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
)
604 if (!WIFEXITED(status
)) {
605 ERR("consumerd termination with error: %d",
608 consumer_data
->pid
= 0;
612 * Cleanup the session daemon's data structures.
614 static void sessiond_cleanup(void)
617 struct ltt_session
*sess
, *stmp
;
620 DBG("Cleanup sessiond");
623 * Close the thread quit pipe. It has already done its job,
624 * since we are now called.
626 utils_close_pipe(thread_quit_pipe
);
629 * If opt_pidfile is undefined, the default file will be wiped when
630 * removing the rundir.
633 ret
= remove(opt_pidfile
);
635 PERROR("remove pidfile %s", opt_pidfile
);
639 DBG("Removing sessiond and consumerd content of directory %s", rundir
);
642 snprintf(path
, PATH_MAX
,
644 rundir
, DEFAULT_LTTNG_SESSIOND_PIDFILE
);
645 DBG("Removing %s", path
);
648 snprintf(path
, PATH_MAX
, "%s/%s", rundir
,
649 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
);
650 DBG("Removing %s", path
);
654 snprintf(path
, PATH_MAX
,
655 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
,
657 DBG("Removing %s", path
);
660 snprintf(path
, PATH_MAX
,
661 DEFAULT_KCONSUMERD_PATH
,
663 DBG("Removing directory %s", path
);
666 /* ust consumerd 32 */
667 snprintf(path
, PATH_MAX
,
668 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
,
670 DBG("Removing %s", path
);
673 snprintf(path
, PATH_MAX
,
674 DEFAULT_USTCONSUMERD32_PATH
,
676 DBG("Removing directory %s", path
);
679 /* ust consumerd 64 */
680 snprintf(path
, PATH_MAX
,
681 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
,
683 DBG("Removing %s", path
);
686 snprintf(path
, PATH_MAX
,
687 DEFAULT_USTCONSUMERD64_PATH
,
689 DBG("Removing directory %s", path
);
692 DBG("Cleaning up all sessions");
694 /* Destroy session list mutex */
695 if (session_list_ptr
!= NULL
) {
696 pthread_mutex_destroy(&session_list_ptr
->lock
);
698 /* Cleanup ALL session */
699 cds_list_for_each_entry_safe(sess
, stmp
,
700 &session_list_ptr
->head
, list
) {
701 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
705 wait_consumer(&kconsumer_data
);
706 wait_consumer(&ustconsumer64_data
);
707 wait_consumer(&ustconsumer32_data
);
709 DBG("Cleaning up all agent apps");
710 agent_app_ht_clean();
712 DBG("Closing all UST sockets");
713 ust_app_clean_list();
714 buffer_reg_destroy_registries();
716 if (is_root
&& !opt_no_kernel
) {
717 DBG2("Closing kernel fd");
718 if (kernel_tracer_fd
>= 0) {
719 ret
= close(kernel_tracer_fd
);
724 DBG("Unloading kernel modules");
725 modprobe_remove_lttng_all();
729 close_consumer_sockets();
732 load_session_destroy_data(load_info
);
737 * Cleanup lock file by deleting it and finaly closing it which will
738 * release the file system lock.
740 if (lockfile_fd
>= 0) {
741 char lockfile_path
[PATH_MAX
];
743 ret
= generate_lock_file_path(lockfile_path
,
744 sizeof(lockfile_path
));
746 ret
= remove(lockfile_path
);
748 PERROR("remove lock file");
750 ret
= close(lockfile_fd
);
752 PERROR("close lock file");
758 * We do NOT rmdir rundir because there are other processes
759 * using it, for instance lttng-relayd, which can start in
760 * parallel with this teardown.
767 * Cleanup the daemon's option data structures.
769 static void sessiond_cleanup_options(void)
771 DBG("Cleaning up options");
774 * If the override option is set, the pointer points to a *non* const
775 * thus freeing it even though the variable type is set to const.
777 if (tracing_group_name_override
) {
778 free((void *) tracing_group_name
);
780 if (consumerd32_bin_override
) {
781 free((void *) consumerd32_bin
);
783 if (consumerd64_bin_override
) {
784 free((void *) consumerd64_bin
);
786 if (consumerd32_libdir_override
) {
787 free((void *) consumerd32_libdir
);
789 if (consumerd64_libdir_override
) {
790 free((void *) consumerd64_libdir
);
794 free(opt_load_session_path
);
795 free(kmod_probes_list
);
796 free(kmod_extra_probes_list
);
798 run_as_destroy_worker();
802 * Send data on a unix socket using the liblttsessiondcomm API.
804 * Return lttcomm error code.
806 static int send_unix_sock(int sock
, void *buf
, size_t len
)
808 /* Check valid length */
813 return lttcomm_send_unix_sock(sock
, buf
, len
);
817 * Free memory of a command context structure.
819 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
821 DBG("Clean command context structure");
823 if ((*cmd_ctx
)->llm
) {
824 free((*cmd_ctx
)->llm
);
826 if ((*cmd_ctx
)->lsm
) {
827 free((*cmd_ctx
)->lsm
);
835 * Notify UST applications using the shm mmap futex.
837 static int notify_ust_apps(int active
)
841 DBG("Notifying applications of session daemon state: %d", active
);
843 /* See shm.c for this call implying mmap, shm and futex calls */
844 wait_shm_mmap
= shm_ust_get_mmap(wait_shm_path
, is_root
);
845 if (wait_shm_mmap
== NULL
) {
849 /* Wake waiting process */
850 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
852 /* Apps notified successfully */
860 * Setup the outgoing data buffer for the response (llm) by allocating the
861 * right amount of memory and copying the original information from the lsm
864 * Return 0 on success, negative value on error.
866 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
,
867 const void *payload_buf
, size_t payload_len
,
868 const void *cmd_header_buf
, size_t cmd_header_len
)
871 const size_t header_len
= sizeof(struct lttcomm_lttng_msg
);
872 const size_t cmd_header_offset
= header_len
;
873 const size_t payload_offset
= cmd_header_offset
+ cmd_header_len
;
874 const size_t total_msg_size
= header_len
+ cmd_header_len
+ payload_len
;
876 cmd_ctx
->llm
= zmalloc(total_msg_size
);
878 if (cmd_ctx
->llm
== NULL
) {
884 /* Copy common data */
885 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
886 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
887 cmd_ctx
->llm
->cmd_header_size
= cmd_header_len
;
888 cmd_ctx
->llm
->data_size
= payload_len
;
889 cmd_ctx
->lttng_msg_size
= total_msg_size
;
891 /* Copy command header */
892 if (cmd_header_len
) {
893 memcpy(((uint8_t *) cmd_ctx
->llm
) + cmd_header_offset
, cmd_header_buf
,
899 memcpy(((uint8_t *) cmd_ctx
->llm
) + payload_offset
, payload_buf
,
908 * Version of setup_lttng_msg() without command header.
910 static int setup_lttng_msg_no_cmd_header(struct command_ctx
*cmd_ctx
,
911 void *payload_buf
, size_t payload_len
)
913 return setup_lttng_msg(cmd_ctx
, payload_buf
, payload_len
, NULL
, 0);
916 * Update the kernel poll set of all channel fd available over all tracing
917 * session. Add the wakeup pipe at the end of the set.
919 static int update_kernel_poll(struct lttng_poll_event
*events
)
922 struct ltt_session
*session
;
923 struct ltt_kernel_channel
*channel
;
925 DBG("Updating kernel poll set");
928 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
929 session_lock(session
);
930 if (session
->kernel_session
== NULL
) {
931 session_unlock(session
);
935 cds_list_for_each_entry(channel
,
936 &session
->kernel_session
->channel_list
.head
, list
) {
937 /* Add channel fd to the kernel poll set */
938 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
940 session_unlock(session
);
943 DBG("Channel fd %d added to kernel set", channel
->fd
);
945 session_unlock(session
);
947 session_unlock_list();
952 session_unlock_list();
957 * Find the channel fd from 'fd' over all tracing session. When found, check
958 * for new channel stream and send those stream fds to the kernel consumer.
960 * Useful for CPU hotplug feature.
962 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
965 struct ltt_session
*session
;
966 struct ltt_kernel_session
*ksess
;
967 struct ltt_kernel_channel
*channel
;
969 DBG("Updating kernel streams for channel fd %d", fd
);
972 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
973 session_lock(session
);
974 if (session
->kernel_session
== NULL
) {
975 session_unlock(session
);
978 ksess
= session
->kernel_session
;
980 cds_list_for_each_entry(channel
,
981 &ksess
->channel_list
.head
, list
) {
982 struct lttng_ht_iter iter
;
983 struct consumer_socket
*socket
;
985 if (channel
->fd
!= fd
) {
988 DBG("Channel found, updating kernel streams");
989 ret
= kernel_open_channel_stream(channel
);
993 /* Update the stream global counter */
994 ksess
->stream_count_global
+= ret
;
997 * Have we already sent fds to the consumer? If yes, it
998 * means that tracing is started so it is safe to send
999 * our updated stream fds.
1001 if (ksess
->consumer_fds_sent
!= 1
1002 || ksess
->consumer
== NULL
) {
1008 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
1009 &iter
.iter
, socket
, node
.node
) {
1010 pthread_mutex_lock(socket
->lock
);
1011 ret
= kernel_consumer_send_channel_stream(socket
,
1013 session
->output_traces
? 1 : 0);
1014 pthread_mutex_unlock(socket
->lock
);
1022 session_unlock(session
);
1024 session_unlock_list();
1028 session_unlock(session
);
1029 session_unlock_list();
1034 * For each tracing session, update newly registered apps. The session list
1035 * lock MUST be acquired before calling this.
1037 static void update_ust_app(int app_sock
)
1039 struct ltt_session
*sess
, *stmp
;
1041 /* Consumer is in an ERROR state. Stop any application update. */
1042 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
1043 /* Stop the update process since the consumer is dead. */
1047 /* For all tracing session(s) */
1048 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
1049 struct ust_app
*app
;
1052 if (!sess
->ust_session
) {
1053 goto unlock_session
;
1057 assert(app_sock
>= 0);
1058 app
= ust_app_find_by_sock(app_sock
);
1061 * Application can be unregistered before so
1062 * this is possible hence simply stopping the
1065 DBG3("UST app update failed to find app sock %d",
1069 ust_app_global_update(sess
->ust_session
, app
);
1073 session_unlock(sess
);
1078 * This thread manage event coming from the kernel.
1080 * Features supported in this thread:
1083 static void *thread_manage_kernel(void *data
)
1085 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
1086 uint32_t revents
, nb_fd
;
1088 struct lttng_poll_event events
;
1090 DBG("[thread] Thread manage kernel started");
1092 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
1095 * This first step of the while is to clean this structure which could free
1096 * non NULL pointers so initialize it before the loop.
1098 lttng_poll_init(&events
);
1100 if (testpoint(sessiond_thread_manage_kernel
)) {
1101 goto error_testpoint
;
1104 health_code_update();
1106 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
1107 goto error_testpoint
;
1111 health_code_update();
1113 if (update_poll_flag
== 1) {
1114 /* Clean events object. We are about to populate it again. */
1115 lttng_poll_clean(&events
);
1117 ret
= sessiond_set_thread_pollset(&events
, 2);
1119 goto error_poll_create
;
1122 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
1127 /* This will add the available kernel channel if any. */
1128 ret
= update_kernel_poll(&events
);
1132 update_poll_flag
= 0;
1135 DBG("Thread kernel polling");
1137 /* Poll infinite value of time */
1139 health_poll_entry();
1140 ret
= lttng_poll_wait(&events
, -1);
1141 DBG("Thread kernel return from poll on %d fds",
1142 LTTNG_POLL_GETNB(&events
));
1146 * Restart interrupted system call.
1148 if (errno
== EINTR
) {
1152 } else if (ret
== 0) {
1153 /* Should not happen since timeout is infinite */
1154 ERR("Return value of poll is 0 with an infinite timeout.\n"
1155 "This should not have happened! Continuing...");
1161 for (i
= 0; i
< nb_fd
; i
++) {
1162 /* Fetch once the poll data */
1163 revents
= LTTNG_POLL_GETEV(&events
, i
);
1164 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1166 health_code_update();
1169 /* No activity for this FD (poll implementation). */
1173 /* Thread quit pipe has been closed. Killing thread. */
1174 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1180 /* Check for data on kernel pipe */
1181 if (revents
& LPOLLIN
) {
1182 if (pollfd
== kernel_poll_pipe
[0]) {
1183 (void) lttng_read(kernel_poll_pipe
[0],
1186 * Ret value is useless here, if this pipe gets any actions an
1187 * update is required anyway.
1189 update_poll_flag
= 1;
1193 * New CPU detected by the kernel. Adding kernel stream to
1194 * kernel session and updating the kernel consumer
1196 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1202 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1203 update_poll_flag
= 1;
1206 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1214 lttng_poll_clean(&events
);
1217 utils_close_pipe(kernel_poll_pipe
);
1218 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1221 ERR("Health error occurred in %s", __func__
);
1222 WARN("Kernel thread died unexpectedly. "
1223 "Kernel tracing can continue but CPU hotplug is disabled.");
1225 health_unregister(health_sessiond
);
1226 DBG("Kernel thread dying");
1231 * Signal pthread condition of the consumer data that the thread.
1233 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1235 pthread_mutex_lock(&data
->cond_mutex
);
1238 * The state is set before signaling. It can be any value, it's the waiter
1239 * job to correctly interpret this condition variable associated to the
1240 * consumer pthread_cond.
1242 * A value of 0 means that the corresponding thread of the consumer data
1243 * was not started. 1 indicates that the thread has started and is ready
1244 * for action. A negative value means that there was an error during the
1247 data
->consumer_thread_is_ready
= state
;
1248 (void) pthread_cond_signal(&data
->cond
);
1250 pthread_mutex_unlock(&data
->cond_mutex
);
1254 * This thread manage the consumer error sent back to the session daemon.
1256 static void *thread_manage_consumer(void *data
)
1258 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1259 uint32_t revents
, nb_fd
;
1260 enum lttcomm_return_code code
;
1261 struct lttng_poll_event events
;
1262 struct consumer_data
*consumer_data
= data
;
1263 struct consumer_socket
*cmd_socket_wrapper
= NULL
;
1265 DBG("[thread] Manage consumer started");
1267 rcu_register_thread();
1268 rcu_thread_online();
1270 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1272 health_code_update();
1275 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1276 * metadata_sock. Nothing more will be added to this poll set.
1278 ret
= sessiond_set_thread_pollset(&events
, 3);
1284 * The error socket here is already in a listening state which was done
1285 * just before spawning this thread to avoid a race between the consumer
1286 * daemon exec trying to connect and the listen() call.
1288 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1293 health_code_update();
1295 /* Infinite blocking call, waiting for transmission */
1297 health_poll_entry();
1299 if (testpoint(sessiond_thread_manage_consumer
)) {
1303 ret
= lttng_poll_wait(&events
, -1);
1307 * Restart interrupted system call.
1309 if (errno
== EINTR
) {
1317 for (i
= 0; i
< nb_fd
; i
++) {
1318 /* Fetch once the poll data */
1319 revents
= LTTNG_POLL_GETEV(&events
, i
);
1320 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1322 health_code_update();
1325 /* No activity for this FD (poll implementation). */
1329 /* Thread quit pipe has been closed. Killing thread. */
1330 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1336 /* Event on the registration socket */
1337 if (pollfd
== consumer_data
->err_sock
) {
1338 if (revents
& LPOLLIN
) {
1340 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1341 ERR("consumer err socket poll error");
1344 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1350 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1356 * Set the CLOEXEC flag. Return code is useless because either way, the
1359 (void) utils_set_fd_cloexec(sock
);
1361 health_code_update();
1363 DBG2("Receiving code from consumer err_sock");
1365 /* Getting status code from kconsumerd */
1366 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1367 sizeof(enum lttcomm_return_code
));
1372 health_code_update();
1373 if (code
!= LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1374 ERR("consumer error when waiting for SOCK_READY : %s",
1375 lttcomm_get_readable_code(-code
));
1379 /* Connect both command and metadata sockets. */
1380 consumer_data
->cmd_sock
=
1381 lttcomm_connect_unix_sock(
1382 consumer_data
->cmd_unix_sock_path
);
1383 consumer_data
->metadata_fd
=
1384 lttcomm_connect_unix_sock(
1385 consumer_data
->cmd_unix_sock_path
);
1386 if (consumer_data
->cmd_sock
< 0 || consumer_data
->metadata_fd
< 0) {
1387 PERROR("consumer connect cmd socket");
1388 /* On error, signal condition and quit. */
1389 signal_consumer_condition(consumer_data
, -1);
1393 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1395 /* Create metadata socket lock. */
1396 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1397 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1398 PERROR("zmalloc pthread mutex");
1401 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1403 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1404 DBG("Consumer metadata socket ready (fd: %d)",
1405 consumer_data
->metadata_fd
);
1408 * Remove the consumerd error sock since we've established a connection.
1410 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1415 /* Add new accepted error socket. */
1416 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1421 /* Add metadata socket that is successfully connected. */
1422 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1423 LPOLLIN
| LPOLLRDHUP
);
1428 health_code_update();
1431 * Transfer the write-end of the channel monitoring pipe to the
1432 * by issuing a SET_CHANNEL_MONITOR_PIPE command.
1434 cmd_socket_wrapper
= consumer_allocate_socket(&consumer_data
->cmd_sock
);
1435 if (!cmd_socket_wrapper
) {
1439 ret
= consumer_send_channel_monitor_pipe(cmd_socket_wrapper
,
1440 consumer_data
->channel_monitor_pipe
);
1444 /* Discard the socket wrapper as it is no longer needed. */
1445 consumer_destroy_socket(cmd_socket_wrapper
);
1446 cmd_socket_wrapper
= NULL
;
1448 /* The thread is completely initialized, signal that it is ready. */
1449 signal_consumer_condition(consumer_data
, 1);
1451 /* Infinite blocking call, waiting for transmission */
1454 health_code_update();
1456 /* Exit the thread because the thread quit pipe has been triggered. */
1458 /* Not a health error. */
1463 health_poll_entry();
1464 ret
= lttng_poll_wait(&events
, -1);
1468 * Restart interrupted system call.
1470 if (errno
== EINTR
) {
1478 for (i
= 0; i
< nb_fd
; i
++) {
1479 /* Fetch once the poll data */
1480 revents
= LTTNG_POLL_GETEV(&events
, i
);
1481 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1483 health_code_update();
1486 /* No activity for this FD (poll implementation). */
1491 * Thread quit pipe has been triggered, flag that we should stop
1492 * but continue the current loop to handle potential data from
1495 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1497 if (pollfd
== sock
) {
1498 /* Event on the consumerd socket */
1499 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1500 && !(revents
& LPOLLIN
)) {
1501 ERR("consumer err socket second poll error");
1504 health_code_update();
1505 /* Wait for any kconsumerd error */
1506 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1507 sizeof(enum lttcomm_return_code
));
1509 ERR("consumer closed the command socket");
1513 ERR("consumer return code : %s",
1514 lttcomm_get_readable_code(-code
));
1517 } else if (pollfd
== consumer_data
->metadata_fd
) {
1518 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1519 && !(revents
& LPOLLIN
)) {
1520 ERR("consumer err metadata socket second poll error");
1523 /* UST metadata requests */
1524 ret
= ust_consumer_metadata_request(
1525 &consumer_data
->metadata_sock
);
1527 ERR("Handling metadata request");
1531 /* No need for an else branch all FDs are tested prior. */
1533 health_code_update();
1539 * We lock here because we are about to close the sockets and some other
1540 * thread might be using them so get exclusive access which will abort all
1541 * other consumer command by other threads.
1543 pthread_mutex_lock(&consumer_data
->lock
);
1545 /* Immediately set the consumerd state to stopped */
1546 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1547 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1548 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1549 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1550 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1552 /* Code flow error... */
1556 if (consumer_data
->err_sock
>= 0) {
1557 ret
= close(consumer_data
->err_sock
);
1561 consumer_data
->err_sock
= -1;
1563 if (consumer_data
->cmd_sock
>= 0) {
1564 ret
= close(consumer_data
->cmd_sock
);
1568 consumer_data
->cmd_sock
= -1;
1570 if (consumer_data
->metadata_sock
.fd_ptr
&&
1571 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1572 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1584 unlink(consumer_data
->err_unix_sock_path
);
1585 unlink(consumer_data
->cmd_unix_sock_path
);
1586 pthread_mutex_unlock(&consumer_data
->lock
);
1588 /* Cleanup metadata socket mutex. */
1589 if (consumer_data
->metadata_sock
.lock
) {
1590 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1591 free(consumer_data
->metadata_sock
.lock
);
1593 lttng_poll_clean(&events
);
1595 if (cmd_socket_wrapper
) {
1596 consumer_destroy_socket(cmd_socket_wrapper
);
1601 ERR("Health error occurred in %s", __func__
);
1603 health_unregister(health_sessiond
);
1604 DBG("consumer thread cleanup completed");
1606 rcu_thread_offline();
1607 rcu_unregister_thread();
1613 * This thread manage application communication.
1615 static void *thread_manage_apps(void *data
)
1617 int i
, ret
, pollfd
, err
= -1;
1619 uint32_t revents
, nb_fd
;
1620 struct lttng_poll_event events
;
1622 DBG("[thread] Manage application started");
1624 rcu_register_thread();
1625 rcu_thread_online();
1627 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1629 if (testpoint(sessiond_thread_manage_apps
)) {
1630 goto error_testpoint
;
1633 health_code_update();
1635 ret
= sessiond_set_thread_pollset(&events
, 2);
1637 goto error_poll_create
;
1640 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1645 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1649 health_code_update();
1652 DBG("Apps thread polling");
1654 /* Inifinite blocking call, waiting for transmission */
1656 health_poll_entry();
1657 ret
= lttng_poll_wait(&events
, -1);
1658 DBG("Apps thread return from poll on %d fds",
1659 LTTNG_POLL_GETNB(&events
));
1663 * Restart interrupted system call.
1665 if (errno
== EINTR
) {
1673 for (i
= 0; i
< nb_fd
; i
++) {
1674 /* Fetch once the poll data */
1675 revents
= LTTNG_POLL_GETEV(&events
, i
);
1676 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1678 health_code_update();
1681 /* No activity for this FD (poll implementation). */
1685 /* Thread quit pipe has been closed. Killing thread. */
1686 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1692 /* Inspect the apps cmd pipe */
1693 if (pollfd
== apps_cmd_pipe
[0]) {
1694 if (revents
& LPOLLIN
) {
1698 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1699 if (size_ret
< sizeof(sock
)) {
1700 PERROR("read apps cmd pipe");
1704 health_code_update();
1707 * Since this is a command socket (write then read),
1708 * we only monitor the error events of the socket.
1710 ret
= lttng_poll_add(&events
, sock
,
1711 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1716 DBG("Apps with sock %d added to poll set", sock
);
1717 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1718 ERR("Apps command pipe error");
1721 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1726 * At this point, we know that a registered application made
1727 * the event at poll_wait.
1729 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1730 /* Removing from the poll set */
1731 ret
= lttng_poll_del(&events
, pollfd
);
1736 /* Socket closed on remote end. */
1737 ust_app_unregister(pollfd
);
1739 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1744 health_code_update();
1750 lttng_poll_clean(&events
);
1753 utils_close_pipe(apps_cmd_pipe
);
1754 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1757 * We don't clean the UST app hash table here since already registered
1758 * applications can still be controlled so let them be until the session
1759 * daemon dies or the applications stop.
1764 ERR("Health error occurred in %s", __func__
);
1766 health_unregister(health_sessiond
);
1767 DBG("Application communication apps thread cleanup complete");
1768 rcu_thread_offline();
1769 rcu_unregister_thread();
1774 * Send a socket to a thread This is called from the dispatch UST registration
1775 * thread once all sockets are set for the application.
1777 * The sock value can be invalid, we don't really care, the thread will handle
1778 * it and make the necessary cleanup if so.
1780 * On success, return 0 else a negative value being the errno message of the
1783 static int send_socket_to_thread(int fd
, int sock
)
1788 * It's possible that the FD is set as invalid with -1 concurrently just
1789 * before calling this function being a shutdown state of the thread.
1796 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1797 if (ret
< sizeof(sock
)) {
1798 PERROR("write apps pipe %d", fd
);
1805 /* All good. Don't send back the write positive ret value. */
1812 * Sanitize the wait queue of the dispatch registration thread meaning removing
1813 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1814 * notify socket is never received.
1816 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1818 int ret
, nb_fd
= 0, i
;
1819 unsigned int fd_added
= 0;
1820 struct lttng_poll_event events
;
1821 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1825 lttng_poll_init(&events
);
1827 /* Just skip everything for an empty queue. */
1828 if (!wait_queue
->count
) {
1832 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1837 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1838 &wait_queue
->head
, head
) {
1839 assert(wait_node
->app
);
1840 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1841 LPOLLHUP
| LPOLLERR
);
1854 * Poll but don't block so we can quickly identify the faulty events and
1855 * clean them afterwards from the wait queue.
1857 ret
= lttng_poll_wait(&events
, 0);
1863 for (i
= 0; i
< nb_fd
; i
++) {
1864 /* Get faulty FD. */
1865 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1866 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1869 /* No activity for this FD (poll implementation). */
1873 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1874 &wait_queue
->head
, head
) {
1875 if (pollfd
== wait_node
->app
->sock
&&
1876 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1877 cds_list_del(&wait_node
->head
);
1878 wait_queue
->count
--;
1879 ust_app_destroy(wait_node
->app
);
1882 * Silence warning of use-after-free in
1883 * cds_list_for_each_entry_safe which uses
1884 * __typeof__(*wait_node).
1889 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1896 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1900 lttng_poll_clean(&events
);
1904 lttng_poll_clean(&events
);
1906 ERR("Unable to sanitize wait queue");
1911 * Dispatch request from the registration threads to the application
1912 * communication thread.
1914 static void *thread_dispatch_ust_registration(void *data
)
1917 struct cds_wfcq_node
*node
;
1918 struct ust_command
*ust_cmd
= NULL
;
1919 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1920 struct ust_reg_wait_queue wait_queue
= {
1924 rcu_register_thread();
1926 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1928 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1929 goto error_testpoint
;
1932 health_code_update();
1934 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1936 DBG("[thread] Dispatch UST command started");
1938 while (!CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1939 health_code_update();
1941 /* Atomically prepare the queue futex */
1942 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1945 struct ust_app
*app
= NULL
;
1949 * Make sure we don't have node(s) that have hung up before receiving
1950 * the notify socket. This is to clean the list in order to avoid
1951 * memory leaks from notify socket that are never seen.
1953 sanitize_wait_queue(&wait_queue
);
1955 health_code_update();
1956 /* Dequeue command for registration */
1957 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1959 DBG("Woken up but nothing in the UST command queue");
1960 /* Continue thread execution */
1964 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1966 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1967 " gid:%d sock:%d name:%s (version %d.%d)",
1968 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1969 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1970 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1971 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1973 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1974 wait_node
= zmalloc(sizeof(*wait_node
));
1976 PERROR("zmalloc wait_node dispatch");
1977 ret
= close(ust_cmd
->sock
);
1979 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1981 lttng_fd_put(LTTNG_FD_APPS
, 1);
1985 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1987 /* Create application object if socket is CMD. */
1988 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1990 if (!wait_node
->app
) {
1991 ret
= close(ust_cmd
->sock
);
1993 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1995 lttng_fd_put(LTTNG_FD_APPS
, 1);
2001 * Add application to the wait queue so we can set the notify
2002 * socket before putting this object in the global ht.
2004 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
2009 * We have to continue here since we don't have the notify
2010 * socket and the application MUST be added to the hash table
2011 * only at that moment.
2016 * Look for the application in the local wait queue and set the
2017 * notify socket if found.
2019 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2020 &wait_queue
.head
, head
) {
2021 health_code_update();
2022 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
2023 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
2024 cds_list_del(&wait_node
->head
);
2026 app
= wait_node
->app
;
2028 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
2034 * With no application at this stage the received socket is
2035 * basically useless so close it before we free the cmd data
2036 * structure for good.
2039 ret
= close(ust_cmd
->sock
);
2041 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
2043 lttng_fd_put(LTTNG_FD_APPS
, 1);
2050 * @session_lock_list
2052 * Lock the global session list so from the register up to the
2053 * registration done message, no thread can see the application
2054 * and change its state.
2056 session_lock_list();
2060 * Add application to the global hash table. This needs to be
2061 * done before the update to the UST registry can locate the
2066 /* Set app version. This call will print an error if needed. */
2067 (void) ust_app_version(app
);
2069 /* Send notify socket through the notify pipe. */
2070 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
2074 session_unlock_list();
2076 * No notify thread, stop the UST tracing. However, this is
2077 * not an internal error of the this thread thus setting
2078 * the health error code to a normal exit.
2085 * Update newly registered application with the tracing
2086 * registry info already enabled information.
2088 update_ust_app(app
->sock
);
2091 * Don't care about return value. Let the manage apps threads
2092 * handle app unregistration upon socket close.
2094 (void) ust_app_register_done(app
);
2097 * Even if the application socket has been closed, send the app
2098 * to the thread and unregistration will take place at that
2101 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
2104 session_unlock_list();
2106 * No apps. thread, stop the UST tracing. However, this is
2107 * not an internal error of the this thread thus setting
2108 * the health error code to a normal exit.
2115 session_unlock_list();
2117 } while (node
!= NULL
);
2119 health_poll_entry();
2120 /* Futex wait on queue. Blocking call on futex() */
2121 futex_nto1_wait(&ust_cmd_queue
.futex
);
2124 /* Normal exit, no error */
2128 /* Clean up wait queue. */
2129 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2130 &wait_queue
.head
, head
) {
2131 cds_list_del(&wait_node
->head
);
2136 /* Empty command queue. */
2138 /* Dequeue command for registration */
2139 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2143 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2144 ret
= close(ust_cmd
->sock
);
2146 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2148 lttng_fd_put(LTTNG_FD_APPS
, 1);
2153 DBG("Dispatch thread dying");
2156 ERR("Health error occurred in %s", __func__
);
2158 health_unregister(health_sessiond
);
2159 rcu_unregister_thread();
2164 * This thread manage application registration.
2166 static void *thread_registration_apps(void *data
)
2168 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2169 uint32_t revents
, nb_fd
;
2170 struct lttng_poll_event events
;
2172 * Get allocated in this thread, enqueued to a global queue, dequeued and
2173 * freed in the manage apps thread.
2175 struct ust_command
*ust_cmd
= NULL
;
2177 DBG("[thread] Manage application registration started");
2179 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2181 if (testpoint(sessiond_thread_registration_apps
)) {
2182 goto error_testpoint
;
2185 ret
= lttcomm_listen_unix_sock(apps_sock
);
2191 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2192 * more will be added to this poll set.
2194 ret
= sessiond_set_thread_pollset(&events
, 2);
2196 goto error_create_poll
;
2199 /* Add the application registration socket */
2200 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2202 goto error_poll_add
;
2205 /* Notify all applications to register */
2206 ret
= notify_ust_apps(1);
2208 ERR("Failed to notify applications or create the wait shared memory.\n"
2209 "Execution continues but there might be problem for already\n"
2210 "running applications that wishes to register.");
2214 DBG("Accepting application registration");
2216 /* Inifinite blocking call, waiting for transmission */
2218 health_poll_entry();
2219 ret
= lttng_poll_wait(&events
, -1);
2223 * Restart interrupted system call.
2225 if (errno
== EINTR
) {
2233 for (i
= 0; i
< nb_fd
; i
++) {
2234 health_code_update();
2236 /* Fetch once the poll data */
2237 revents
= LTTNG_POLL_GETEV(&events
, i
);
2238 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2241 /* No activity for this FD (poll implementation). */
2245 /* Thread quit pipe has been closed. Killing thread. */
2246 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2252 /* Event on the registration socket */
2253 if (pollfd
== apps_sock
) {
2254 if (revents
& LPOLLIN
) {
2255 sock
= lttcomm_accept_unix_sock(apps_sock
);
2261 * Set socket timeout for both receiving and ending.
2262 * app_socket_timeout is in seconds, whereas
2263 * lttcomm_setsockopt_rcv_timeout and
2264 * lttcomm_setsockopt_snd_timeout expect msec as
2267 if (app_socket_timeout
>= 0) {
2268 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2269 app_socket_timeout
* 1000);
2270 (void) lttcomm_setsockopt_snd_timeout(sock
,
2271 app_socket_timeout
* 1000);
2275 * Set the CLOEXEC flag. Return code is useless because
2276 * either way, the show must go on.
2278 (void) utils_set_fd_cloexec(sock
);
2280 /* Create UST registration command for enqueuing */
2281 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2282 if (ust_cmd
== NULL
) {
2283 PERROR("ust command zmalloc");
2292 * Using message-based transmissions to ensure we don't
2293 * have to deal with partially received messages.
2295 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2297 ERR("Exhausted file descriptors allowed for applications.");
2307 health_code_update();
2308 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2311 /* Close socket of the application. */
2316 lttng_fd_put(LTTNG_FD_APPS
, 1);
2320 health_code_update();
2322 ust_cmd
->sock
= sock
;
2325 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2326 " gid:%d sock:%d name:%s (version %d.%d)",
2327 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2328 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2329 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2330 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2333 * Lock free enqueue the registration request. The red pill
2334 * has been taken! This apps will be part of the *system*.
2336 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2339 * Wake the registration queue futex. Implicit memory
2340 * barrier with the exchange in cds_wfcq_enqueue.
2342 futex_nto1_wake(&ust_cmd_queue
.futex
);
2343 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2344 ERR("Register apps socket poll error");
2347 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2356 /* Notify that the registration thread is gone */
2359 if (apps_sock
>= 0) {
2360 ret
= close(apps_sock
);
2370 lttng_fd_put(LTTNG_FD_APPS
, 1);
2372 unlink(apps_unix_sock_path
);
2375 lttng_poll_clean(&events
);
2379 DBG("UST Registration thread cleanup complete");
2382 ERR("Health error occurred in %s", __func__
);
2384 health_unregister(health_sessiond
);
2390 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2391 * exec or it will fails.
2393 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2396 struct timespec timeout
;
2399 * Make sure we set the readiness flag to 0 because we are NOT ready.
2400 * This access to consumer_thread_is_ready does not need to be
2401 * protected by consumer_data.cond_mutex (yet) since the consumer
2402 * management thread has not been started at this point.
2404 consumer_data
->consumer_thread_is_ready
= 0;
2406 /* Setup pthread condition */
2407 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2410 PERROR("pthread_condattr_init consumer data");
2415 * Set the monotonic clock in order to make sure we DO NOT jump in time
2416 * between the clock_gettime() call and the timedwait call. See bug #324
2417 * for a more details and how we noticed it.
2419 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2422 PERROR("pthread_condattr_setclock consumer data");
2426 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2429 PERROR("pthread_cond_init consumer data");
2433 ret
= pthread_create(&consumer_data
->thread
, default_pthread_attr(),
2434 thread_manage_consumer
, consumer_data
);
2437 PERROR("pthread_create consumer");
2442 /* We are about to wait on a pthread condition */
2443 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2445 /* Get time for sem_timedwait absolute timeout */
2446 clock_ret
= lttng_clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2448 * Set the timeout for the condition timed wait even if the clock gettime
2449 * call fails since we might loop on that call and we want to avoid to
2450 * increment the timeout too many times.
2452 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2455 * The following loop COULD be skipped in some conditions so this is why we
2456 * set ret to 0 in order to make sure at least one round of the loop is
2462 * Loop until the condition is reached or when a timeout is reached. Note
2463 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2464 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2465 * possible. This loop does not take any chances and works with both of
2468 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2469 if (clock_ret
< 0) {
2470 PERROR("clock_gettime spawn consumer");
2471 /* Infinite wait for the consumerd thread to be ready */
2472 ret
= pthread_cond_wait(&consumer_data
->cond
,
2473 &consumer_data
->cond_mutex
);
2475 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2476 &consumer_data
->cond_mutex
, &timeout
);
2480 /* Release the pthread condition */
2481 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2485 if (ret
== ETIMEDOUT
) {
2489 * Call has timed out so we kill the kconsumerd_thread and return
2492 ERR("Condition timed out. The consumer thread was never ready."
2494 pth_ret
= pthread_cancel(consumer_data
->thread
);
2496 PERROR("pthread_cancel consumer thread");
2499 PERROR("pthread_cond_wait failed consumer thread");
2501 /* Caller is expecting a negative value on failure. */
2506 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2507 if (consumer_data
->pid
== 0) {
2508 ERR("Consumerd did not start");
2509 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2512 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2521 * Join consumer thread
2523 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2527 /* Consumer pid must be a real one. */
2528 if (consumer_data
->pid
> 0) {
2530 ret
= kill(consumer_data
->pid
, SIGTERM
);
2532 PERROR("Error killing consumer daemon");
2535 return pthread_join(consumer_data
->thread
, &status
);
2542 * Fork and exec a consumer daemon (consumerd).
2544 * Return pid if successful else -1.
2546 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2550 const char *consumer_to_use
;
2551 const char *verbosity
;
2554 DBG("Spawning consumerd");
2561 if (opt_verbose_consumer
) {
2562 verbosity
= "--verbose";
2563 } else if (lttng_opt_quiet
) {
2564 verbosity
= "--quiet";
2569 switch (consumer_data
->type
) {
2570 case LTTNG_CONSUMER_KERNEL
:
2572 * Find out which consumerd to execute. We will first try the
2573 * 64-bit path, then the sessiond's installation directory, and
2574 * fallback on the 32-bit one,
2576 DBG3("Looking for a kernel consumer at these locations:");
2577 DBG3(" 1) %s", consumerd64_bin
);
2578 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, CONSUMERD_FILE
);
2579 DBG3(" 3) %s", consumerd32_bin
);
2580 if (stat(consumerd64_bin
, &st
) == 0) {
2581 DBG3("Found location #1");
2582 consumer_to_use
= consumerd64_bin
;
2583 } else if (stat(INSTALL_BIN_PATH
"/" CONSUMERD_FILE
, &st
) == 0) {
2584 DBG3("Found location #2");
2585 consumer_to_use
= INSTALL_BIN_PATH
"/" CONSUMERD_FILE
;
2586 } else if (stat(consumerd32_bin
, &st
) == 0) {
2587 DBG3("Found location #3");
2588 consumer_to_use
= consumerd32_bin
;
2590 DBG("Could not find any valid consumerd executable");
2594 DBG("Using kernel consumer at: %s", consumer_to_use
);
2595 ret
= execl(consumer_to_use
,
2596 "lttng-consumerd", verbosity
, "-k",
2597 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2598 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2599 "--group", tracing_group_name
,
2602 case LTTNG_CONSUMER64_UST
:
2604 char *tmpnew
= NULL
;
2606 if (consumerd64_libdir
[0] != '\0') {
2610 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2614 tmplen
= strlen("LD_LIBRARY_PATH=")
2615 + strlen(consumerd64_libdir
) + 1 /* : */ + strlen(tmp
);
2616 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2621 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2622 strcat(tmpnew
, consumerd64_libdir
);
2623 if (tmp
[0] != '\0') {
2624 strcat(tmpnew
, ":");
2625 strcat(tmpnew
, tmp
);
2627 ret
= putenv(tmpnew
);
2634 DBG("Using 64-bit UST consumer at: %s", consumerd64_bin
);
2635 ret
= execl(consumerd64_bin
, "lttng-consumerd", verbosity
, "-u",
2636 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2637 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2638 "--group", tracing_group_name
,
2640 if (consumerd64_libdir
[0] != '\0') {
2645 case LTTNG_CONSUMER32_UST
:
2647 char *tmpnew
= NULL
;
2649 if (consumerd32_libdir
[0] != '\0') {
2653 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2657 tmplen
= strlen("LD_LIBRARY_PATH=")
2658 + strlen(consumerd32_libdir
) + 1 /* : */ + strlen(tmp
);
2659 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2664 strcpy(tmpnew
, "LD_LIBRARY_PATH=");
2665 strcat(tmpnew
, consumerd32_libdir
);
2666 if (tmp
[0] != '\0') {
2667 strcat(tmpnew
, ":");
2668 strcat(tmpnew
, tmp
);
2670 ret
= putenv(tmpnew
);
2677 DBG("Using 32-bit UST consumer at: %s", consumerd32_bin
);
2678 ret
= execl(consumerd32_bin
, "lttng-consumerd", verbosity
, "-u",
2679 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2680 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2681 "--group", tracing_group_name
,
2683 if (consumerd32_libdir
[0] != '\0') {
2689 PERROR("unknown consumer type");
2693 PERROR("Consumer execl()");
2695 /* Reaching this point, we got a failure on our execl(). */
2697 } else if (pid
> 0) {
2700 PERROR("start consumer fork");
2708 * Spawn the consumerd daemon and session daemon thread.
2710 static int start_consumerd(struct consumer_data
*consumer_data
)
2715 * Set the listen() state on the socket since there is a possible race
2716 * between the exec() of the consumer daemon and this call if place in the
2717 * consumer thread. See bug #366 for more details.
2719 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2724 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2725 if (consumer_data
->pid
!= 0) {
2726 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2730 ret
= spawn_consumerd(consumer_data
);
2732 ERR("Spawning consumerd failed");
2733 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2737 /* Setting up the consumer_data pid */
2738 consumer_data
->pid
= ret
;
2739 DBG2("Consumer pid %d", consumer_data
->pid
);
2740 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2742 DBG2("Spawning consumer control thread");
2743 ret
= spawn_consumer_thread(consumer_data
);
2745 ERR("Fatal error spawning consumer control thread");
2753 /* Cleanup already created sockets on error. */
2754 if (consumer_data
->err_sock
>= 0) {
2757 err
= close(consumer_data
->err_sock
);
2759 PERROR("close consumer data error socket");
2766 * Setup necessary data for kernel tracer action.
2768 static int init_kernel_tracer(void)
2772 /* Modprobe lttng kernel modules */
2773 ret
= modprobe_lttng_control();
2778 /* Open debugfs lttng */
2779 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2780 if (kernel_tracer_fd
< 0) {
2781 DBG("Failed to open %s", module_proc_lttng
);
2786 /* Validate kernel version */
2787 ret
= kernel_validate_version(kernel_tracer_fd
);
2792 ret
= modprobe_lttng_data();
2797 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2801 modprobe_remove_lttng_control();
2802 ret
= close(kernel_tracer_fd
);
2806 kernel_tracer_fd
= -1;
2807 return LTTNG_ERR_KERN_VERSION
;
2810 ret
= close(kernel_tracer_fd
);
2816 modprobe_remove_lttng_control();
2819 WARN("No kernel tracer available");
2820 kernel_tracer_fd
= -1;
2822 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2824 return LTTNG_ERR_KERN_NA
;
2830 * Copy consumer output from the tracing session to the domain session. The
2831 * function also applies the right modification on a per domain basis for the
2832 * trace files destination directory.
2834 * Should *NOT* be called with RCU read-side lock held.
2836 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2839 const char *dir_name
;
2840 struct consumer_output
*consumer
;
2843 assert(session
->consumer
);
2846 case LTTNG_DOMAIN_KERNEL
:
2847 DBG3("Copying tracing session consumer output in kernel session");
2849 * XXX: We should audit the session creation and what this function
2850 * does "extra" in order to avoid a destroy since this function is used
2851 * in the domain session creation (kernel and ust) only. Same for UST
2854 if (session
->kernel_session
->consumer
) {
2855 consumer_output_put(session
->kernel_session
->consumer
);
2857 session
->kernel_session
->consumer
=
2858 consumer_copy_output(session
->consumer
);
2859 /* Ease our life a bit for the next part */
2860 consumer
= session
->kernel_session
->consumer
;
2861 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2863 case LTTNG_DOMAIN_JUL
:
2864 case LTTNG_DOMAIN_LOG4J
:
2865 case LTTNG_DOMAIN_PYTHON
:
2866 case LTTNG_DOMAIN_UST
:
2867 DBG3("Copying tracing session consumer output in UST session");
2868 if (session
->ust_session
->consumer
) {
2869 consumer_output_put(session
->ust_session
->consumer
);
2871 session
->ust_session
->consumer
=
2872 consumer_copy_output(session
->consumer
);
2873 /* Ease our life a bit for the next part */
2874 consumer
= session
->ust_session
->consumer
;
2875 dir_name
= DEFAULT_UST_TRACE_DIR
;
2878 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2882 /* Append correct directory to subdir */
2883 strncat(consumer
->subdir
, dir_name
,
2884 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2885 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2894 * Create an UST session and add it to the session ust list.
2896 * Should *NOT* be called with RCU read-side lock held.
2898 static int create_ust_session(struct ltt_session
*session
,
2899 struct lttng_domain
*domain
)
2902 struct ltt_ust_session
*lus
= NULL
;
2906 assert(session
->consumer
);
2908 switch (domain
->type
) {
2909 case LTTNG_DOMAIN_JUL
:
2910 case LTTNG_DOMAIN_LOG4J
:
2911 case LTTNG_DOMAIN_PYTHON
:
2912 case LTTNG_DOMAIN_UST
:
2915 ERR("Unknown UST domain on create session %d", domain
->type
);
2916 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2920 DBG("Creating UST session");
2922 lus
= trace_ust_create_session(session
->id
);
2924 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2928 lus
->uid
= session
->uid
;
2929 lus
->gid
= session
->gid
;
2930 lus
->output_traces
= session
->output_traces
;
2931 lus
->snapshot_mode
= session
->snapshot_mode
;
2932 lus
->live_timer_interval
= session
->live_timer
;
2933 session
->ust_session
= lus
;
2934 if (session
->shm_path
[0]) {
2935 strncpy(lus
->root_shm_path
, session
->shm_path
,
2936 sizeof(lus
->root_shm_path
));
2937 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2938 strncpy(lus
->shm_path
, session
->shm_path
,
2939 sizeof(lus
->shm_path
));
2940 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2941 strncat(lus
->shm_path
, "/ust",
2942 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2944 /* Copy session output to the newly created UST session */
2945 ret
= copy_session_consumer(domain
->type
, session
);
2946 if (ret
!= LTTNG_OK
) {
2954 session
->ust_session
= NULL
;
2959 * Create a kernel tracer session then create the default channel.
2961 static int create_kernel_session(struct ltt_session
*session
)
2965 DBG("Creating kernel session");
2967 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2969 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2973 /* Code flow safety */
2974 assert(session
->kernel_session
);
2976 /* Copy session output to the newly created Kernel session */
2977 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2978 if (ret
!= LTTNG_OK
) {
2982 /* Create directory(ies) on local filesystem. */
2983 if (session
->kernel_session
->consumer
->type
== CONSUMER_DST_LOCAL
&&
2984 strlen(session
->kernel_session
->consumer
->dst
.trace_path
) > 0) {
2985 ret
= run_as_mkdir_recursive(
2986 session
->kernel_session
->consumer
->dst
.trace_path
,
2987 S_IRWXU
| S_IRWXG
, session
->uid
, session
->gid
);
2989 if (errno
!= EEXIST
) {
2990 ERR("Trace directory creation error");
2996 session
->kernel_session
->uid
= session
->uid
;
2997 session
->kernel_session
->gid
= session
->gid
;
2998 session
->kernel_session
->output_traces
= session
->output_traces
;
2999 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
3004 trace_kernel_destroy_session(session
->kernel_session
);
3005 session
->kernel_session
= NULL
;
3010 * Count number of session permitted by uid/gid.
3012 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
3015 struct ltt_session
*session
;
3017 DBG("Counting number of available session for UID %d GID %d",
3019 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
3021 * Only list the sessions the user can control.
3023 if (!session_access_ok(session
, uid
, gid
)) {
3032 * Process the command requested by the lttng client within the command
3033 * context structure. This function make sure that the return structure (llm)
3034 * is set and ready for transmission before returning.
3036 * Return any error encountered or 0 for success.
3038 * "sock" is only used for special-case var. len data.
3040 * Should *NOT* be called with RCU read-side lock held.
3042 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
3046 int need_tracing_session
= 1;
3049 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
3051 assert(!rcu_read_ongoing());
3055 switch (cmd_ctx
->lsm
->cmd_type
) {
3056 case LTTNG_CREATE_SESSION
:
3057 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3058 case LTTNG_CREATE_SESSION_LIVE
:
3059 case LTTNG_DESTROY_SESSION
:
3060 case LTTNG_LIST_SESSIONS
:
3061 case LTTNG_LIST_DOMAINS
:
3062 case LTTNG_START_TRACE
:
3063 case LTTNG_STOP_TRACE
:
3064 case LTTNG_DATA_PENDING
:
3065 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3066 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3067 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3068 case LTTNG_SNAPSHOT_RECORD
:
3069 case LTTNG_SAVE_SESSION
:
3070 case LTTNG_SET_SESSION_SHM_PATH
:
3071 case LTTNG_REGENERATE_METADATA
:
3072 case LTTNG_REGENERATE_STATEDUMP
:
3073 case LTTNG_REGISTER_TRIGGER
:
3074 case LTTNG_UNREGISTER_TRIGGER
:
3081 if (opt_no_kernel
&& need_domain
3082 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
3084 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3086 ret
= LTTNG_ERR_KERN_NA
;
3091 /* Deny register consumer if we already have a spawned consumer. */
3092 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
3093 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3094 if (kconsumer_data
.pid
> 0) {
3095 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3096 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3099 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3103 * Check for command that don't needs to allocate a returned payload. We do
3104 * this here so we don't have to make the call for no payload at each
3107 switch(cmd_ctx
->lsm
->cmd_type
) {
3108 case LTTNG_LIST_SESSIONS
:
3109 case LTTNG_LIST_TRACEPOINTS
:
3110 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3111 case LTTNG_LIST_DOMAINS
:
3112 case LTTNG_LIST_CHANNELS
:
3113 case LTTNG_LIST_EVENTS
:
3114 case LTTNG_LIST_SYSCALLS
:
3115 case LTTNG_LIST_TRACKER_PIDS
:
3116 case LTTNG_DATA_PENDING
:
3119 /* Setup lttng message with no payload */
3120 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0);
3122 /* This label does not try to unlock the session */
3123 goto init_setup_error
;
3127 /* Commands that DO NOT need a session. */
3128 switch (cmd_ctx
->lsm
->cmd_type
) {
3129 case LTTNG_CREATE_SESSION
:
3130 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3131 case LTTNG_CREATE_SESSION_LIVE
:
3132 case LTTNG_LIST_SESSIONS
:
3133 case LTTNG_LIST_TRACEPOINTS
:
3134 case LTTNG_LIST_SYSCALLS
:
3135 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3136 case LTTNG_SAVE_SESSION
:
3137 case LTTNG_REGISTER_TRIGGER
:
3138 case LTTNG_UNREGISTER_TRIGGER
:
3139 need_tracing_session
= 0;
3142 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3144 * We keep the session list lock across _all_ commands
3145 * for now, because the per-session lock does not
3146 * handle teardown properly.
3148 session_lock_list();
3149 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3150 if (cmd_ctx
->session
== NULL
) {
3151 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3154 /* Acquire lock for the session */
3155 session_lock(cmd_ctx
->session
);
3161 * Commands that need a valid session but should NOT create one if none
3162 * exists. Instead of creating one and destroying it when the command is
3163 * handled, process that right before so we save some round trip in useless
3166 switch (cmd_ctx
->lsm
->cmd_type
) {
3167 case LTTNG_DISABLE_CHANNEL
:
3168 case LTTNG_DISABLE_EVENT
:
3169 switch (cmd_ctx
->lsm
->domain
.type
) {
3170 case LTTNG_DOMAIN_KERNEL
:
3171 if (!cmd_ctx
->session
->kernel_session
) {
3172 ret
= LTTNG_ERR_NO_CHANNEL
;
3176 case LTTNG_DOMAIN_JUL
:
3177 case LTTNG_DOMAIN_LOG4J
:
3178 case LTTNG_DOMAIN_PYTHON
:
3179 case LTTNG_DOMAIN_UST
:
3180 if (!cmd_ctx
->session
->ust_session
) {
3181 ret
= LTTNG_ERR_NO_CHANNEL
;
3186 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3198 * Check domain type for specific "pre-action".
3200 switch (cmd_ctx
->lsm
->domain
.type
) {
3201 case LTTNG_DOMAIN_KERNEL
:
3203 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3207 /* Kernel tracer check */
3208 if (kernel_tracer_fd
== -1) {
3209 /* Basically, load kernel tracer modules */
3210 ret
= init_kernel_tracer();
3216 /* Consumer is in an ERROR state. Report back to client */
3217 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3218 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3222 /* Need a session for kernel command */
3223 if (need_tracing_session
) {
3224 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3225 ret
= create_kernel_session(cmd_ctx
->session
);
3227 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3232 /* Start the kernel consumer daemon */
3233 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3234 if (kconsumer_data
.pid
== 0 &&
3235 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3236 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3237 ret
= start_consumerd(&kconsumer_data
);
3239 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3242 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3244 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3248 * The consumer was just spawned so we need to add the socket to
3249 * the consumer output of the session if exist.
3251 ret
= consumer_create_socket(&kconsumer_data
,
3252 cmd_ctx
->session
->kernel_session
->consumer
);
3259 case LTTNG_DOMAIN_JUL
:
3260 case LTTNG_DOMAIN_LOG4J
:
3261 case LTTNG_DOMAIN_PYTHON
:
3262 case LTTNG_DOMAIN_UST
:
3264 if (!ust_app_supported()) {
3265 ret
= LTTNG_ERR_NO_UST
;
3268 /* Consumer is in an ERROR state. Report back to client */
3269 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3270 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3274 if (need_tracing_session
) {
3275 /* Create UST session if none exist. */
3276 if (cmd_ctx
->session
->ust_session
== NULL
) {
3277 ret
= create_ust_session(cmd_ctx
->session
,
3278 &cmd_ctx
->lsm
->domain
);
3279 if (ret
!= LTTNG_OK
) {
3284 /* Start the UST consumer daemons */
3286 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3287 if (consumerd64_bin
[0] != '\0' &&
3288 ustconsumer64_data
.pid
== 0 &&
3289 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3290 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3291 ret
= start_consumerd(&ustconsumer64_data
);
3293 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3294 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3298 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3299 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3301 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3305 * Setup socket for consumer 64 bit. No need for atomic access
3306 * since it was set above and can ONLY be set in this thread.
3308 ret
= consumer_create_socket(&ustconsumer64_data
,
3309 cmd_ctx
->session
->ust_session
->consumer
);
3315 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3316 if (consumerd32_bin
[0] != '\0' &&
3317 ustconsumer32_data
.pid
== 0 &&
3318 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3319 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3320 ret
= start_consumerd(&ustconsumer32_data
);
3322 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3323 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3327 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3328 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3330 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3334 * Setup socket for consumer 64 bit. No need for atomic access
3335 * since it was set above and can ONLY be set in this thread.
3337 ret
= consumer_create_socket(&ustconsumer32_data
,
3338 cmd_ctx
->session
->ust_session
->consumer
);
3350 /* Validate consumer daemon state when start/stop trace command */
3351 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3352 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3353 switch (cmd_ctx
->lsm
->domain
.type
) {
3354 case LTTNG_DOMAIN_NONE
:
3356 case LTTNG_DOMAIN_JUL
:
3357 case LTTNG_DOMAIN_LOG4J
:
3358 case LTTNG_DOMAIN_PYTHON
:
3359 case LTTNG_DOMAIN_UST
:
3360 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3361 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3365 case LTTNG_DOMAIN_KERNEL
:
3366 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3367 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3372 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3378 * Check that the UID or GID match that of the tracing session.
3379 * The root user can interact with all sessions.
3381 if (need_tracing_session
) {
3382 if (!session_access_ok(cmd_ctx
->session
,
3383 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3384 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3385 ret
= LTTNG_ERR_EPERM
;
3391 * Send relayd information to consumer as soon as we have a domain and a
3394 if (cmd_ctx
->session
&& need_domain
) {
3396 * Setup relayd if not done yet. If the relayd information was already
3397 * sent to the consumer, this call will gracefully return.
3399 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3400 if (ret
!= LTTNG_OK
) {
3405 /* Process by command type */
3406 switch (cmd_ctx
->lsm
->cmd_type
) {
3407 case LTTNG_ADD_CONTEXT
:
3410 * An LTTNG_ADD_CONTEXT command might have a supplementary
3411 * payload if the context being added is an application context.
3413 if (cmd_ctx
->lsm
->u
.context
.ctx
.ctx
==
3414 LTTNG_EVENT_CONTEXT_APP_CONTEXT
) {
3415 char *provider_name
= NULL
, *context_name
= NULL
;
3416 size_t provider_name_len
=
3417 cmd_ctx
->lsm
->u
.context
.provider_name_len
;
3418 size_t context_name_len
=
3419 cmd_ctx
->lsm
->u
.context
.context_name_len
;
3421 if (provider_name_len
== 0 || context_name_len
== 0) {
3423 * Application provider and context names MUST
3426 ret
= -LTTNG_ERR_INVALID
;
3430 provider_name
= zmalloc(provider_name_len
+ 1);
3431 if (!provider_name
) {
3432 ret
= -LTTNG_ERR_NOMEM
;
3435 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
=
3438 context_name
= zmalloc(context_name_len
+ 1);
3439 if (!context_name
) {
3440 ret
= -LTTNG_ERR_NOMEM
;
3441 goto error_add_context
;
3443 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
=
3446 ret
= lttcomm_recv_unix_sock(sock
, provider_name
,
3449 goto error_add_context
;
3452 ret
= lttcomm_recv_unix_sock(sock
, context_name
,
3455 goto error_add_context
;
3460 * cmd_add_context assumes ownership of the provider and context
3463 ret
= cmd_add_context(cmd_ctx
->session
,
3464 cmd_ctx
->lsm
->domain
.type
,
3465 cmd_ctx
->lsm
->u
.context
.channel_name
,
3466 &cmd_ctx
->lsm
->u
.context
.ctx
,
3467 kernel_poll_pipe
[1]);
3469 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
= NULL
;
3470 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
= NULL
;
3472 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
);
3473 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
);
3479 case LTTNG_DISABLE_CHANNEL
:
3481 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3482 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3485 case LTTNG_DISABLE_EVENT
:
3489 * FIXME: handle filter; for now we just receive the filter's
3490 * bytecode along with the filter expression which are sent by
3491 * liblttng-ctl and discard them.
3493 * This fixes an issue where the client may block while sending
3494 * the filter payload and encounter an error because the session
3495 * daemon closes the socket without ever handling this data.
3497 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3498 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3501 char data
[LTTNG_FILTER_MAX_LEN
];
3503 DBG("Discarding disable event command payload of size %zu", count
);
3505 ret
= lttcomm_recv_unix_sock(sock
, data
,
3506 count
> sizeof(data
) ? sizeof(data
) : count
);
3511 count
-= (size_t) ret
;
3514 /* FIXME: passing packed structure to non-packed pointer */
3515 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3516 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3517 &cmd_ctx
->lsm
->u
.disable
.event
);
3520 case LTTNG_ENABLE_CHANNEL
:
3522 cmd_ctx
->lsm
->u
.channel
.chan
.attr
.extended
.ptr
=
3523 (struct lttng_channel_extended
*) &cmd_ctx
->lsm
->u
.channel
.extended
;
3524 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3525 &cmd_ctx
->lsm
->u
.channel
.chan
,
3526 kernel_poll_pipe
[1]);
3529 case LTTNG_TRACK_PID
:
3531 ret
= cmd_track_pid(cmd_ctx
->session
,
3532 cmd_ctx
->lsm
->domain
.type
,
3533 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3536 case LTTNG_UNTRACK_PID
:
3538 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3539 cmd_ctx
->lsm
->domain
.type
,
3540 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3543 case LTTNG_ENABLE_EVENT
:
3545 struct lttng_event_exclusion
*exclusion
= NULL
;
3546 struct lttng_filter_bytecode
*bytecode
= NULL
;
3547 char *filter_expression
= NULL
;
3549 /* Handle exclusion events and receive it from the client. */
3550 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3551 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3553 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3554 (count
* LTTNG_SYMBOL_NAME_LEN
));
3556 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3560 DBG("Receiving var len exclusion event list from client ...");
3561 exclusion
->count
= count
;
3562 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3563 count
* LTTNG_SYMBOL_NAME_LEN
);
3565 DBG("Nothing recv() from client var len data... continuing");
3568 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3573 /* Get filter expression from client. */
3574 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3575 size_t expression_len
=
3576 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3578 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3579 ret
= LTTNG_ERR_FILTER_INVAL
;
3584 filter_expression
= zmalloc(expression_len
);
3585 if (!filter_expression
) {
3587 ret
= LTTNG_ERR_FILTER_NOMEM
;
3591 /* Receive var. len. data */
3592 DBG("Receiving var len filter's expression from client ...");
3593 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3596 DBG("Nothing recv() from client car len data... continuing");
3598 free(filter_expression
);
3600 ret
= LTTNG_ERR_FILTER_INVAL
;
3605 /* Handle filter and get bytecode from client. */
3606 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3607 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3609 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3610 ret
= LTTNG_ERR_FILTER_INVAL
;
3611 free(filter_expression
);
3616 bytecode
= zmalloc(bytecode_len
);
3618 free(filter_expression
);
3620 ret
= LTTNG_ERR_FILTER_NOMEM
;
3624 /* Receive var. len. data */
3625 DBG("Receiving var len filter's bytecode from client ...");
3626 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3628 DBG("Nothing recv() from client car len data... continuing");
3630 free(filter_expression
);
3633 ret
= LTTNG_ERR_FILTER_INVAL
;
3637 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3638 free(filter_expression
);
3641 ret
= LTTNG_ERR_FILTER_INVAL
;
3646 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3647 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3648 &cmd_ctx
->lsm
->u
.enable
.event
,
3649 filter_expression
, bytecode
, exclusion
,
3650 kernel_poll_pipe
[1]);
3653 case LTTNG_LIST_TRACEPOINTS
:
3655 struct lttng_event
*events
;
3658 session_lock_list();
3659 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3660 session_unlock_list();
3661 if (nb_events
< 0) {
3662 /* Return value is a negative lttng_error_code. */
3668 * Setup lttng message with payload size set to the event list size in
3669 * bytes and then copy list into the llm payload.
3671 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3672 sizeof(struct lttng_event
) * nb_events
);
3682 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3684 struct lttng_event_field
*fields
;
3687 session_lock_list();
3688 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3690 session_unlock_list();
3691 if (nb_fields
< 0) {
3692 /* Return value is a negative lttng_error_code. */
3698 * Setup lttng message with payload size set to the event list size in
3699 * bytes and then copy list into the llm payload.
3701 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, fields
,
3702 sizeof(struct lttng_event_field
) * nb_fields
);
3712 case LTTNG_LIST_SYSCALLS
:
3714 struct lttng_event
*events
;
3717 nb_events
= cmd_list_syscalls(&events
);
3718 if (nb_events
< 0) {
3719 /* Return value is a negative lttng_error_code. */
3725 * Setup lttng message with payload size set to the event list size in
3726 * bytes and then copy list into the llm payload.
3728 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3729 sizeof(struct lttng_event
) * nb_events
);
3739 case LTTNG_LIST_TRACKER_PIDS
:
3741 int32_t *pids
= NULL
;
3744 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3745 cmd_ctx
->lsm
->domain
.type
, &pids
);
3747 /* Return value is a negative lttng_error_code. */
3753 * Setup lttng message with payload size set to the event list size in
3754 * bytes and then copy list into the llm payload.
3756 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, pids
,
3757 sizeof(int32_t) * nr_pids
);
3767 case LTTNG_SET_CONSUMER_URI
:
3770 struct lttng_uri
*uris
;
3772 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3773 len
= nb_uri
* sizeof(struct lttng_uri
);
3776 ret
= LTTNG_ERR_INVALID
;
3780 uris
= zmalloc(len
);
3782 ret
= LTTNG_ERR_FATAL
;
3786 /* Receive variable len data */
3787 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3788 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3790 DBG("No URIs received from client... continuing");
3792 ret
= LTTNG_ERR_SESSION_FAIL
;
3797 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3799 if (ret
!= LTTNG_OK
) {
3806 case LTTNG_START_TRACE
:
3808 ret
= cmd_start_trace(cmd_ctx
->session
);
3811 case LTTNG_STOP_TRACE
:
3813 ret
= cmd_stop_trace(cmd_ctx
->session
);
3816 case LTTNG_CREATE_SESSION
:
3819 struct lttng_uri
*uris
= NULL
;
3821 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3822 len
= nb_uri
* sizeof(struct lttng_uri
);
3825 uris
= zmalloc(len
);
3827 ret
= LTTNG_ERR_FATAL
;
3831 /* Receive variable len data */
3832 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3833 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3835 DBG("No URIs received from client... continuing");
3837 ret
= LTTNG_ERR_SESSION_FAIL
;
3842 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3843 DBG("Creating session with ONE network URI is a bad call");
3844 ret
= LTTNG_ERR_SESSION_FAIL
;
3850 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3851 &cmd_ctx
->creds
, 0);
3857 case LTTNG_DESTROY_SESSION
:
3859 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3861 /* Set session to NULL so we do not unlock it after free. */
3862 cmd_ctx
->session
= NULL
;
3865 case LTTNG_LIST_DOMAINS
:
3868 struct lttng_domain
*domains
= NULL
;
3870 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3872 /* Return value is a negative lttng_error_code. */
3877 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, domains
,
3878 nb_dom
* sizeof(struct lttng_domain
));
3888 case LTTNG_LIST_CHANNELS
:
3890 ssize_t payload_size
;
3891 struct lttng_channel
*channels
= NULL
;
3893 payload_size
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3894 cmd_ctx
->session
, &channels
);
3895 if (payload_size
< 0) {
3896 /* Return value is a negative lttng_error_code. */
3897 ret
= -payload_size
;
3901 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, channels
,
3912 case LTTNG_LIST_EVENTS
:
3915 struct lttng_event
*events
= NULL
;
3916 struct lttcomm_event_command_header cmd_header
;
3919 memset(&cmd_header
, 0, sizeof(cmd_header
));
3920 /* Extended infos are included at the end of events */
3921 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
,
3922 cmd_ctx
->session
, cmd_ctx
->lsm
->u
.list
.channel_name
,
3923 &events
, &total_size
);
3926 /* Return value is a negative lttng_error_code. */
3931 cmd_header
.nb_events
= nb_event
;
3932 ret
= setup_lttng_msg(cmd_ctx
, events
, total_size
,
3933 &cmd_header
, sizeof(cmd_header
));
3943 case LTTNG_LIST_SESSIONS
:
3945 unsigned int nr_sessions
;
3946 void *sessions_payload
;
3949 session_lock_list();
3950 nr_sessions
= lttng_sessions_count(
3951 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3952 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3953 payload_len
= sizeof(struct lttng_session
) * nr_sessions
;
3954 sessions_payload
= zmalloc(payload_len
);
3956 if (!sessions_payload
) {
3957 session_unlock_list();
3962 cmd_list_lttng_sessions(sessions_payload
,
3963 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3964 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3965 session_unlock_list();
3967 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, sessions_payload
,
3969 free(sessions_payload
);
3978 case LTTNG_REGISTER_CONSUMER
:
3980 struct consumer_data
*cdata
;
3982 switch (cmd_ctx
->lsm
->domain
.type
) {
3983 case LTTNG_DOMAIN_KERNEL
:
3984 cdata
= &kconsumer_data
;
3987 ret
= LTTNG_ERR_UND
;
3991 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3992 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3995 case LTTNG_DATA_PENDING
:
3998 uint8_t pending_ret_byte
;
4000 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
4005 * This function may returns 0 or 1 to indicate whether or not
4006 * there is data pending. In case of error, it should return an
4007 * LTTNG_ERR code. However, some code paths may still return
4008 * a nondescript error code, which we handle by returning an
4011 if (pending_ret
== 0 || pending_ret
== 1) {
4013 * ret will be set to LTTNG_OK at the end of
4016 } else if (pending_ret
< 0) {
4017 ret
= LTTNG_ERR_UNK
;
4024 pending_ret_byte
= (uint8_t) pending_ret
;
4026 /* 1 byte to return whether or not data is pending */
4027 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
,
4028 &pending_ret_byte
, 1);
4037 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
4039 struct lttcomm_lttng_output_id reply
;
4041 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
4042 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
4043 if (ret
!= LTTNG_OK
) {
4047 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &reply
,
4053 /* Copy output list into message payload */
4057 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
4059 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
4060 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
4063 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
4066 struct lttng_snapshot_output
*outputs
= NULL
;
4068 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
4069 if (nb_output
< 0) {
4074 assert((nb_output
> 0 && outputs
) || nb_output
== 0);
4075 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, outputs
,
4076 nb_output
* sizeof(struct lttng_snapshot_output
));
4086 case LTTNG_SNAPSHOT_RECORD
:
4088 ret
= cmd_snapshot_record(cmd_ctx
->session
,
4089 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
4090 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
4093 case LTTNG_CREATE_SESSION_SNAPSHOT
:
4096 struct lttng_uri
*uris
= NULL
;
4098 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4099 len
= nb_uri
* sizeof(struct lttng_uri
);
4102 uris
= zmalloc(len
);
4104 ret
= LTTNG_ERR_FATAL
;
4108 /* Receive variable len data */
4109 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4110 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4112 DBG("No URIs received from client... continuing");
4114 ret
= LTTNG_ERR_SESSION_FAIL
;
4119 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4120 DBG("Creating session with ONE network URI is a bad call");
4121 ret
= LTTNG_ERR_SESSION_FAIL
;
4127 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
4128 nb_uri
, &cmd_ctx
->creds
);
4132 case LTTNG_CREATE_SESSION_LIVE
:
4135 struct lttng_uri
*uris
= NULL
;
4137 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4138 len
= nb_uri
* sizeof(struct lttng_uri
);
4141 uris
= zmalloc(len
);
4143 ret
= LTTNG_ERR_FATAL
;
4147 /* Receive variable len data */
4148 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4149 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4151 DBG("No URIs received from client... continuing");
4153 ret
= LTTNG_ERR_SESSION_FAIL
;
4158 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4159 DBG("Creating session with ONE network URI is a bad call");
4160 ret
= LTTNG_ERR_SESSION_FAIL
;
4166 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4167 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4171 case LTTNG_SAVE_SESSION
:
4173 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4177 case LTTNG_SET_SESSION_SHM_PATH
:
4179 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4180 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4183 case LTTNG_REGENERATE_METADATA
:
4185 ret
= cmd_regenerate_metadata(cmd_ctx
->session
);
4188 case LTTNG_REGENERATE_STATEDUMP
:
4190 ret
= cmd_regenerate_statedump(cmd_ctx
->session
);
4193 case LTTNG_REGISTER_TRIGGER
:
4195 ret
= cmd_register_trigger(cmd_ctx
, sock
,
4196 notification_thread_handle
);
4199 case LTTNG_UNREGISTER_TRIGGER
:
4201 ret
= cmd_unregister_trigger(cmd_ctx
, sock
,
4202 notification_thread_handle
);
4206 ret
= LTTNG_ERR_UND
;
4211 if (cmd_ctx
->llm
== NULL
) {
4212 DBG("Missing llm structure. Allocating one.");
4213 if (setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0) < 0) {
4217 /* Set return code */
4218 cmd_ctx
->llm
->ret_code
= ret
;
4220 if (cmd_ctx
->session
) {
4221 session_unlock(cmd_ctx
->session
);
4223 if (need_tracing_session
) {
4224 session_unlock_list();
4227 assert(!rcu_read_ongoing());
4232 * Thread managing health check socket.
4234 static void *thread_manage_health(void *data
)
4236 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4237 uint32_t revents
, nb_fd
;
4238 struct lttng_poll_event events
;
4239 struct health_comm_msg msg
;
4240 struct health_comm_reply reply
;
4242 DBG("[thread] Manage health check started");
4244 rcu_register_thread();
4246 /* We might hit an error path before this is created. */
4247 lttng_poll_init(&events
);
4249 /* Create unix socket */
4250 sock
= lttcomm_create_unix_sock(health_unix_sock_path
);
4252 ERR("Unable to create health check Unix socket");
4257 /* lttng health client socket path permissions */
4258 ret
= chown(health_unix_sock_path
, 0,
4259 utils_get_group_id(tracing_group_name
));
4261 ERR("Unable to set group on %s", health_unix_sock_path
);
4266 ret
= chmod(health_unix_sock_path
,
4267 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4269 ERR("Unable to set permissions on %s", health_unix_sock_path
);
4276 * Set the CLOEXEC flag. Return code is useless because either way, the
4279 (void) utils_set_fd_cloexec(sock
);
4281 ret
= lttcomm_listen_unix_sock(sock
);
4287 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4288 * more will be added to this poll set.
4290 ret
= sessiond_set_thread_pollset(&events
, 2);
4295 /* Add the application registration socket */
4296 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4301 sessiond_notify_ready();
4304 DBG("Health check ready");
4306 /* Inifinite blocking call, waiting for transmission */
4308 ret
= lttng_poll_wait(&events
, -1);
4311 * Restart interrupted system call.
4313 if (errno
== EINTR
) {
4321 for (i
= 0; i
< nb_fd
; i
++) {
4322 /* Fetch once the poll data */
4323 revents
= LTTNG_POLL_GETEV(&events
, i
);
4324 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4327 /* No activity for this FD (poll implementation). */
4331 /* Thread quit pipe has been closed. Killing thread. */
4332 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4338 /* Event on the registration socket */
4339 if (pollfd
== sock
) {
4340 if (revents
& LPOLLIN
) {
4342 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4343 ERR("Health socket poll error");
4346 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4352 new_sock
= lttcomm_accept_unix_sock(sock
);
4358 * Set the CLOEXEC flag. Return code is useless because either way, the
4361 (void) utils_set_fd_cloexec(new_sock
);
4363 DBG("Receiving data from client for health...");
4364 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4366 DBG("Nothing recv() from client... continuing");
4367 ret
= close(new_sock
);
4374 rcu_thread_online();
4376 memset(&reply
, 0, sizeof(reply
));
4377 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4379 * health_check_state returns 0 if health is
4382 if (!health_check_state(health_sessiond
, i
)) {
4383 reply
.ret_code
|= 1ULL << i
;
4387 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4389 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4391 ERR("Failed to send health data back to client");
4394 /* End of transmission */
4395 ret
= close(new_sock
);
4404 ERR("Health error occurred in %s", __func__
);
4406 DBG("Health check thread dying");
4407 unlink(health_unix_sock_path
);
4415 lttng_poll_clean(&events
);
4417 rcu_unregister_thread();
4422 * This thread manage all clients request using the unix client socket for
4425 static void *thread_manage_clients(void *data
)
4427 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4429 uint32_t revents
, nb_fd
;
4430 struct command_ctx
*cmd_ctx
= NULL
;
4431 struct lttng_poll_event events
;
4433 DBG("[thread] Manage client started");
4435 rcu_register_thread();
4437 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4439 health_code_update();
4441 ret
= lttcomm_listen_unix_sock(client_sock
);
4447 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4448 * more will be added to this poll set.
4450 ret
= sessiond_set_thread_pollset(&events
, 2);
4452 goto error_create_poll
;
4455 /* Add the application registration socket */
4456 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4461 sessiond_notify_ready();
4462 ret
= sem_post(&load_info
->message_thread_ready
);
4464 PERROR("sem_post message_thread_ready");
4468 /* This testpoint is after we signal readiness to the parent. */
4469 if (testpoint(sessiond_thread_manage_clients
)) {
4473 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4477 health_code_update();
4480 DBG("Accepting client command ...");
4482 /* Inifinite blocking call, waiting for transmission */
4484 health_poll_entry();
4485 ret
= lttng_poll_wait(&events
, -1);
4489 * Restart interrupted system call.
4491 if (errno
== EINTR
) {
4499 for (i
= 0; i
< nb_fd
; i
++) {
4500 /* Fetch once the poll data */
4501 revents
= LTTNG_POLL_GETEV(&events
, i
);
4502 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4504 health_code_update();
4507 /* No activity for this FD (poll implementation). */
4511 /* Thread quit pipe has been closed. Killing thread. */
4512 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4518 /* Event on the registration socket */
4519 if (pollfd
== client_sock
) {
4520 if (revents
& LPOLLIN
) {
4522 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4523 ERR("Client socket poll error");
4526 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4532 DBG("Wait for client response");
4534 health_code_update();
4536 sock
= lttcomm_accept_unix_sock(client_sock
);
4542 * Set the CLOEXEC flag. Return code is useless because either way, the
4545 (void) utils_set_fd_cloexec(sock
);
4547 /* Set socket option for credentials retrieval */
4548 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4553 /* Allocate context command to process the client request */
4554 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4555 if (cmd_ctx
== NULL
) {
4556 PERROR("zmalloc cmd_ctx");
4560 /* Allocate data buffer for reception */
4561 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4562 if (cmd_ctx
->lsm
== NULL
) {
4563 PERROR("zmalloc cmd_ctx->lsm");
4567 cmd_ctx
->llm
= NULL
;
4568 cmd_ctx
->session
= NULL
;
4570 health_code_update();
4573 * Data is received from the lttng client. The struct
4574 * lttcomm_session_msg (lsm) contains the command and data request of
4577 DBG("Receiving data from client ...");
4578 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4579 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4581 DBG("Nothing recv() from client... continuing");
4587 clean_command_ctx(&cmd_ctx
);
4591 health_code_update();
4593 // TODO: Validate cmd_ctx including sanity check for
4594 // security purpose.
4596 rcu_thread_online();
4598 * This function dispatch the work to the kernel or userspace tracer
4599 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4600 * informations for the client. The command context struct contains
4601 * everything this function may needs.
4603 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4604 rcu_thread_offline();
4612 * TODO: Inform client somehow of the fatal error. At
4613 * this point, ret < 0 means that a zmalloc failed
4614 * (ENOMEM). Error detected but still accept
4615 * command, unless a socket error has been
4618 clean_command_ctx(&cmd_ctx
);
4622 health_code_update();
4624 DBG("Sending response (size: %d, retcode: %s (%d))",
4625 cmd_ctx
->lttng_msg_size
,
4626 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4627 cmd_ctx
->llm
->ret_code
);
4628 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4630 ERR("Failed to send data back to client");
4633 /* End of transmission */
4640 clean_command_ctx(&cmd_ctx
);
4642 health_code_update();
4654 lttng_poll_clean(&events
);
4655 clean_command_ctx(&cmd_ctx
);
4659 unlink(client_unix_sock_path
);
4660 if (client_sock
>= 0) {
4661 ret
= close(client_sock
);
4669 ERR("Health error occurred in %s", __func__
);
4672 health_unregister(health_sessiond
);
4674 DBG("Client thread dying");
4676 rcu_unregister_thread();
4679 * Since we are creating the consumer threads, we own them, so we need
4680 * to join them before our thread exits.
4682 ret
= join_consumer_thread(&kconsumer_data
);
4685 PERROR("join_consumer");
4688 ret
= join_consumer_thread(&ustconsumer32_data
);
4691 PERROR("join_consumer ust32");
4694 ret
= join_consumer_thread(&ustconsumer64_data
);
4697 PERROR("join_consumer ust64");
4702 static int string_match(const char *str1
, const char *str2
)
4704 return (str1
&& str2
) && !strcmp(str1
, str2
);
4708 * Take an option from the getopt output and set it in the right variable to be
4711 * Return 0 on success else a negative value.
4713 static int set_option(int opt
, const char *arg
, const char *optname
)
4717 if (string_match(optname
, "client-sock") || opt
== 'c') {
4718 if (!arg
|| *arg
== '\0') {
4722 if (lttng_is_setuid_setgid()) {
4723 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4724 "-c, --client-sock");
4726 snprintf(client_unix_sock_path
, PATH_MAX
, "%s", arg
);
4728 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4729 if (!arg
|| *arg
== '\0') {
4733 if (lttng_is_setuid_setgid()) {
4734 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4737 snprintf(apps_unix_sock_path
, PATH_MAX
, "%s", arg
);
4739 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4741 } else if (string_match(optname
, "background") || opt
== 'b') {
4743 } else if (string_match(optname
, "group") || opt
== 'g') {
4744 if (!arg
|| *arg
== '\0') {
4748 if (lttng_is_setuid_setgid()) {
4749 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4753 * If the override option is set, the pointer points to a
4754 * *non* const thus freeing it even though the variable type is
4757 if (tracing_group_name_override
) {
4758 free((void *) tracing_group_name
);
4760 tracing_group_name
= strdup(arg
);
4761 if (!tracing_group_name
) {
4765 tracing_group_name_override
= 1;
4767 } else if (string_match(optname
, "help") || opt
== 'h') {
4768 ret
= utils_show_help(8, "lttng-sessiond", help_msg
);
4770 ERR("Cannot show --help for `lttng-sessiond`");
4773 exit(ret
? EXIT_FAILURE
: EXIT_SUCCESS
);
4774 } else if (string_match(optname
, "version") || opt
== 'V') {
4775 fprintf(stdout
, "%s\n", VERSION
);
4777 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4779 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4780 if (!arg
|| *arg
== '\0') {
4784 if (lttng_is_setuid_setgid()) {
4785 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4786 "--kconsumerd-err-sock");
4788 snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4790 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4791 if (!arg
|| *arg
== '\0') {
4795 if (lttng_is_setuid_setgid()) {
4796 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4797 "--kconsumerd-cmd-sock");
4799 snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4801 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4802 if (!arg
|| *arg
== '\0') {
4806 if (lttng_is_setuid_setgid()) {
4807 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4808 "--ustconsumerd64-err-sock");
4810 snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4812 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4813 if (!arg
|| *arg
== '\0') {
4817 if (lttng_is_setuid_setgid()) {
4818 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4819 "--ustconsumerd64-cmd-sock");
4821 snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4823 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4824 if (!arg
|| *arg
== '\0') {
4828 if (lttng_is_setuid_setgid()) {
4829 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4830 "--ustconsumerd32-err-sock");
4832 snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
, "%s", arg
);
4834 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4835 if (!arg
|| *arg
== '\0') {
4839 if (lttng_is_setuid_setgid()) {
4840 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4841 "--ustconsumerd32-cmd-sock");
4843 snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
, "%s", arg
);
4845 } else if (string_match(optname
, "no-kernel")) {
4847 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4848 lttng_opt_quiet
= 1;
4849 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4850 /* Verbose level can increase using multiple -v */
4852 /* Value obtained from config file */
4853 lttng_opt_verbose
= config_parse_value(arg
);
4855 /* -v used on command line */
4856 lttng_opt_verbose
++;
4858 /* Clamp value to [0, 3] */
4859 lttng_opt_verbose
= lttng_opt_verbose
< 0 ? 0 :
4860 (lttng_opt_verbose
<= 3 ? lttng_opt_verbose
: 3);
4861 } else if (string_match(optname
, "verbose-consumer")) {
4863 opt_verbose_consumer
= config_parse_value(arg
);
4865 opt_verbose_consumer
++;
4867 } else if (string_match(optname
, "consumerd32-path")) {
4868 if (!arg
|| *arg
== '\0') {
4872 if (lttng_is_setuid_setgid()) {
4873 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4874 "--consumerd32-path");
4876 if (consumerd32_bin_override
) {
4877 free((void *) consumerd32_bin
);
4879 consumerd32_bin
= strdup(arg
);
4880 if (!consumerd32_bin
) {
4884 consumerd32_bin_override
= 1;
4886 } else if (string_match(optname
, "consumerd32-libdir")) {
4887 if (!arg
|| *arg
== '\0') {
4891 if (lttng_is_setuid_setgid()) {
4892 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4893 "--consumerd32-libdir");
4895 if (consumerd32_libdir_override
) {
4896 free((void *) consumerd32_libdir
);
4898 consumerd32_libdir
= strdup(arg
);
4899 if (!consumerd32_libdir
) {
4903 consumerd32_libdir_override
= 1;
4905 } else if (string_match(optname
, "consumerd64-path")) {
4906 if (!arg
|| *arg
== '\0') {
4910 if (lttng_is_setuid_setgid()) {
4911 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4912 "--consumerd64-path");
4914 if (consumerd64_bin_override
) {
4915 free((void *) consumerd64_bin
);
4917 consumerd64_bin
= strdup(arg
);
4918 if (!consumerd64_bin
) {
4922 consumerd64_bin_override
= 1;
4924 } else if (string_match(optname
, "consumerd64-libdir")) {
4925 if (!arg
|| *arg
== '\0') {
4929 if (lttng_is_setuid_setgid()) {
4930 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4931 "--consumerd64-libdir");
4933 if (consumerd64_libdir_override
) {
4934 free((void *) consumerd64_libdir
);
4936 consumerd64_libdir
= strdup(arg
);
4937 if (!consumerd64_libdir
) {
4941 consumerd64_libdir_override
= 1;
4943 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4944 if (!arg
|| *arg
== '\0') {
4948 if (lttng_is_setuid_setgid()) {
4949 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4953 opt_pidfile
= strdup(arg
);
4959 } else if (string_match(optname
, "agent-tcp-port")) {
4960 if (!arg
|| *arg
== '\0') {
4964 if (lttng_is_setuid_setgid()) {
4965 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4966 "--agent-tcp-port");
4971 v
= strtoul(arg
, NULL
, 0);
4972 if (errno
!= 0 || !isdigit(arg
[0])) {
4973 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4976 if (v
== 0 || v
>= 65535) {
4977 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
4980 agent_tcp_port
= (uint32_t) v
;
4981 DBG3("Agent TCP port set to non default: %u", agent_tcp_port
);
4983 } else if (string_match(optname
, "load") || opt
== 'l') {
4984 if (!arg
|| *arg
== '\0') {
4988 if (lttng_is_setuid_setgid()) {
4989 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4992 free(opt_load_session_path
);
4993 opt_load_session_path
= strdup(arg
);
4994 if (!opt_load_session_path
) {
4999 } else if (string_match(optname
, "kmod-probes")) {
5000 if (!arg
|| *arg
== '\0') {
5004 if (lttng_is_setuid_setgid()) {
5005 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5008 free(kmod_probes_list
);
5009 kmod_probes_list
= strdup(arg
);
5010 if (!kmod_probes_list
) {
5015 } else if (string_match(optname
, "extra-kmod-probes")) {
5016 if (!arg
|| *arg
== '\0') {
5020 if (lttng_is_setuid_setgid()) {
5021 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5022 "--extra-kmod-probes");
5024 free(kmod_extra_probes_list
);
5025 kmod_extra_probes_list
= strdup(arg
);
5026 if (!kmod_extra_probes_list
) {
5031 } else if (string_match(optname
, "config") || opt
== 'f') {
5032 /* This is handled in set_options() thus silent skip. */
5035 /* Unknown option or other error.
5036 * Error is printed by getopt, just return */
5041 if (ret
== -EINVAL
) {
5042 const char *opt_name
= "unknown";
5045 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
5047 if (opt
== long_options
[i
].val
) {
5048 opt_name
= long_options
[i
].name
;
5053 WARN("Invalid argument provided for option \"%s\", using default value.",
5061 * config_entry_handler_cb used to handle options read from a config file.
5062 * See config_entry_handler_cb comment in common/config/session-config.h for the
5063 * return value conventions.
5065 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
5069 if (!entry
|| !entry
->name
|| !entry
->value
) {
5074 /* Check if the option is to be ignored */
5075 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
5076 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
5081 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
5084 /* Ignore if not fully matched. */
5085 if (strcmp(entry
->name
, long_options
[i
].name
)) {
5090 * If the option takes no argument on the command line, we have to
5091 * check if the value is "true". We support non-zero numeric values,
5094 if (!long_options
[i
].has_arg
) {
5095 ret
= config_parse_value(entry
->value
);
5098 WARN("Invalid configuration value \"%s\" for option %s",
5099 entry
->value
, entry
->name
);
5101 /* False, skip boolean config option. */
5106 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
5110 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
5117 * daemon configuration loading and argument parsing
5119 static int set_options(int argc
, char **argv
)
5121 int ret
= 0, c
= 0, option_index
= 0;
5122 int orig_optopt
= optopt
, orig_optind
= optind
;
5124 const char *config_path
= NULL
;
5126 optstring
= utils_generate_optstring(long_options
,
5127 sizeof(long_options
) / sizeof(struct option
));
5133 /* Check for the --config option */
5134 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5135 &option_index
)) != -1) {
5139 } else if (c
!= 'f') {
5140 /* if not equal to --config option. */
5144 if (lttng_is_setuid_setgid()) {
5145 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5148 config_path
= utils_expand_path(optarg
);
5150 ERR("Failed to resolve path: %s", optarg
);
5155 ret
= config_get_section_entries(config_path
, config_section_name
,
5156 config_entry_handler
, NULL
);
5159 ERR("Invalid configuration option at line %i", ret
);
5165 /* Reset getopt's global state */
5166 optopt
= orig_optopt
;
5167 optind
= orig_optind
;
5171 * getopt_long() will not set option_index if it encounters a
5174 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5181 * Pass NULL as the long option name if popt left the index
5184 ret
= set_option(c
, optarg
,
5185 option_index
< 0 ? NULL
:
5186 long_options
[option_index
].name
);
5198 * Creates the two needed socket by the daemon.
5199 * apps_sock - The communication socket for all UST apps.
5200 * client_sock - The communication of the cli tool (lttng).
5202 static int init_daemon_socket(void)
5207 old_umask
= umask(0);
5209 /* Create client tool unix socket */
5210 client_sock
= lttcomm_create_unix_sock(client_unix_sock_path
);
5211 if (client_sock
< 0) {
5212 ERR("Create unix sock failed: %s", client_unix_sock_path
);
5217 /* Set the cloexec flag */
5218 ret
= utils_set_fd_cloexec(client_sock
);
5220 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5221 "Continuing but note that the consumer daemon will have a "
5222 "reference to this socket on exec()", client_sock
);
5225 /* File permission MUST be 660 */
5226 ret
= chmod(client_unix_sock_path
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5228 ERR("Set file permissions failed: %s", client_unix_sock_path
);
5233 /* Create the application unix socket */
5234 apps_sock
= lttcomm_create_unix_sock(apps_unix_sock_path
);
5235 if (apps_sock
< 0) {
5236 ERR("Create unix sock failed: %s", apps_unix_sock_path
);
5241 /* Set the cloexec flag */
5242 ret
= utils_set_fd_cloexec(apps_sock
);
5244 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5245 "Continuing but note that the consumer daemon will have a "
5246 "reference to this socket on exec()", apps_sock
);
5249 /* File permission MUST be 666 */
5250 ret
= chmod(apps_unix_sock_path
,
5251 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5253 ERR("Set file permissions failed: %s", apps_unix_sock_path
);
5258 DBG3("Session daemon client socket %d and application socket %d created",
5259 client_sock
, apps_sock
);
5267 * Check if the global socket is available, and if a daemon is answering at the
5268 * other side. If yes, error is returned.
5270 static int check_existing_daemon(void)
5272 /* Is there anybody out there ? */
5273 if (lttng_session_daemon_alive()) {
5281 * Set the tracing group gid onto the client socket.
5283 * Race window between mkdir and chown is OK because we are going from more
5284 * permissive (root.root) to less permissive (root.tracing).
5286 static int set_permissions(char *rundir
)
5291 gid
= utils_get_group_id(tracing_group_name
);
5293 /* Set lttng run dir */
5294 ret
= chown(rundir
, 0, gid
);
5296 ERR("Unable to set group on %s", rundir
);
5301 * Ensure all applications and tracing group can search the run
5302 * dir. Allow everyone to read the directory, since it does not
5303 * buy us anything to hide its content.
5305 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5307 ERR("Unable to set permissions on %s", rundir
);
5311 /* lttng client socket path */
5312 ret
= chown(client_unix_sock_path
, 0, gid
);
5314 ERR("Unable to set group on %s", client_unix_sock_path
);
5318 /* kconsumer error socket path */
5319 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5321 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5325 /* 64-bit ustconsumer error socket path */
5326 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5328 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5332 /* 32-bit ustconsumer compat32 error socket path */
5333 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5335 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5339 DBG("All permissions are set");
5345 * Create the lttng run directory needed for all global sockets and pipe.
5347 static int create_lttng_rundir(const char *rundir
)
5351 DBG3("Creating LTTng run directory: %s", rundir
);
5353 ret
= mkdir(rundir
, S_IRWXU
);
5355 if (errno
!= EEXIST
) {
5356 ERR("Unable to create %s", rundir
);
5368 * Setup sockets and directory needed by the kconsumerd communication with the
5371 static int set_consumer_sockets(struct consumer_data
*consumer_data
,
5375 char path
[PATH_MAX
];
5377 switch (consumer_data
->type
) {
5378 case LTTNG_CONSUMER_KERNEL
:
5379 snprintf(path
, PATH_MAX
, DEFAULT_KCONSUMERD_PATH
, rundir
);
5381 case LTTNG_CONSUMER64_UST
:
5382 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD64_PATH
, rundir
);
5384 case LTTNG_CONSUMER32_UST
:
5385 snprintf(path
, PATH_MAX
, DEFAULT_USTCONSUMERD32_PATH
, rundir
);
5388 ERR("Consumer type unknown");
5393 DBG2("Creating consumer directory: %s", path
);
5395 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5397 if (errno
!= EEXIST
) {
5399 ERR("Failed to create %s", path
);
5405 ret
= chown(path
, 0, utils_get_group_id(tracing_group_name
));
5407 ERR("Unable to set group on %s", path
);
5413 /* Create the kconsumerd error unix socket */
5414 consumer_data
->err_sock
=
5415 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5416 if (consumer_data
->err_sock
< 0) {
5417 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5423 * Set the CLOEXEC flag. Return code is useless because either way, the
5426 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5428 PERROR("utils_set_fd_cloexec");
5429 /* continue anyway */
5432 /* File permission MUST be 660 */
5433 ret
= chmod(consumer_data
->err_unix_sock_path
,
5434 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5436 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5446 * Signal handler for the daemon
5448 * Simply stop all worker threads, leaving main() return gracefully after
5449 * joining all threads and calling cleanup().
5451 static void sighandler(int sig
)
5455 DBG("SIGINT caught");
5459 DBG("SIGTERM caught");
5463 CMM_STORE_SHARED(recv_child_signal
, 1);
5471 * Setup signal handler for :
5472 * SIGINT, SIGTERM, SIGPIPE
5474 static int set_signal_handler(void)
5477 struct sigaction sa
;
5480 if ((ret
= sigemptyset(&sigset
)) < 0) {
5481 PERROR("sigemptyset");
5485 sa
.sa_mask
= sigset
;
5488 sa
.sa_handler
= sighandler
;
5489 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5490 PERROR("sigaction");
5494 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5495 PERROR("sigaction");
5499 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5500 PERROR("sigaction");
5504 sa
.sa_handler
= SIG_IGN
;
5505 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5506 PERROR("sigaction");
5510 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5516 * Set open files limit to unlimited. This daemon can open a large number of
5517 * file descriptors in order to consume multiple kernel traces.
5519 static void set_ulimit(void)
5524 /* The kernel does not allow an infinite limit for open files */
5525 lim
.rlim_cur
= 65535;
5526 lim
.rlim_max
= 65535;
5528 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5530 PERROR("failed to set open files limit");
5535 * Write pidfile using the rundir and opt_pidfile.
5537 static int write_pidfile(void)
5540 char pidfile_path
[PATH_MAX
];
5545 if (lttng_strncpy(pidfile_path
, opt_pidfile
, sizeof(pidfile_path
))) {
5550 /* Build pidfile path from rundir and opt_pidfile. */
5551 ret
= snprintf(pidfile_path
, sizeof(pidfile_path
), "%s/"
5552 DEFAULT_LTTNG_SESSIOND_PIDFILE
, rundir
);
5554 PERROR("snprintf pidfile path");
5560 * Create pid file in rundir.
5562 ret
= utils_create_pid_file(getpid(), pidfile_path
);
5568 * Create lockfile using the rundir and return its fd.
5570 static int create_lockfile(void)
5573 char lockfile_path
[PATH_MAX
];
5575 ret
= generate_lock_file_path(lockfile_path
, sizeof(lockfile_path
));
5580 ret
= utils_create_lock_file(lockfile_path
);
5586 * Write agent TCP port using the rundir.
5588 static int write_agent_port(void)
5591 char path
[PATH_MAX
];
5595 ret
= snprintf(path
, sizeof(path
), "%s/"
5596 DEFAULT_LTTNG_SESSIOND_AGENTPORT_FILE
, rundir
);
5598 PERROR("snprintf agent port path");
5603 * Create TCP agent port file in rundir.
5605 ret
= utils_create_pid_file(agent_tcp_port
, path
);
5614 int main(int argc
, char **argv
)
5616 int ret
= 0, retval
= 0;
5618 const char *home_path
, *env_app_timeout
;
5619 struct lttng_pipe
*ust32_channel_monitor_pipe
= NULL
,
5620 *ust64_channel_monitor_pipe
= NULL
,
5621 *kernel_channel_monitor_pipe
= NULL
;
5623 init_kernel_workarounds();
5625 rcu_register_thread();
5627 if (set_signal_handler()) {
5629 goto exit_set_signal_handler
;
5632 setup_consumerd_path();
5634 page_size
= sysconf(_SC_PAGESIZE
);
5635 if (page_size
< 0) {
5636 PERROR("sysconf _SC_PAGESIZE");
5637 page_size
= LONG_MAX
;
5638 WARN("Fallback page size to %ld", page_size
);
5642 * Parse arguments and load the daemon configuration file.
5644 * We have an exit_options exit path to free memory reserved by
5645 * set_options. This is needed because the rest of sessiond_cleanup()
5646 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5647 * depends on set_options.
5650 if (set_options(argc
, argv
)) {
5656 if (opt_daemon
|| opt_background
) {
5659 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5667 * We are in the child. Make sure all other file descriptors are
5668 * closed, in case we are called with more opened file
5669 * descriptors than the standard ones.
5671 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5676 if (run_as_create_worker(argv
[0]) < 0) {
5677 goto exit_create_run_as_worker_cleanup
;
5681 * Starting from here, we can create threads. This needs to be after
5682 * lttng_daemonize due to RCU.
5686 * Initialize the health check subsystem. This call should set the
5687 * appropriate time values.
5689 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5690 if (!health_sessiond
) {
5691 PERROR("health_app_create error");
5693 goto exit_health_sessiond_cleanup
;
5696 /* Create thread to clean up RCU hash tables */
5697 if (init_ht_cleanup_thread(&ht_cleanup_thread
)) {
5699 goto exit_ht_cleanup
;
5702 /* Create thread quit pipe */
5703 if (init_thread_quit_pipe()) {
5705 goto exit_init_data
;
5708 /* Check if daemon is UID = 0 */
5709 is_root
= !getuid();
5712 rundir
= strdup(DEFAULT_LTTNG_RUNDIR
);
5715 goto exit_init_data
;
5718 /* Create global run dir with root access */
5719 if (create_lttng_rundir(rundir
)) {
5721 goto exit_init_data
;
5724 if (strlen(apps_unix_sock_path
) == 0) {
5725 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5726 DEFAULT_GLOBAL_APPS_UNIX_SOCK
);
5729 goto exit_init_data
;
5733 if (strlen(client_unix_sock_path
) == 0) {
5734 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5735 DEFAULT_GLOBAL_CLIENT_UNIX_SOCK
);
5738 goto exit_init_data
;
5742 /* Set global SHM for ust */
5743 if (strlen(wait_shm_path
) == 0) {
5744 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5745 DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH
);
5748 goto exit_init_data
;
5752 if (strlen(health_unix_sock_path
) == 0) {
5753 ret
= snprintf(health_unix_sock_path
,
5754 sizeof(health_unix_sock_path
),
5755 DEFAULT_GLOBAL_HEALTH_UNIX_SOCK
);
5758 goto exit_init_data
;
5762 /* Setup kernel consumerd path */
5763 ret
= snprintf(kconsumer_data
.err_unix_sock_path
, PATH_MAX
,
5764 DEFAULT_KCONSUMERD_ERR_SOCK_PATH
, rundir
);
5767 goto exit_init_data
;
5769 ret
= snprintf(kconsumer_data
.cmd_unix_sock_path
, PATH_MAX
,
5770 DEFAULT_KCONSUMERD_CMD_SOCK_PATH
, rundir
);
5773 goto exit_init_data
;
5776 DBG2("Kernel consumer err path: %s",
5777 kconsumer_data
.err_unix_sock_path
);
5778 DBG2("Kernel consumer cmd path: %s",
5779 kconsumer_data
.cmd_unix_sock_path
);
5780 kernel_channel_monitor_pipe
= lttng_pipe_open(0);
5781 if (!kernel_channel_monitor_pipe
) {
5782 ERR("Failed to create kernel consumer channel monitor pipe");
5784 goto exit_init_data
;
5786 kconsumer_data
.channel_monitor_pipe
=
5787 lttng_pipe_release_writefd(
5788 kernel_channel_monitor_pipe
);
5789 if (kconsumer_data
.channel_monitor_pipe
< 0) {
5791 goto exit_init_data
;
5794 home_path
= utils_get_home_dir();
5795 if (home_path
== NULL
) {
5796 /* TODO: Add --socket PATH option */
5797 ERR("Can't get HOME directory for sockets creation.");
5799 goto exit_init_data
;
5803 * Create rundir from home path. This will create something like
5806 ret
= asprintf(&rundir
, DEFAULT_LTTNG_HOME_RUNDIR
, home_path
);
5809 goto exit_init_data
;
5812 if (create_lttng_rundir(rundir
)) {
5814 goto exit_init_data
;
5817 if (strlen(apps_unix_sock_path
) == 0) {
5818 ret
= snprintf(apps_unix_sock_path
, PATH_MAX
,
5819 DEFAULT_HOME_APPS_UNIX_SOCK
,
5823 goto exit_init_data
;
5827 /* Set the cli tool unix socket path */
5828 if (strlen(client_unix_sock_path
) == 0) {
5829 ret
= snprintf(client_unix_sock_path
, PATH_MAX
,
5830 DEFAULT_HOME_CLIENT_UNIX_SOCK
,
5834 goto exit_init_data
;
5838 /* Set global SHM for ust */
5839 if (strlen(wait_shm_path
) == 0) {
5840 ret
= snprintf(wait_shm_path
, PATH_MAX
,
5841 DEFAULT_HOME_APPS_WAIT_SHM_PATH
,
5845 goto exit_init_data
;
5849 /* Set health check Unix path */
5850 if (strlen(health_unix_sock_path
) == 0) {
5851 ret
= snprintf(health_unix_sock_path
,
5852 sizeof(health_unix_sock_path
),
5853 DEFAULT_HOME_HEALTH_UNIX_SOCK
,
5857 goto exit_init_data
;
5862 lockfile_fd
= create_lockfile();
5863 if (lockfile_fd
< 0) {
5865 goto exit_init_data
;
5868 /* Set consumer initial state */
5869 kernel_consumerd_state
= CONSUMER_STOPPED
;
5870 ust_consumerd_state
= CONSUMER_STOPPED
;
5872 DBG("Client socket path %s", client_unix_sock_path
);
5873 DBG("Application socket path %s", apps_unix_sock_path
);
5874 DBG("Application wait path %s", wait_shm_path
);
5875 DBG("LTTng run directory path: %s", rundir
);
5877 /* 32 bits consumerd path setup */
5878 ret
= snprintf(ustconsumer32_data
.err_unix_sock_path
, PATH_MAX
,
5879 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH
, rundir
);
5881 PERROR("snprintf 32-bit consumer error socket path");
5883 goto exit_init_data
;
5885 ret
= snprintf(ustconsumer32_data
.cmd_unix_sock_path
, PATH_MAX
,
5886 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH
, rundir
);
5888 PERROR("snprintf 32-bit consumer command socket path");
5890 goto exit_init_data
;
5893 DBG2("UST consumer 32 bits err path: %s",
5894 ustconsumer32_data
.err_unix_sock_path
);
5895 DBG2("UST consumer 32 bits cmd path: %s",
5896 ustconsumer32_data
.cmd_unix_sock_path
);
5897 ust32_channel_monitor_pipe
= lttng_pipe_open(0);
5898 if (!ust32_channel_monitor_pipe
) {
5899 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5901 goto exit_init_data
;
5903 ustconsumer32_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
5904 ust32_channel_monitor_pipe
);
5905 if (ustconsumer32_data
.channel_monitor_pipe
< 0) {
5907 goto exit_init_data
;
5910 /* 64 bits consumerd path setup */
5911 ret
= snprintf(ustconsumer64_data
.err_unix_sock_path
, PATH_MAX
,
5912 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH
, rundir
);
5914 PERROR("snprintf 64-bit consumer error socket path");
5916 goto exit_init_data
;
5918 ret
= snprintf(ustconsumer64_data
.cmd_unix_sock_path
, PATH_MAX
,
5919 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH
, rundir
);
5921 PERROR("snprintf 64-bit consumer command socket path");
5923 goto exit_init_data
;
5926 DBG2("UST consumer 64 bits err path: %s",
5927 ustconsumer64_data
.err_unix_sock_path
);
5928 DBG2("UST consumer 64 bits cmd path: %s",
5929 ustconsumer64_data
.cmd_unix_sock_path
);
5930 ust64_channel_monitor_pipe
= lttng_pipe_open(0);
5931 if (!ust64_channel_monitor_pipe
) {
5932 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5934 goto exit_init_data
;
5936 ustconsumer64_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
5937 ust64_channel_monitor_pipe
);
5938 if (ustconsumer64_data
.channel_monitor_pipe
< 0) {
5940 goto exit_init_data
;
5944 * See if daemon already exist.
5946 if (check_existing_daemon()) {
5947 ERR("Already running daemon.\n");
5949 * We do not goto exit because we must not cleanup()
5950 * because a daemon is already running.
5953 goto exit_init_data
;
5957 * Init UST app hash table. Alloc hash table before this point since
5958 * cleanup() can get called after that point.
5960 if (ust_app_ht_alloc()) {
5961 ERR("Failed to allocate UST app hash table");
5963 goto exit_init_data
;
5967 * Initialize agent app hash table. We allocate the hash table here
5968 * since cleanup() can get called after this point.
5970 if (agent_app_ht_alloc()) {
5971 ERR("Failed to allocate Agent app hash table");
5973 goto exit_init_data
;
5977 * These actions must be executed as root. We do that *after* setting up
5978 * the sockets path because we MUST make the check for another daemon using
5979 * those paths *before* trying to set the kernel consumer sockets and init
5983 if (set_consumer_sockets(&kconsumer_data
, rundir
)) {
5985 goto exit_init_data
;
5988 /* Setup kernel tracer */
5989 if (!opt_no_kernel
) {
5990 init_kernel_tracer();
5991 if (kernel_tracer_fd
>= 0) {
5992 ret
= syscall_init_table();
5994 ERR("Unable to populate syscall table. "
5995 "Syscall tracing won't work "
5996 "for this session daemon.");
6001 /* Set ulimit for open files */
6004 /* init lttng_fd tracking must be done after set_ulimit. */
6007 if (set_consumer_sockets(&ustconsumer64_data
, rundir
)) {
6009 goto exit_init_data
;
6012 if (set_consumer_sockets(&ustconsumer32_data
, rundir
)) {
6014 goto exit_init_data
;
6017 /* Setup the needed unix socket */
6018 if (init_daemon_socket()) {
6020 goto exit_init_data
;
6023 /* Set credentials to socket */
6024 if (is_root
&& set_permissions(rundir
)) {
6026 goto exit_init_data
;
6029 /* Get parent pid if -S, --sig-parent is specified. */
6030 if (opt_sig_parent
) {
6034 /* Setup the kernel pipe for waking up the kernel thread */
6035 if (is_root
&& !opt_no_kernel
) {
6036 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
6038 goto exit_init_data
;
6042 /* Setup the thread apps communication pipe. */
6043 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
6045 goto exit_init_data
;
6048 /* Setup the thread apps notify communication pipe. */
6049 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
6051 goto exit_init_data
;
6054 /* Initialize global buffer per UID and PID registry. */
6055 buffer_reg_init_uid_registry();
6056 buffer_reg_init_pid_registry();
6058 /* Init UST command queue. */
6059 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
6062 * Get session list pointer. This pointer MUST NOT be free'd. This list
6063 * is statically declared in session.c
6065 session_list_ptr
= session_get_list();
6069 /* Check for the application socket timeout env variable. */
6070 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
6071 if (env_app_timeout
) {
6072 app_socket_timeout
= atoi(env_app_timeout
);
6074 app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
6077 ret
= write_pidfile();
6079 ERR("Error in write_pidfile");
6081 goto exit_init_data
;
6083 ret
= write_agent_port();
6085 ERR("Error in write_agent_port");
6087 goto exit_init_data
;
6090 /* Initialize communication library */
6092 /* Initialize TCP timeout values */
6093 lttcomm_inet_init();
6095 if (load_session_init_data(&load_info
) < 0) {
6097 goto exit_init_data
;
6099 load_info
->path
= opt_load_session_path
;
6101 /* Create health-check thread. */
6102 ret
= pthread_create(&health_thread
, default_pthread_attr(),
6103 thread_manage_health
, (void *) NULL
);
6106 PERROR("pthread_create health");
6111 /* notification_thread_data acquires the pipes' read side. */
6112 notification_thread_handle
= notification_thread_handle_create(
6113 ust32_channel_monitor_pipe
,
6114 ust64_channel_monitor_pipe
,
6115 kernel_channel_monitor_pipe
);
6116 if (!notification_thread_handle
) {
6118 ERR("Failed to create notification thread shared data");
6120 goto exit_notification
;
6123 /* Create notification thread. */
6124 ret
= pthread_create(¬ification_thread
, default_pthread_attr(),
6125 thread_notification
, notification_thread_handle
);
6128 PERROR("pthread_create notification");
6131 goto exit_notification
;
6134 /* Create thread to manage the client socket */
6135 ret
= pthread_create(&client_thread
, default_pthread_attr(),
6136 thread_manage_clients
, (void *) NULL
);
6139 PERROR("pthread_create clients");
6145 /* Create thread to dispatch registration */
6146 ret
= pthread_create(&dispatch_thread
, default_pthread_attr(),
6147 thread_dispatch_ust_registration
, (void *) NULL
);
6150 PERROR("pthread_create dispatch");
6156 /* Create thread to manage application registration. */
6157 ret
= pthread_create(®_apps_thread
, default_pthread_attr(),
6158 thread_registration_apps
, (void *) NULL
);
6161 PERROR("pthread_create registration");
6167 /* Create thread to manage application socket */
6168 ret
= pthread_create(&apps_thread
, default_pthread_attr(),
6169 thread_manage_apps
, (void *) NULL
);
6172 PERROR("pthread_create apps");
6178 /* Create thread to manage application notify socket */
6179 ret
= pthread_create(&apps_notify_thread
, default_pthread_attr(),
6180 ust_thread_manage_notify
, (void *) NULL
);
6183 PERROR("pthread_create notify");
6186 goto exit_apps_notify
;
6189 /* Create agent registration thread. */
6190 ret
= pthread_create(&agent_reg_thread
, default_pthread_attr(),
6191 agent_thread_manage_registration
, (void *) NULL
);
6194 PERROR("pthread_create agent");
6197 goto exit_agent_reg
;
6200 /* Don't start this thread if kernel tracing is not requested nor root */
6201 if (is_root
&& !opt_no_kernel
) {
6202 /* Create kernel thread to manage kernel event */
6203 ret
= pthread_create(&kernel_thread
, default_pthread_attr(),
6204 thread_manage_kernel
, (void *) NULL
);
6207 PERROR("pthread_create kernel");
6214 /* Create session loading thread. */
6215 ret
= pthread_create(&load_session_thread
, default_pthread_attr(),
6216 thread_load_session
, load_info
);
6219 PERROR("pthread_create load_session_thread");
6222 goto exit_load_session
;
6226 * This is where we start awaiting program completion (e.g. through
6227 * signal that asks threads to teardown).
6230 ret
= pthread_join(load_session_thread
, &status
);
6233 PERROR("pthread_join load_session_thread");
6238 if (is_root
&& !opt_no_kernel
) {
6239 ret
= pthread_join(kernel_thread
, &status
);
6242 PERROR("pthread_join");
6248 ret
= pthread_join(agent_reg_thread
, &status
);
6251 PERROR("pthread_join agent");
6256 ret
= pthread_join(apps_notify_thread
, &status
);
6259 PERROR("pthread_join apps notify");
6264 ret
= pthread_join(apps_thread
, &status
);
6267 PERROR("pthread_join apps");
6272 ret
= pthread_join(reg_apps_thread
, &status
);
6275 PERROR("pthread_join");
6281 * Join dispatch thread after joining reg_apps_thread to ensure
6282 * we don't leak applications in the queue.
6284 ret
= pthread_join(dispatch_thread
, &status
);
6287 PERROR("pthread_join");
6292 ret
= pthread_join(client_thread
, &status
);
6295 PERROR("pthread_join");
6301 ret
= pthread_join(health_thread
, &status
);
6304 PERROR("pthread_join health thread");
6311 * Wait for all pending call_rcu work to complete before tearing
6312 * down data structures. call_rcu worker may be trying to
6313 * perform lookups in those structures.
6317 * sessiond_cleanup() is called when no other thread is running, except
6318 * the ht_cleanup thread, which is needed to destroy the hash tables.
6320 rcu_thread_online();
6324 * Ensure all prior call_rcu are done. call_rcu callbacks may push
6325 * hash tables to the ht_cleanup thread. Therefore, we ensure that
6326 * the queue is empty before shutting down the clean-up thread.
6331 * The teardown of the notification system is performed after the
6332 * session daemon's teardown in order to allow it to be notified
6333 * of the active session and channels at the moment of the teardown.
6335 if (notification_thread_handle
) {
6336 notification_thread_command_quit(notification_thread_handle
);
6337 notification_thread_handle_destroy(notification_thread_handle
);
6340 ret
= pthread_join(notification_thread
, &status
);
6343 PERROR("pthread_join notification thread");
6347 rcu_thread_offline();
6348 rcu_unregister_thread();
6350 ret
= fini_ht_cleanup_thread(&ht_cleanup_thread
);
6354 lttng_pipe_destroy(ust32_channel_monitor_pipe
);
6355 lttng_pipe_destroy(ust64_channel_monitor_pipe
);
6356 lttng_pipe_destroy(kernel_channel_monitor_pipe
);
6359 health_app_destroy(health_sessiond
);
6360 exit_health_sessiond_cleanup
:
6361 exit_create_run_as_worker_cleanup
:
6364 sessiond_cleanup_options();
6366 exit_set_signal_handler
: