN("Invalid configuration value \"%s\" for option %s",
5067 entry
->value
, entry
->name
);
5069 /* False, skip boolean config option. */
5074 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
5078 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
5085 * daemon configuration loading and argument parsing
5087 static int set_options(int argc
, char **argv
)
5089 int ret
= 0, c
= 0, option_index
= 0;
5090 int orig_optopt
= optopt
, orig_optind
= optind
;
5092 const char *config_path
= NULL
;
5094 optstring
= utils_generate_optstring(long_options
,
5095 sizeof(long_options
) / sizeof(struct option
));
5101 /* Check for the --config option */
5102 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5103 &option_index
)) != -1) {
5107 } else if (c
!= 'f') {
5108 /* if not equal to --config option. */
5112 if (lttng_is_setuid_setgid()) {
5113 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5116 config_path
= utils_expand_path(optarg
);
5118 ERR("Failed to resolve path: %s", optarg
);
5123 ret
= config_get_section_entries(config_path
, config_section_name
,
5124 config_entry_handler
, NULL
);
5127 ERR("Invalid configuration option at line %i", ret
);
5133 /* Reset getopt's global state */
5134 optopt
= orig_optopt
;
5135 optind
= orig_optind
;
5139 * getopt_long() will not set option_index if it encounters a
5142 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5149 * Pass NULL as the long option name if popt left the index
5152 ret
= set_option(c
, optarg
,
5153 option_index
< 0 ? NULL
:
5154 long_options
[option_index
].name
);
5166 * Creates the two needed socket by the daemon.
5167 * apps_sock - The communication socket for all UST apps.
5168 * client_sock - The communication of the cli tool (lttng).
5170 static int init_daemon_socket(void)
5175 old_umask
= umask(0);
5177 /* Create client tool unix socket */
5178 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
5179 if (client_sock
< 0) {
5180 ERR("Create unix sock failed: %s", client_unix_sock_path
);
5185 /* Set the cloexec flag */
5186 ret
= utils_set_fd_cloexec(client_sock
);
5188 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5189 "Continuing but note that the consumer daemon will have a "
5190 "reference to this socket on exec()", client_sock
);
5193 /* File permission MUST be 660 */
5194 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5196 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5201 /* Create the application unix socket */
5202 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5203 if (apps_sock
< 0) {
5204 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5209 /* Set the cloexec flag */
5210 ret
= utils_set_fd_cloexec(apps_sock
);
5212 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5213 "Continuing but note that the consumer daemon will have a "
5214 "reference to this socket on exec()", apps_sock
);
5217 /* File permission MUST be 666 */
5218 ret
= chmod(apps_unix_sock_path
,
5219 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5221 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5226 DBG3("Session daemon client socket %d and application socket %d created",
5227 client_sock
, apps_sock
);
5235 * Check if the global socket is available, and if a daemon is answering at the
5236 * other side. If yes, error is returned.
5238 static int check_existing_daemon(void)
5240 /* Is there anybody out there ? */
5241 if (lttng_session_daemon_alive()) {
5249 * Set the tracing group gid onto the client socket.
5251 * Race window between mkdir and chown is OK because we are going from more
5252 * permissive (root.root) to less permissive (root.tracing).
5254 static int set_permissions(char *rundir
)
5259 gid
= utils_get_group_id(tracing_group_name
);
5261 /* Set lttng run dir */
5262 ret
= chown(rundir
, 0, gid
);
5264 ERR("Unable to set group on %s", rundir
);
5269 * Ensure all applications and tracing group can search the run
5270 * dir. Allow everyone to read the directory, since it does not
5271 * buy us anything to hide its content.
5273 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5275 ERR("Unable to set permissions on %s", rundir
);
5279 /* lttng client socket path */
5280 ret
= chown(client_unix_sock_path
, 0, gid
);
5282 ERR("Unable to set group on %s", client_unix_sock_path
);
5286 /* kconsumer error socket path */
5287 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5289 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5293 /* 64-bit ustconsumer error socket path */
5294 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5296 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5300 /* 32-bit ustconsumer compat32 error socket path */
5301 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5303 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5307 DBG("All permissions are set");
5313 * Create the lttng run directory needed for all global sockets and pipe.
5315 static int create_lttng_rundir(const char *rundir
)
5319 DBG3("Creating LTTng run directory: %s", rundir
);
5321 ret
= mkdir(rundir
, S_IRWXU
);
5323 if (errno
!= EEXIST
) {
5324 ERR("Unable to create %s", rundir
);
5336 * Setup sockets and directory needed by the kconsumerd communication with the
5339 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5343 char path
[PATH_MAX
];
5345 switch (consumer_data
->type
) {
5346 case LTTNG_CONSUMER_KERNEL
:
5347 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5349 case LTTNG_CONSUMER64_UST
:
5350 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5352 case LTTNG_CONSUMER32_UST
:
5353 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5356 ERR("Consumer type unknown");
5361 DBG2("Creating consumer directory: %s", path
);
5363 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5365 if (errno
!= EEXIST
) {
5367 ERR("Failed to create %s", path
);
5373 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5375 ERR("Unable to set group on %s", path
);
5381 /* Create the kconsumerd error unix socket */
5382 consumer_data
->err_sock
=
5383 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5384 if (consumer_data
->err_sock
< 0) {
5385 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5391 * Set the CLOEXEC flag. Return code is useless because either way, the
5394 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5396 PERROR("utils_set_fd_cloexec");
5397 /* continue anyway */
5400 /* File permission MUST be 660 */
5401 ret
= chmod(consumer_data
->err_unix_sock_path
,
5402 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5404 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5414 * Signal handler for the daemon
5416 * Simply stop all worker threads, leaving main() return gracefully after
5417 * joining all threads and calling cleanup().
5419 static void sighandler(int sig
)
5423 DBG("SIGINT caught");
5427 DBG("SIGTERM caught");
5431 CMM_STORE_SHARED(recv_child_signal
, 1);
5439 * Setup signal handler for :
5440 * SIGINT, SIGTERM, SIGPIPE
5442 static int set_signal_handler(void)
5445 struct sigaction sa
;
5448 if ((ret
= sigemptyset(&sigset
)) < 0) {
5449 PERROR("sigemptyset");
5453 sa
.sa_mask
= sigset
;
5456 sa
.sa_handler
= sighandler
;
5457 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5458 PERROR("sigaction");
5462 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5463 PERROR("sigaction");
5467 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5468 PERROR("sigaction");
5472 sa
.sa_handler
= SIG_IGN
;
5473 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5474 PERROR("sigaction");
5478 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5484 * Set open files limit to unlimited. This daemon can open a large number of
5485 * file descriptors in order to consumer multiple kernel traces.
5487 static void set_ulimit(void)
5492 /* The kernel does not allowed an infinite limit for open files */
5493 lim
.rlim_cur
= 65535;
5494 lim
.rlim_max
= 65535;
5496 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5498 PERROR("failed to set open files limit");
5503 * Write pidfile using the rundir and opt_pidfile.
5505 static int write_pidfile(void)
5508 char pidfile_path
[PATH_MAX
];
5513 if (lttng_strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
))) {
5518 /* Build pidfile path from rundir and opt_pidfile. */
5519 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5520 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5522 PERROR("snprintf pidfile path");
5528 * Create pid file in rundir.
5530 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5536 * Create lockfile using the rundir and return its fd.
5538 static int create_lockfile(void)
5541 char lockfile_path
[PATH_MAX
];
5543 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5548 ret
= utils_create_lock_file(lockfile_path
);
5554 * Write agent TCP port using the rundir.
5556 static int write_agent_port(void)
5559 char path
[PATH_MAX
];
5563 ret
= snprintf(path
, sizeof(path
), "%s/"
5564 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5566 PERROR("snprintf agent port path");
5571 * Create TCP agent port file in rundir.
5573 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5582 int main(int argc
, char **argv
)
5584 int ret
= 0, retval
= 0;
5586 const char *home_path
, *env_app_timeout
;
5588 init_kernel_workarounds();
5590 rcu_register_thread();
5592 if (set_signal_handler()) {
5594 goto exit_set_signal_handler
;
5597 setup_consumerd_path();
5599 page_size
= sysconf(_SC_PAGESIZE
);
5600 if (page_size
< 0) {
5601 PERROR("sysconf _SC_PAGESIZE");
5602 page_size
= LONG_MAX
;
5603 WARN("Fallback page size to %ld", page_size
);
5607 * Parse arguments and load the daemon configuration file.
5609 * We have an exit_options exit path to free memory reserved by
5610 * set_options. This is needed because the rest of sessiond_cleanup()
5611 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5612 * depends on set_options.
5615 if (set_options(argc
, argv
)) {
5621 if (opt_daemon
|| opt_background
) {
5624 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5632 * We are in the child. Make sure all other file descriptors are
5633 * closed, in case we are called with more opened file
5634 * descriptors than the standard ones.
5636 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5641 if (run_as_create_worker(argv
[0]) < 0) {
5642 goto exit_create_run_as_worker_cleanup
;
5646 * Starting from here, we can create threads. This needs to be after
5647 * lttng_daemonize due to RCU.
5651 * Initialize the health check subsystem. This call should set the
5652 * appropriate time values.
5654 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5655 if (!health_sessiond
) {
5656 PERROR("health_app_create error");
5658 goto exit_health_sessiond_cleanup
;
5661 if (init_ht_cleanup_quit_pipe()) {
5663 goto exit_ht_cleanup_quit_pipe
;
5666 /* Setup the thread ht_cleanup communication pipe. */
5667 if (utils_create_pipe_cloexec(ht_cleanup_pipe
)) {
5669 goto exit_ht_cleanup_pipe
;
5672 /* Set up max poll set size */
5673 if (lttng_poll_set_max_size()) {
5675 goto exit_set_max_size
;
5678 /* Create thread to clean up RCU hash tables */
5679 ret
= pthread_create(&ht_cleanup_thread
, NULL
,
5680 thread_ht_cleanup
, (void *) NULL
);
5683 PERROR("pthread_create ht_cleanup");
5685 goto exit_ht_cleanup
;
5688 /* Create thread quit pipe */
5689 if (init_thread_quit_pipe()) {
5691 goto exit_init_data
;
5694 /* Check if daemon is UID = 0 */
5695 is_root
= !getuid();
5698 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5701 goto exit_init_data
;
5704 /* Create global run dir with root access */
5705 if (create_lttng_rundir(rundir
)) {
5707 goto exit_init_data
;
5710 if (strlen(apps_unix_sock_path
) == 0) {
5711 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5712 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5715 goto exit_init_data
;
5719 if (strlen(client_unix_sock_path
) == 0) {
5720 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5721 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5724 goto exit_init_data
;
5728 /* Set global SHM for ust */
5729 if (strlen(wait_shm_path
) == 0) {
5730 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5731 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5734 goto exit_init_data
;
5738 if (strlen(health_unix_sock_path
) == 0) {
5739 ret
= snprintf(health_unix_sock_path
,
5740 sizeof(health_unix_sock_path
),
5741 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5744 goto exit_init_data
;
5748 /* Setup kernel consumerd path */
5749 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5750 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5753 goto exit_init_data
;
5755 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5756 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5759 goto exit_init_data
;
5762 DBG2("Kernel consumer err path: %s",
5763 kconsumer_data
.err_unix_sock_path
);
5764 DBG2("Kernel consumer cmd path: %s",
5765 kconsumer_data
.cmd_unix_sock_path
);
5767 home_path
= utils_get_home_dir();
5768 if (home_path
== NULL
) {
5769 /* TODO: Add --socket PATH option */
5770 ERR("Can't get HOME directory for sockets creation.");
5772 goto exit_init_data
;
5776 * Create rundir from home path. This will create something like
5779 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5782 goto exit_init_data
;
5785 if (create_lttng_rundir(rundir
)) {
5787 goto exit_init_data
;
5790 if (strlen(apps_unix_sock_path
) == 0) {
5791 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5792 DEFAULT_HOME_APPS_UNIX_SOCK
,
5796 goto exit_init_data
;
5800 /* Set the cli tool unix socket path */
5801 if (strlen(client_unix_sock_path
) == 0) {
5802 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5803 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5807 goto exit_init_data
;
5811 /* Set global SHM for ust */
5812 if (strlen(wait_shm_path
) == 0) {
5813 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5814 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5818 goto exit_init_data
;
5822 /* Set health check Unix path */
5823 if (strlen(health_unix_sock_path
) == 0) {
5824 ret
= snprintf(health_unix_sock_path
,
5825 sizeof(health_unix_sock_path
),
5826 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5830 goto exit_init_data
;
5835 lockfile_fd
= create_lockfile();
5836 if (lockfile_fd
< 0) {
5838 goto exit_init_data
;
5841 /* Set consumer initial state */
5842 kernel_consumerd_state
= CONSUMER_STOPPED
;
5843 ust_consumerd_state
= CONSUMER_STOPPED
;
5845 DBG("Client socket path %s", client_unix_sock_path
);
5846 DBG("Application socket path %s", apps_unix_sock_path
);
5847 DBG("Application wait path %s", wait_shm_path
);
5848 DBG("LTTng run directory path: %s", rundir
);
5850 /* 32 bits consumerd path setup */
5851 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5852 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5854 PERROR("snprintf 32-bit consumer error socket path");
5856 goto exit_init_data
;
5858 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5859 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5861 PERROR("snprintf 32-bit consumer command socket path");
5863 goto exit_init_data
;
5866 DBG2("UST consumer 32 bits err path: %s",
5867 ustconsumer32_data
.err_unix_sock_path
);
5868 DBG2("UST consumer 32 bits cmd path: %s",
5869 ustconsumer32_data
.cmd_unix_sock_path
);
5871 /* 64 bits consumerd path setup */
5872 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5873 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5875 PERROR("snprintf 64-bit consumer error socket path");
5877 goto exit_init_data
;
5879 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5880 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5882 PERROR("snprintf 64-bit consumer command socket path");
5884 goto exit_init_data
;
5887 DBG2("UST consumer 64 bits err path: %s",
5888 ustconsumer64_data
.err_unix_sock_path
);
5889 DBG2("UST consumer 64 bits cmd path: %s",
5890 ustconsumer64_data
.cmd_unix_sock_path
);
5893 * See if daemon already exist.
5895 if (check_existing_daemon()) {
5896 ERR("Already running daemon.\n");
5898 * We do not goto exit because we must not cleanup()
5899 * because a daemon is already running.
5902 goto exit_init_data
;
5906 * Init UST app hash table. Alloc hash table before this point since
5907 * cleanup() can get called after that point.
5909 if (ust_app_ht_alloc()) {
5910 ERR("Failed to allocate UST app hash table");
5912 goto exit_init_data
;
5916 * Initialize agent app hash table. We allocate the hash table here
5917 * since cleanup() can get called after this point.
5919 if (agent_app_ht_alloc()) {
5920 ERR("Failed to allocate Agent app hash table");
5922 goto exit_init_data
;
5926 * These actions must be executed as root. We do that *after* setting up
5927 * the sockets path because we MUST make the check for another daemon using
5928 * those paths *before* trying to set the kernel consumer sockets and init
5932 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5934 goto exit_init_data
;
5937 /* Setup kernel tracer */
5938 if (!opt_no_kernel
) {
5939 init_kernel_tracer();
5940 if (kernel_tracer_fd
>= 0) {
5941 ret
= syscall_init_table();
5943 ERR("Unable to populate syscall table. "
5944 "Syscall tracing won't work "
5945 "for this session daemon.");
5950 /* Set ulimit for open files */
5953 /* init lttng_fd tracking must be done after set_ulimit. */
5956 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
5958 goto exit_init_data
;
5961 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
5963 goto exit_init_data
;
5966 /* Setup the needed unix socket */
5967 if (init_daemon_socket()) {
5969 goto exit_init_data
;
5972 /* Set credentials to socket */
5973 if (is_root
&& set_permissions(rundir
)) {
5975 goto exit_init_data
;
5978 /* Get parent pid if -S, --sig-parent is specified. */
5979 if (opt_sig_parent
) {
5983 /* Setup the kernel pipe for waking up the kernel thread */
5984 if (is_root
&& !opt_no_kernel
) {
5985 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5987 goto exit_init_data
;
5991 /* Setup the thread apps communication pipe. */
5992 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
5994 goto exit_init_data
;
5997 /* Setup the thread apps notify communication pipe. */
5998 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
6000 goto exit_init_data
;
6003 /* Initialize global buffer per UID and PID registry. */
6004 buffer_reg_init_uid_registry();
6005 buffer_reg_init_pid_registry();
6007 /* Init UST command queue. */
6008 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
6011 * Get session list pointer. This pointer MUST NOT be free'd. This list
6012 * is statically declared in session.c
6014 session_list_ptr
= session_get_list();
6018 /* Check for the application socket timeout env variable. */
6019 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
6020 if (env_app_timeout
) {
6021 app_socket_timeout
= atoi(env_app_timeout
);
6023 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
6026 ret
= write_pidfile();
6028 ERR("Error in write_pidfile");
6030 goto exit_init_data
;
6032 ret
= write_agent_port();
6034 ERR("Error in write_agent_port");
6036 goto exit_init_data
;
6039 /* Initialize communication library */
6041 /* Initialize TCP timeout values */
6042 lttcomm_inet_init();
6044 if (load_session_init_data(&load_info
) < 0) {
6046 goto exit_init_data
;
6048 load_info
->path
= opt_load_session_path
;
6050 /* Create health-check thread */
6051 ret
= pthread_create(&health_thread
, NULL
,
6052 thread_manage_health
, (void *) NULL
);
6055 PERROR("pthread_create health");
6060 /* Create thread to manage the client socket */
6061 ret
= pthread_create(&client_thread
, NULL
,
6062 thread_manage_clients
, (void *) NULL
);
6065 PERROR("pthread_create clients");
6070 /* Create thread to dispatch registration */
6071 ret
= pthread_create(&dispatch_thread
, NULL
,
6072 thread_dispatch_ust_registration
, (void *) NULL
);
6075 PERROR("pthread_create dispatch");
6080 /* Create thread to manage application registration. */
6081 ret
= pthread_create(®_apps_thread
, NULL
,
6082 thread_registration_apps
, (void *) NULL
);
6085 PERROR("pthread_create registration");
6090 /* Create thread to manage application socket */
6091 ret
= pthread_create(&apps_thread
, NULL
,
6092 thread_manage_apps
, (void *) NULL
);
6095 PERROR("pthread_create apps");
6100 /* Create thread to manage application notify socket */
6101 ret
= pthread_create(&apps_notify_thread
, NULL
,
6102 ust_thread_manage_notify
, (void *) NULL
);
6105 PERROR("pthread_create notify");
6107 goto exit_apps_notify
;
6110 /* Create agent registration thread. */
6111 ret
= pthread_create(&agent_reg_thread
, NULL
,
6112 agent_thread_manage_registration
, (void *) NULL
);
6115 PERROR("pthread_create agent");
6117 goto exit_agent_reg
;
6120 /* Don't start this thread if kernel tracing is not requested nor root */
6121 if (is_root
&& !opt_no_kernel
) {
6122 /* Create kernel thread to manage kernel event */
6123 ret
= pthread_create(&kernel_thread
, NULL
,
6124 thread_manage_kernel
, (void *) NULL
);
6127 PERROR("pthread_create kernel");
6133 /* Create session loading thread. */
6134 ret
= pthread_create(&load_session_thread
, NULL
, thread_load_session
,
6138 PERROR("pthread_create load_session_thread");
6140 goto exit_load_session
;
6144 * This is where we start awaiting program completion (e.g. through
6145 * signal that asks threads to teardown).
6148 ret
= pthread_join(load_session_thread
, &status
);
6151 PERROR("pthread_join load_session_thread");
6156 if (is_root
&& !opt_no_kernel
) {
6157 ret
= pthread_join(kernel_thread
, &status
);
6160 PERROR("pthread_join");
6166 ret
= pthread_join(agent_reg_thread
, &status
);
6169 PERROR("pthread_join agent");
6174 ret
= pthread_join(apps_notify_thread
, &status
);
6177 PERROR("pthread_join apps notify");
6182 ret
= pthread_join(apps_thread
, &status
);
6185 PERROR("pthread_join apps");
6190 ret
= pthread_join(reg_apps_thread
, &status
);
6193 PERROR("pthread_join");
6199 * Join dispatch thread after joining reg_apps_thread to ensure
6200 * we don't leak applications in the queue.
6202 ret
= pthread_join(dispatch_thread
, &status
);
6205 PERROR("pthread_join");
6210 ret
= pthread_join(client_thread
, &status
);
6213 PERROR("pthread_join");
6218 ret
= pthread_join(health_thread
, &status
);
6221 PERROR("pthread_join health thread");
6228 * Wait for all pending call_rcu work to complete before tearing
6229 * down data structures. call_rcu worker may be trying to
6230 * perform lookups in those structures.
6234 * sessiond_cleanup() is called when no other thread is running, except
6235 * the ht_cleanup thread, which is needed to destroy the hash tables.
6237 rcu_thread_online();
6239 rcu_thread_offline();
6240 rcu_unregister_thread();
6243 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6244 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6245 * the queue is empty before shutting down the clean-up thread.
6249 ret
= notify_thread_pipe(ht_cleanup_quit_pipe
[1]);
6251 ERR("write error on ht_cleanup quit pipe");
6255 ret
= pthread_join(ht_cleanup_thread
, &status
);
6258 PERROR("pthread_join ht cleanup thread");
6264 utils_close_pipe(ht_cleanup_pipe
);
6265 exit_ht_cleanup_pipe
:
6268 * Close the ht_cleanup quit pipe.
6270 utils_close_pipe(ht_cleanup_quit_pipe
);
6271 exit_ht_cleanup_quit_pipe
:
6273 health_app_destroy(health_sessiond
);
6274 exit_health_sessiond_cleanup
:
6275 exit_create_run_as_worker_cleanup
:
6278 sessiond_cleanup_options();
6280 exit_set_signal_handler
: