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"
74 #include "notification-thread.h"
75 #include "notification-thread-commands.h"
78 #include "ht-cleanup.h"
80 #define CONSUMERD_FILE "lttng-consumerd"
82 static const char *help_msg
=
83 #ifdef LTTNG_EMBED_HELP
84 #include <lttng-sessiond.8.h>
91 const char *tracing_group_name
= DEFAULT_TRACING_GROUP
;
92 static int tracing_group_name_override
;
93 static char *opt_pidfile
;
94 static int opt_sig_parent
;
95 static int opt_verbose_consumer
;
96 static int opt_daemon
, opt_background
;
97 static int opt_no_kernel
;
98 static char *opt_load_session_path
;
99 static pid_t ppid
; /* Parent PID for --sig-parent option */
100 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
102 static int lockfile_fd
= -1;
104 /* Set to 1 when a SIGUSR1 signal is received. */
105 static int recv_child_signal
;
108 * Consumer daemon specific control data. Every value not initialized here is
109 * set to 0 by the static definition.
111 static struct consumer_data kconsumer_data
= {
112 .type
= LTTNG_CONSUMER_KERNEL
,
113 .err_unix_sock_path
= DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
114 .cmd_unix_sock_path
= DEFAULT_KCONSUMERD_CMD_SOCK_PATH
,
117 .channel_monitor_pipe
= -1,
118 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
119 .lock
= PTHREAD_MUTEX_INITIALIZER
,
120 .cond
= PTHREAD_COND_INITIALIZER
,
121 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
123 static struct consumer_data ustconsumer64_data
= {
124 .type
= LTTNG_CONSUMER64_UST
,
125 .err_unix_sock_path
= DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
126 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
,
129 .channel_monitor_pipe
= -1,
130 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
131 .lock
= PTHREAD_MUTEX_INITIALIZER
,
132 .cond
= PTHREAD_COND_INITIALIZER
,
133 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
135 static struct consumer_data ustconsumer32_data
= {
136 .type
= LTTNG_CONSUMER32_UST
,
137 .err_unix_sock_path
= DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
138 .cmd_unix_sock_path
= DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
,
141 .channel_monitor_pipe
= -1,
142 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
143 .lock
= PTHREAD_MUTEX_INITIALIZER
,
144 .cond
= PTHREAD_COND_INITIALIZER
,
145 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
148 /* Command line options */
149 static const struct option long_options
[] = {
150 { "client-sock", required_argument
, 0, 'c' },
151 { "apps-sock", required_argument
, 0, 'a' },
152 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
153 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
154 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
155 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
156 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
157 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
158 { "consumerd32-path", required_argument
, 0, '\0' },
159 { "consumerd32-libdir", required_argument
, 0, '\0' },
160 { "consumerd64-path", required_argument
, 0, '\0' },
161 { "consumerd64-libdir", required_argument
, 0, '\0' },
162 { "daemonize", no_argument
, 0, 'd' },
163 { "background", no_argument
, 0, 'b' },
164 { "sig-parent", no_argument
, 0, 'S' },
165 { "help", no_argument
, 0, 'h' },
166 { "group", required_argument
, 0, 'g' },
167 { "version", no_argument
, 0, 'V' },
168 { "quiet", no_argument
, 0, 'q' },
169 { "verbose", no_argument
, 0, 'v' },
170 { "verbose-consumer", no_argument
, 0, '\0' },
171 { "no-kernel", no_argument
, 0, '\0' },
172 { "pidfile", required_argument
, 0, 'p' },
173 { "agent-tcp-port", required_argument
, 0, '\0' },
174 { "config", required_argument
, 0, 'f' },
175 { "load", required_argument
, 0, 'l' },
176 { "kmod-probes", required_argument
, 0, '\0' },
177 { "extra-kmod-probes", required_argument
, 0, '\0' },
181 /* Command line options to ignore from configuration file */
182 static const char *config_ignore_options
[] = { "help", "version", "config" };
184 /* Shared between threads */
185 static int dispatch_thread_exit
;
187 /* Global application Unix socket path */
188 static char apps_unix_sock_path
[PATH_MAX
];
189 /* Global client Unix socket path */
190 static char client_unix_sock_path
[PATH_MAX
];
191 /* global wait shm path for UST */
192 static char wait_shm_path
[PATH_MAX
];
193 /* Global health check unix path */
194 static char health_unix_sock_path
[PATH_MAX
];
196 /* Sockets and FDs */
197 static int client_sock
= -1;
198 static int apps_sock
= -1;
199 int kernel_tracer_fd
= -1;
200 static int kernel_poll_pipe
[2] = { -1, -1 };
203 * Quit pipe for all threads. This permits a single cancellation point
204 * for all threads when receiving an event on the pipe.
206 static int thread_quit_pipe
[2] = { -1, -1 };
209 * This pipe is used to inform the thread managing application communication
210 * that a command is queued and ready to be processed.
212 static int apps_cmd_pipe
[2] = { -1, -1 };
214 int apps_cmd_notify_pipe
[2] = { -1, -1 };
216 /* Pthread, Mutexes and Semaphores */
217 static pthread_t apps_thread
;
218 static pthread_t apps_notify_thread
;
219 static pthread_t reg_apps_thread
;
220 static pthread_t client_thread
;
221 static pthread_t kernel_thread
;
222 static pthread_t dispatch_thread
;
223 static pthread_t health_thread
;
224 static pthread_t ht_cleanup_thread
;
225 static pthread_t agent_reg_thread
;
226 static pthread_t load_session_thread
;
227 static pthread_t notification_thread
;
230 * UST registration command queue. This queue is tied with a futex and uses a N
231 * wakers / 1 waiter implemented and detailed in futex.c/.h
233 * The thread_registration_apps and thread_dispatch_ust_registration uses this
234 * queue along with the wait/wake scheme. The thread_manage_apps receives down
235 * the line new application socket and monitors it for any I/O error or clean
236 * close that triggers an unregistration of the application.
238 static struct ust_cmd_queue ust_cmd_queue
;
241 * Pointer initialized before thread creation.
243 * This points to the tracing session list containing the session count and a
244 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
245 * MUST NOT be taken if you call a public function in session.c.
247 * The lock is nested inside the structure: session_list_ptr->lock. Please use
248 * session_lock_list and session_unlock_list for lock acquisition.
250 static struct ltt_session_list
*session_list_ptr
;
252 int ust_consumerd64_fd
= -1;
253 int ust_consumerd32_fd
= -1;
255 static const char *consumerd32_bin
= CONFIG_CONSUMERD32_BIN
;
256 static const char *consumerd64_bin
= CONFIG_CONSUMERD64_BIN
;
257 static const char *consumerd32_libdir
= CONFIG_CONSUMERD32_LIBDIR
;
258 static const char *consumerd64_libdir
= CONFIG_CONSUMERD64_LIBDIR
;
259 static int consumerd32_bin_override
;
260 static int consumerd64_bin_override
;
261 static int consumerd32_libdir_override
;
262 static int consumerd64_libdir_override
;
264 static const char *module_proc_lttng
= "/proc/lttng";
267 * Consumer daemon state which is changed when spawning it, killing it or in
268 * case of a fatal error.
270 enum consumerd_state
{
271 CONSUMER_STARTED
= 1,
272 CONSUMER_STOPPED
= 2,
277 * This consumer daemon state is used to validate if a client command will be
278 * able to reach the consumer. If not, the client is informed. For instance,
279 * doing a "lttng start" when the consumer state is set to ERROR will return an
280 * error to the client.
282 * The following example shows a possible race condition of this scheme:
284 * consumer thread error happens
286 * client cmd checks state -> still OK
287 * consumer thread exit, sets error
288 * client cmd try to talk to consumer
291 * However, since the consumer is a different daemon, we have no way of making
292 * sure the command will reach it safely even with this state flag. This is why
293 * we consider that up to the state validation during command processing, the
294 * command is safe. After that, we can not guarantee the correctness of the
295 * client request vis-a-vis the consumer.
297 static enum consumerd_state ust_consumerd_state
;
298 static enum consumerd_state kernel_consumerd_state
;
301 * Socket timeout for receiving and sending in seconds.
303 static int app_socket_timeout
;
305 /* Set in main() with the current page size. */
308 /* Application health monitoring */
309 struct health_app
*health_sessiond
;
311 /* Agent TCP port for registration. Used by the agent thread. */
312 unsigned int agent_tcp_port
= DEFAULT_AGENT_TCP_PORT
;
314 /* Am I root or not. */
315 int is_root
; /* Set to 1 if the daemon is running as root */
317 const char * const config_section_name
= "sessiond";
319 /* Load session thread information to operate. */
320 struct load_session_thread_data
*load_info
;
322 /* Notification thread handle. */
323 struct notification_thread_handle
*notification_thread_handle
;
325 /* Global hash tables */
326 struct lttng_ht
*agent_apps_ht_by_sock
= NULL
;
329 * Whether sessiond is ready for commands/notification channel/health check
331 * NR_LTTNG_SESSIOND_READY must match the number of calls to
332 * sessiond_notify_ready().
334 #define NR_LTTNG_SESSIOND_READY 4
335 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
337 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
339 return (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) ? 1 : 0;
342 /* Notify parents that we are ready for cmd and health check */
344 void sessiond_notify_ready(void)
346 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
348 * Notify parent pid that we are ready to accept command
349 * for client side. This ppid is the one from the
350 * external process that spawned us.
352 if (opt_sig_parent
) {
357 * Notify the parent of the fork() process that we are
360 if (opt_daemon
|| opt_background
) {
361 kill(child_ppid
, SIGUSR1
);
367 void setup_consumerd_path(void)
369 const char *bin
, *libdir
;
372 * Allow INSTALL_BIN_PATH to be used as a target path for the
373 * native architecture size consumer if CONFIG_CONSUMER*_PATH
374 * has not been defined.
376 #if (CAA_BITS_PER_LONG == 32)
377 if (!consumerd32_bin
[0]) {
378 consumerd32_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
380 if (!consumerd32_libdir
[0]) {
381 consumerd32_libdir
= INSTALL_LIB_PATH
;
383 #elif (CAA_BITS_PER_LONG == 64)
384 if (!consumerd64_bin
[0]) {
385 consumerd64_bin
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
387 if (!consumerd64_libdir
[0]) {
388 consumerd64_libdir
= INSTALL_LIB_PATH
;
391 #error "Unknown bitness"
395 * runtime env. var. overrides the build default.
397 bin
= lttng_secure_getenv("LTTNG_CONSUMERD32_BIN");
399 consumerd32_bin
= bin
;
401 bin
= lttng_secure_getenv("LTTNG_CONSUMERD64_BIN");
403 consumerd64_bin
= bin
;
405 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD32_LIBDIR");
407 consumerd32_libdir
= libdir
;
409 libdir
= lttng_secure_getenv("LTTNG_CONSUMERD64_LIBDIR");
411 consumerd64_libdir
= libdir
;
416 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
423 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
429 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
441 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
443 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
445 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
449 * Init thread quit pipe.
451 * Return -1 on error or 0 if all pipes are created.
453 static int __init_thread_quit_pipe(int *a_pipe
)
459 PERROR("thread quit pipe");
463 for (i
= 0; i
< 2; i
++) {
464 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
475 static int init_thread_quit_pipe(void)
477 return __init_thread_quit_pipe(thread_quit_pipe
);
481 * Stop all threads by closing the thread quit pipe.
483 static void stop_threads(void)
487 /* Stopping all threads */
488 DBG("Terminating all threads");
489 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
491 ERR("write error on thread quit pipe");
494 /* Dispatch thread */
495 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
496 futex_nto1_wake(&ust_cmd_queue
.futex
);
500 * Close every consumer sockets.
502 static void close_consumer_sockets(void)
506 if (kconsumer_data
.err_sock
>= 0) {
507 ret
= close(kconsumer_data
.err_sock
);
509 PERROR("kernel consumer err_sock close");
512 if (ustconsumer32_data
.err_sock
>= 0) {
513 ret
= close(ustconsumer32_data
.err_sock
);
515 PERROR("UST consumerd32 err_sock close");
518 if (ustconsumer64_data
.err_sock
>= 0) {
519 ret
= close(ustconsumer64_data
.err_sock
);
521 PERROR("UST consumerd64 err_sock close");
524 if (kconsumer_data
.cmd_sock
>= 0) {
525 ret
= close(kconsumer_data
.cmd_sock
);
527 PERROR("kernel consumer cmd_sock close");
530 if (ustconsumer32_data
.cmd_sock
>= 0) {
531 ret
= close(ustconsumer32_data
.cmd_sock
);
533 PERROR("UST consumerd32 cmd_sock close");
536 if (ustconsumer64_data
.cmd_sock
>= 0) {
537 ret
= close(ustconsumer64_data
.cmd_sock
);
539 PERROR("UST consumerd64 cmd_sock close");
542 if (kconsumer_data
.channel_monitor_pipe
>= 0) {
543 ret
= close(kconsumer_data
.channel_monitor_pipe
);
545 PERROR("kernel consumer channel monitor pipe close");
548 if (ustconsumer32_data
.channel_monitor_pipe
>= 0) {
549 ret
= close(ustconsumer32_data
.channel_monitor_pipe
);
551 PERROR("UST consumerd32 channel monitor pipe close");
554 if (ustconsumer64_data
.channel_monitor_pipe
>= 0) {
555 ret
= close(ustconsumer64_data
.channel_monitor_pipe
);
557 PERROR("UST consumerd64 channel monitor pipe close");
563 * Generate the full lock file path using the rundir.
565 * Return the snprintf() return value thus a negative value is an error.
567 static int generate_lock_file_path(char *path
, size_t len
)
574 /* Build lockfile path from rundir. */
575 ret
= snprintf(path
, len
, "%s/" DEFAULT_LTTNG_SESSIOND_LOCKFILE
, rundir
);
577 PERROR("snprintf lockfile path");
584 * Wait on consumer process termination.
586 * Need to be called with the consumer data lock held or from a context
587 * ensuring no concurrent access to data (e.g: cleanup).
589 static void wait_consumer(struct consumer_data
*consumer_data
)
594 if (consumer_data
->pid
<= 0) {
598 DBG("Waiting for complete teardown of consumerd (PID: %d)",
600 ret
= waitpid(consumer_data
->pid
, &status
, 0);
602 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
603 } else if (!WIFEXITED(status
)) {
604 ERR("consumerd termination with error: %d",
607 consumer_data
->pid
= 0;
611 * Cleanup the session daemon's data structures.
613 static void sessiond_cleanup(void)
616 struct ltt_session
*sess
, *stmp
;
619 DBG("Cleanup sessiond");
622 * Close the thread quit pipe. It has already done its job,
623 * since we are now called.
625 utils_close_pipe(thread_quit_pipe
);
628 * If opt_pidfile is undefined, the default file will be wiped when
629 * removing the rundir.
632 ret
= remove(opt_pidfile
);
634 PERROR("remove pidfile %s", opt_pidfile
);
638 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
641 snprintf(path
, PATH_MAX
,
643 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
644 DBG("Removing %s", path
);
647 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
648 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
);
649 DBG("Removing %s", path
);
653 snprintf(path
, PATH_MAX
,
654 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
656 DBG("Removing %s", path
);
659 snprintf(path
, PATH_MAX
,
660 DEFAULT_KCONSUMERD_PATH
,
662 DBG("Removing directory %s", path
);
665 /* ust consumerd 32 */
666 snprintf(path
, PATH_MAX
,
667 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
669 DBG("Removing %s", path
);
672 snprintf(path
, PATH_MAX
,
673 DEFAULT_USTCONSUMERD32_PATH
,
675 DBG("Removing directory %s", path
);
678 /* ust consumerd 64 */
679 snprintf(path
, PATH_MAX
,
680 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
682 DBG("Removing %s", path
);
685 snprintf(path
, PATH_MAX
,
686 DEFAULT_USTCONSUMERD64_PATH
,
688 DBG("Removing directory %s", path
);
691 DBG("Cleaning up all sessions");
693 /* Destroy session list mutex */
694 if (session_list_ptr
!= NULL
) {
695 pthread_mutex_destroy(&session_list_ptr
->lock
);
697 /* Cleanup ALL session */
698 cds_list_for_each_entry_safe(sess
, stmp
,
699 &session_list_ptr
->head
, list
) {
700 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
704 wait_consumer(&kconsumer_data
);
705 wait_consumer(&ustconsumer64_data
);
706 wait_consumer(&ustconsumer32_data
);
708 DBG("Cleaning up all agent apps");
709 agent_app_ht_clean();
711 DBG("Closing all UST sockets");
712 ust_app_clean_list();
713 buffer_reg_destroy_registries();
715 if (is_root
&& !opt_no_kernel
) {
716 DBG2("Closing kernel fd");
717 if (kernel_tracer_fd
>= 0) {
718 ret
= close(kernel_tracer_fd
);
723 DBG("Unloading kernel modules");
724 modprobe_remove_lttng_all();
728 close_consumer_sockets();
731 load_session_destroy_data(load_info
);
736 * Cleanup lock file by deleting it and finaly closing it which will
737 * release the file system lock.
739 if (lockfile_fd
>= 0) {
740 char lockfile_path
[PATH_MAX
];
742 ret
= generate_lock_file_path(lockfile_path
,
743 sizeof(lockfile_path
));
745 ret
= remove(lockfile_path
);
747 PERROR("remove lock file");
749 ret
= close(lockfile_fd
);
751 PERROR("close lock file");
757 * We do NOT rmdir rundir because there are other processes
758 * using it, for instance lttng-relayd, which can start in
759 * parallel with this teardown.
766 * Cleanup the daemon's option data structures.
768 static void sessiond_cleanup_options(void)
770 DBG("Cleaning up options");
773 * If the override option is set, the pointer points to a *non* const
774 * thus freeing it even though the variable type is set to const.
776 if (tracing_group_name_override
) {
777 free((void *) tracing_group_name
);
779 if (consumerd32_bin_override
) {
780 free((void *) consumerd32_bin
);
782 if (consumerd64_bin_override
) {
783 free((void *) consumerd64_bin
);
785 if (consumerd32_libdir_override
) {
786 free((void *) consumerd32_libdir
);
788 if (consumerd64_libdir_override
) {
789 free((void *) consumerd64_libdir
);
793 free(opt_load_session_path
);
794 free(kmod_probes_list
);
795 free(kmod_extra_probes_list
);
797 run_as_destroy_worker();
801 * Send data on a unix socket using the liblttsessiondcomm API.
803 * Return lttcomm error code.
805 static int send_unix_sock(int sock
, void *buf
, size_t len
)
807 /* Check valid length */
812 return lttcomm_send_unix_sock(sock
, buf
, len
);
816 * Free memory of a command context structure.
818 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
820 DBG("Clean command context structure");
822 if ((*cmd_ctx
)->llm
) {
823 free((*cmd_ctx
)->llm
);
825 if ((*cmd_ctx
)->lsm
) {
826 free((*cmd_ctx
)->lsm
);
834 * Notify UST applications using the shm mmap futex.
836 static int notify_ust_apps(int active
)
840 DBG("Notifying applications of session daemon state: %d", active
);
842 /* See shm.c for this call implying mmap, shm and futex calls */
843 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
844 if (wait_shm_mmap
== NULL
) {
848 /* Wake waiting process */
849 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
851 /* Apps notified successfully */
859 * Setup the outgoing data buffer for the response (llm) by allocating the
860 * right amount of memory and copying the original information from the lsm
863 * Return 0 on success, negative value on error.
865 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
,
866 const void *payload_buf
, size_t payload_len
,
867 const void *cmd_header_buf
, size_t cmd_header_len
)
870 const size_t header_len
= sizeof(struct lttcomm_lttng_msg
);
871 const size_t cmd_header_offset
= header_len
;
872 const size_t payload_offset
= cmd_header_offset
+ cmd_header_len
;
873 const size_t total_msg_size
= header_len
+ cmd_header_len
+ payload_len
;
875 cmd_ctx
->llm
= zmalloc(total_msg_size
);
877 if (cmd_ctx
->llm
== NULL
) {
883 /* Copy common data */
884 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
885 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
886 cmd_ctx
->llm
->cmd_header_size
= cmd_header_len
;
887 cmd_ctx
->llm
->data_size
= payload_len
;
888 cmd_ctx
->lttng_msg_size
= total_msg_size
;
890 /* Copy command header */
891 if (cmd_header_len
) {
892 memcpy(((uint8_t *) cmd_ctx
->llm
) + cmd_header_offset
, cmd_header_buf
,
898 memcpy(((uint8_t *) cmd_ctx
->llm
) + payload_offset
, payload_buf
,
907 * Version of setup_lttng_msg() without command header.
909 static int setup_lttng_msg_no_cmd_header(struct command_ctx
*cmd_ctx
,
910 void *payload_buf
, size_t payload_len
)
912 return setup_lttng_msg(cmd_ctx
, payload_buf
, payload_len
, NULL
, 0);
915 * Update the kernel poll set of all channel fd available over all tracing
916 * session. Add the wakeup pipe at the end of the set.
918 static int update_kernel_poll(struct lttng_poll_event
*events
)
921 struct ltt_session
*session
;
922 struct ltt_kernel_channel
*channel
;
924 DBG("Updating kernel poll set");
927 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
928 session_lock(session
);
929 if (session
->kernel_session
== NULL
) {
930 session_unlock(session
);
934 cds_list_for_each_entry(channel
,
935 &session
->kernel_session
->channel_list
.head
, list
) {
936 /* Add channel fd to the kernel poll set */
937 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
939 session_unlock(session
);
942 DBG("Channel fd %d added to kernel set", channel
->fd
);
944 session_unlock(session
);
946 session_unlock_list();
951 session_unlock_list();
956 * Find the channel fd from 'fd' over all tracing session. When found, check
957 * for new channel stream and send those stream fds to the kernel consumer.
959 * Useful for CPU hotplug feature.
961 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
964 struct ltt_session
*session
;
965 struct ltt_kernel_session
*ksess
;
966 struct ltt_kernel_channel
*channel
;
968 DBG("Updating kernel streams for channel fd %d", fd
);
971 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
972 session_lock(session
);
973 if (session
->kernel_session
== NULL
) {
974 session_unlock(session
);
977 ksess
= session
->kernel_session
;
979 cds_list_for_each_entry(channel
,
980 &ksess
->channel_list
.head
, list
) {
981 struct lttng_ht_iter iter
;
982 struct consumer_socket
*socket
;
984 if (channel
->fd
!= fd
) {
987 DBG("Channel found, updating kernel streams");
988 ret
= kernel_open_channel_stream(channel
);
992 /* Update the stream global counter */
993 ksess
->stream_count_global
+= ret
;
996 * Have we already sent fds to the consumer? If yes, it
997 * means that tracing is started so it is safe to send
998 * our updated stream fds.
1000 if (ksess
->consumer_fds_sent
!= 1
1001 || ksess
->consumer
== NULL
) {
1007 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
1008 &iter
.iter
, socket
, node
.node
) {
1009 pthread_mutex_lock(socket
->lock
);
1010 ret
= kernel_consumer_send_channel_stream(socket
,
1012 session
->output_traces
? 1 : 0);
1013 pthread_mutex_unlock(socket
->lock
);
1021 session_unlock(session
);
1023 session_unlock_list();
1027 session_unlock(session
);
1028 session_unlock_list();
1033 * For each tracing session, update newly registered apps. The session list
1034 * lock MUST be acquired before calling this.
1036 static void update_ust_app(int app_sock
)
1038 struct ltt_session
*sess
, *stmp
;
1040 /* Consumer is in an ERROR state. Stop any application update. */
1041 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
1042 /* Stop the update process since the consumer is dead. */
1046 /* For all tracing session(s) */
1047 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
1048 struct ust_app
*app
;
1051 if (!sess
->ust_session
) {
1052 goto unlock_session
;
1056 assert(app_sock
>= 0);
1057 app
= ust_app_find_by_sock(app_sock
);
1060 * Application can be unregistered before so
1061 * this is possible hence simply stopping the
1064 DBG3("UST app update failed to find app sock %d",
1068 ust_app_global_update(sess
->ust_session
, app
);
1072 session_unlock(sess
);
1077 * This thread manage event coming from the kernel.
1079 * Features supported in this thread:
1082 static void *thread_manage_kernel(void *data
)
1084 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1085 uint32_t revents
, nb_fd
;
1087 struct lttng_poll_event events
;
1089 DBG("[thread] Thread manage kernel started");
1091 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1094 * This first step of the while is to clean this structure which could free
1095 * non NULL pointers so initialize it before the loop.
1097 lttng_poll_init(&events
);
1099 if (testpoint(sessiond_thread_manage_kernel
)) {
1100 goto error_testpoint
;
1103 health_code_update();
1105 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1106 goto error_testpoint
;
1110 health_code_update();
1112 if (update_poll_flag
== 1) {
1113 /* Clean events object. We are about to populate it again. */
1114 lttng_poll_clean(&events
);
1116 ret
= sessiond_set_thread_pollset(&events
, 2);
1118 goto error_poll_create
;
1121 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1126 /* This will add the available kernel channel if any. */
1127 ret
= update_kernel_poll(&events
);
1131 update_poll_flag
= 0;
1134 DBG("Thread kernel polling");
1136 /* Poll infinite value of time */
1138 health_poll_entry();
1139 ret
= lttng_poll_wait(&events
, -1);
1140 DBG("Thread kernel return from poll on %d fds",
1141 LTTNG_POLL_GETNB(&events
));
1145 * Restart interrupted system call.
1147 if (errno
== EINTR
) {
1151 } else if (ret
== 0) {
1152 /* Should not happen since timeout is infinite */
1153 ERR("Return value of poll is 0 with an infinite timeout.\n"
1154 "This should not have happened! Continuing...");
1160 for (i
= 0; i
< nb_fd
; i
++) {
1161 /* Fetch once the poll data */
1162 revents
= LTTNG_POLL_GETEV(&events
, i
);
1163 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1165 health_code_update();
1168 /* No activity for this FD (poll implementation). */
1172 /* Thread quit pipe has been closed. Killing thread. */
1173 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1179 /* Check for data on kernel pipe */
1180 if (revents
& LPOLLIN
) {
1181 if (pollfd
== kernel_poll_pipe
[0]) {
1182 (void) lttng_read(kernel_poll_pipe
[0],
1185 * Ret value is useless here, if this pipe gets any actions an
1186 * update is required anyway.
1188 update_poll_flag
= 1;
1192 * New CPU detected by the kernel. Adding kernel stream to
1193 * kernel session and updating the kernel consumer
1195 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1201 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1202 update_poll_flag
= 1;
1205 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1213 lttng_poll_clean(&events
);
1216 utils_close_pipe(kernel_poll_pipe
);
1217 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1220 ERR("Health error occurred in %s", __func__
);
1221 WARN("Kernel thread died unexpectedly. "
1222 "Kernel tracing can continue but CPU hotplug is disabled.");
1224 health_unregister(health_sessiond
);
1225 DBG("Kernel thread dying");
1230 * Signal pthread condition of the consumer data that the thread.
1232 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1234 pthread_mutex_lock(&data
->cond_mutex
);
1237 * The state is set before signaling. It can be any value, it's the waiter
1238 * job to correctly interpret this condition variable associated to the
1239 * consumer pthread_cond.
1241 * A value of 0 means that the corresponding thread of the consumer data
1242 * was not started. 1 indicates that the thread has started and is ready
1243 * for action. A negative value means that there was an error during the
1246 data
->consumer_thread_is_ready
= state
;
1247 (void) pthread_cond_signal(&data
->cond
);
1249 pthread_mutex_unlock(&data
->cond_mutex
);
1253 * This thread manage the consumer error sent back to the session daemon.
1255 static void *thread_manage_consumer(void *data
)
1257 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1258 uint32_t revents
, nb_fd
;
1259 enum lttcomm_return_code code
;
1260 struct lttng_poll_event events
;
1261 struct consumer_data
*consumer_data
= data
;
1262 struct consumer_socket
*cmd_socket_wrapper
= NULL
;
1264 DBG("[thread] Manage consumer started");
1266 rcu_register_thread();
1267 rcu_thread_online();
1269 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1271 health_code_update();
1274 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1275 * metadata_sock. Nothing more will be added to this poll set.
1277 ret
= sessiond_set_thread_pollset(&events
, 3);
1283 * The error socket here is already in a listening state which was done
1284 * just before spawning this thread to avoid a race between the consumer
1285 * daemon exec trying to connect and the listen() call.
1287 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1292 health_code_update();
1294 /* Infinite blocking call, waiting for transmission */
1296 health_poll_entry();
1298 if (testpoint(sessiond_thread_manage_consumer
)) {
1302 ret
= lttng_poll_wait(&events
, -1);
1306 * Restart interrupted system call.
1308 if (errno
== EINTR
) {
1316 for (i
= 0; i
< nb_fd
; i
++) {
1317 /* Fetch once the poll data */
1318 revents
= LTTNG_POLL_GETEV(&events
, i
);
1319 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1321 health_code_update();
1324 /* No activity for this FD (poll implementation). */
1328 /* Thread quit pipe has been closed. Killing thread. */
1329 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1335 /* Event on the registration socket */
1336 if (pollfd
== consumer_data
->err_sock
) {
1337 if (revents
& LPOLLIN
) {
1339 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1340 ERR("consumer err socket poll error");
1343 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1349 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1355 * Set the CLOEXEC flag. Return code is useless because either way, the
1358 (void) utils_set_fd_cloexec(sock
);
1360 health_code_update();
1362 DBG2("Receiving code from consumer err_sock");
1364 /* Getting status code from kconsumerd */
1365 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1366 sizeof(enum lttcomm_return_code
));
1371 health_code_update();
1372 if (code
!= LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1373 ERR("consumer error when waiting for SOCK_READY : %s",
1374 lttcomm_get_readable_code(-code
));
1378 /* Connect both command and metadata sockets. */
1379 consumer_data
->cmd_sock
=
1380 lttcomm_connect_unix_sock(
1381 consumer_data
->cmd_unix_sock_path
);
1382 consumer_data
->metadata_fd
=
1383 lttcomm_connect_unix_sock(
1384 consumer_data
->cmd_unix_sock_path
);
1385 if (consumer_data
->cmd_sock
< 0 || consumer_data
->metadata_fd
< 0) {
1386 PERROR("consumer connect cmd socket");
1387 /* On error, signal condition and quit. */
1388 signal_consumer_condition(consumer_data
, -1);
1392 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1394 /* Create metadata socket lock. */
1395 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1396 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1397 PERROR("zmalloc pthread mutex");
1400 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1402 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1403 DBG("Consumer metadata socket ready (fd: %d)",
1404 consumer_data
->metadata_fd
);
1407 * Remove the consumerd error sock since we've established a connection.
1409 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1414 /* Add new accepted error socket. */
1415 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1420 /* Add metadata socket that is successfully connected. */
1421 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1422 LPOLLIN
| LPOLLRDHUP
);
1427 health_code_update();
1430 * Transfer the write-end of the channel monitoring pipe to the
1431 * by issuing a SET_CHANNEL_MONITOR_PIPE command.
1433 cmd_socket_wrapper
= consumer_allocate_socket(&consumer_data
->cmd_sock
);
1434 if (!cmd_socket_wrapper
) {
1438 ret
= consumer_send_channel_monitor_pipe(cmd_socket_wrapper
,
1439 consumer_data
->channel_monitor_pipe
);
1443 /* Discard the socket wrapper as it is no longer needed. */
1444 consumer_destroy_socket(cmd_socket_wrapper
);
1445 cmd_socket_wrapper
= NULL
;
1447 /* The thread is completely initialized, signal that it is ready. */
1448 signal_consumer_condition(consumer_data
, 1);
1450 /* Infinite blocking call, waiting for transmission */
1453 health_code_update();
1455 /* Exit the thread because the thread quit pipe has been triggered. */
1457 /* Not a health error. */
1462 health_poll_entry();
1463 ret
= lttng_poll_wait(&events
, -1);
1467 * Restart interrupted system call.
1469 if (errno
== EINTR
) {
1477 for (i
= 0; i
< nb_fd
; i
++) {
1478 /* Fetch once the poll data */
1479 revents
= LTTNG_POLL_GETEV(&events
, i
);
1480 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1482 health_code_update();
1485 /* No activity for this FD (poll implementation). */
1490 * Thread quit pipe has been triggered, flag that we should stop
1491 * but continue the current loop to handle potential data from
1494 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1496 if (pollfd
== sock
) {
1497 /* Event on the consumerd socket */
1498 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1499 && !(revents
& LPOLLIN
)) {
1500 ERR("consumer err socket second poll error");
1503 health_code_update();
1504 /* Wait for any kconsumerd error */
1505 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1506 sizeof(enum lttcomm_return_code
));
1508 ERR("consumer closed the command socket");
1512 ERR("consumer return code : %s",
1513 lttcomm_get_readable_code(-code
));
1516 } else if (pollfd
== consumer_data
->metadata_fd
) {
1517 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1518 && !(revents
& LPOLLIN
)) {
1519 ERR("consumer err metadata socket second poll error");
1522 /* UST metadata requests */
1523 ret
= ust_consumer_metadata_request(
1524 &consumer_data
->metadata_sock
);
1526 ERR("Handling metadata request");
1530 /* No need for an else branch all FDs are tested prior. */
1532 health_code_update();
1538 * We lock here because we are about to close the sockets and some other
1539 * thread might be using them so get exclusive access which will abort all
1540 * other consumer command by other threads.
1542 pthread_mutex_lock(&consumer_data
->lock
);
1544 /* Immediately set the consumerd state to stopped */
1545 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1546 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1547 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1548 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1549 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1551 /* Code flow error... */
1555 if (consumer_data
->err_sock
>= 0) {
1556 ret
= close(consumer_data
->err_sock
);
1560 consumer_data
->err_sock
= -1;
1562 if (consumer_data
->cmd_sock
>= 0) {
1563 ret
= close(consumer_data
->cmd_sock
);
1567 consumer_data
->cmd_sock
= -1;
1569 if (consumer_data
->metadata_sock
.fd_ptr
&&
1570 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1571 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1583 unlink(consumer_data
->err_unix_sock_path
);
1584 unlink(consumer_data
->cmd_unix_sock_path
);
1585 pthread_mutex_unlock(&consumer_data
->lock
);
1587 /* Cleanup metadata socket mutex. */
1588 if (consumer_data
->metadata_sock
.lock
) {
1589 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1590 free(consumer_data
->metadata_sock
.lock
);
1592 lttng_poll_clean(&events
);
1594 if (cmd_socket_wrapper
) {
1595 consumer_destroy_socket(cmd_socket_wrapper
);
1600 ERR("Health error occurred in %s", __func__
);
1602 health_unregister(health_sessiond
);
1603 DBG("consumer thread cleanup completed");
1605 rcu_thread_offline();
1606 rcu_unregister_thread();
1612 * This thread manage application communication.
1614 static void *thread_manage_apps(void *data
)
1616 int i
, ret
, pollfd
, err
= -1;
1618 uint32_t revents
, nb_fd
;
1619 struct lttng_poll_event events
;
1621 DBG("[thread] Manage application started");
1623 rcu_register_thread();
1624 rcu_thread_online();
1626 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1628 if (testpoint(sessiond_thread_manage_apps
)) {
1629 goto error_testpoint
;
1632 health_code_update();
1634 ret
= sessiond_set_thread_pollset(&events
, 2);
1636 goto error_poll_create
;
1639 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1644 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1648 health_code_update();
1651 DBG("Apps thread polling");
1653 /* Inifinite blocking call, waiting for transmission */
1655 health_poll_entry();
1656 ret
= lttng_poll_wait(&events
, -1);
1657 DBG("Apps thread return from poll on %d fds",
1658 LTTNG_POLL_GETNB(&events
));
1662 * Restart interrupted system call.
1664 if (errno
== EINTR
) {
1672 for (i
= 0; i
< nb_fd
; i
++) {
1673 /* Fetch once the poll data */
1674 revents
= LTTNG_POLL_GETEV(&events
, i
);
1675 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1677 health_code_update();
1680 /* No activity for this FD (poll implementation). */
1684 /* Thread quit pipe has been closed. Killing thread. */
1685 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1691 /* Inspect the apps cmd pipe */
1692 if (pollfd
== apps_cmd_pipe
[0]) {
1693 if (revents
& LPOLLIN
) {
1697 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1698 if (size_ret
< sizeof(sock
)) {
1699 PERROR("read apps cmd pipe");
1703 health_code_update();
1706 * Since this is a command socket (write then read),
1707 * we only monitor the error events of the socket.
1709 ret
= lttng_poll_add(&events
, sock
,
1710 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1715 DBG("Apps with sock %d added to poll set", sock
);
1716 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1717 ERR("Apps command pipe error");
1720 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1725 * At this point, we know that a registered application made
1726 * the event at poll_wait.
1728 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1729 /* Removing from the poll set */
1730 ret
= lttng_poll_del(&events
, pollfd
);
1735 /* Socket closed on remote end. */
1736 ust_app_unregister(pollfd
);
1738 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1743 health_code_update();
1749 lttng_poll_clean(&events
);
1752 utils_close_pipe(apps_cmd_pipe
);
1753 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1756 * We don't clean the UST app hash table here since already registered
1757 * applications can still be controlled so let them be until the session
1758 * daemon dies or the applications stop.
1763 ERR("Health error occurred in %s", __func__
);
1765 health_unregister(health_sessiond
);
1766 DBG("Application communication apps thread cleanup complete");
1767 rcu_thread_offline();
1768 rcu_unregister_thread();
1773 * Send a socket to a thread This is called from the dispatch UST registration
1774 * thread once all sockets are set for the application.
1776 * The sock value can be invalid, we don't really care, the thread will handle
1777 * it and make the necessary cleanup if so.
1779 * On success, return 0 else a negative value being the errno message of the
1782 static int send_socket_to_thread(int fd
, int sock
)
1787 * It's possible that the FD is set as invalid with -1 concurrently just
1788 * before calling this function being a shutdown state of the thread.
1795 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1796 if (ret
< sizeof(sock
)) {
1797 PERROR("write apps pipe %d", fd
);
1804 /* All good. Don't send back the write positive ret value. */
1811 * Sanitize the wait queue of the dispatch registration thread meaning removing
1812 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1813 * notify socket is never received.
1815 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1817 int ret
, nb_fd
= 0, i
;
1818 unsigned int fd_added
= 0;
1819 struct lttng_poll_event events
;
1820 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1824 lttng_poll_init(&events
);
1826 /* Just skip everything for an empty queue. */
1827 if (!wait_queue
->count
) {
1831 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1836 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1837 &wait_queue
->head
, head
) {
1838 assert(wait_node
->app
);
1839 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1840 LPOLLHUP
| LPOLLERR
);
1853 * Poll but don't block so we can quickly identify the faulty events and
1854 * clean them afterwards from the wait queue.
1856 ret
= lttng_poll_wait(&events
, 0);
1862 for (i
= 0; i
< nb_fd
; i
++) {
1863 /* Get faulty FD. */
1864 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1865 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1868 /* No activity for this FD (poll implementation). */
1872 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1873 &wait_queue
->head
, head
) {
1874 if (pollfd
== wait_node
->app
->sock
&&
1875 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1876 cds_list_del(&wait_node
->head
);
1877 wait_queue
->count
--;
1878 ust_app_destroy(wait_node
->app
);
1881 * Silence warning of use-after-free in
1882 * cds_list_for_each_entry_safe which uses
1883 * __typeof__(*wait_node).
1888 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1895 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1899 lttng_poll_clean(&events
);
1903 lttng_poll_clean(&events
);
1905 ERR("Unable to sanitize wait queue");
1910 * Dispatch request from the registration threads to the application
1911 * communication thread.
1913 static void *thread_dispatch_ust_registration(void *data
)
1916 struct cds_wfcq_node
*node
;
1917 struct ust_command
*ust_cmd
= NULL
;
1918 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1919 struct ust_reg_wait_queue wait_queue
= {
1923 rcu_register_thread();
1925 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1927 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1928 goto error_testpoint
;
1931 health_code_update();
1933 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1935 DBG("[thread] Dispatch UST command started");
1938 health_code_update();
1940 /* Atomically prepare the queue futex */
1941 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1943 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1948 struct ust_app
*app
= NULL
;
1952 * Make sure we don't have node(s) that have hung up before receiving
1953 * the notify socket. This is to clean the list in order to avoid
1954 * memory leaks from notify socket that are never seen.
1956 sanitize_wait_queue(&wait_queue
);
1958 health_code_update();
1959 /* Dequeue command for registration */
1960 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1962 DBG("Woken up but nothing in the UST command queue");
1963 /* Continue thread execution */
1967 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1969 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1970 " gid:%d sock:%d name:%s (version %d.%d)",
1971 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1972 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1973 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1974 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1976 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1977 wait_node
= zmalloc(sizeof(*wait_node
));
1979 PERROR("zmalloc wait_node dispatch");
1980 ret
= close(ust_cmd
->sock
);
1982 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1984 lttng_fd_put(LTTNG_FD_APPS
, 1);
1988 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1990 /* Create application object if socket is CMD. */
1991 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1993 if (!wait_node
->app
) {
1994 ret
= close(ust_cmd
->sock
);
1996 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1998 lttng_fd_put(LTTNG_FD_APPS
, 1);
2004 * Add application to the wait queue so we can set the notify
2005 * socket before putting this object in the global ht.
2007 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
2012 * We have to continue here since we don't have the notify
2013 * socket and the application MUST be added to the hash table
2014 * only at that moment.
2019 * Look for the application in the local wait queue and set the
2020 * notify socket if found.
2022 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2023 &wait_queue
.head
, head
) {
2024 health_code_update();
2025 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
2026 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
2027 cds_list_del(&wait_node
->head
);
2029 app
= wait_node
->app
;
2031 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
2037 * With no application at this stage the received socket is
2038 * basically useless so close it before we free the cmd data
2039 * structure for good.
2042 ret
= close(ust_cmd
->sock
);
2044 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
2046 lttng_fd_put(LTTNG_FD_APPS
, 1);
2053 * @session_lock_list
2055 * Lock the global session list so from the register up to the
2056 * registration done message, no thread can see the application
2057 * and change its state.
2059 session_lock_list();
2063 * Add application to the global hash table. This needs to be
2064 * done before the update to the UST registry can locate the
2069 /* Set app version. This call will print an error if needed. */
2070 (void) ust_app_version(app
);
2072 /* Send notify socket through the notify pipe. */
2073 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
2077 session_unlock_list();
2079 * No notify thread, stop the UST tracing. However, this is
2080 * not an internal error of the this thread thus setting
2081 * the health error code to a normal exit.
2088 * Update newly registered application with the tracing
2089 * registry info already enabled information.
2091 update_ust_app(app
->sock
);
2094 * Don't care about return value. Let the manage apps threads
2095 * handle app unregistration upon socket close.
2097 (void) ust_app_register_done(app
);
2100 * Even if the application socket has been closed, send the app
2101 * to the thread and unregistration will take place at that
2104 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2107 session_unlock_list();
2109 * No apps. thread, stop the UST tracing. However, this is
2110 * not an internal error of the this thread thus setting
2111 * the health error code to a normal exit.
2118 session_unlock_list();
2120 } while (node
!= NULL
);
2122 health_poll_entry();
2123 /* Futex wait on queue. Blocking call on futex() */
2124 futex_nto1_wait(&ust_cmd_queue
.futex
);
2127 /* Normal exit, no error */
2131 /* Clean up wait queue. */
2132 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2133 &wait_queue
.head
, head
) {
2134 cds_list_del(&wait_node
->head
);
2139 /* Empty command queue. */
2141 /* Dequeue command for registration */
2142 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2146 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2147 ret
= close(ust_cmd
->sock
);
2149 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2151 lttng_fd_put(LTTNG_FD_APPS
, 1);
2156 DBG("Dispatch thread dying");
2159 ERR("Health error occurred in %s", __func__
);
2161 health_unregister(health_sessiond
);
2162 rcu_unregister_thread();
2167 * This thread manage application registration.
2169 static void *thread_registration_apps(void *data
)
2171 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2172 uint32_t revents
, nb_fd
;
2173 struct lttng_poll_event events
;
2175 * Get allocated in this thread, enqueued to a global queue, dequeued and
2176 * freed in the manage apps thread.
2178 struct ust_command
*ust_cmd
= NULL
;
2180 DBG("[thread] Manage application registration started");
2182 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2184 if (testpoint(sessiond_thread_registration_apps
)) {
2185 goto error_testpoint
;
2188 ret
= lttcomm_listen_unix_sock(apps_sock
);
2194 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2195 * more will be added to this poll set.
2197 ret
= sessiond_set_thread_pollset(&events
, 2);
2199 goto error_create_poll
;
2202 /* Add the application registration socket */
2203 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2205 goto error_poll_add
;
2208 /* Notify all applications to register */
2209 ret
= notify_ust_apps(1);
2211 ERR("Failed to notify applications or create the wait shared memory.\n"
2212 "Execution continues but there might be problem for already\n"
2213 "running applications that wishes to register.");
2217 DBG("Accepting application registration");
2219 /* Inifinite blocking call, waiting for transmission */
2221 health_poll_entry();
2222 ret
= lttng_poll_wait(&events
, -1);
2226 * Restart interrupted system call.
2228 if (errno
== EINTR
) {
2236 for (i
= 0; i
< nb_fd
; i
++) {
2237 health_code_update();
2239 /* Fetch once the poll data */
2240 revents
= LTTNG_POLL_GETEV(&events
, i
);
2241 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2244 /* No activity for this FD (poll implementation). */
2248 /* Thread quit pipe has been closed. Killing thread. */
2249 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2255 /* Event on the registration socket */
2256 if (pollfd
== apps_sock
) {
2257 if (revents
& LPOLLIN
) {
2258 sock
= lttcomm_accept_unix_sock(apps_sock
);
2264 * Set socket timeout for both receiving and ending.
2265 * app_socket_timeout is in seconds, whereas
2266 * lttcomm_setsockopt_rcv_timeout and
2267 * lttcomm_setsockopt_snd_timeout expect msec as
2270 if (app_socket_timeout
>= 0) {
2271 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2272 app_socket_timeout
* 1000);
2273 (void) lttcomm_setsockopt_snd_timeout(sock
,
2274 app_socket_timeout
* 1000);
2278 * Set the CLOEXEC flag. Return code is useless because
2279 * either way, the show must go on.
2281 (void) utils_set_fd_cloexec(sock
);
2283 /* Create UST registration command for enqueuing */
2284 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2285 if (ust_cmd
== NULL
) {
2286 PERROR("ust command zmalloc");
2295 * Using message-based transmissions to ensure we don't
2296 * have to deal with partially received messages.
2298 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2300 ERR("Exhausted file descriptors allowed for applications.");
2310 health_code_update();
2311 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2314 /* Close socket of the application. */
2319 lttng_fd_put(LTTNG_FD_APPS
, 1);
2323 health_code_update();
2325 ust_cmd
->sock
= sock
;
2328 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2329 " gid:%d sock:%d name:%s (version %d.%d)",
2330 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2331 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2332 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2333 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2336 * Lock free enqueue the registration request. The red pill
2337 * has been taken! This apps will be part of the *system*.
2339 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2342 * Wake the registration queue futex. Implicit memory
2343 * barrier with the exchange in cds_wfcq_enqueue.
2345 futex_nto1_wake(&ust_cmd_queue
.futex
);
2346 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2347 ERR("Register apps socket poll error");
2350 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2359 /* Notify that the registration thread is gone */
2362 if (apps_sock
>= 0) {
2363 ret
= close(apps_sock
);
2373 lttng_fd_put(LTTNG_FD_APPS
, 1);
2375 unlink(apps_unix_sock_path
);
2378 lttng_poll_clean(&events
);
2382 DBG("UST Registration thread cleanup complete");
2385 ERR("Health error occurred in %s", __func__
);
2387 health_unregister(health_sessiond
);
2393 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2394 * exec or it will fails.
2396 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2399 struct timespec timeout
;
2402 * Make sure we set the readiness flag to 0 because we are NOT ready.
2403 * This access to consumer_thread_is_ready does not need to be
2404 * protected by consumer_data.cond_mutex (yet) since the consumer
2405 * management thread has not been started at this point.
2407 consumer_data
->consumer_thread_is_ready
= 0;
2409 /* Setup pthread condition */
2410 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2413 PERROR("pthread_condattr_init consumer data");
2418 * Set the monotonic clock in order to make sure we DO NOT jump in time
2419 * between the clock_gettime() call and the timedwait call. See bug #324
2420 * for a more details and how we noticed it.
2422 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2425 PERROR("pthread_condattr_setclock consumer data");
2429 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2432 PERROR("pthread_cond_init consumer data");
2436 ret
= pthread_create(&consumer_data
->thread
, default_pthread_attr(),
2437 thread_manage_consumer
, consumer_data
);
2440 PERROR("pthread_create consumer");
2445 /* We are about to wait on a pthread condition */
2446 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2448 /* Get time for sem_timedwait absolute timeout */
2449 clock_ret
= lttng_clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2451 * Set the timeout for the condition timed wait even if the clock gettime
2452 * call fails since we might loop on that call and we want to avoid to
2453 * increment the timeout too many times.
2455 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2458 * The following loop COULD be skipped in some conditions so this is why we
2459 * set ret to 0 in order to make sure at least one round of the loop is
2465 * Loop until the condition is reached or when a timeout is reached. Note
2466 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2467 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2468 * possible. This loop does not take any chances and works with both of
2471 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2472 if (clock_ret
< 0) {
2473 PERROR("clock_gettime spawn consumer");
2474 /* Infinite wait for the consumerd thread to be ready */
2475 ret
= pthread_cond_wait(&consumer_data
->cond
,
2476 &consumer_data
->cond_mutex
);
2478 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2479 &consumer_data
->cond_mutex
, &timeout
);
2483 /* Release the pthread condition */
2484 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2488 if (ret
== ETIMEDOUT
) {
2492 * Call has timed out so we kill the kconsumerd_thread and return
2495 ERR("Condition timed out. The consumer thread was never ready."
2497 pth_ret
= pthread_cancel(consumer_data
->thread
);
2499 PERROR("pthread_cancel consumer thread");
2502 PERROR("pthread_cond_wait failed consumer thread");
2504 /* Caller is expecting a negative value on failure. */
2509 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2510 if (consumer_data
->pid
== 0) {
2511 ERR("Consumerd did not start");
2512 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2515 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2524 * Join consumer thread
2526 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2530 /* Consumer pid must be a real one. */
2531 if (consumer_data
->pid
> 0) {
2533 ret
= kill(consumer_data
->pid
, SIGTERM
);
2535 PERROR("Error killing consumer daemon");
2538 return pthread_join(consumer_data
->thread
, &status
);
2545 * Fork and exec a consumer daemon (consumerd).
2547 * Return pid if successful else -1.
2549 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2553 const char *consumer_to_use
;
2554 const char *verbosity
;
2557 DBG("Spawning consumerd");
2564 if (opt_verbose_consumer
) {
2565 verbosity
= "--verbose";
2566 } else if (lttng_opt_quiet
) {
2567 verbosity
= "--quiet";
2572 switch (consumer_data
->type
) {
2573 case LTTNG_CONSUMER_KERNEL
:
2575 * Find out which consumerd to execute. We will first try the
2576 * 64-bit path, then the sessiond's installation directory, and
2577 * fallback on the 32-bit one,
2579 DBG3("Looking for a kernel consumer at these locations:");
2580 DBG3(" 1) %s", consumerd64_bin
);
2581 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2582 DBG3(" 3) %s", consumerd32_bin
);
2583 if (stat(consumerd64_bin
, &st
) == 0) {
2584 DBG3("Found location #1");
2585 consumer_to_use
= consumerd64_bin
;
2586 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2587 DBG3("Found location #2");
2588 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2589 } else if (stat(consumerd32_bin
, &st
) == 0) {
2590 DBG3("Found location #3");
2591 consumer_to_use
= consumerd32_bin
;
2593 DBG("Could not find any valid consumerd executable");
2597 DBG("Using kernel consumer at: %s", consumer_to_use
);
2598 ret
= execl(consumer_to_use
,
2599 "lttng-consumerd", verbosity
, "-k",
2600 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2601 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2602 "--group", tracing_group_name
,
2605 case LTTNG_CONSUMER64_UST
:
2607 char *tmpnew
= NULL
;
2609 if (consumerd64_libdir
[0] != '\0') {
2613 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2617 tmplen
= strlen("LD_LIBRARY_PATH=")
2618 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2619 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2624 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2625 strcat(tmpnew
, consumerd64_libdir
);
2626 if (tmp
[0] != '\0') {
2627 strcat(tmpnew
, ":");
2628 strcat(tmpnew
, tmp
);
2630 ret
= putenv(tmpnew
);
2637 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2638 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2639 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2640 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2641 "--group", tracing_group_name
,
2643 if (consumerd64_libdir
[0] != '\0') {
2648 case LTTNG_CONSUMER32_UST
:
2650 char *tmpnew
= NULL
;
2652 if (consumerd32_libdir
[0] != '\0') {
2656 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2660 tmplen
= strlen("LD_LIBRARY_PATH=")
2661 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2662 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2667 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2668 strcat(tmpnew
, consumerd32_libdir
);
2669 if (tmp
[0] != '\0') {
2670 strcat(tmpnew
, ":");
2671 strcat(tmpnew
, tmp
);
2673 ret
= putenv(tmpnew
);
2680 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2681 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2682 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2683 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2684 "--group", tracing_group_name
,
2686 if (consumerd32_libdir
[0] != '\0') {
2692 PERROR("unknown consumer type");
2696 PERROR("Consumer execl()");
2698 /* Reaching this point, we got a failure on our execl(). */
2700 } else if (pid
> 0) {
2703 PERROR("start consumer fork");
2711 * Spawn the consumerd daemon and session daemon thread.
2713 static int start_consumerd(struct consumer_data
*consumer_data
)
2718 * Set the listen() state on the socket since there is a possible race
2719 * between the exec() of the consumer daemon and this call if place in the
2720 * consumer thread. See bug #366 for more details.
2722 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2727 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2728 if (consumer_data
->pid
!= 0) {
2729 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2733 ret
= spawn_consumerd(consumer_data
);
2735 ERR("Spawning consumerd failed");
2736 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2740 /* Setting up the consumer_data pid */
2741 consumer_data
->pid
= ret
;
2742 DBG2("Consumer pid %d", consumer_data
->pid
);
2743 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2745 DBG2("Spawning consumer control thread");
2746 ret
= spawn_consumer_thread(consumer_data
);
2748 ERR("Fatal error spawning consumer control thread");
2756 /* Cleanup already created sockets on error. */
2757 if (consumer_data
->err_sock
>= 0) {
2760 err
= close(consumer_data
->err_sock
);
2762 PERROR("close consumer data error socket");
2769 * Setup necessary data for kernel tracer action.
2771 static int init_kernel_tracer(void)
2775 /* Modprobe lttng kernel modules */
2776 ret
= modprobe_lttng_control();
2781 /* Open debugfs lttng */
2782 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2783 if (kernel_tracer_fd
< 0) {
2784 DBG("Failed to open %s", module_proc_lttng
);
2789 /* Validate kernel version */
2790 ret
= kernel_validate_version(kernel_tracer_fd
);
2795 ret
= modprobe_lttng_data();
2800 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2804 modprobe_remove_lttng_control();
2805 ret
= close(kernel_tracer_fd
);
2809 kernel_tracer_fd
= -1;
2810 return LTTNG_ERR_KERN_VERSION
;
2813 ret
= close(kernel_tracer_fd
);
2819 modprobe_remove_lttng_control();
2822 WARN("No kernel tracer available");
2823 kernel_tracer_fd
= -1;
2825 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2827 return LTTNG_ERR_KERN_NA
;
2833 * Copy consumer output from the tracing session to the domain session. The
2834 * function also applies the right modification on a per domain basis for the
2835 * trace files destination directory.
2837 * Should *NOT* be called with RCU read-side lock held.
2839 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2842 const char *dir_name
;
2843 struct consumer_output
*consumer
;
2846 assert(session
->consumer
);
2849 case LTTNG_DOMAIN_KERNEL
:
2850 DBG3("Copying tracing session consumer output in kernel session");
2852 * XXX: We should audit the session creation and what this function
2853 * does "extra" in order to avoid a destroy since this function is used
2854 * in the domain session creation (kernel and ust) only. Same for UST
2857 if (session
->kernel_session
->consumer
) {
2858 consumer_output_put(session
->kernel_session
->consumer
);
2860 session
->kernel_session
->consumer
=
2861 consumer_copy_output(session
->consumer
);
2862 /* Ease our life a bit for the next part */
2863 consumer
= session
->kernel_session
->consumer
;
2864 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2866 case LTTNG_DOMAIN_JUL
:
2867 case LTTNG_DOMAIN_LOG4J
:
2868 case LTTNG_DOMAIN_PYTHON
:
2869 case LTTNG_DOMAIN_UST
:
2870 DBG3("Copying tracing session consumer output in UST session");
2871 if (session
->ust_session
->consumer
) {
2872 consumer_output_put(session
->ust_session
->consumer
);
2874 session
->ust_session
->consumer
=
2875 consumer_copy_output(session
->consumer
);
2876 /* Ease our life a bit for the next part */
2877 consumer
= session
->ust_session
->consumer
;
2878 dir_name
= DEFAULT_UST_TRACE_DIR
;
2881 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2885 /* Append correct directory to subdir */
2886 strncat(consumer
->subdir
, dir_name
,
2887 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2888 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2897 * Create an UST session and add it to the session ust list.
2899 * Should *NOT* be called with RCU read-side lock held.
2901 static int create_ust_session(struct ltt_session
*session
,
2902 struct lttng_domain
*domain
)
2905 struct ltt_ust_session
*lus
= NULL
;
2909 assert(session
->consumer
);
2911 switch (domain
->type
) {
2912 case LTTNG_DOMAIN_JUL
:
2913 case LTTNG_DOMAIN_LOG4J
:
2914 case LTTNG_DOMAIN_PYTHON
:
2915 case LTTNG_DOMAIN_UST
:
2918 ERR("Unknown UST domain on create session %d", domain
->type
);
2919 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2923 DBG("Creating UST session");
2925 lus
= trace_ust_create_session(session
->id
);
2927 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2931 lus
->uid
= session
->uid
;
2932 lus
->gid
= session
->gid
;
2933 lus
->output_traces
= session
->output_traces
;
2934 lus
->snapshot_mode
= session
->snapshot_mode
;
2935 lus
->live_timer_interval
= session
->live_timer
;
2936 session
->ust_session
= lus
;
2937 if (session
->shm_path
[0]) {
2938 strncpy(lus
->root_shm_path
, session
->shm_path
,
2939 sizeof(lus
->root_shm_path
));
2940 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2941 strncpy(lus
->shm_path
, session
->shm_path
,
2942 sizeof(lus
->shm_path
));
2943 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2944 strncat(lus
->shm_path
, "/ust",
2945 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2947 /* Copy session output to the newly created UST session */
2948 ret
= copy_session_consumer(domain
->type
, session
);
2949 if (ret
!= LTTNG_OK
) {
2957 session
->ust_session
= NULL
;
2962 * Create a kernel tracer session then create the default channel.
2964 static int create_kernel_session(struct ltt_session
*session
)
2968 DBG("Creating kernel session");
2970 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2972 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2976 /* Code flow safety */
2977 assert(session
->kernel_session
);
2979 /* Copy session output to the newly created Kernel session */
2980 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2981 if (ret
!= LTTNG_OK
) {
2985 /* Create directory(ies) on local filesystem. */
2986 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2987 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2988 ret
= run_as_mkdir_recursive(
2989 session
->kernel_session
->consumer
->dst
.trace_path
,
2990 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2992 if (errno
!= EEXIST
) {
2993 ERR("Trace directory creation error");
2999 session
->kernel_session
->uid
= session
->uid
;
3000 session
->kernel_session
->gid
= session
->gid
;
3001 session
->kernel_session
->output_traces
= session
->output_traces
;
3002 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
3007 trace_kernel_destroy_session(session
->kernel_session
);
3008 session
->kernel_session
= NULL
;
3013 * Count number of session permitted by uid/gid.
3015 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
3018 struct ltt_session
*session
;
3020 DBG("Counting number of available session for UID %d GID %d",
3022 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
3024 * Only list the sessions the user can control.
3026 if (!session_access_ok(session
, uid
, gid
)) {
3035 * Process the command requested by the lttng client within the command
3036 * context structure. This function make sure that the return structure (llm)
3037 * is set and ready for transmission before returning.
3039 * Return any error encountered or 0 for success.
3041 * "sock" is only used for special-case var. len data.
3043 * Should *NOT* be called with RCU read-side lock held.
3045 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
3049 int need_tracing_session
= 1;
3052 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
3054 assert(!rcu_read_ongoing());
3058 switch (cmd_ctx
->lsm
->cmd_type
) {
3059 case LTTNG_CREATE_SESSION
:
3060 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3061 case LTTNG_CREATE_SESSION_LIVE
:
3062 case LTTNG_DESTROY_SESSION
:
3063 case LTTNG_LIST_SESSIONS
:
3064 case LTTNG_LIST_DOMAINS
:
3065 case LTTNG_START_TRACE
:
3066 case LTTNG_STOP_TRACE
:
3067 case LTTNG_DATA_PENDING
:
3068 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3069 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3070 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3071 case LTTNG_SNAPSHOT_RECORD
:
3072 case LTTNG_SAVE_SESSION
:
3073 case LTTNG_SET_SESSION_SHM_PATH
:
3074 case LTTNG_REGENERATE_METADATA
:
3075 case LTTNG_REGENERATE_STATEDUMP
:
3076 case LTTNG_REGISTER_TRIGGER
:
3077 case LTTNG_UNREGISTER_TRIGGER
:
3084 if (opt_no_kernel
&& need_domain
3085 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3087 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3089 ret
= LTTNG_ERR_KERN_NA
;
3094 /* Deny register consumer if we already have a spawned consumer. */
3095 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3096 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3097 if (kconsumer_data
.pid
> 0) {
3098 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3099 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3102 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3106 * Check for command that don't needs to allocate a returned payload. We do
3107 * this here so we don't have to make the call for no payload at each
3110 switch(cmd_ctx
->lsm
->cmd_type
) {
3111 case LTTNG_LIST_SESSIONS
:
3112 case LTTNG_LIST_TRACEPOINTS
:
3113 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3114 case LTTNG_LIST_DOMAINS
:
3115 case LTTNG_LIST_CHANNELS
:
3116 case LTTNG_LIST_EVENTS
:
3117 case LTTNG_LIST_SYSCALLS
:
3118 case LTTNG_LIST_TRACKER_PIDS
:
3119 case LTTNG_DATA_PENDING
:
3122 /* Setup lttng message with no payload */
3123 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0);
3125 /* This label does not try to unlock the session */
3126 goto init_setup_error
;
3130 /* Commands that DO NOT need a session. */
3131 switch (cmd_ctx
->lsm
->cmd_type
) {
3132 case LTTNG_CREATE_SESSION
:
3133 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3134 case LTTNG_CREATE_SESSION_LIVE
:
3135 case LTTNG_LIST_SESSIONS
:
3136 case LTTNG_LIST_TRACEPOINTS
:
3137 case LTTNG_LIST_SYSCALLS
:
3138 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3139 case LTTNG_SAVE_SESSION
:
3140 case LTTNG_REGISTER_TRIGGER
:
3141 case LTTNG_UNREGISTER_TRIGGER
:
3142 need_tracing_session
= 0;
3145 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3147 * We keep the session list lock across _all_ commands
3148 * for now, because the per-session lock does not
3149 * handle teardown properly.
3151 session_lock_list();
3152 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3153 if (cmd_ctx
->session
== NULL
) {
3154 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3157 /* Acquire lock for the session */
3158 session_lock(cmd_ctx
->session
);
3164 * Commands that need a valid session but should NOT create one if none
3165 * exists. Instead of creating one and destroying it when the command is
3166 * handled, process that right before so we save some round trip in useless
3169 switch (cmd_ctx
->lsm
->cmd_type
) {
3170 case LTTNG_DISABLE_CHANNEL
:
3171 case LTTNG_DISABLE_EVENT
:
3172 switch (cmd_ctx
->lsm
->domain
.type
) {
3173 case LTTNG_DOMAIN_KERNEL
:
3174 if (!cmd_ctx
->session
->kernel_session
) {
3175 ret
= LTTNG_ERR_NO_CHANNEL
;
3179 case LTTNG_DOMAIN_JUL
:
3180 case LTTNG_DOMAIN_LOG4J
:
3181 case LTTNG_DOMAIN_PYTHON
:
3182 case LTTNG_DOMAIN_UST
:
3183 if (!cmd_ctx
->session
->ust_session
) {
3184 ret
= LTTNG_ERR_NO_CHANNEL
;
3189 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3201 * Check domain type for specific "pre-action".
3203 switch (cmd_ctx
->lsm
->domain
.type
) {
3204 case LTTNG_DOMAIN_KERNEL
:
3206 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3210 /* Kernel tracer check */
3211 if (kernel_tracer_fd
== -1) {
3212 /* Basically, load kernel tracer modules */
3213 ret
= init_kernel_tracer();
3219 /* Consumer is in an ERROR state. Report back to client */
3220 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3221 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3225 /* Need a session for kernel command */
3226 if (need_tracing_session
) {
3227 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3228 ret
= create_kernel_session(cmd_ctx
->session
);
3230 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3235 /* Start the kernel consumer daemon */
3236 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3237 if (kconsumer_data
.pid
== 0 &&
3238 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3239 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3240 ret
= start_consumerd(&kconsumer_data
);
3242 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3245 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3247 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3251 * The consumer was just spawned so we need to add the socket to
3252 * the consumer output of the session if exist.
3254 ret
= consumer_create_socket(&kconsumer_data
,
3255 cmd_ctx
->session
->kernel_session
->consumer
);
3262 case LTTNG_DOMAIN_JUL
:
3263 case LTTNG_DOMAIN_LOG4J
:
3264 case LTTNG_DOMAIN_PYTHON
:
3265 case LTTNG_DOMAIN_UST
:
3267 if (!ust_app_supported()) {
3268 ret
= LTTNG_ERR_NO_UST
;
3271 /* Consumer is in an ERROR state. Report back to client */
3272 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3273 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3277 if (need_tracing_session
) {
3278 /* Create UST session if none exist. */
3279 if (cmd_ctx
->session
->ust_session
== NULL
) {
3280 ret
= create_ust_session(cmd_ctx
->session
,
3281 &cmd_ctx
->lsm
->domain
);
3282 if (ret
!= LTTNG_OK
) {
3287 /* Start the UST consumer daemons */
3289 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3290 if (consumerd64_bin
[0] != '\0' &&
3291 ustconsumer64_data
.pid
== 0 &&
3292 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3293 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3294 ret
= start_consumerd(&ustconsumer64_data
);
3296 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3297 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3301 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3302 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3304 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3308 * Setup socket for consumer 64 bit. No need for atomic access
3309 * since it was set above and can ONLY be set in this thread.
3311 ret
= consumer_create_socket(&ustconsumer64_data
,
3312 cmd_ctx
->session
->ust_session
->consumer
);
3318 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3319 if (consumerd32_bin
[0] != '\0' &&
3320 ustconsumer32_data
.pid
== 0 &&
3321 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3322 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3323 ret
= start_consumerd(&ustconsumer32_data
);
3325 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3326 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3330 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3331 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3333 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3337 * Setup socket for consumer 64 bit. No need for atomic access
3338 * since it was set above and can ONLY be set in this thread.
3340 ret
= consumer_create_socket(&ustconsumer32_data
,
3341 cmd_ctx
->session
->ust_session
->consumer
);
3353 /* Validate consumer daemon state when start/stop trace command */
3354 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3355 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3356 switch (cmd_ctx
->lsm
->domain
.type
) {
3357 case LTTNG_DOMAIN_NONE
:
3359 case LTTNG_DOMAIN_JUL
:
3360 case LTTNG_DOMAIN_LOG4J
:
3361 case LTTNG_DOMAIN_PYTHON
:
3362 case LTTNG_DOMAIN_UST
:
3363 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3364 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3368 case LTTNG_DOMAIN_KERNEL
:
3369 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3370 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3375 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3381 * Check that the UID or GID match that of the tracing session.
3382 * The root user can interact with all sessions.
3384 if (need_tracing_session
) {
3385 if (!session_access_ok(cmd_ctx
->session
,
3386 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3387 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3388 ret
= LTTNG_ERR_EPERM
;
3394 * Send relayd information to consumer as soon as we have a domain and a
3397 if (cmd_ctx
->session
&& need_domain
) {
3399 * Setup relayd if not done yet. If the relayd information was already
3400 * sent to the consumer, this call will gracefully return.
3402 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3403 if (ret
!= LTTNG_OK
) {
3408 /* Process by command type */
3409 switch (cmd_ctx
->lsm
->cmd_type
) {
3410 case LTTNG_ADD_CONTEXT
:
3413 * An LTTNG_ADD_CONTEXT command might have a supplementary
3414 * payload if the context being added is an application context.
3416 if (cmd_ctx
->lsm
->u
.context
.ctx
.ctx
==
3417 LTTNG_EVENT_CONTEXT_APP_CONTEXT
) {
3418 char *provider_name
= NULL
, *context_name
= NULL
;
3419 size_t provider_name_len
=
3420 cmd_ctx
->lsm
->u
.context
.provider_name_len
;
3421 size_t context_name_len
=
3422 cmd_ctx
->lsm
->u
.context
.context_name_len
;
3424 if (provider_name_len
== 0 || context_name_len
== 0) {
3426 * Application provider and context names MUST
3429 ret
= -LTTNG_ERR_INVALID
;
3433 provider_name
= zmalloc(provider_name_len
+ 1);
3434 if (!provider_name
) {
3435 ret
= -LTTNG_ERR_NOMEM
;
3438 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
=
3441 context_name
= zmalloc(context_name_len
+ 1);
3442 if (!context_name
) {
3443 ret
= -LTTNG_ERR_NOMEM
;
3444 goto error_add_context
;
3446 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
=
3449 ret
= lttcomm_recv_unix_sock(sock
, provider_name
,
3452 goto error_add_context
;
3455 ret
= lttcomm_recv_unix_sock(sock
, context_name
,
3458 goto error_add_context
;
3463 * cmd_add_context assumes ownership of the provider and context
3466 ret
= cmd_add_context(cmd_ctx
->session
,
3467 cmd_ctx
->lsm
->domain
.type
,
3468 cmd_ctx
->lsm
->u
.context
.channel_name
,
3469 &cmd_ctx
->lsm
->u
.context
.ctx
,
3470 kernel_poll_pipe
[1]);
3472 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
= NULL
;
3473 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
= NULL
;
3475 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
);
3476 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
);
3482 case LTTNG_DISABLE_CHANNEL
:
3484 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3485 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3488 case LTTNG_DISABLE_EVENT
:
3492 * FIXME: handle filter; for now we just receive the filter's
3493 * bytecode along with the filter expression which are sent by
3494 * liblttng-ctl and discard them.
3496 * This fixes an issue where the client may block while sending
3497 * the filter payload and encounter an error because the session
3498 * daemon closes the socket without ever handling this data.
3500 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3501 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3504 char data
[LTTNG_FILTER_MAX_LEN
];
3506 DBG("Discarding disable event command payload of size %zu", count
);
3508 ret
= lttcomm_recv_unix_sock(sock
, data
,
3509 count
> sizeof(data
) ? sizeof(data
) : count
);
3514 count
-= (size_t) ret
;
3517 /* FIXME: passing packed structure to non-packed pointer */
3518 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3519 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3520 &cmd_ctx
->lsm
->u
.disable
.event
);
3523 case LTTNG_ENABLE_CHANNEL
:
3525 cmd_ctx
->lsm
->u
.channel
.chan
.attr
.extended
.ptr
=
3526 (struct lttng_channel_extended
*) &cmd_ctx
->lsm
->u
.channel
.extended
;
3527 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3528 &cmd_ctx
->lsm
->u
.channel
.chan
,
3529 kernel_poll_pipe
[1]);
3532 case LTTNG_TRACK_PID
:
3534 ret
= cmd_track_pid(cmd_ctx
->session
,
3535 cmd_ctx
->lsm
->domain
.type
,
3536 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3539 case LTTNG_UNTRACK_PID
:
3541 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3542 cmd_ctx
->lsm
->domain
.type
,
3543 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3546 case LTTNG_ENABLE_EVENT
:
3548 struct lttng_event_exclusion
*exclusion
= NULL
;
3549 struct lttng_filter_bytecode
*bytecode
= NULL
;
3550 char *filter_expression
= NULL
;
3552 /* Handle exclusion events and receive it from the client. */
3553 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3554 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3556 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3557 (count
* LTTNG_SYMBOL_NAME_LEN
));
3559 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3563 DBG("Receiving var len exclusion event list from client ...");
3564 exclusion
->count
= count
;
3565 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3566 count
* LTTNG_SYMBOL_NAME_LEN
);
3568 DBG("Nothing recv() from client var len data... continuing");
3571 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3576 /* Get filter expression from client. */
3577 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3578 size_t expression_len
=
3579 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3581 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3582 ret
= LTTNG_ERR_FILTER_INVAL
;
3587 filter_expression
= zmalloc(expression_len
);
3588 if (!filter_expression
) {
3590 ret
= LTTNG_ERR_FILTER_NOMEM
;
3594 /* Receive var. len. data */
3595 DBG("Receiving var len filter's expression from client ...");
3596 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3599 DBG("Nothing recv() from client car len data... continuing");
3601 free(filter_expression
);
3603 ret
= LTTNG_ERR_FILTER_INVAL
;
3608 /* Handle filter and get bytecode from client. */
3609 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3610 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3612 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3613 ret
= LTTNG_ERR_FILTER_INVAL
;
3614 free(filter_expression
);
3619 bytecode
= zmalloc(bytecode_len
);
3621 free(filter_expression
);
3623 ret
= LTTNG_ERR_FILTER_NOMEM
;
3627 /* Receive var. len. data */
3628 DBG("Receiving var len filter's bytecode from client ...");
3629 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3631 DBG("Nothing recv() from client car len data... continuing");
3633 free(filter_expression
);
3636 ret
= LTTNG_ERR_FILTER_INVAL
;
3640 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3641 free(filter_expression
);
3644 ret
= LTTNG_ERR_FILTER_INVAL
;
3649 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3650 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3651 &cmd_ctx
->lsm
->u
.enable
.event
,
3652 filter_expression
, bytecode
, exclusion
,
3653 kernel_poll_pipe
[1]);
3656 case LTTNG_LIST_TRACEPOINTS
:
3658 struct lttng_event
*events
;
3661 session_lock_list();
3662 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3663 session_unlock_list();
3664 if (nb_events
< 0) {
3665 /* Return value is a negative lttng_error_code. */
3671 * Setup lttng message with payload size set to the event list size in
3672 * bytes and then copy list into the llm payload.
3674 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3675 sizeof(struct lttng_event
) * nb_events
);
3685 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3687 struct lttng_event_field
*fields
;
3690 session_lock_list();
3691 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3693 session_unlock_list();
3694 if (nb_fields
< 0) {
3695 /* Return value is a negative lttng_error_code. */
3701 * Setup lttng message with payload size set to the event list size in
3702 * bytes and then copy list into the llm payload.
3704 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, fields
,
3705 sizeof(struct lttng_event_field
) * nb_fields
);
3715 case LTTNG_LIST_SYSCALLS
:
3717 struct lttng_event
*events
;
3720 nb_events
= cmd_list_syscalls(&events
);
3721 if (nb_events
< 0) {
3722 /* Return value is a negative lttng_error_code. */
3728 * Setup lttng message with payload size set to the event list size in
3729 * bytes and then copy list into the llm payload.
3731 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3732 sizeof(struct lttng_event
) * nb_events
);
3742 case LTTNG_LIST_TRACKER_PIDS
:
3744 int32_t *pids
= NULL
;
3747 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3748 cmd_ctx
->lsm
->domain
.type
, &pids
);
3750 /* Return value is a negative lttng_error_code. */
3756 * Setup lttng message with payload size set to the event list size in
3757 * bytes and then copy list into the llm payload.
3759 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, pids
,
3760 sizeof(int32_t) * nr_pids
);
3770 case LTTNG_SET_CONSUMER_URI
:
3773 struct lttng_uri
*uris
;
3775 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3776 len
= nb_uri
* sizeof(struct lttng_uri
);
3779 ret
= LTTNG_ERR_INVALID
;
3783 uris
= zmalloc(len
);
3785 ret
= LTTNG_ERR_FATAL
;
3789 /* Receive variable len data */
3790 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3791 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3793 DBG("No URIs received from client... continuing");
3795 ret
= LTTNG_ERR_SESSION_FAIL
;
3800 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3802 if (ret
!= LTTNG_OK
) {
3809 case LTTNG_START_TRACE
:
3811 ret
= cmd_start_trace(cmd_ctx
->session
);
3814 case LTTNG_STOP_TRACE
:
3816 ret
= cmd_stop_trace(cmd_ctx
->session
);
3819 case LTTNG_CREATE_SESSION
:
3822 struct lttng_uri
*uris
= NULL
;
3824 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3825 len
= nb_uri
* sizeof(struct lttng_uri
);
3828 uris
= zmalloc(len
);
3830 ret
= LTTNG_ERR_FATAL
;
3834 /* Receive variable len data */
3835 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3836 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3838 DBG("No URIs received from client... continuing");
3840 ret
= LTTNG_ERR_SESSION_FAIL
;
3845 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3846 DBG("Creating session with ONE network URI is a bad call");
3847 ret
= LTTNG_ERR_SESSION_FAIL
;
3853 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3854 &cmd_ctx
->creds
, 0);
3860 case LTTNG_DESTROY_SESSION
:
3862 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3864 /* Set session to NULL so we do not unlock it after free. */
3865 cmd_ctx
->session
= NULL
;
3868 case LTTNG_LIST_DOMAINS
:
3871 struct lttng_domain
*domains
= NULL
;
3873 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3875 /* Return value is a negative lttng_error_code. */
3880 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, domains
,
3881 nb_dom
* sizeof(struct lttng_domain
));
3891 case LTTNG_LIST_CHANNELS
:
3893 ssize_t payload_size
;
3894 struct lttng_channel
*channels
= NULL
;
3896 payload_size
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3897 cmd_ctx
->session
, &channels
);
3898 if (payload_size
< 0) {
3899 /* Return value is a negative lttng_error_code. */
3900 ret
= -payload_size
;
3904 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, channels
,
3915 case LTTNG_LIST_EVENTS
:
3918 struct lttng_event
*events
= NULL
;
3919 struct lttcomm_event_command_header cmd_header
;
3922 memset(&cmd_header
, 0, sizeof(cmd_header
));
3923 /* Extended infos are included at the end of events */
3924 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
,
3925 cmd_ctx
->session
, cmd_ctx
->lsm
->u
.list
.channel_name
,
3926 &events
, &total_size
);
3929 /* Return value is a negative lttng_error_code. */
3934 cmd_header
.nb_events
= nb_event
;
3935 ret
= setup_lttng_msg(cmd_ctx
, events
, total_size
,
3936 &cmd_header
, sizeof(cmd_header
));
3946 case LTTNG_LIST_SESSIONS
:
3948 unsigned int nr_sessions
;
3949 void *sessions_payload
;
3952 session_lock_list();
3953 nr_sessions
= lttng_sessions_count(
3954 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3955 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3956 payload_len
= sizeof(struct lttng_session
) * nr_sessions
;
3957 sessions_payload
= zmalloc(payload_len
);
3959 if (!sessions_payload
) {
3960 session_unlock_list();
3965 cmd_list_lttng_sessions(sessions_payload
,
3966 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3967 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3968 session_unlock_list();
3970 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, sessions_payload
,
3972 free(sessions_payload
);
3981 case LTTNG_REGISTER_CONSUMER
:
3983 struct consumer_data
*cdata
;
3985 switch (cmd_ctx
->lsm
->domain
.type
) {
3986 case LTTNG_DOMAIN_KERNEL
:
3987 cdata
= &kconsumer_data
;
3990 ret
= LTTNG_ERR_UND
;
3994 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3995 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3998 case LTTNG_DATA_PENDING
:
4001 uint8_t pending_ret_byte
;
4003 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
4008 * This function may returns 0 or 1 to indicate whether or not
4009 * there is data pending. In case of error, it should return an
4010 * LTTNG_ERR code. However, some code paths may still return
4011 * a nondescript error code, which we handle by returning an
4014 if (pending_ret
== 0 || pending_ret
== 1) {
4016 * ret will be set to LTTNG_OK at the end of
4019 } else if (pending_ret
< 0) {
4020 ret
= LTTNG_ERR_UNK
;
4027 pending_ret_byte
= (uint8_t) pending_ret
;
4029 /* 1 byte to return whether or not data is pending */
4030 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
,
4031 &pending_ret_byte
, 1);
4040 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
4042 struct lttcomm_lttng_output_id reply
;
4044 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
4045 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
4046 if (ret
!= LTTNG_OK
) {
4050 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &reply
,
4056 /* Copy output list into message payload */
4060 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
4062 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
4063 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
4066 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
4069 struct lttng_snapshot_output
*outputs
= NULL
;
4071 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
4072 if (nb_output
< 0) {
4077 assert((nb_output
> 0 && outputs
) || nb_output
== 0);
4078 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, outputs
,
4079 nb_output
* sizeof(struct lttng_snapshot_output
));
4089 case LTTNG_SNAPSHOT_RECORD
:
4091 ret
= cmd_snapshot_record(cmd_ctx
->session
,
4092 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
4093 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
4096 case LTTNG_CREATE_SESSION_SNAPSHOT
:
4099 struct lttng_uri
*uris
= NULL
;
4101 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4102 len
= nb_uri
* sizeof(struct lttng_uri
);
4105 uris
= zmalloc(len
);
4107 ret
= LTTNG_ERR_FATAL
;
4111 /* Receive variable len data */
4112 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4113 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4115 DBG("No URIs received from client... continuing");
4117 ret
= LTTNG_ERR_SESSION_FAIL
;
4122 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4123 DBG("Creating session with ONE network URI is a bad call");
4124 ret
= LTTNG_ERR_SESSION_FAIL
;
4130 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
4131 nb_uri
, &cmd_ctx
->creds
);
4135 case LTTNG_CREATE_SESSION_LIVE
:
4138 struct lttng_uri
*uris
= NULL
;
4140 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4141 len
= nb_uri
* sizeof(struct lttng_uri
);
4144 uris
= zmalloc(len
);
4146 ret
= LTTNG_ERR_FATAL
;
4150 /* Receive variable len data */
4151 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4152 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4154 DBG("No URIs received from client... continuing");
4156 ret
= LTTNG_ERR_SESSION_FAIL
;
4161 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4162 DBG("Creating session with ONE network URI is a bad call");
4163 ret
= LTTNG_ERR_SESSION_FAIL
;
4169 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4170 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4174 case LTTNG_SAVE_SESSION
:
4176 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4180 case LTTNG_SET_SESSION_SHM_PATH
:
4182 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4183 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4186 case LTTNG_REGENERATE_METADATA
:
4188 ret
= cmd_regenerate_metadata(cmd_ctx
->session
);
4191 case LTTNG_REGENERATE_STATEDUMP
:
4193 ret
= cmd_regenerate_statedump(cmd_ctx
->session
);
4196 case LTTNG_REGISTER_TRIGGER
:
4198 ret
= cmd_register_trigger(cmd_ctx
, sock
,
4199 notification_thread_handle
);
4202 case LTTNG_UNREGISTER_TRIGGER
:
4204 ret
= cmd_unregister_trigger(cmd_ctx
, sock
,
4205 notification_thread_handle
);
4209 ret
= LTTNG_ERR_UND
;
4214 if (cmd_ctx
->llm
== NULL
) {
4215 DBG("Missing llm structure. Allocating one.");
4216 if (setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0) < 0) {
4220 /* Set return code */
4221 cmd_ctx
->llm
->ret_code
= ret
;
4223 if (cmd_ctx
->session
) {
4224 session_unlock(cmd_ctx
->session
);
4226 if (need_tracing_session
) {
4227 session_unlock_list();
4230 assert(!rcu_read_ongoing());
4235 * Thread managing health check socket.
4237 static void *thread_manage_health(void *data
)
4239 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4240 uint32_t revents
, nb_fd
;
4241 struct lttng_poll_event events
;
4242 struct health_comm_msg msg
;
4243 struct health_comm_reply reply
;
4245 DBG("[thread] Manage health check started");
4247 rcu_register_thread();
4249 /* We might hit an error path before this is created. */
4250 lttng_poll_init(&events
);
4252 /* Create unix socket */
4253 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4255 ERR("Unable to create health check Unix socket");
4260 /* lttng health client socket path permissions */
4261 ret
= chown(health_unix_sock_path
, 0,
4262 utils_get_group_id(tracing_group_name
));
4264 ERR("Unable to set group on %s", health_unix_sock_path
);
4269 ret
= chmod(health_unix_sock_path
,
4270 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4272 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4279 * Set the CLOEXEC flag. Return code is useless because either way, the
4282 (void) utils_set_fd_cloexec(sock
);
4284 ret
= lttcomm_listen_unix_sock(sock
);
4290 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4291 * more will be added to this poll set.
4293 ret
= sessiond_set_thread_pollset(&events
, 2);
4298 /* Add the application registration socket */
4299 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4304 sessiond_notify_ready();
4307 DBG("Health check ready");
4309 /* Inifinite blocking call, waiting for transmission */
4311 ret
= lttng_poll_wait(&events
, -1);
4314 * Restart interrupted system call.
4316 if (errno
== EINTR
) {
4324 for (i
= 0; i
< nb_fd
; i
++) {
4325 /* Fetch once the poll data */
4326 revents
= LTTNG_POLL_GETEV(&events
, i
);
4327 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4330 /* No activity for this FD (poll implementation). */
4334 /* Thread quit pipe has been closed. Killing thread. */
4335 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4341 /* Event on the registration socket */
4342 if (pollfd
== sock
) {
4343 if (revents
& LPOLLIN
) {
4345 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4346 ERR("Health socket poll error");
4349 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4355 new_sock
= lttcomm_accept_unix_sock(sock
);
4361 * Set the CLOEXEC flag. Return code is useless because either way, the
4364 (void) utils_set_fd_cloexec(new_sock
);
4366 DBG("Receiving data from client for health...");
4367 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4369 DBG("Nothing recv() from client... continuing");
4370 ret
= close(new_sock
);
4377 rcu_thread_online();
4379 memset(&reply
, 0, sizeof(reply
));
4380 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4382 * health_check_state returns 0 if health is
4385 if (!health_check_state(health_sessiond
, i
)) {
4386 reply
.ret_code
|= 1ULL << i
;
4390 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4392 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4394 ERR("Failed to send health data back to client");
4397 /* End of transmission */
4398 ret
= close(new_sock
);
4407 ERR("Health error occurred in %s", __func__
);
4409 DBG("Health check thread dying");
4410 unlink(health_unix_sock_path
);
4418 lttng_poll_clean(&events
);
4420 rcu_unregister_thread();
4425 * This thread manage all clients request using the unix client socket for
4428 static void *thread_manage_clients(void *data
)
4430 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4432 uint32_t revents
, nb_fd
;
4433 struct command_ctx
*cmd_ctx
= NULL
;
4434 struct lttng_poll_event events
;
4436 DBG("[thread] Manage client started");
4438 rcu_register_thread();
4440 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4442 health_code_update();
4444 ret
= lttcomm_listen_unix_sock(client_sock
);
4450 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4451 * more will be added to this poll set.
4453 ret
= sessiond_set_thread_pollset(&events
, 2);
4455 goto error_create_poll
;
4458 /* Add the application registration socket */
4459 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4464 sessiond_notify_ready();
4465 ret
= sem_post(&load_info
->message_thread_ready
);
4467 PERROR("sem_post message_thread_ready");
4471 /* This testpoint is after we signal readiness to the parent. */
4472 if (testpoint(sessiond_thread_manage_clients
)) {
4476 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4480 health_code_update();
4483 DBG("Accepting client command ...");
4485 /* Inifinite blocking call, waiting for transmission */
4487 health_poll_entry();
4488 ret
= lttng_poll_wait(&events
, -1);
4492 * Restart interrupted system call.
4494 if (errno
== EINTR
) {
4502 for (i
= 0; i
< nb_fd
; i
++) {
4503 /* Fetch once the poll data */
4504 revents
= LTTNG_POLL_GETEV(&events
, i
);
4505 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4507 health_code_update();
4510 /* No activity for this FD (poll implementation). */
4514 /* Thread quit pipe has been closed. Killing thread. */
4515 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4521 /* Event on the registration socket */
4522 if (pollfd
== client_sock
) {
4523 if (revents
& LPOLLIN
) {
4525 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4526 ERR("Client socket poll error");
4529 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4535 DBG("Wait for client response");
4537 health_code_update();
4539 sock
= lttcomm_accept_unix_sock(client_sock
);
4545 * Set the CLOEXEC flag. Return code is useless because either way, the
4548 (void) utils_set_fd_cloexec(sock
);
4550 /* Set socket option for credentials retrieval */
4551 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4556 /* Allocate context command to process the client request */
4557 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4558 if (cmd_ctx
== NULL
) {
4559 PERROR("zmalloc cmd_ctx");
4563 /* Allocate data buffer for reception */
4564 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4565 if (cmd_ctx
->lsm
== NULL
) {
4566 PERROR("zmalloc cmd_ctx->lsm");
4570 cmd_ctx
->llm
= NULL
;
4571 cmd_ctx
->session
= NULL
;
4573 health_code_update();
4576 * Data is received from the lttng client. The struct
4577 * lttcomm_session_msg (lsm) contains the command and data request of
4580 DBG("Receiving data from client ...");
4581 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4582 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4584 DBG("Nothing recv() from client... continuing");
4590 clean_command_ctx(&cmd_ctx
);
4594 health_code_update();
4596 // TODO: Validate cmd_ctx including sanity check for
4597 // security purpose.
4599 rcu_thread_online();
4601 * This function dispatch the work to the kernel or userspace tracer
4602 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4603 * informations for the client. The command context struct contains
4604 * everything this function may needs.
4606 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4607 rcu_thread_offline();
4615 * TODO: Inform client somehow of the fatal error. At
4616 * this point, ret < 0 means that a zmalloc failed
4617 * (ENOMEM). Error detected but still accept
4618 * command, unless a socket error has been
4621 clean_command_ctx(&cmd_ctx
);
4625 health_code_update();
4627 DBG("Sending response (size: %d, retcode: %s (%d))",
4628 cmd_ctx
->lttng_msg_size
,
4629 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4630 cmd_ctx
->llm
->ret_code
);
4631 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4633 ERR("Failed to send data back to client");
4636 /* End of transmission */
4643 clean_command_ctx(&cmd_ctx
);
4645 health_code_update();
4657 lttng_poll_clean(&events
);
4658 clean_command_ctx(&cmd_ctx
);
4662 unlink(client_unix_sock_path
);
4663 if (client_sock
>= 0) {
4664 ret
= close(client_sock
);
4672 ERR("Health error occurred in %s", __func__
);
4675 health_unregister(health_sessiond
);
4677 DBG("Client thread dying");
4679 rcu_unregister_thread();
4682 * Since we are creating the consumer threads, we own them, so we need
4683 * to join them before our thread exits.
4685 ret
= join_consumer_thread(&kconsumer_data
);
4688 PERROR("join_consumer");
4691 ret
= join_consumer_thread(&ustconsumer32_data
);
4694 PERROR("join_consumer ust32");
4697 ret
= join_consumer_thread(&ustconsumer64_data
);
4700 PERROR("join_consumer ust64");
4705 static int string_match(const char *str1
, const char *str2
)
4707 return (str1
&& str2
) && !strcmp(str1
, str2
);
4711 * Take an option from the getopt output and set it in the right variable to be
4714 * Return 0 on success else a negative value.
4716 static int set_option(int opt
, const char *arg
, const char *optname
)
4720 if (string_match(optname
, "client-sock") || opt
== 'c') {
4721 if (!arg
|| *arg
== '\0') {
4725 if (lttng_is_setuid_setgid()) {
4726 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4727 "-c, --client-sock");
4729 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4731 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4732 if (!arg
|| *arg
== '\0') {
4736 if (lttng_is_setuid_setgid()) {
4737 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4740 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4742 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4744 } else if (string_match(optname
, "background") || opt
== 'b') {
4746 } else if (string_match(optname
, "group") || opt
== 'g') {
4747 if (!arg
|| *arg
== '\0') {
4751 if (lttng_is_setuid_setgid()) {
4752 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4756 * If the override option is set, the pointer points to a
4757 * *non* const thus freeing it even though the variable type is
4760 if (tracing_group_name_override
) {
4761 free((void *) tracing_group_name
);
4763 tracing_group_name
= strdup(arg
);
4764 if (!tracing_group_name
) {
4768 tracing_group_name_override
= 1;
4770 } else if (string_match(optname
, "help") || opt
== 'h') {
4771 ret
= utils_show_help(8, "lttng-sessiond", help_msg
);
4773 ERR("Cannot show --help for `lttng-sessiond`");
4776 exit(ret
? EXIT_FAILURE
: EXIT_SUCCESS
);
4777 } else if (string_match(optname
, "version") || opt
== 'V') {
4778 fprintf(stdout
, "%s\n", VERSION
);
4780 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4782 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4783 if (!arg
|| *arg
== '\0') {
4787 if (lttng_is_setuid_setgid()) {
4788 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4789 "--kconsumerd-err-sock");
4791 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4793 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4794 if (!arg
|| *arg
== '\0') {
4798 if (lttng_is_setuid_setgid()) {
4799 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4800 "--kconsumerd-cmd-sock");
4802 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4804 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4805 if (!arg
|| *arg
== '\0') {
4809 if (lttng_is_setuid_setgid()) {
4810 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4811 "--ustconsumerd64-err-sock");
4813 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4815 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4816 if (!arg
|| *arg
== '\0') {
4820 if (lttng_is_setuid_setgid()) {
4821 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4822 "--ustconsumerd64-cmd-sock");
4824 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4826 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4827 if (!arg
|| *arg
== '\0') {
4831 if (lttng_is_setuid_setgid()) {
4832 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4833 "--ustconsumerd32-err-sock");
4835 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4837 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4838 if (!arg
|| *arg
== '\0') {
4842 if (lttng_is_setuid_setgid()) {
4843 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4844 "--ustconsumerd32-cmd-sock");
4846 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4848 } else if (string_match(optname
, "no-kernel")) {
4850 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4851 lttng_opt_quiet
= 1;
4852 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4853 /* Verbose level can increase using multiple -v */
4855 /* Value obtained from config file */
4856 lttng_opt_verbose
= config_parse_value(arg
);
4858 /* -v used on command line */
4859 lttng_opt_verbose
++;
4861 /* Clamp value to [0, 3] */
4862 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4863 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4864 } else if (string_match(optname
, "verbose-consumer")) {
4866 opt_verbose_consumer
= config_parse_value(arg
);
4868 opt_verbose_consumer
++;
4870 } else if (string_match(optname
, "consumerd32-path")) {
4871 if (!arg
|| *arg
== '\0') {
4875 if (lttng_is_setuid_setgid()) {
4876 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4877 "--consumerd32-path");
4879 if (consumerd32_bin_override
) {
4880 free((void *) consumerd32_bin
);
4882 consumerd32_bin
= strdup(arg
);
4883 if (!consumerd32_bin
) {
4887 consumerd32_bin_override
= 1;
4889 } else if (string_match(optname
, "consumerd32-libdir")) {
4890 if (!arg
|| *arg
== '\0') {
4894 if (lttng_is_setuid_setgid()) {
4895 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4896 "--consumerd32-libdir");
4898 if (consumerd32_libdir_override
) {
4899 free((void *) consumerd32_libdir
);
4901 consumerd32_libdir
= strdup(arg
);
4902 if (!consumerd32_libdir
) {
4906 consumerd32_libdir_override
= 1;
4908 } else if (string_match(optname
, "consumerd64-path")) {
4909 if (!arg
|| *arg
== '\0') {
4913 if (lttng_is_setuid_setgid()) {
4914 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4915 "--consumerd64-path");
4917 if (consumerd64_bin_override
) {
4918 free((void *) consumerd64_bin
);
4920 consumerd64_bin
= strdup(arg
);
4921 if (!consumerd64_bin
) {
4925 consumerd64_bin_override
= 1;
4927 } else if (string_match(optname
, "consumerd64-libdir")) {
4928 if (!arg
|| *arg
== '\0') {
4932 if (lttng_is_setuid_setgid()) {
4933 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4934 "--consumerd64-libdir");
4936 if (consumerd64_libdir_override
) {
4937 free((void *) consumerd64_libdir
);
4939 consumerd64_libdir
= strdup(arg
);
4940 if (!consumerd64_libdir
) {
4944 consumerd64_libdir_override
= 1;
4946 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4947 if (!arg
|| *arg
== '\0') {
4951 if (lttng_is_setuid_setgid()) {
4952 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4956 opt_pidfile
= strdup(arg
);
4962 } else if (string_match(optname
, "agent-tcp-port")) {
4963 if (!arg
|| *arg
== '\0') {
4967 if (lttng_is_setuid_setgid()) {
4968 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4969 "--agent-tcp-port");
4974 v
= strtoul(arg
, NULL
, 0);
4975 if (errno
!= 0 || !isdigit(arg
[0])) {
4976 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4979 if (v
== 0 || v
>= 65535) {
4980 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4983 agent_tcp_port
= (uint32_t) v
;
4984 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4986 } else if (string_match(optname
, "load") || opt
== 'l') {
4987 if (!arg
|| *arg
== '\0') {
4991 if (lttng_is_setuid_setgid()) {
4992 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4995 free(opt_load_session_path
);
4996 opt_load_session_path
= strdup(arg
);
4997 if (!opt_load_session_path
) {
5002 } else if (string_match(optname
, "kmod-probes")) {
5003 if (!arg
|| *arg
== '\0') {
5007 if (lttng_is_setuid_setgid()) {
5008 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5011 free(kmod_probes_list
);
5012 kmod_probes_list
= strdup(arg
);
5013 if (!kmod_probes_list
) {
5018 } else if (string_match(optname
, "extra-kmod-probes")) {
5019 if (!arg
|| *arg
== '\0') {
5023 if (lttng_is_setuid_setgid()) {
5024 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5025 "--extra-kmod-probes");
5027 free(kmod_extra_probes_list
);
5028 kmod_extra_probes_list
= strdup(arg
);
5029 if (!kmod_extra_probes_list
) {
5034 } else if (string_match(optname
, "config") || opt
== 'f') {
5035 /* This is handled in set_options() thus silent skip. */
5038 /* Unknown option or other error.
5039 * Error is printed by getopt, just return */
5044 if (ret
== -EINVAL
) {
5045 const char *opt_name
= "unknown";
5048 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
5050 if (opt
== long_options
[i
].val
) {
5051 opt_name
= long_options
[i
].name
;
5056 WARN("Invalid argument provided for option \"%s\", using default value.",
5064 * config_entry_handler_cb used to handle options read from a config file.
5065 * See config_entry_handler_cb comment in common/config/session-config.h for the
5066 * return value conventions.
5068 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
5072 if (!entry
|| !entry
->name
|| !entry
->value
) {
5077 /* Check if the option is to be ignored */
5078 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
5079 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
5084 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
5087 /* Ignore if not fully matched. */
5088 if (strcmp(entry
->name
, long_options
[i
].name
)) {
5093 * If the option takes no argument on the command line, we have to
5094 * check if the value is "true". We support non-zero numeric values,
5097 if (!long_options
[i
].has_arg
) {
5098 ret
= config_parse_value(entry
->value
);
5101 WARN("Invalid configuration value \"%s\" for option %s",
5102 entry
->value
, entry
->name
);
5104 /* False, skip boolean config option. */
5109 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
5113 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
5120 * daemon configuration loading and argument parsing
5122 static int set_options(int argc
, char **argv
)
5124 int ret
= 0, c
= 0, option_index
= 0;
5125 int orig_optopt
= optopt
, orig_optind
= optind
;
5127 const char *config_path
= NULL
;
5129 optstring
= utils_generate_optstring(long_options
,
5130 sizeof(long_options
) / sizeof(struct option
));
5136 /* Check for the --config option */
5137 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5138 &option_index
)) != -1) {
5142 } else if (c
!= 'f') {
5143 /* if not equal to --config option. */
5147 if (lttng_is_setuid_setgid()) {
5148 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5151 config_path
= utils_expand_path(optarg
);
5153 ERR("Failed to resolve path: %s", optarg
);
5158 ret
= config_get_section_entries(config_path
, config_section_name
,
5159 config_entry_handler
, NULL
);
5162 ERR("Invalid configuration option at line %i", ret
);
5168 /* Reset getopt's global state */
5169 optopt
= orig_optopt
;
5170 optind
= orig_optind
;
5174 * getopt_long() will not set option_index if it encounters a
5177 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5184 * Pass NULL as the long option name if popt left the index
5187 ret
= set_option(c
, optarg
,
5188 option_index
< 0 ? NULL
:
5189 long_options
[option_index
].name
);
5201 * Creates the two needed socket by the daemon.
5202 * apps_sock - The communication socket for all UST apps.
5203 * client_sock - The communication of the cli tool (lttng).
5205 static int init_daemon_socket(void)
5210 old_umask
= umask(0);
5212 /* Create client tool unix socket */
5213 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
5214 if (client_sock
< 0) {
5215 ERR("Create unix sock failed: %s", client_unix_sock_path
);
5220 /* Set the cloexec flag */
5221 ret
= utils_set_fd_cloexec(client_sock
);
5223 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5224 "Continuing but note that the consumer daemon will have a "
5225 "reference to this socket on exec()", client_sock
);
5228 /* File permission MUST be 660 */
5229 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5231 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5236 /* Create the application unix socket */
5237 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5238 if (apps_sock
< 0) {
5239 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5244 /* Set the cloexec flag */
5245 ret
= utils_set_fd_cloexec(apps_sock
);
5247 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5248 "Continuing but note that the consumer daemon will have a "
5249 "reference to this socket on exec()", apps_sock
);
5252 /* File permission MUST be 666 */
5253 ret
= chmod(apps_unix_sock_path
,
5254 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5256 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5261 DBG3("Session daemon client socket %d and application socket %d created",
5262 client_sock
, apps_sock
);
5270 * Check if the global socket is available, and if a daemon is answering at the
5271 * other side. If yes, error is returned.
5273 static int check_existing_daemon(void)
5275 /* Is there anybody out there ? */
5276 if (lttng_session_daemon_alive()) {
5284 * Set the tracing group gid onto the client socket.
5286 * Race window between mkdir and chown is OK because we are going from more
5287 * permissive (root.root) to less permissive (root.tracing).
5289 static int set_permissions(char *rundir
)
5294 gid
= utils_get_group_id(tracing_group_name
);
5296 /* Set lttng run dir */
5297 ret
= chown(rundir
, 0, gid
);
5299 ERR("Unable to set group on %s", rundir
);
5304 * Ensure all applications and tracing group can search the run
5305 * dir. Allow everyone to read the directory, since it does not
5306 * buy us anything to hide its content.
5308 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5310 ERR("Unable to set permissions on %s", rundir
);
5314 /* lttng client socket path */
5315 ret
= chown(client_unix_sock_path
, 0, gid
);
5317 ERR("Unable to set group on %s", client_unix_sock_path
);
5321 /* kconsumer error socket path */
5322 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5324 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5328 /* 64-bit ustconsumer error socket path */
5329 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5331 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5335 /* 32-bit ustconsumer compat32 error socket path */
5336 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5338 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5342 DBG("All permissions are set");
5348 * Create the lttng run directory needed for all global sockets and pipe.
5350 static int create_lttng_rundir(const char *rundir
)
5354 DBG3("Creating LTTng run directory: %s", rundir
);
5356 ret
= mkdir(rundir
, S_IRWXU
);
5358 if (errno
!= EEXIST
) {
5359 ERR("Unable to create %s", rundir
);
5371 * Setup sockets and directory needed by the kconsumerd communication with the
5374 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5378 char path
[PATH_MAX
];
5380 switch (consumer_data
->type
) {
5381 case LTTNG_CONSUMER_KERNEL
:
5382 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5384 case LTTNG_CONSUMER64_UST
:
5385 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5387 case LTTNG_CONSUMER32_UST
:
5388 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5391 ERR("Consumer type unknown");
5396 DBG2("Creating consumer directory: %s", path
);
5398 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5400 if (errno
!= EEXIST
) {
5402 ERR("Failed to create %s", path
);
5408 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5410 ERR("Unable to set group on %s", path
);
5416 /* Create the kconsumerd error unix socket */
5417 consumer_data
->err_sock
=
5418 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5419 if (consumer_data
->err_sock
< 0) {
5420 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5426 * Set the CLOEXEC flag. Return code is useless because either way, the
5429 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5431 PERROR("utils_set_fd_cloexec");
5432 /* continue anyway */
5435 /* File permission MUST be 660 */
5436 ret
= chmod(consumer_data
->err_unix_sock_path
,
5437 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5439 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5449 * Signal handler for the daemon
5451 * Simply stop all worker threads, leaving main() return gracefully after
5452 * joining all threads and calling cleanup().
5454 static void sighandler(int sig
)
5458 DBG("SIGINT caught");
5462 DBG("SIGTERM caught");
5466 CMM_STORE_SHARED(recv_child_signal
, 1);
5474 * Setup signal handler for :
5475 * SIGINT, SIGTERM, SIGPIPE
5477 static int set_signal_handler(void)
5480 struct sigaction sa
;
5483 if ((ret
= sigemptyset(&sigset
)) < 0) {
5484 PERROR("sigemptyset");
5488 sa
.sa_mask
= sigset
;
5491 sa
.sa_handler
= sighandler
;
5492 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5493 PERROR("sigaction");
5497 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5498 PERROR("sigaction");
5502 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5503 PERROR("sigaction");
5507 sa
.sa_handler
= SIG_IGN
;
5508 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5509 PERROR("sigaction");
5513 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5519 * Set open files limit to unlimited. This daemon can open a large number of
5520 * file descriptors in order to consume multiple kernel traces.
5522 static void set_ulimit(void)
5527 /* The kernel does not allow an infinite limit for open files */
5528 lim
.rlim_cur
= 65535;
5529 lim
.rlim_max
= 65535;
5531 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5533 PERROR("failed to set open files limit");
5538 * Write pidfile using the rundir and opt_pidfile.
5540 static int write_pidfile(void)
5543 char pidfile_path
[PATH_MAX
];
5548 if (lttng_strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
))) {
5553 /* Build pidfile path from rundir and opt_pidfile. */
5554 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5555 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5557 PERROR("snprintf pidfile path");
5563 * Create pid file in rundir.
5565 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5571 * Create lockfile using the rundir and return its fd.
5573 static int create_lockfile(void)
5576 char lockfile_path
[PATH_MAX
];
5578 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5583 ret
= utils_create_lock_file(lockfile_path
);
5589 * Write agent TCP port using the rundir.
5591 static int write_agent_port(void)
5594 char path
[PATH_MAX
];
5598 ret
= snprintf(path
, sizeof(path
), "%s/"
5599 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5601 PERROR("snprintf agent port path");
5606 * Create TCP agent port file in rundir.
5608 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5617 int main(int argc
, char **argv
)
5619 int ret
= 0, retval
= 0;
5621 const char *home_path
, *env_app_timeout
;
5622 struct lttng_pipe
*ust32_channel_monitor_pipe
= NULL
,
5623 *ust64_channel_monitor_pipe
= NULL
,
5624 *kernel_channel_monitor_pipe
= NULL
;
5625 bool notification_thread_running
= false;
5627 init_kernel_workarounds();
5629 rcu_register_thread();
5631 if (set_signal_handler()) {
5633 goto exit_set_signal_handler
;
5636 setup_consumerd_path();
5638 page_size
= sysconf(_SC_PAGESIZE
);
5639 if (page_size
< 0) {
5640 PERROR("sysconf _SC_PAGESIZE");
5641 page_size
= LONG_MAX
;
5642 WARN("Fallback page size to %ld", page_size
);
5646 * Parse arguments and load the daemon configuration file.
5648 * We have an exit_options exit path to free memory reserved by
5649 * set_options. This is needed because the rest of sessiond_cleanup()
5650 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5651 * depends on set_options.
5654 if (set_options(argc
, argv
)) {
5660 if (opt_daemon
|| opt_background
) {
5663 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5671 * We are in the child. Make sure all other file descriptors are
5672 * closed, in case we are called with more opened file
5673 * descriptors than the standard ones.
5675 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5680 if (run_as_create_worker(argv
[0]) < 0) {
5681 goto exit_create_run_as_worker_cleanup
;
5685 * Starting from here, we can create threads. This needs to be after
5686 * lttng_daemonize due to RCU.
5690 * Initialize the health check subsystem. This call should set the
5691 * appropriate time values.
5693 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5694 if (!health_sessiond
) {
5695 PERROR("health_app_create error");
5697 goto exit_health_sessiond_cleanup
;
5700 /* Create thread to clean up RCU hash tables */
5701 if (init_ht_cleanup_thread(&ht_cleanup_thread
)) {
5703 goto exit_ht_cleanup
;
5706 /* Create thread quit pipe */
5707 if (init_thread_quit_pipe()) {
5709 goto exit_init_data
;
5712 /* Check if daemon is UID = 0 */
5713 is_root
= !getuid();
5716 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5719 goto exit_init_data
;
5722 /* Create global run dir with root access */
5723 if (create_lttng_rundir(rundir
)) {
5725 goto exit_init_data
;
5728 if (strlen(apps_unix_sock_path
) == 0) {
5729 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5730 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5733 goto exit_init_data
;
5737 if (strlen(client_unix_sock_path
) == 0) {
5738 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5739 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5742 goto exit_init_data
;
5746 /* Set global SHM for ust */
5747 if (strlen(wait_shm_path
) == 0) {
5748 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5749 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5752 goto exit_init_data
;
5756 if (strlen(health_unix_sock_path
) == 0) {
5757 ret
= snprintf(health_unix_sock_path
,
5758 sizeof(health_unix_sock_path
),
5759 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5762 goto exit_init_data
;
5766 /* Setup kernel consumerd path */
5767 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5768 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5771 goto exit_init_data
;
5773 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5774 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5777 goto exit_init_data
;
5780 DBG2("Kernel consumer err path: %s",
5781 kconsumer_data
.err_unix_sock_path
);
5782 DBG2("Kernel consumer cmd path: %s",
5783 kconsumer_data
.cmd_unix_sock_path
);
5784 kernel_channel_monitor_pipe
= lttng_pipe_open(0);
5785 if (!kernel_channel_monitor_pipe
) {
5786 ERR("Failed to create kernel consumer channel monitor pipe");
5788 goto exit_init_data
;
5790 kconsumer_data
.channel_monitor_pipe
=
5791 lttng_pipe_release_writefd(
5792 kernel_channel_monitor_pipe
);
5793 if (kconsumer_data
.channel_monitor_pipe
< 0) {
5795 goto exit_init_data
;
5798 home_path
= utils_get_home_dir();
5799 if (home_path
== NULL
) {
5800 /* TODO: Add --socket PATH option */
5801 ERR("Can't get HOME directory for sockets creation.");
5803 goto exit_init_data
;
5807 * Create rundir from home path. This will create something like
5810 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5813 goto exit_init_data
;
5816 if (create_lttng_rundir(rundir
)) {
5818 goto exit_init_data
;
5821 if (strlen(apps_unix_sock_path
) == 0) {
5822 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5823 DEFAULT_HOME_APPS_UNIX_SOCK
,
5827 goto exit_init_data
;
5831 /* Set the cli tool unix socket path */
5832 if (strlen(client_unix_sock_path
) == 0) {
5833 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5834 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5838 goto exit_init_data
;
5842 /* Set global SHM for ust */
5843 if (strlen(wait_shm_path
) == 0) {
5844 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5845 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5849 goto exit_init_data
;
5853 /* Set health check Unix path */
5854 if (strlen(health_unix_sock_path
) == 0) {
5855 ret
= snprintf(health_unix_sock_path
,
5856 sizeof(health_unix_sock_path
),
5857 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5861 goto exit_init_data
;
5866 lockfile_fd
= create_lockfile();
5867 if (lockfile_fd
< 0) {
5869 goto exit_init_data
;
5872 /* Set consumer initial state */
5873 kernel_consumerd_state
= CONSUMER_STOPPED
;
5874 ust_consumerd_state
= CONSUMER_STOPPED
;
5876 DBG("Client socket path %s", client_unix_sock_path
);
5877 DBG("Application socket path %s", apps_unix_sock_path
);
5878 DBG("Application wait path %s", wait_shm_path
);
5879 DBG("LTTng run directory path: %s", rundir
);
5881 /* 32 bits consumerd path setup */
5882 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5883 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5885 PERROR("snprintf 32-bit consumer error socket path");
5887 goto exit_init_data
;
5889 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5890 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5892 PERROR("snprintf 32-bit consumer command socket path");
5894 goto exit_init_data
;
5897 DBG2("UST consumer 32 bits err path: %s",
5898 ustconsumer32_data
.err_unix_sock_path
);
5899 DBG2("UST consumer 32 bits cmd path: %s",
5900 ustconsumer32_data
.cmd_unix_sock_path
);
5901 ust32_channel_monitor_pipe
= lttng_pipe_open(0);
5902 if (!ust32_channel_monitor_pipe
) {
5903 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5905 goto exit_init_data
;
5907 ustconsumer32_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
5908 ust32_channel_monitor_pipe
);
5909 if (ustconsumer32_data
.channel_monitor_pipe
< 0) {
5911 goto exit_init_data
;
5914 /* 64 bits consumerd path setup */
5915 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5916 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5918 PERROR("snprintf 64-bit consumer error socket path");
5920 goto exit_init_data
;
5922 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5923 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5925 PERROR("snprintf 64-bit consumer command socket path");
5927 goto exit_init_data
;
5930 DBG2("UST consumer 64 bits err path: %s",
5931 ustconsumer64_data
.err_unix_sock_path
);
5932 DBG2("UST consumer 64 bits cmd path: %s",
5933 ustconsumer64_data
.cmd_unix_sock_path
);
5934 ust64_channel_monitor_pipe
= lttng_pipe_open(0);
5935 if (!ust64_channel_monitor_pipe
) {
5936 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5938 goto exit_init_data
;
5940 ustconsumer64_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
5941 ust64_channel_monitor_pipe
);
5942 if (ustconsumer64_data
.channel_monitor_pipe
< 0) {
5944 goto exit_init_data
;
5948 * See if daemon already exist.
5950 if (check_existing_daemon()) {
5951 ERR("Already running daemon.\n");
5953 * We do not goto exit because we must not cleanup()
5954 * because a daemon is already running.
5957 goto exit_init_data
;
5961 * Init UST app hash table. Alloc hash table before this point since
5962 * cleanup() can get called after that point.
5964 if (ust_app_ht_alloc()) {
5965 ERR("Failed to allocate UST app hash table");
5967 goto exit_init_data
;
5971 * Initialize agent app hash table. We allocate the hash table here
5972 * since cleanup() can get called after this point.
5974 if (agent_app_ht_alloc()) {
5975 ERR("Failed to allocate Agent app hash table");
5977 goto exit_init_data
;
5981 * These actions must be executed as root. We do that *after* setting up
5982 * the sockets path because we MUST make the check for another daemon using
5983 * those paths *before* trying to set the kernel consumer sockets and init
5987 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5989 goto exit_init_data
;
5992 /* Setup kernel tracer */
5993 if (!opt_no_kernel
) {
5994 init_kernel_tracer();
5995 if (kernel_tracer_fd
>= 0) {
5996 ret
= syscall_init_table();
5998 ERR("Unable to populate syscall table. "
5999 "Syscall tracing won't work "
6000 "for this session daemon.");
6005 /* Set ulimit for open files */
6008 /* init lttng_fd tracking must be done after set_ulimit. */
6011 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
6013 goto exit_init_data
;
6016 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
6018 goto exit_init_data
;
6021 /* Setup the needed unix socket */
6022 if (init_daemon_socket()) {
6024 goto exit_init_data
;
6027 /* Set credentials to socket */
6028 if (is_root
&& set_permissions(rundir
)) {
6030 goto exit_init_data
;
6033 /* Get parent pid if -S, --sig-parent is specified. */
6034 if (opt_sig_parent
) {
6038 /* Setup the kernel pipe for waking up the kernel thread */
6039 if (is_root
&& !opt_no_kernel
) {
6040 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
6042 goto exit_init_data
;
6046 /* Setup the thread apps communication pipe. */
6047 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
6049 goto exit_init_data
;
6052 /* Setup the thread apps notify communication pipe. */
6053 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
6055 goto exit_init_data
;
6058 /* Initialize global buffer per UID and PID registry. */
6059 buffer_reg_init_uid_registry();
6060 buffer_reg_init_pid_registry();
6062 /* Init UST command queue. */
6063 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
6066 * Get session list pointer. This pointer MUST NOT be free'd. This list
6067 * is statically declared in session.c
6069 session_list_ptr
= session_get_list();
6073 /* Check for the application socket timeout env variable. */
6074 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
6075 if (env_app_timeout
) {
6076 app_socket_timeout
= atoi(env_app_timeout
);
6078 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
6081 ret
= write_pidfile();
6083 ERR("Error in write_pidfile");
6085 goto exit_init_data
;
6087 ret
= write_agent_port();
6089 ERR("Error in write_agent_port");
6091 goto exit_init_data
;
6094 /* Initialize communication library */
6096 /* Initialize TCP timeout values */
6097 lttcomm_inet_init();
6099 if (load_session_init_data(&load_info
) < 0) {
6101 goto exit_init_data
;
6103 load_info
->path
= opt_load_session_path
;
6105 /* Create health-check thread. */
6106 ret
= pthread_create(&health_thread
, default_pthread_attr(),
6107 thread_manage_health
, (void *) NULL
);
6110 PERROR("pthread_create health");
6115 /* notification_thread_data acquires the pipes' read side. */
6116 notification_thread_handle
= notification_thread_handle_create(
6117 ust32_channel_monitor_pipe
,
6118 ust64_channel_monitor_pipe
,
6119 kernel_channel_monitor_pipe
);
6120 if (!notification_thread_handle
) {
6122 ERR("Failed to create notification thread shared data");
6124 goto exit_notification
;
6127 /* Create notification thread. */
6128 ret
= pthread_create(¬ification_thread
, default_pthread_attr(),
6129 thread_notification
, notification_thread_handle
);
6132 PERROR("pthread_create notification");
6135 goto exit_notification
;
6137 notification_thread_running
= true;
6139 /* Create thread to manage the client socket */
6140 ret
= pthread_create(&client_thread
, default_pthread_attr(),
6141 thread_manage_clients
, (void *) NULL
);
6144 PERROR("pthread_create clients");
6150 /* Create thread to dispatch registration */
6151 ret
= pthread_create(&dispatch_thread
, default_pthread_attr(),
6152 thread_dispatch_ust_registration
, (void *) NULL
);
6155 PERROR("pthread_create dispatch");
6161 /* Create thread to manage application registration. */
6162 ret
= pthread_create(®_apps_thread
, default_pthread_attr(),
6163 thread_registration_apps
, (void *) NULL
);
6166 PERROR("pthread_create registration");
6172 /* Create thread to manage application socket */
6173 ret
= pthread_create(&apps_thread
, default_pthread_attr(),
6174 thread_manage_apps
, (void *) NULL
);
6177 PERROR("pthread_create apps");
6183 /* Create thread to manage application notify socket */
6184 ret
= pthread_create(&apps_notify_thread
, default_pthread_attr(),
6185 ust_thread_manage_notify
, (void *) NULL
);
6188 PERROR("pthread_create notify");
6191 goto exit_apps_notify
;
6194 /* Create agent registration thread. */
6195 ret
= pthread_create(&agent_reg_thread
, default_pthread_attr(),
6196 agent_thread_manage_registration
, (void *) NULL
);
6199 PERROR("pthread_create agent");
6202 goto exit_agent_reg
;
6205 /* Don't start this thread if kernel tracing is not requested nor root */
6206 if (is_root
&& !opt_no_kernel
) {
6207 /* Create kernel thread to manage kernel event */
6208 ret
= pthread_create(&kernel_thread
, default_pthread_attr(),
6209 thread_manage_kernel
, (void *) NULL
);
6212 PERROR("pthread_create kernel");
6219 /* Create session loading thread. */
6220 ret
= pthread_create(&load_session_thread
, default_pthread_attr(),
6221 thread_load_session
, load_info
);
6224 PERROR("pthread_create load_session_thread");
6227 goto exit_load_session
;
6231 * This is where we start awaiting program completion (e.g. through
6232 * signal that asks threads to teardown).
6235 ret
= pthread_join(load_session_thread
, &status
);
6238 PERROR("pthread_join load_session_thread");
6243 if (is_root
&& !opt_no_kernel
) {
6244 ret
= pthread_join(kernel_thread
, &status
);
6247 PERROR("pthread_join");
6253 ret
= pthread_join(agent_reg_thread
, &status
);
6256 PERROR("pthread_join agent");
6261 ret
= pthread_join(apps_notify_thread
, &status
);
6264 PERROR("pthread_join apps notify");
6269 ret
= pthread_join(apps_thread
, &status
);
6272 PERROR("pthread_join apps");
6277 ret
= pthread_join(reg_apps_thread
, &status
);
6280 PERROR("pthread_join");
6286 * Join dispatch thread after joining reg_apps_thread to ensure
6287 * we don't leak applications in the queue.
6289 ret
= pthread_join(dispatch_thread
, &status
);
6292 PERROR("pthread_join");
6297 ret
= pthread_join(client_thread
, &status
);
6300 PERROR("pthread_join");
6306 ret
= pthread_join(health_thread
, &status
);
6309 PERROR("pthread_join health thread");
6316 * Wait for all pending call_rcu work to complete before tearing
6317 * down data structures. call_rcu worker may be trying to
6318 * perform lookups in those structures.
6322 * sessiond_cleanup() is called when no other thread is running, except
6323 * the ht_cleanup thread, which is needed to destroy the hash tables.
6325 rcu_thread_online();
6329 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6330 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6331 * the queue is empty before shutting down the clean-up thread.
6336 * The teardown of the notification system is performed after the
6337 * session daemon's teardown in order to allow it to be notified
6338 * of the active session and channels at the moment of the teardown.
6340 if (notification_thread_handle
) {
6341 if (notification_thread_running
) {
6342 notification_thread_command_quit(
6343 notification_thread_handle
);
6344 ret
= pthread_join(notification_thread
, &status
);
6347 PERROR("pthread_join notification thread");
6351 notification_thread_handle_destroy(notification_thread_handle
);
6354 rcu_thread_offline();
6355 rcu_unregister_thread();
6357 ret
= fini_ht_cleanup_thread(&ht_cleanup_thread
);
6361 lttng_pipe_destroy(ust32_channel_monitor_pipe
);
6362 lttng_pipe_destroy(ust64_channel_monitor_pipe
);
6363 lttng_pipe_destroy(kernel_channel_monitor_pipe
);
6366 health_app_destroy(health_sessiond
);
6367 exit_health_sessiond_cleanup
:
6368 exit_create_run_as_worker_cleanup
:
6371 sessiond_cleanup_options();
6373 exit_set_signal_handler
: