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"
76 #include "rotation-thread.h"
79 #include "ht-cleanup.h"
80 #include "sessiond-config.h"
81 #include "sessiond-timer.h"
83 static const char *help_msg
=
84 #ifdef LTTNG_EMBED_HELP
85 #include <lttng-sessiond.8.h>
92 static pid_t ppid
; /* Parent PID for --sig-parent option */
93 static pid_t child_ppid
; /* Internal parent PID use with daemonize. */
94 static int lockfile_fd
= -1;
96 /* Set to 1 when a SIGUSR1 signal is received. */
97 static int recv_child_signal
;
99 static struct lttng_kernel_tracer_version kernel_tracer_version
;
100 static struct lttng_kernel_tracer_abi_version kernel_tracer_abi_version
;
103 * Consumer daemon specific control data. Every value not initialized here is
104 * set to 0 by the static definition.
106 static struct consumer_data kconsumer_data
= {
107 .type
= LTTNG_CONSUMER_KERNEL
,
110 .channel_monitor_pipe
= -1,
111 .channel_rotate_pipe
= -1,
112 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
113 .lock
= PTHREAD_MUTEX_INITIALIZER
,
114 .cond
= PTHREAD_COND_INITIALIZER
,
115 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
117 static struct consumer_data ustconsumer64_data
= {
118 .type
= LTTNG_CONSUMER64_UST
,
121 .channel_monitor_pipe
= -1,
122 .channel_rotate_pipe
= -1,
123 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
124 .lock
= PTHREAD_MUTEX_INITIALIZER
,
125 .cond
= PTHREAD_COND_INITIALIZER
,
126 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
128 static struct consumer_data ustconsumer32_data
= {
129 .type
= LTTNG_CONSUMER32_UST
,
132 .channel_monitor_pipe
= -1,
133 .channel_rotate_pipe
= -1,
134 .pid_mutex
= PTHREAD_MUTEX_INITIALIZER
,
135 .lock
= PTHREAD_MUTEX_INITIALIZER
,
136 .cond
= PTHREAD_COND_INITIALIZER
,
137 .cond_mutex
= PTHREAD_MUTEX_INITIALIZER
,
140 /* Command line options */
141 static const struct option long_options
[] = {
142 { "client-sock", required_argument
, 0, 'c' },
143 { "apps-sock", required_argument
, 0, 'a' },
144 { "kconsumerd-cmd-sock", required_argument
, 0, '\0' },
145 { "kconsumerd-err-sock", required_argument
, 0, '\0' },
146 { "ustconsumerd32-cmd-sock", required_argument
, 0, '\0' },
147 { "ustconsumerd32-err-sock", required_argument
, 0, '\0' },
148 { "ustconsumerd64-cmd-sock", required_argument
, 0, '\0' },
149 { "ustconsumerd64-err-sock", required_argument
, 0, '\0' },
150 { "consumerd32-path", required_argument
, 0, '\0' },
151 { "consumerd32-libdir", required_argument
, 0, '\0' },
152 { "consumerd64-path", required_argument
, 0, '\0' },
153 { "consumerd64-libdir", required_argument
, 0, '\0' },
154 { "daemonize", no_argument
, 0, 'd' },
155 { "background", no_argument
, 0, 'b' },
156 { "sig-parent", no_argument
, 0, 'S' },
157 { "help", no_argument
, 0, 'h' },
158 { "group", required_argument
, 0, 'g' },
159 { "version", no_argument
, 0, 'V' },
160 { "quiet", no_argument
, 0, 'q' },
161 { "verbose", no_argument
, 0, 'v' },
162 { "verbose-consumer", no_argument
, 0, '\0' },
163 { "no-kernel", no_argument
, 0, '\0' },
164 { "pidfile", required_argument
, 0, 'p' },
165 { "agent-tcp-port", required_argument
, 0, '\0' },
166 { "config", required_argument
, 0, 'f' },
167 { "load", required_argument
, 0, 'l' },
168 { "kmod-probes", required_argument
, 0, '\0' },
169 { "extra-kmod-probes", required_argument
, 0, '\0' },
173 struct sessiond_config config
;
175 /* Command line options to ignore from configuration file */
176 static const char *config_ignore_options
[] = { "help", "version", "config" };
178 /* Shared between threads */
179 static int dispatch_thread_exit
;
181 /* Sockets and FDs */
182 static int client_sock
= -1;
183 static int apps_sock
= -1;
184 int kernel_tracer_fd
= -1;
185 static int kernel_poll_pipe
[2] = { -1, -1 };
188 * Quit pipe for all threads. This permits a single cancellation point
189 * for all threads when receiving an event on the pipe.
191 static int thread_quit_pipe
[2] = { -1, -1 };
194 * This pipe is used to inform the thread managing application communication
195 * that a command is queued and ready to be processed.
197 static int apps_cmd_pipe
[2] = { -1, -1 };
199 int apps_cmd_notify_pipe
[2] = { -1, -1 };
201 /* Pthread, Mutexes and Semaphores */
202 static pthread_t apps_thread
;
203 static pthread_t apps_notify_thread
;
204 static pthread_t reg_apps_thread
;
205 static pthread_t client_thread
;
206 static pthread_t kernel_thread
;
207 static pthread_t dispatch_thread
;
208 static pthread_t health_thread
;
209 static pthread_t ht_cleanup_thread
;
210 static pthread_t agent_reg_thread
;
211 static pthread_t load_session_thread
;
212 static pthread_t notification_thread
;
213 static pthread_t rotation_thread
;
214 static pthread_t timer_thread
;
217 * UST registration command queue. This queue is tied with a futex and uses a N
218 * wakers / 1 waiter implemented and detailed in futex.c/.h
220 * The thread_registration_apps and thread_dispatch_ust_registration uses this
221 * queue along with the wait/wake scheme. The thread_manage_apps receives down
222 * the line new application socket and monitors it for any I/O error or clean
223 * close that triggers an unregistration of the application.
225 static struct ust_cmd_queue ust_cmd_queue
;
228 * Pointer initialized before thread creation.
230 * This points to the tracing session list containing the session count and a
231 * mutex lock. The lock MUST be taken if you iterate over the list. The lock
232 * MUST NOT be taken if you call a public function in session.c.
234 * The lock is nested inside the structure: session_list_ptr->lock. Please use
235 * session_lock_list and session_unlock_list for lock acquisition.
237 static struct ltt_session_list
*session_list_ptr
;
239 int ust_consumerd64_fd
= -1;
240 int ust_consumerd32_fd
= -1;
242 static const char *module_proc_lttng
= "/proc/lttng";
245 * Consumer daemon state which is changed when spawning it, killing it or in
246 * case of a fatal error.
248 enum consumerd_state
{
249 CONSUMER_STARTED
= 1,
250 CONSUMER_STOPPED
= 2,
255 * This consumer daemon state is used to validate if a client command will be
256 * able to reach the consumer. If not, the client is informed. For instance,
257 * doing a "lttng start" when the consumer state is set to ERROR will return an
258 * error to the client.
260 * The following example shows a possible race condition of this scheme:
262 * consumer thread error happens
264 * client cmd checks state -> still OK
265 * consumer thread exit, sets error
266 * client cmd try to talk to consumer
269 * However, since the consumer is a different daemon, we have no way of making
270 * sure the command will reach it safely even with this state flag. This is why
271 * we consider that up to the state validation during command processing, the
272 * command is safe. After that, we can not guarantee the correctness of the
273 * client request vis-a-vis the consumer.
275 static enum consumerd_state ust_consumerd_state
;
276 static enum consumerd_state kernel_consumerd_state
;
278 /* Set in main() with the current page size. */
281 /* Application health monitoring */
282 struct health_app
*health_sessiond
;
284 /* Am I root or not. */
285 int is_root
; /* Set to 1 if the daemon is running as root */
287 const char * const config_section_name
= "sessiond";
289 /* Load session thread information to operate. */
290 struct load_session_thread_data
*load_info
;
292 /* Notification thread handle. */
293 struct notification_thread_handle
*notification_thread_handle
;
295 /* Rotation thread handle. */
296 struct rotation_thread_handle
*rotation_thread_handle
;
298 /* Global hash tables */
299 struct lttng_ht
*agent_apps_ht_by_sock
= NULL
;
302 * Whether sessiond is ready for commands/notification channel/health check
304 * NR_LTTNG_SESSIOND_READY must match the number of calls to
305 * sessiond_notify_ready().
307 #define NR_LTTNG_SESSIOND_READY 5
308 int lttng_sessiond_ready
= NR_LTTNG_SESSIOND_READY
;
310 int sessiond_check_thread_quit_pipe(int fd
, uint32_t events
)
312 return (fd
== thread_quit_pipe
[0] && (events
& LPOLLIN
)) ? 1 : 0;
315 /* Notify parents that we are ready for cmd and health check */
317 void sessiond_notify_ready(void)
319 if (uatomic_sub_return(<tng_sessiond_ready
, 1) == 0) {
321 * Notify parent pid that we are ready to accept command
322 * for client side. This ppid is the one from the
323 * external process that spawned us.
325 if (config
.sig_parent
) {
330 * Notify the parent of the fork() process that we are
333 if (config
.daemonize
|| config
.background
) {
334 kill(child_ppid
, SIGUSR1
);
340 int __sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
,
347 ret
= lttng_poll_create(events
, size
, LTTNG_CLOEXEC
);
353 ret
= lttng_poll_add(events
, a_pipe
[0], LPOLLIN
| LPOLLERR
);
365 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
367 int sessiond_set_thread_pollset(struct lttng_poll_event
*events
, size_t size
)
369 return __sessiond_set_thread_pollset(events
, size
, thread_quit_pipe
);
373 * Init thread quit pipe.
375 * Return -1 on error or 0 if all pipes are created.
377 static int __init_thread_quit_pipe(int *a_pipe
)
383 PERROR("thread quit pipe");
387 for (i
= 0; i
< 2; i
++) {
388 ret
= fcntl(a_pipe
[i
], F_SETFD
, FD_CLOEXEC
);
399 static int init_thread_quit_pipe(void)
401 return __init_thread_quit_pipe(thread_quit_pipe
);
405 * Stop all threads by closing the thread quit pipe.
407 static void stop_threads(void)
411 /* Stopping all threads */
412 DBG("Terminating all threads");
413 ret
= notify_thread_pipe(thread_quit_pipe
[1]);
415 ERR("write error on thread quit pipe");
418 /* Dispatch thread */
419 CMM_STORE_SHARED(dispatch_thread_exit
, 1);
420 futex_nto1_wake(&ust_cmd_queue
.futex
);
424 * Close every consumer sockets.
426 static void close_consumer_sockets(void)
430 if (kconsumer_data
.err_sock
>= 0) {
431 ret
= close(kconsumer_data
.err_sock
);
433 PERROR("kernel consumer err_sock close");
436 if (ustconsumer32_data
.err_sock
>= 0) {
437 ret
= close(ustconsumer32_data
.err_sock
);
439 PERROR("UST consumerd32 err_sock close");
442 if (ustconsumer64_data
.err_sock
>= 0) {
443 ret
= close(ustconsumer64_data
.err_sock
);
445 PERROR("UST consumerd64 err_sock close");
448 if (kconsumer_data
.cmd_sock
>= 0) {
449 ret
= close(kconsumer_data
.cmd_sock
);
451 PERROR("kernel consumer cmd_sock close");
454 if (ustconsumer32_data
.cmd_sock
>= 0) {
455 ret
= close(ustconsumer32_data
.cmd_sock
);
457 PERROR("UST consumerd32 cmd_sock close");
460 if (ustconsumer64_data
.cmd_sock
>= 0) {
461 ret
= close(ustconsumer64_data
.cmd_sock
);
463 PERROR("UST consumerd64 cmd_sock close");
466 if (kconsumer_data
.channel_monitor_pipe
>= 0) {
467 ret
= close(kconsumer_data
.channel_monitor_pipe
);
469 PERROR("kernel consumer channel monitor pipe close");
472 if (ustconsumer32_data
.channel_monitor_pipe
>= 0) {
473 ret
= close(ustconsumer32_data
.channel_monitor_pipe
);
475 PERROR("UST consumerd32 channel monitor pipe close");
478 if (ustconsumer64_data
.channel_monitor_pipe
>= 0) {
479 ret
= close(ustconsumer64_data
.channel_monitor_pipe
);
481 PERROR("UST consumerd64 channel monitor pipe close");
484 if (kconsumer_data
.channel_rotate_pipe
>= 0) {
485 ret
= close(kconsumer_data
.channel_rotate_pipe
);
487 PERROR("kernel consumer channel rotate pipe close");
490 if (ustconsumer32_data
.channel_rotate_pipe
>= 0) {
491 ret
= close(ustconsumer32_data
.channel_rotate_pipe
);
493 PERROR("UST consumerd32 channel rotate pipe close");
496 if (ustconsumer64_data
.channel_rotate_pipe
>= 0) {
497 ret
= close(ustconsumer64_data
.channel_rotate_pipe
);
499 PERROR("UST consumerd64 channel rotate pipe close");
505 * Wait on consumer process termination.
507 * Need to be called with the consumer data lock held or from a context
508 * ensuring no concurrent access to data (e.g: cleanup).
510 static void wait_consumer(struct consumer_data
*consumer_data
)
515 if (consumer_data
->pid
<= 0) {
519 DBG("Waiting for complete teardown of consumerd (PID: %d)",
521 ret
= waitpid(consumer_data
->pid
, &status
, 0);
523 PERROR("consumerd waitpid pid: %d", consumer_data
->pid
)
524 } else if (!WIFEXITED(status
)) {
525 ERR("consumerd termination with error: %d",
528 consumer_data
->pid
= 0;
532 * Cleanup the session daemon's data structures.
534 static void sessiond_cleanup(void)
537 struct ltt_session
*sess
, *stmp
;
539 DBG("Cleanup sessiond");
542 * Close the thread quit pipe. It has already done its job,
543 * since we are now called.
545 utils_close_pipe(thread_quit_pipe
);
548 * If config.pid_file_path.value is undefined, the default file will be
549 * wiped when removing the rundir.
551 if (config
.pid_file_path
.value
) {
552 ret
= remove(config
.pid_file_path
.value
);
554 PERROR("remove pidfile %s", config
.pid_file_path
.value
);
558 DBG("Removing sessiond and consumerd content of directory %s",
559 config
.rundir
.value
);
562 DBG("Removing %s", config
.pid_file_path
.value
);
563 (void) unlink(config
.pid_file_path
.value
);
565 DBG("Removing %s", config
.agent_port_file_path
.value
);
566 (void) unlink(config
.agent_port_file_path
.value
);
569 DBG("Removing %s", kconsumer_data
.err_unix_sock_path
);
570 (void) unlink(kconsumer_data
.err_unix_sock_path
);
572 DBG("Removing directory %s", config
.kconsumerd_path
.value
);
573 (void) rmdir(config
.kconsumerd_path
.value
);
575 /* ust consumerd 32 */
576 DBG("Removing %s", config
.consumerd32_err_unix_sock_path
.value
);
577 (void) unlink(config
.consumerd32_err_unix_sock_path
.value
);
579 DBG("Removing directory %s", config
.consumerd32_path
.value
);
580 (void) rmdir(config
.consumerd32_path
.value
);
582 /* ust consumerd 64 */
583 DBG("Removing %s", config
.consumerd64_err_unix_sock_path
.value
);
584 (void) unlink(config
.consumerd64_err_unix_sock_path
.value
);
586 DBG("Removing directory %s", config
.consumerd64_path
.value
);
587 (void) rmdir(config
.consumerd64_path
.value
);
589 DBG("Cleaning up all sessions");
591 /* Destroy session list mutex */
592 if (session_list_ptr
!= NULL
) {
593 pthread_mutex_destroy(&session_list_ptr
->lock
);
595 /* Cleanup ALL session */
596 cds_list_for_each_entry_safe(sess
, stmp
,
597 &session_list_ptr
->head
, list
) {
598 cmd_destroy_session(sess
, kernel_poll_pipe
[1]);
602 wait_consumer(&kconsumer_data
);
603 wait_consumer(&ustconsumer64_data
);
604 wait_consumer(&ustconsumer32_data
);
606 DBG("Cleaning up all agent apps");
607 agent_app_ht_clean();
609 DBG("Closing all UST sockets");
610 ust_app_clean_list();
611 buffer_reg_destroy_registries();
613 if (is_root
&& !config
.no_kernel
) {
614 DBG2("Closing kernel fd");
615 if (kernel_tracer_fd
>= 0) {
616 ret
= close(kernel_tracer_fd
);
621 DBG("Unloading kernel modules");
622 modprobe_remove_lttng_all();
626 close_consumer_sockets();
629 load_session_destroy_data(load_info
);
634 * Cleanup lock file by deleting it and finaly closing it which will
635 * release the file system lock.
637 if (lockfile_fd
>= 0) {
638 ret
= remove(config
.lock_file_path
.value
);
640 PERROR("remove lock file");
642 ret
= close(lockfile_fd
);
644 PERROR("close lock file");
649 * We do NOT rmdir rundir because there are other processes
650 * using it, for instance lttng-relayd, which can start in
651 * parallel with this teardown.
656 * Cleanup the daemon's option data structures.
658 static void sessiond_cleanup_options(void)
660 DBG("Cleaning up options");
662 sessiond_config_fini(&config
);
664 run_as_destroy_worker();
668 * Send data on a unix socket using the liblttsessiondcomm API.
670 * Return lttcomm error code.
672 static int send_unix_sock(int sock
, void *buf
, size_t len
)
674 /* Check valid length */
679 return lttcomm_send_unix_sock(sock
, buf
, len
);
683 * Free memory of a command context structure.
685 static void clean_command_ctx(struct command_ctx
**cmd_ctx
)
687 DBG("Clean command context structure");
689 if ((*cmd_ctx
)->llm
) {
690 free((*cmd_ctx
)->llm
);
692 if ((*cmd_ctx
)->lsm
) {
693 free((*cmd_ctx
)->lsm
);
701 * Notify UST applications using the shm mmap futex.
703 static int notify_ust_apps(int active
)
707 DBG("Notifying applications of session daemon state: %d", active
);
709 /* See shm.c for this call implying mmap, shm and futex calls */
710 wait_shm_mmap
= shm_ust_get_mmap(config
.wait_shm_path
.value
, is_root
);
711 if (wait_shm_mmap
== NULL
) {
715 /* Wake waiting process */
716 futex_wait_update((int32_t *) wait_shm_mmap
, active
);
718 /* Apps notified successfully */
726 * Setup the outgoing data buffer for the response (llm) by allocating the
727 * right amount of memory and copying the original information from the lsm
730 * Return 0 on success, negative value on error.
732 static int setup_lttng_msg(struct command_ctx
*cmd_ctx
,
733 const void *payload_buf
, size_t payload_len
,
734 const void *cmd_header_buf
, size_t cmd_header_len
)
737 const size_t header_len
= sizeof(struct lttcomm_lttng_msg
);
738 const size_t cmd_header_offset
= header_len
;
739 const size_t payload_offset
= cmd_header_offset
+ cmd_header_len
;
740 const size_t total_msg_size
= header_len
+ cmd_header_len
+ payload_len
;
742 cmd_ctx
->llm
= zmalloc(total_msg_size
);
744 if (cmd_ctx
->llm
== NULL
) {
750 /* Copy common data */
751 cmd_ctx
->llm
->cmd_type
= cmd_ctx
->lsm
->cmd_type
;
752 cmd_ctx
->llm
->pid
= cmd_ctx
->lsm
->domain
.attr
.pid
;
753 cmd_ctx
->llm
->cmd_header_size
= cmd_header_len
;
754 cmd_ctx
->llm
->data_size
= payload_len
;
755 cmd_ctx
->lttng_msg_size
= total_msg_size
;
757 /* Copy command header */
758 if (cmd_header_len
) {
759 memcpy(((uint8_t *) cmd_ctx
->llm
) + cmd_header_offset
, cmd_header_buf
,
765 memcpy(((uint8_t *) cmd_ctx
->llm
) + payload_offset
, payload_buf
,
774 * Version of setup_lttng_msg() without command header.
776 static int setup_lttng_msg_no_cmd_header(struct command_ctx
*cmd_ctx
,
777 void *payload_buf
, size_t payload_len
)
779 return setup_lttng_msg(cmd_ctx
, payload_buf
, payload_len
, NULL
, 0);
782 * Update the kernel poll set of all channel fd available over all tracing
783 * session. Add the wakeup pipe at the end of the set.
785 static int update_kernel_poll(struct lttng_poll_event
*events
)
788 struct ltt_session
*session
;
789 struct ltt_kernel_channel
*channel
;
791 DBG("Updating kernel poll set");
794 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
795 session_lock(session
);
796 if (session
->kernel_session
== NULL
) {
797 session_unlock(session
);
801 cds_list_for_each_entry(channel
,
802 &session
->kernel_session
->channel_list
.head
, list
) {
803 /* Add channel fd to the kernel poll set */
804 ret
= lttng_poll_add(events
, channel
->fd
, LPOLLIN
| LPOLLRDNORM
);
806 session_unlock(session
);
809 DBG("Channel fd %d added to kernel set", channel
->fd
);
811 session_unlock(session
);
813 session_unlock_list();
818 session_unlock_list();
823 * Find the channel fd from 'fd' over all tracing session. When found, check
824 * for new channel stream and send those stream fds to the kernel consumer.
826 * Useful for CPU hotplug feature.
828 static int update_kernel_stream(struct consumer_data
*consumer_data
, int fd
)
831 struct ltt_session
*session
;
832 struct ltt_kernel_session
*ksess
;
833 struct ltt_kernel_channel
*channel
;
835 DBG("Updating kernel streams for channel fd %d", fd
);
838 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
839 session_lock(session
);
840 if (session
->kernel_session
== NULL
) {
841 session_unlock(session
);
844 ksess
= session
->kernel_session
;
846 cds_list_for_each_entry(channel
,
847 &ksess
->channel_list
.head
, list
) {
848 struct lttng_ht_iter iter
;
849 struct consumer_socket
*socket
;
851 if (channel
->fd
!= fd
) {
854 DBG("Channel found, updating kernel streams");
855 ret
= kernel_open_channel_stream(channel
);
859 /* Update the stream global counter */
860 ksess
->stream_count_global
+= ret
;
863 * Have we already sent fds to the consumer? If yes, it
864 * means that tracing is started so it is safe to send
865 * our updated stream fds.
867 if (ksess
->consumer_fds_sent
!= 1
868 || ksess
->consumer
== NULL
) {
874 cds_lfht_for_each_entry(ksess
->consumer
->socks
->ht
,
875 &iter
.iter
, socket
, node
.node
) {
876 pthread_mutex_lock(socket
->lock
);
877 ret
= kernel_consumer_send_channel_stream(socket
,
879 session
->output_traces
? 1 : 0);
880 pthread_mutex_unlock(socket
->lock
);
888 session_unlock(session
);
890 session_unlock_list();
894 session_unlock(session
);
895 session_unlock_list();
900 * For each tracing session, update newly registered apps. The session list
901 * lock MUST be acquired before calling this.
903 static void update_ust_app(int app_sock
)
905 struct ltt_session
*sess
, *stmp
;
907 /* Consumer is in an ERROR state. Stop any application update. */
908 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
909 /* Stop the update process since the consumer is dead. */
913 /* For all tracing session(s) */
914 cds_list_for_each_entry_safe(sess
, stmp
, &session_list_ptr
->head
, list
) {
918 if (!sess
->ust_session
) {
923 assert(app_sock
>= 0);
924 app
= ust_app_find_by_sock(app_sock
);
927 * Application can be unregistered before so
928 * this is possible hence simply stopping the
931 DBG3("UST app update failed to find app sock %d",
935 ust_app_global_update(sess
->ust_session
, app
);
939 session_unlock(sess
);
944 * This thread manage event coming from the kernel.
946 * Features supported in this thread:
949 static void *thread_manage_kernel(void *data
)
951 int ret
, i
, pollfd
, update_poll_flag
= 1, err
= -1;
952 uint32_t revents
, nb_fd
;
954 struct lttng_poll_event events
;
956 DBG("[thread] Thread manage kernel started");
958 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_KERNEL
);
961 * This first step of the while is to clean this structure which could free
962 * non NULL pointers so initialize it before the loop.
964 lttng_poll_init(&events
);
966 if (testpoint(sessiond_thread_manage_kernel
)) {
967 goto error_testpoint
;
970 health_code_update();
972 if (testpoint(sessiond_thread_manage_kernel_before_loop
)) {
973 goto error_testpoint
;
977 health_code_update();
979 if (update_poll_flag
== 1) {
980 /* Clean events object. We are about to populate it again. */
981 lttng_poll_clean(&events
);
983 ret
= sessiond_set_thread_pollset(&events
, 2);
985 goto error_poll_create
;
988 ret
= lttng_poll_add(&events
, kernel_poll_pipe
[0], LPOLLIN
);
993 /* This will add the available kernel channel if any. */
994 ret
= update_kernel_poll(&events
);
998 update_poll_flag
= 0;
1001 DBG("Thread kernel polling");
1003 /* Poll infinite value of time */
1005 health_poll_entry();
1006 ret
= lttng_poll_wait(&events
, -1);
1007 DBG("Thread kernel return from poll on %d fds",
1008 LTTNG_POLL_GETNB(&events
));
1012 * Restart interrupted system call.
1014 if (errno
== EINTR
) {
1018 } else if (ret
== 0) {
1019 /* Should not happen since timeout is infinite */
1020 ERR("Return value of poll is 0 with an infinite timeout.\n"
1021 "This should not have happened! Continuing...");
1027 for (i
= 0; i
< nb_fd
; i
++) {
1028 /* Fetch once the poll data */
1029 revents
= LTTNG_POLL_GETEV(&events
, i
);
1030 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1032 health_code_update();
1035 /* No activity for this FD (poll implementation). */
1039 /* Thread quit pipe has been closed. Killing thread. */
1040 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1046 /* Check for data on kernel pipe */
1047 if (revents
& LPOLLIN
) {
1048 if (pollfd
== kernel_poll_pipe
[0]) {
1049 (void) lttng_read(kernel_poll_pipe
[0],
1052 * Ret value is useless here, if this pipe gets any actions an
1053 * update is required anyway.
1055 update_poll_flag
= 1;
1059 * New CPU detected by the kernel. Adding kernel stream to
1060 * kernel session and updating the kernel consumer
1062 ret
= update_kernel_stream(&kconsumer_data
, pollfd
);
1068 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1069 update_poll_flag
= 1;
1072 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1080 lttng_poll_clean(&events
);
1083 utils_close_pipe(kernel_poll_pipe
);
1084 kernel_poll_pipe
[0] = kernel_poll_pipe
[1] = -1;
1087 ERR("Health error occurred in %s", __func__
);
1088 WARN("Kernel thread died unexpectedly. "
1089 "Kernel tracing can continue but CPU hotplug is disabled.");
1091 health_unregister(health_sessiond
);
1092 DBG("Kernel thread dying");
1097 * Signal pthread condition of the consumer data that the thread.
1099 static void signal_consumer_condition(struct consumer_data
*data
, int state
)
1101 pthread_mutex_lock(&data
->cond_mutex
);
1104 * The state is set before signaling. It can be any value, it's the waiter
1105 * job to correctly interpret this condition variable associated to the
1106 * consumer pthread_cond.
1108 * A value of 0 means that the corresponding thread of the consumer data
1109 * was not started. 1 indicates that the thread has started and is ready
1110 * for action. A negative value means that there was an error during the
1113 data
->consumer_thread_is_ready
= state
;
1114 (void) pthread_cond_signal(&data
->cond
);
1116 pthread_mutex_unlock(&data
->cond_mutex
);
1120 * This thread manage the consumer error sent back to the session daemon.
1122 static void *thread_manage_consumer(void *data
)
1124 int sock
= -1, i
, ret
, pollfd
, err
= -1, should_quit
= 0;
1125 uint32_t revents
, nb_fd
;
1126 enum lttcomm_return_code code
;
1127 struct lttng_poll_event events
;
1128 struct consumer_data
*consumer_data
= data
;
1129 struct consumer_socket
*cmd_socket_wrapper
= NULL
;
1131 DBG("[thread] Manage consumer started");
1133 rcu_register_thread();
1134 rcu_thread_online();
1136 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CONSUMER
);
1138 health_code_update();
1141 * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the
1142 * metadata_sock. Nothing more will be added to this poll set.
1144 ret
= sessiond_set_thread_pollset(&events
, 3);
1150 * The error socket here is already in a listening state which was done
1151 * just before spawning this thread to avoid a race between the consumer
1152 * daemon exec trying to connect and the listen() call.
1154 ret
= lttng_poll_add(&events
, consumer_data
->err_sock
, LPOLLIN
| LPOLLRDHUP
);
1159 health_code_update();
1161 /* Infinite blocking call, waiting for transmission */
1163 health_poll_entry();
1165 if (testpoint(sessiond_thread_manage_consumer
)) {
1169 ret
= lttng_poll_wait(&events
, -1);
1173 * Restart interrupted system call.
1175 if (errno
== EINTR
) {
1183 for (i
= 0; i
< nb_fd
; i
++) {
1184 /* Fetch once the poll data */
1185 revents
= LTTNG_POLL_GETEV(&events
, i
);
1186 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1188 health_code_update();
1191 /* No activity for this FD (poll implementation). */
1195 /* Thread quit pipe has been closed. Killing thread. */
1196 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1202 /* Event on the registration socket */
1203 if (pollfd
== consumer_data
->err_sock
) {
1204 if (revents
& LPOLLIN
) {
1206 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1207 ERR("consumer err socket poll error");
1210 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1216 sock
= lttcomm_accept_unix_sock(consumer_data
->err_sock
);
1222 * Set the CLOEXEC flag. Return code is useless because either way, the
1225 (void) utils_set_fd_cloexec(sock
);
1227 health_code_update();
1229 DBG2("Receiving code from consumer err_sock");
1231 /* Getting status code from kconsumerd */
1232 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1233 sizeof(enum lttcomm_return_code
));
1238 health_code_update();
1239 if (code
!= LTTCOMM_CONSUMERD_COMMAND_SOCK_READY
) {
1240 ERR("consumer error when waiting for SOCK_READY : %s",
1241 lttcomm_get_readable_code(-code
));
1245 /* Connect both command and metadata sockets. */
1246 consumer_data
->cmd_sock
=
1247 lttcomm_connect_unix_sock(
1248 consumer_data
->cmd_unix_sock_path
);
1249 consumer_data
->metadata_fd
=
1250 lttcomm_connect_unix_sock(
1251 consumer_data
->cmd_unix_sock_path
);
1252 if (consumer_data
->cmd_sock
< 0 || consumer_data
->metadata_fd
< 0) {
1253 PERROR("consumer connect cmd socket");
1254 /* On error, signal condition and quit. */
1255 signal_consumer_condition(consumer_data
, -1);
1259 consumer_data
->metadata_sock
.fd_ptr
= &consumer_data
->metadata_fd
;
1261 /* Create metadata socket lock. */
1262 consumer_data
->metadata_sock
.lock
= zmalloc(sizeof(pthread_mutex_t
));
1263 if (consumer_data
->metadata_sock
.lock
== NULL
) {
1264 PERROR("zmalloc pthread mutex");
1267 pthread_mutex_init(consumer_data
->metadata_sock
.lock
, NULL
);
1269 DBG("Consumer command socket ready (fd: %d", consumer_data
->cmd_sock
);
1270 DBG("Consumer metadata socket ready (fd: %d)",
1271 consumer_data
->metadata_fd
);
1274 * Remove the consumerd error sock since we've established a connection.
1276 ret
= lttng_poll_del(&events
, consumer_data
->err_sock
);
1281 /* Add new accepted error socket. */
1282 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLRDHUP
);
1287 /* Add metadata socket that is successfully connected. */
1288 ret
= lttng_poll_add(&events
, consumer_data
->metadata_fd
,
1289 LPOLLIN
| LPOLLRDHUP
);
1294 health_code_update();
1297 * Transfer the write-end of the channel monitoring and rotate pipe
1298 * to the consumer by issuing a SET_CHANNEL_MONITOR_PIPE and
1299 * SET_CHANNEL_ROTATE_PIPE commands.
1301 cmd_socket_wrapper
= consumer_allocate_socket(&consumer_data
->cmd_sock
);
1302 if (!cmd_socket_wrapper
) {
1305 cmd_socket_wrapper
->lock
= &consumer_data
->lock
;
1307 ret
= consumer_send_channel_monitor_pipe(cmd_socket_wrapper
,
1308 consumer_data
->channel_monitor_pipe
);
1313 ret
= consumer_send_channel_rotate_pipe(cmd_socket_wrapper
,
1314 consumer_data
->channel_rotate_pipe
);
1319 /* Discard the socket wrapper as it is no longer needed. */
1320 consumer_destroy_socket(cmd_socket_wrapper
);
1321 cmd_socket_wrapper
= NULL
;
1323 /* The thread is completely initialized, signal that it is ready. */
1324 signal_consumer_condition(consumer_data
, 1);
1326 /* Infinite blocking call, waiting for transmission */
1329 health_code_update();
1331 /* Exit the thread because the thread quit pipe has been triggered. */
1333 /* Not a health error. */
1338 health_poll_entry();
1339 ret
= lttng_poll_wait(&events
, -1);
1343 * Restart interrupted system call.
1345 if (errno
== EINTR
) {
1353 for (i
= 0; i
< nb_fd
; i
++) {
1354 /* Fetch once the poll data */
1355 revents
= LTTNG_POLL_GETEV(&events
, i
);
1356 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1358 health_code_update();
1361 /* No activity for this FD (poll implementation). */
1366 * Thread quit pipe has been triggered, flag that we should stop
1367 * but continue the current loop to handle potential data from
1370 should_quit
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1372 if (pollfd
== sock
) {
1373 /* Event on the consumerd socket */
1374 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1375 && !(revents
& LPOLLIN
)) {
1376 ERR("consumer err socket second poll error");
1379 health_code_update();
1380 /* Wait for any kconsumerd error */
1381 ret
= lttcomm_recv_unix_sock(sock
, &code
,
1382 sizeof(enum lttcomm_return_code
));
1384 ERR("consumer closed the command socket");
1388 ERR("consumer return code : %s",
1389 lttcomm_get_readable_code(-code
));
1392 } else if (pollfd
== consumer_data
->metadata_fd
) {
1393 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)
1394 && !(revents
& LPOLLIN
)) {
1395 ERR("consumer err metadata socket second poll error");
1398 /* UST metadata requests */
1399 ret
= ust_consumer_metadata_request(
1400 &consumer_data
->metadata_sock
);
1402 ERR("Handling metadata request");
1406 /* No need for an else branch all FDs are tested prior. */
1408 health_code_update();
1414 * We lock here because we are about to close the sockets and some other
1415 * thread might be using them so get exclusive access which will abort all
1416 * other consumer command by other threads.
1418 pthread_mutex_lock(&consumer_data
->lock
);
1420 /* Immediately set the consumerd state to stopped */
1421 if (consumer_data
->type
== LTTNG_CONSUMER_KERNEL
) {
1422 uatomic_set(&kernel_consumerd_state
, CONSUMER_ERROR
);
1423 } else if (consumer_data
->type
== LTTNG_CONSUMER64_UST
||
1424 consumer_data
->type
== LTTNG_CONSUMER32_UST
) {
1425 uatomic_set(&ust_consumerd_state
, CONSUMER_ERROR
);
1427 /* Code flow error... */
1431 if (consumer_data
->err_sock
>= 0) {
1432 ret
= close(consumer_data
->err_sock
);
1436 consumer_data
->err_sock
= -1;
1438 if (consumer_data
->cmd_sock
>= 0) {
1439 ret
= close(consumer_data
->cmd_sock
);
1443 consumer_data
->cmd_sock
= -1;
1445 if (consumer_data
->metadata_sock
.fd_ptr
&&
1446 *consumer_data
->metadata_sock
.fd_ptr
>= 0) {
1447 ret
= close(*consumer_data
->metadata_sock
.fd_ptr
);
1459 unlink(consumer_data
->err_unix_sock_path
);
1460 unlink(consumer_data
->cmd_unix_sock_path
);
1461 pthread_mutex_unlock(&consumer_data
->lock
);
1463 /* Cleanup metadata socket mutex. */
1464 if (consumer_data
->metadata_sock
.lock
) {
1465 pthread_mutex_destroy(consumer_data
->metadata_sock
.lock
);
1466 free(consumer_data
->metadata_sock
.lock
);
1468 lttng_poll_clean(&events
);
1470 if (cmd_socket_wrapper
) {
1471 consumer_destroy_socket(cmd_socket_wrapper
);
1476 ERR("Health error occurred in %s", __func__
);
1478 health_unregister(health_sessiond
);
1479 DBG("consumer thread cleanup completed");
1481 rcu_thread_offline();
1482 rcu_unregister_thread();
1488 * This thread receives application command sockets (FDs) on the
1489 * apps_cmd_pipe and waits (polls) on them until they are closed
1490 * or an error occurs.
1492 * At that point, it flushes the data (tracing and metadata) associated
1493 * with this application and tears down ust app sessions and other
1494 * associated data structures through ust_app_unregister().
1496 * Note that this thread never sends commands to the applications
1497 * through the command sockets; it merely listens for hang-ups
1498 * and errors on those sockets and cleans-up as they occur.
1500 static void *thread_manage_apps(void *data
)
1502 int i
, ret
, pollfd
, err
= -1;
1504 uint32_t revents
, nb_fd
;
1505 struct lttng_poll_event events
;
1507 DBG("[thread] Manage application started");
1509 rcu_register_thread();
1510 rcu_thread_online();
1512 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_MANAGE
);
1514 if (testpoint(sessiond_thread_manage_apps
)) {
1515 goto error_testpoint
;
1518 health_code_update();
1520 ret
= sessiond_set_thread_pollset(&events
, 2);
1522 goto error_poll_create
;
1525 ret
= lttng_poll_add(&events
, apps_cmd_pipe
[0], LPOLLIN
| LPOLLRDHUP
);
1530 if (testpoint(sessiond_thread_manage_apps_before_loop
)) {
1534 health_code_update();
1537 DBG("Apps thread polling");
1539 /* Inifinite blocking call, waiting for transmission */
1541 health_poll_entry();
1542 ret
= lttng_poll_wait(&events
, -1);
1543 DBG("Apps thread return from poll on %d fds",
1544 LTTNG_POLL_GETNB(&events
));
1548 * Restart interrupted system call.
1550 if (errno
== EINTR
) {
1558 for (i
= 0; i
< nb_fd
; i
++) {
1559 /* Fetch once the poll data */
1560 revents
= LTTNG_POLL_GETEV(&events
, i
);
1561 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1563 health_code_update();
1566 /* No activity for this FD (poll implementation). */
1570 /* Thread quit pipe has been closed. Killing thread. */
1571 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
1577 /* Inspect the apps cmd pipe */
1578 if (pollfd
== apps_cmd_pipe
[0]) {
1579 if (revents
& LPOLLIN
) {
1583 size_ret
= lttng_read(apps_cmd_pipe
[0], &sock
, sizeof(sock
));
1584 if (size_ret
< sizeof(sock
)) {
1585 PERROR("read apps cmd pipe");
1589 health_code_update();
1592 * Since this is a command socket (write then read),
1593 * we only monitor the error events of the socket.
1595 ret
= lttng_poll_add(&events
, sock
,
1596 LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
);
1601 DBG("Apps with sock %d added to poll set", sock
);
1602 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1603 ERR("Apps command pipe error");
1606 ERR("Unknown poll events %u for sock %d", revents
, pollfd
);
1611 * At this point, we know that a registered application made
1612 * the event at poll_wait.
1614 if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
1615 /* Removing from the poll set */
1616 ret
= lttng_poll_del(&events
, pollfd
);
1621 /* Socket closed on remote end. */
1622 ust_app_unregister(pollfd
);
1624 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1629 health_code_update();
1635 lttng_poll_clean(&events
);
1638 utils_close_pipe(apps_cmd_pipe
);
1639 apps_cmd_pipe
[0] = apps_cmd_pipe
[1] = -1;
1642 * We don't clean the UST app hash table here since already registered
1643 * applications can still be controlled so let them be until the session
1644 * daemon dies or the applications stop.
1649 ERR("Health error occurred in %s", __func__
);
1651 health_unregister(health_sessiond
);
1652 DBG("Application communication apps thread cleanup complete");
1653 rcu_thread_offline();
1654 rcu_unregister_thread();
1659 * Send a socket to a thread This is called from the dispatch UST registration
1660 * thread once all sockets are set for the application.
1662 * The sock value can be invalid, we don't really care, the thread will handle
1663 * it and make the necessary cleanup if so.
1665 * On success, return 0 else a negative value being the errno message of the
1668 static int send_socket_to_thread(int fd
, int sock
)
1673 * It's possible that the FD is set as invalid with -1 concurrently just
1674 * before calling this function being a shutdown state of the thread.
1681 ret
= lttng_write(fd
, &sock
, sizeof(sock
));
1682 if (ret
< sizeof(sock
)) {
1683 PERROR("write apps pipe %d", fd
);
1690 /* All good. Don't send back the write positive ret value. */
1697 * Sanitize the wait queue of the dispatch registration thread meaning removing
1698 * invalid nodes from it. This is to avoid memory leaks for the case the UST
1699 * notify socket is never received.
1701 static void sanitize_wait_queue(struct ust_reg_wait_queue
*wait_queue
)
1703 int ret
, nb_fd
= 0, i
;
1704 unsigned int fd_added
= 0;
1705 struct lttng_poll_event events
;
1706 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1710 lttng_poll_init(&events
);
1712 /* Just skip everything for an empty queue. */
1713 if (!wait_queue
->count
) {
1717 ret
= lttng_poll_create(&events
, wait_queue
->count
, LTTNG_CLOEXEC
);
1722 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1723 &wait_queue
->head
, head
) {
1724 assert(wait_node
->app
);
1725 ret
= lttng_poll_add(&events
, wait_node
->app
->sock
,
1726 LPOLLHUP
| LPOLLERR
);
1739 * Poll but don't block so we can quickly identify the faulty events and
1740 * clean them afterwards from the wait queue.
1742 ret
= lttng_poll_wait(&events
, 0);
1748 for (i
= 0; i
< nb_fd
; i
++) {
1749 /* Get faulty FD. */
1750 uint32_t revents
= LTTNG_POLL_GETEV(&events
, i
);
1751 int pollfd
= LTTNG_POLL_GETFD(&events
, i
);
1754 /* No activity for this FD (poll implementation). */
1758 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1759 &wait_queue
->head
, head
) {
1760 if (pollfd
== wait_node
->app
->sock
&&
1761 (revents
& (LPOLLHUP
| LPOLLERR
))) {
1762 cds_list_del(&wait_node
->head
);
1763 wait_queue
->count
--;
1764 ust_app_destroy(wait_node
->app
);
1767 * Silence warning of use-after-free in
1768 * cds_list_for_each_entry_safe which uses
1769 * __typeof__(*wait_node).
1774 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
1781 DBG("Wait queue sanitized, %d node were cleaned up", nb_fd
);
1785 lttng_poll_clean(&events
);
1789 lttng_poll_clean(&events
);
1791 ERR("Unable to sanitize wait queue");
1796 * Dispatch request from the registration threads to the application
1797 * communication thread.
1799 static void *thread_dispatch_ust_registration(void *data
)
1802 struct cds_wfcq_node
*node
;
1803 struct ust_command
*ust_cmd
= NULL
;
1804 struct ust_reg_wait_node
*wait_node
= NULL
, *tmp_wait_node
;
1805 struct ust_reg_wait_queue wait_queue
= {
1809 rcu_register_thread();
1811 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG_DISPATCH
);
1813 if (testpoint(sessiond_thread_app_reg_dispatch
)) {
1814 goto error_testpoint
;
1817 health_code_update();
1819 CDS_INIT_LIST_HEAD(&wait_queue
.head
);
1821 DBG("[thread] Dispatch UST command started");
1824 health_code_update();
1826 /* Atomically prepare the queue futex */
1827 futex_nto1_prepare(&ust_cmd_queue
.futex
);
1829 if (CMM_LOAD_SHARED(dispatch_thread_exit
)) {
1834 struct ust_app
*app
= NULL
;
1838 * Make sure we don't have node(s) that have hung up before receiving
1839 * the notify socket. This is to clean the list in order to avoid
1840 * memory leaks from notify socket that are never seen.
1842 sanitize_wait_queue(&wait_queue
);
1844 health_code_update();
1845 /* Dequeue command for registration */
1846 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
1848 DBG("Woken up but nothing in the UST command queue");
1849 /* Continue thread execution */
1853 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
1855 DBG("Dispatching UST registration pid:%d ppid:%d uid:%d"
1856 " gid:%d sock:%d name:%s (version %d.%d)",
1857 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
1858 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
1859 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
1860 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
1862 if (ust_cmd
->reg_msg
.type
== USTCTL_SOCKET_CMD
) {
1863 wait_node
= zmalloc(sizeof(*wait_node
));
1865 PERROR("zmalloc wait_node dispatch");
1866 ret
= close(ust_cmd
->sock
);
1868 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1870 lttng_fd_put(LTTNG_FD_APPS
, 1);
1874 CDS_INIT_LIST_HEAD(&wait_node
->head
);
1876 /* Create application object if socket is CMD. */
1877 wait_node
->app
= ust_app_create(&ust_cmd
->reg_msg
,
1879 if (!wait_node
->app
) {
1880 ret
= close(ust_cmd
->sock
);
1882 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1884 lttng_fd_put(LTTNG_FD_APPS
, 1);
1890 * Add application to the wait queue so we can set the notify
1891 * socket before putting this object in the global ht.
1893 cds_list_add(&wait_node
->head
, &wait_queue
.head
);
1898 * We have to continue here since we don't have the notify
1899 * socket and the application MUST be added to the hash table
1900 * only at that moment.
1905 * Look for the application in the local wait queue and set the
1906 * notify socket if found.
1908 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
1909 &wait_queue
.head
, head
) {
1910 health_code_update();
1911 if (wait_node
->app
->pid
== ust_cmd
->reg_msg
.pid
) {
1912 wait_node
->app
->notify_sock
= ust_cmd
->sock
;
1913 cds_list_del(&wait_node
->head
);
1915 app
= wait_node
->app
;
1917 DBG3("UST app notify socket %d is set", ust_cmd
->sock
);
1923 * With no application at this stage the received socket is
1924 * basically useless so close it before we free the cmd data
1925 * structure for good.
1928 ret
= close(ust_cmd
->sock
);
1930 PERROR("close ust sock dispatch %d", ust_cmd
->sock
);
1932 lttng_fd_put(LTTNG_FD_APPS
, 1);
1939 * @session_lock_list
1941 * Lock the global session list so from the register up to the
1942 * registration done message, no thread can see the application
1943 * and change its state.
1945 session_lock_list();
1949 * Add application to the global hash table. This needs to be
1950 * done before the update to the UST registry can locate the
1955 /* Set app version. This call will print an error if needed. */
1956 (void) ust_app_version(app
);
1958 /* Send notify socket through the notify pipe. */
1959 ret
= send_socket_to_thread(apps_cmd_notify_pipe
[1],
1963 session_unlock_list();
1965 * No notify thread, stop the UST tracing. However, this is
1966 * not an internal error of the this thread thus setting
1967 * the health error code to a normal exit.
1974 * Update newly registered application with the tracing
1975 * registry info already enabled information.
1977 update_ust_app(app
->sock
);
1980 * Don't care about return value. Let the manage apps threads
1981 * handle app unregistration upon socket close.
1983 (void) ust_app_register_done(app
);
1986 * Even if the application socket has been closed, send the app
1987 * to the thread and unregistration will take place at that
1990 ret
= send_socket_to_thread(apps_cmd_pipe
[1], app
->sock
);
1993 session_unlock_list();
1995 * No apps. thread, stop the UST tracing. However, this is
1996 * not an internal error of the this thread thus setting
1997 * the health error code to a normal exit.
2004 session_unlock_list();
2006 } while (node
!= NULL
);
2008 health_poll_entry();
2009 /* Futex wait on queue. Blocking call on futex() */
2010 futex_nto1_wait(&ust_cmd_queue
.futex
);
2013 /* Normal exit, no error */
2017 /* Clean up wait queue. */
2018 cds_list_for_each_entry_safe(wait_node
, tmp_wait_node
,
2019 &wait_queue
.head
, head
) {
2020 cds_list_del(&wait_node
->head
);
2025 /* Empty command queue. */
2027 /* Dequeue command for registration */
2028 node
= cds_wfcq_dequeue_blocking(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
2032 ust_cmd
= caa_container_of(node
, struct ust_command
, node
);
2033 ret
= close(ust_cmd
->sock
);
2035 PERROR("close ust sock exit dispatch %d", ust_cmd
->sock
);
2037 lttng_fd_put(LTTNG_FD_APPS
, 1);
2042 DBG("Dispatch thread dying");
2045 ERR("Health error occurred in %s", __func__
);
2047 health_unregister(health_sessiond
);
2048 rcu_unregister_thread();
2053 * This thread manage application registration.
2055 static void *thread_registration_apps(void *data
)
2057 int sock
= -1, i
, ret
, pollfd
, err
= -1;
2058 uint32_t revents
, nb_fd
;
2059 struct lttng_poll_event events
;
2061 * Get allocated in this thread, enqueued to a global queue, dequeued and
2062 * freed in the manage apps thread.
2064 struct ust_command
*ust_cmd
= NULL
;
2066 DBG("[thread] Manage application registration started");
2068 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_APP_REG
);
2070 if (testpoint(sessiond_thread_registration_apps
)) {
2071 goto error_testpoint
;
2074 ret
= lttcomm_listen_unix_sock(apps_sock
);
2080 * Pass 2 as size here for the thread quit pipe and apps socket. Nothing
2081 * more will be added to this poll set.
2083 ret
= sessiond_set_thread_pollset(&events
, 2);
2085 goto error_create_poll
;
2088 /* Add the application registration socket */
2089 ret
= lttng_poll_add(&events
, apps_sock
, LPOLLIN
| LPOLLRDHUP
);
2091 goto error_poll_add
;
2094 /* Notify all applications to register */
2095 ret
= notify_ust_apps(1);
2097 ERR("Failed to notify applications or create the wait shared memory.\n"
2098 "Execution continues but there might be problem for already\n"
2099 "running applications that wishes to register.");
2103 DBG("Accepting application registration");
2105 /* Inifinite blocking call, waiting for transmission */
2107 health_poll_entry();
2108 ret
= lttng_poll_wait(&events
, -1);
2112 * Restart interrupted system call.
2114 if (errno
== EINTR
) {
2122 for (i
= 0; i
< nb_fd
; i
++) {
2123 health_code_update();
2125 /* Fetch once the poll data */
2126 revents
= LTTNG_POLL_GETEV(&events
, i
);
2127 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
2130 /* No activity for this FD (poll implementation). */
2134 /* Thread quit pipe has been closed. Killing thread. */
2135 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
2141 /* Event on the registration socket */
2142 if (pollfd
== apps_sock
) {
2143 if (revents
& LPOLLIN
) {
2144 sock
= lttcomm_accept_unix_sock(apps_sock
);
2150 * Set socket timeout for both receiving and ending.
2151 * app_socket_timeout is in seconds, whereas
2152 * lttcomm_setsockopt_rcv_timeout and
2153 * lttcomm_setsockopt_snd_timeout expect msec as
2156 if (config
.app_socket_timeout
>= 0) {
2157 (void) lttcomm_setsockopt_rcv_timeout(sock
,
2158 config
.app_socket_timeout
* 1000);
2159 (void) lttcomm_setsockopt_snd_timeout(sock
,
2160 config
.app_socket_timeout
* 1000);
2164 * Set the CLOEXEC flag. Return code is useless because
2165 * either way, the show must go on.
2167 (void) utils_set_fd_cloexec(sock
);
2169 /* Create UST registration command for enqueuing */
2170 ust_cmd
= zmalloc(sizeof(struct ust_command
));
2171 if (ust_cmd
== NULL
) {
2172 PERROR("ust command zmalloc");
2181 * Using message-based transmissions to ensure we don't
2182 * have to deal with partially received messages.
2184 ret
= lttng_fd_get(LTTNG_FD_APPS
, 1);
2186 ERR("Exhausted file descriptors allowed for applications.");
2196 health_code_update();
2197 ret
= ust_app_recv_registration(sock
, &ust_cmd
->reg_msg
);
2200 /* Close socket of the application. */
2205 lttng_fd_put(LTTNG_FD_APPS
, 1);
2209 health_code_update();
2211 ust_cmd
->sock
= sock
;
2214 DBG("UST registration received with pid:%d ppid:%d uid:%d"
2215 " gid:%d sock:%d name:%s (version %d.%d)",
2216 ust_cmd
->reg_msg
.pid
, ust_cmd
->reg_msg
.ppid
,
2217 ust_cmd
->reg_msg
.uid
, ust_cmd
->reg_msg
.gid
,
2218 ust_cmd
->sock
, ust_cmd
->reg_msg
.name
,
2219 ust_cmd
->reg_msg
.major
, ust_cmd
->reg_msg
.minor
);
2222 * Lock free enqueue the registration request. The red pill
2223 * has been taken! This apps will be part of the *system*.
2225 cds_wfcq_enqueue(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
, &ust_cmd
->node
);
2228 * Wake the registration queue futex. Implicit memory
2229 * barrier with the exchange in cds_wfcq_enqueue.
2231 futex_nto1_wake(&ust_cmd_queue
.futex
);
2232 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
2233 ERR("Register apps socket poll error");
2236 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
2245 /* Notify that the registration thread is gone */
2248 if (apps_sock
>= 0) {
2249 ret
= close(apps_sock
);
2259 lttng_fd_put(LTTNG_FD_APPS
, 1);
2261 unlink(config
.apps_unix_sock_path
.value
);
2264 lttng_poll_clean(&events
);
2268 DBG("UST Registration thread cleanup complete");
2271 ERR("Health error occurred in %s", __func__
);
2273 health_unregister(health_sessiond
);
2279 * Start the thread_manage_consumer. This must be done after a lttng-consumerd
2280 * exec or it will fails.
2282 static int spawn_consumer_thread(struct consumer_data
*consumer_data
)
2285 struct timespec timeout
;
2288 * Make sure we set the readiness flag to 0 because we are NOT ready.
2289 * This access to consumer_thread_is_ready does not need to be
2290 * protected by consumer_data.cond_mutex (yet) since the consumer
2291 * management thread has not been started at this point.
2293 consumer_data
->consumer_thread_is_ready
= 0;
2295 /* Setup pthread condition */
2296 ret
= pthread_condattr_init(&consumer_data
->condattr
);
2299 PERROR("pthread_condattr_init consumer data");
2304 * Set the monotonic clock in order to make sure we DO NOT jump in time
2305 * between the clock_gettime() call and the timedwait call. See bug #324
2306 * for a more details and how we noticed it.
2308 ret
= pthread_condattr_setclock(&consumer_data
->condattr
, CLOCK_MONOTONIC
);
2311 PERROR("pthread_condattr_setclock consumer data");
2315 ret
= pthread_cond_init(&consumer_data
->cond
, &consumer_data
->condattr
);
2318 PERROR("pthread_cond_init consumer data");
2322 ret
= pthread_create(&consumer_data
->thread
, default_pthread_attr(),
2323 thread_manage_consumer
, consumer_data
);
2326 PERROR("pthread_create consumer");
2331 /* We are about to wait on a pthread condition */
2332 pthread_mutex_lock(&consumer_data
->cond_mutex
);
2334 /* Get time for sem_timedwait absolute timeout */
2335 clock_ret
= lttng_clock_gettime(CLOCK_MONOTONIC
, &timeout
);
2337 * Set the timeout for the condition timed wait even if the clock gettime
2338 * call fails since we might loop on that call and we want to avoid to
2339 * increment the timeout too many times.
2341 timeout
.tv_sec
+= DEFAULT_SEM_WAIT_TIMEOUT
;
2344 * The following loop COULD be skipped in some conditions so this is why we
2345 * set ret to 0 in order to make sure at least one round of the loop is
2351 * Loop until the condition is reached or when a timeout is reached. Note
2352 * that the pthread_cond_timedwait(P) man page specifies that EINTR can NOT
2353 * be returned but the pthread_cond(3), from the glibc-doc, says that it is
2354 * possible. This loop does not take any chances and works with both of
2357 while (!consumer_data
->consumer_thread_is_ready
&& ret
!= ETIMEDOUT
) {
2358 if (clock_ret
< 0) {
2359 PERROR("clock_gettime spawn consumer");
2360 /* Infinite wait for the consumerd thread to be ready */
2361 ret
= pthread_cond_wait(&consumer_data
->cond
,
2362 &consumer_data
->cond_mutex
);
2364 ret
= pthread_cond_timedwait(&consumer_data
->cond
,
2365 &consumer_data
->cond_mutex
, &timeout
);
2369 /* Release the pthread condition */
2370 pthread_mutex_unlock(&consumer_data
->cond_mutex
);
2374 if (ret
== ETIMEDOUT
) {
2378 * Call has timed out so we kill the kconsumerd_thread and return
2381 ERR("Condition timed out. The consumer thread was never ready."
2383 pth_ret
= pthread_cancel(consumer_data
->thread
);
2385 PERROR("pthread_cancel consumer thread");
2388 PERROR("pthread_cond_wait failed consumer thread");
2390 /* Caller is expecting a negative value on failure. */
2395 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2396 if (consumer_data
->pid
== 0) {
2397 ERR("Consumerd did not start");
2398 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2401 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2410 * Join consumer thread
2412 static int join_consumer_thread(struct consumer_data
*consumer_data
)
2416 /* Consumer pid must be a real one. */
2417 if (consumer_data
->pid
> 0) {
2419 ret
= kill(consumer_data
->pid
, SIGTERM
);
2421 PERROR("Error killing consumer daemon");
2424 return pthread_join(consumer_data
->thread
, &status
);
2431 * Fork and exec a consumer daemon (consumerd).
2433 * Return pid if successful else -1.
2435 static pid_t
spawn_consumerd(struct consumer_data
*consumer_data
)
2439 const char *consumer_to_use
;
2440 const char *verbosity
;
2443 DBG("Spawning consumerd");
2450 if (config
.verbose_consumer
) {
2451 verbosity
= "--verbose";
2452 } else if (lttng_opt_quiet
) {
2453 verbosity
= "--quiet";
2458 switch (consumer_data
->type
) {
2459 case LTTNG_CONSUMER_KERNEL
:
2461 * Find out which consumerd to execute. We will first try the
2462 * 64-bit path, then the sessiond's installation directory, and
2463 * fallback on the 32-bit one,
2465 DBG3("Looking for a kernel consumer at these locations:");
2466 DBG3(" 1) %s", config
.consumerd64_bin_path
.value
? : "NULL");
2467 DBG3(" 2) %s/%s", INSTALL_BIN_PATH
, DEFAULT_CONSUMERD_FILE
);
2468 DBG3(" 3) %s", config
.consumerd32_bin_path
.value
? : "NULL");
2469 if (stat(config
.consumerd64_bin_path
.value
, &st
) == 0) {
2470 DBG3("Found location #1");
2471 consumer_to_use
= config
.consumerd64_bin_path
.value
;
2472 } else if (stat(INSTALL_BIN_PATH
"/" DEFAULT_CONSUMERD_FILE
, &st
) == 0) {
2473 DBG3("Found location #2");
2474 consumer_to_use
= INSTALL_BIN_PATH
"/" DEFAULT_CONSUMERD_FILE
;
2475 } else if (stat(config
.consumerd32_bin_path
.value
, &st
) == 0) {
2476 DBG3("Found location #3");
2477 consumer_to_use
= config
.consumerd32_bin_path
.value
;
2479 DBG("Could not find any valid consumerd executable");
2483 DBG("Using kernel consumer at: %s", consumer_to_use
);
2484 (void) execl(consumer_to_use
,
2485 "lttng-consumerd", verbosity
, "-k",
2486 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2487 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2488 "--group", config
.tracing_group_name
.value
,
2491 case LTTNG_CONSUMER64_UST
:
2493 if (config
.consumerd64_lib_dir
.value
) {
2498 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2502 tmplen
= strlen(config
.consumerd64_lib_dir
.value
) + 1 /* : */ + strlen(tmp
);
2503 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2508 strcat(tmpnew
, config
.consumerd64_lib_dir
.value
);
2509 if (tmp
[0] != '\0') {
2510 strcat(tmpnew
, ":");
2511 strcat(tmpnew
, tmp
);
2513 ret
= setenv("LD_LIBRARY_PATH", tmpnew
, 1);
2520 DBG("Using 64-bit UST consumer at: %s", config
.consumerd64_bin_path
.value
);
2521 (void) execl(config
.consumerd64_bin_path
.value
, "lttng-consumerd", verbosity
, "-u",
2522 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2523 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2524 "--group", config
.tracing_group_name
.value
,
2528 case LTTNG_CONSUMER32_UST
:
2530 if (config
.consumerd32_lib_dir
.value
) {
2535 tmp
= lttng_secure_getenv("LD_LIBRARY_PATH");
2539 tmplen
= strlen(config
.consumerd32_lib_dir
.value
) + 1 /* : */ + strlen(tmp
);
2540 tmpnew
= zmalloc(tmplen
+ 1 /* \0 */);
2545 strcat(tmpnew
, config
.consumerd32_lib_dir
.value
);
2546 if (tmp
[0] != '\0') {
2547 strcat(tmpnew
, ":");
2548 strcat(tmpnew
, tmp
);
2550 ret
= setenv("LD_LIBRARY_PATH", tmpnew
, 1);
2557 DBG("Using 32-bit UST consumer at: %s", config
.consumerd32_bin_path
.value
);
2558 (void) execl(config
.consumerd32_bin_path
.value
, "lttng-consumerd", verbosity
, "-u",
2559 "--consumerd-cmd-sock", consumer_data
->cmd_unix_sock_path
,
2560 "--consumerd-err-sock", consumer_data
->err_unix_sock_path
,
2561 "--group", config
.tracing_group_name
.value
,
2566 ERR("unknown consumer type");
2570 PERROR("Consumer execl()");
2572 /* Reaching this point, we got a failure on our execl(). */
2574 } else if (pid
> 0) {
2577 PERROR("start consumer fork");
2585 * Spawn the consumerd daemon and session daemon thread.
2587 static int start_consumerd(struct consumer_data
*consumer_data
)
2592 * Set the listen() state on the socket since there is a possible race
2593 * between the exec() of the consumer daemon and this call if place in the
2594 * consumer thread. See bug #366 for more details.
2596 ret
= lttcomm_listen_unix_sock(consumer_data
->err_sock
);
2601 pthread_mutex_lock(&consumer_data
->pid_mutex
);
2602 if (consumer_data
->pid
!= 0) {
2603 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2607 ret
= spawn_consumerd(consumer_data
);
2609 ERR("Spawning consumerd failed");
2610 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2614 /* Setting up the consumer_data pid */
2615 consumer_data
->pid
= ret
;
2616 DBG2("Consumer pid %d", consumer_data
->pid
);
2617 pthread_mutex_unlock(&consumer_data
->pid_mutex
);
2619 DBG2("Spawning consumer control thread");
2620 ret
= spawn_consumer_thread(consumer_data
);
2622 ERR("Fatal error spawning consumer control thread");
2630 /* Cleanup already created sockets on error. */
2631 if (consumer_data
->err_sock
>= 0) {
2634 err
= close(consumer_data
->err_sock
);
2636 PERROR("close consumer data error socket");
2643 * Setup necessary data for kernel tracer action.
2645 static int init_kernel_tracer(void)
2649 /* Modprobe lttng kernel modules */
2650 ret
= modprobe_lttng_control();
2655 /* Open debugfs lttng */
2656 kernel_tracer_fd
= open(module_proc_lttng
, O_RDWR
);
2657 if (kernel_tracer_fd
< 0) {
2658 DBG("Failed to open %s", module_proc_lttng
);
2662 /* Validate kernel version */
2663 ret
= kernel_validate_version(kernel_tracer_fd
, &kernel_tracer_version
,
2664 &kernel_tracer_abi_version
);
2669 ret
= modprobe_lttng_data();
2674 ret
= kernel_supports_ring_buffer_snapshot_sample_positions(
2681 WARN("Kernel tracer does not support buffer monitoring. "
2682 "The monitoring timer of channels in the kernel domain "
2683 "will be set to 0 (disabled).");
2686 DBG("Kernel tracer fd %d", kernel_tracer_fd
);
2690 modprobe_remove_lttng_control();
2691 ret
= close(kernel_tracer_fd
);
2695 kernel_tracer_fd
= -1;
2696 return LTTNG_ERR_KERN_VERSION
;
2699 ret
= close(kernel_tracer_fd
);
2705 modprobe_remove_lttng_control();
2708 WARN("No kernel tracer available");
2709 kernel_tracer_fd
= -1;
2711 return LTTNG_ERR_NEED_ROOT_SESSIOND
;
2713 return LTTNG_ERR_KERN_NA
;
2719 * Copy consumer output from the tracing session to the domain session. The
2720 * function also applies the right modification on a per domain basis for the
2721 * trace files destination directory.
2723 * Should *NOT* be called with RCU read-side lock held.
2725 static int copy_session_consumer(int domain
, struct ltt_session
*session
)
2728 const char *dir_name
;
2729 struct consumer_output
*consumer
;
2732 assert(session
->consumer
);
2735 case LTTNG_DOMAIN_KERNEL
:
2736 DBG3("Copying tracing session consumer output in kernel session");
2738 * XXX: We should audit the session creation and what this function
2739 * does "extra" in order to avoid a destroy since this function is used
2740 * in the domain session creation (kernel and ust) only. Same for UST
2743 if (session
->kernel_session
->consumer
) {
2744 consumer_output_put(session
->kernel_session
->consumer
);
2746 session
->kernel_session
->consumer
=
2747 consumer_copy_output(session
->consumer
);
2748 /* Ease our life a bit for the next part */
2749 consumer
= session
->kernel_session
->consumer
;
2750 dir_name
= DEFAULT_KERNEL_TRACE_DIR
;
2752 case LTTNG_DOMAIN_JUL
:
2753 case LTTNG_DOMAIN_LOG4J
:
2754 case LTTNG_DOMAIN_PYTHON
:
2755 case LTTNG_DOMAIN_UST
:
2756 DBG3("Copying tracing session consumer output in UST session");
2757 if (session
->ust_session
->consumer
) {
2758 consumer_output_put(session
->ust_session
->consumer
);
2760 session
->ust_session
->consumer
=
2761 consumer_copy_output(session
->consumer
);
2762 /* Ease our life a bit for the next part */
2763 consumer
= session
->ust_session
->consumer
;
2764 dir_name
= DEFAULT_UST_TRACE_DIR
;
2767 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2771 /* Append correct directory to subdir */
2772 strncat(consumer
->subdir
, dir_name
,
2773 sizeof(consumer
->subdir
) - strlen(consumer
->subdir
) - 1);
2774 DBG3("Copy session consumer subdir %s", consumer
->subdir
);
2783 * Create an UST session and add it to the session ust list.
2785 * Should *NOT* be called with RCU read-side lock held.
2787 static int create_ust_session(struct ltt_session
*session
,
2788 struct lttng_domain
*domain
)
2791 struct ltt_ust_session
*lus
= NULL
;
2795 assert(session
->consumer
);
2797 switch (domain
->type
) {
2798 case LTTNG_DOMAIN_JUL
:
2799 case LTTNG_DOMAIN_LOG4J
:
2800 case LTTNG_DOMAIN_PYTHON
:
2801 case LTTNG_DOMAIN_UST
:
2804 ERR("Unknown UST domain on create session %d", domain
->type
);
2805 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
2809 DBG("Creating UST session");
2811 lus
= trace_ust_create_session(session
->id
);
2813 ret
= LTTNG_ERR_UST_SESS_FAIL
;
2817 lus
->uid
= session
->uid
;
2818 lus
->gid
= session
->gid
;
2819 lus
->output_traces
= session
->output_traces
;
2820 lus
->snapshot_mode
= session
->snapshot_mode
;
2821 lus
->live_timer_interval
= session
->live_timer
;
2822 session
->ust_session
= lus
;
2823 if (session
->shm_path
[0]) {
2824 strncpy(lus
->root_shm_path
, session
->shm_path
,
2825 sizeof(lus
->root_shm_path
));
2826 lus
->root_shm_path
[sizeof(lus
->root_shm_path
) - 1] = '\0';
2827 strncpy(lus
->shm_path
, session
->shm_path
,
2828 sizeof(lus
->shm_path
));
2829 lus
->shm_path
[sizeof(lus
->shm_path
) - 1] = '\0';
2830 strncat(lus
->shm_path
, "/ust",
2831 sizeof(lus
->shm_path
) - strlen(lus
->shm_path
) - 1);
2833 /* Copy session output to the newly created UST session */
2834 ret
= copy_session_consumer(domain
->type
, session
);
2835 if (ret
!= LTTNG_OK
) {
2843 session
->ust_session
= NULL
;
2848 * Create a kernel tracer session then create the default channel.
2850 static int create_kernel_session(struct ltt_session
*session
)
2854 DBG("Creating kernel session");
2856 ret
= kernel_create_session(session
, kernel_tracer_fd
);
2858 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
2862 /* Code flow safety */
2863 assert(session
->kernel_session
);
2865 /* Copy session output to the newly created Kernel session */
2866 ret
= copy_session_consumer(LTTNG_DOMAIN_KERNEL
, session
);
2867 if (ret
!= LTTNG_OK
) {
2871 session
->kernel_session
->uid
= session
->uid
;
2872 session
->kernel_session
->gid
= session
->gid
;
2873 session
->kernel_session
->output_traces
= session
->output_traces
;
2874 session
->kernel_session
->snapshot_mode
= session
->snapshot_mode
;
2879 trace_kernel_destroy_session(session
->kernel_session
);
2880 session
->kernel_session
= NULL
;
2885 * Count number of session permitted by uid/gid.
2887 static unsigned int lttng_sessions_count(uid_t uid
, gid_t gid
)
2890 struct ltt_session
*session
;
2892 DBG("Counting number of available session for UID %d GID %d",
2894 cds_list_for_each_entry(session
, &session_list_ptr
->head
, list
) {
2896 * Only list the sessions the user can control.
2898 if (!session_access_ok(session
, uid
, gid
)) {
2907 * Check if the current kernel tracer supports the session rotation feature.
2908 * Return 1 if it does, 0 otherwise.
2910 static int check_rotate_compatible(void)
2914 if (kernel_tracer_version
.major
!= 2 || kernel_tracer_version
.minor
< 11) {
2915 DBG("Kernel tracer version is not compatible with the rotation feature");
2923 * Process the command requested by the lttng client within the command
2924 * context structure. This function make sure that the return structure (llm)
2925 * is set and ready for transmission before returning.
2927 * Return any error encountered or 0 for success.
2929 * "sock" is only used for special-case var. len data.
2931 * Should *NOT* be called with RCU read-side lock held.
2933 static int process_client_msg(struct command_ctx
*cmd_ctx
, int sock
,
2937 int need_tracing_session
= 1;
2940 DBG("Processing client command %d", cmd_ctx
->lsm
->cmd_type
);
2942 assert(!rcu_read_ongoing());
2946 switch (cmd_ctx
->lsm
->cmd_type
) {
2947 case LTTNG_CREATE_SESSION
:
2948 case LTTNG_CREATE_SESSION_SNAPSHOT
:
2949 case LTTNG_CREATE_SESSION_LIVE
:
2950 case LTTNG_DESTROY_SESSION
:
2951 case LTTNG_LIST_SESSIONS
:
2952 case LTTNG_LIST_DOMAINS
:
2953 case LTTNG_START_TRACE
:
2954 case LTTNG_STOP_TRACE
:
2955 case LTTNG_DATA_PENDING
:
2956 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
2957 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
2958 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
2959 case LTTNG_SNAPSHOT_RECORD
:
2960 case LTTNG_SAVE_SESSION
:
2961 case LTTNG_SET_SESSION_SHM_PATH
:
2962 case LTTNG_REGENERATE_METADATA
:
2963 case LTTNG_REGENERATE_STATEDUMP
:
2964 case LTTNG_REGISTER_TRIGGER
:
2965 case LTTNG_UNREGISTER_TRIGGER
:
2966 case LTTNG_ROTATE_SESSION
:
2967 case LTTNG_ROTATION_GET_INFO
:
2968 case LTTNG_SESSION_GET_CURRENT_OUTPUT
:
2969 case LTTNG_ROTATION_SET_SCHEDULE
:
2976 if (config
.no_kernel
&& need_domain
2977 && cmd_ctx
->lsm
->domain
.type
== LTTNG_DOMAIN_KERNEL
) {
2979 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
2981 ret
= LTTNG_ERR_KERN_NA
;
2986 /* Deny register consumer if we already have a spawned consumer. */
2987 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_REGISTER_CONSUMER
) {
2988 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
2989 if (kconsumer_data
.pid
> 0) {
2990 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
2991 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2994 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
2998 * Check for command that don't needs to allocate a returned payload. We do
2999 * this here so we don't have to make the call for no payload at each
3002 switch(cmd_ctx
->lsm
->cmd_type
) {
3003 case LTTNG_LIST_SESSIONS
:
3004 case LTTNG_LIST_TRACEPOINTS
:
3005 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3006 case LTTNG_LIST_DOMAINS
:
3007 case LTTNG_LIST_CHANNELS
:
3008 case LTTNG_LIST_EVENTS
:
3009 case LTTNG_LIST_SYSCALLS
:
3010 case LTTNG_LIST_TRACKER_PIDS
:
3011 case LTTNG_DATA_PENDING
:
3012 case LTTNG_ROTATE_SESSION
:
3013 case LTTNG_ROTATION_GET_INFO
:
3016 /* Setup lttng message with no payload */
3017 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0);
3019 /* This label does not try to unlock the session */
3020 goto init_setup_error
;
3024 /* Commands that DO NOT need a session. */
3025 switch (cmd_ctx
->lsm
->cmd_type
) {
3026 case LTTNG_CREATE_SESSION
:
3027 case LTTNG_CREATE_SESSION_SNAPSHOT
:
3028 case LTTNG_CREATE_SESSION_LIVE
:
3029 case LTTNG_LIST_SESSIONS
:
3030 case LTTNG_LIST_TRACEPOINTS
:
3031 case LTTNG_LIST_SYSCALLS
:
3032 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3033 case LTTNG_SAVE_SESSION
:
3034 case LTTNG_REGISTER_TRIGGER
:
3035 case LTTNG_UNREGISTER_TRIGGER
:
3036 need_tracing_session
= 0;
3039 DBG("Getting session %s by name", cmd_ctx
->lsm
->session
.name
);
3041 * We keep the session list lock across _all_ commands
3042 * for now, because the per-session lock does not
3043 * handle teardown properly.
3045 session_lock_list();
3046 cmd_ctx
->session
= session_find_by_name(cmd_ctx
->lsm
->session
.name
);
3047 if (cmd_ctx
->session
== NULL
) {
3048 ret
= LTTNG_ERR_SESS_NOT_FOUND
;
3051 /* Acquire lock for the session */
3052 session_lock(cmd_ctx
->session
);
3058 * Commands that need a valid session but should NOT create one if none
3059 * exists. Instead of creating one and destroying it when the command is
3060 * handled, process that right before so we save some round trip in useless
3063 switch (cmd_ctx
->lsm
->cmd_type
) {
3064 case LTTNG_DISABLE_CHANNEL
:
3065 case LTTNG_DISABLE_EVENT
:
3066 switch (cmd_ctx
->lsm
->domain
.type
) {
3067 case LTTNG_DOMAIN_KERNEL
:
3068 if (!cmd_ctx
->session
->kernel_session
) {
3069 ret
= LTTNG_ERR_NO_CHANNEL
;
3073 case LTTNG_DOMAIN_JUL
:
3074 case LTTNG_DOMAIN_LOG4J
:
3075 case LTTNG_DOMAIN_PYTHON
:
3076 case LTTNG_DOMAIN_UST
:
3077 if (!cmd_ctx
->session
->ust_session
) {
3078 ret
= LTTNG_ERR_NO_CHANNEL
;
3083 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3095 * Check domain type for specific "pre-action".
3097 switch (cmd_ctx
->lsm
->domain
.type
) {
3098 case LTTNG_DOMAIN_KERNEL
:
3100 ret
= LTTNG_ERR_NEED_ROOT_SESSIOND
;
3104 /* Kernel tracer check */
3105 if (kernel_tracer_fd
== -1) {
3106 /* Basically, load kernel tracer modules */
3107 ret
= init_kernel_tracer();
3113 /* Consumer is in an ERROR state. Report back to client */
3114 if (uatomic_read(&kernel_consumerd_state
) == CONSUMER_ERROR
) {
3115 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3119 /* Need a session for kernel command */
3120 if (need_tracing_session
) {
3121 if (cmd_ctx
->session
->kernel_session
== NULL
) {
3122 ret
= create_kernel_session(cmd_ctx
->session
);
3124 ret
= LTTNG_ERR_KERN_SESS_FAIL
;
3129 /* Start the kernel consumer daemon */
3130 pthread_mutex_lock(&kconsumer_data
.pid_mutex
);
3131 if (kconsumer_data
.pid
== 0 &&
3132 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3133 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3134 ret
= start_consumerd(&kconsumer_data
);
3136 ret
= LTTNG_ERR_KERN_CONSUMER_FAIL
;
3139 uatomic_set(&kernel_consumerd_state
, CONSUMER_STARTED
);
3141 pthread_mutex_unlock(&kconsumer_data
.pid_mutex
);
3145 * The consumer was just spawned so we need to add the socket to
3146 * the consumer output of the session if exist.
3148 ret
= consumer_create_socket(&kconsumer_data
,
3149 cmd_ctx
->session
->kernel_session
->consumer
);
3156 case LTTNG_DOMAIN_JUL
:
3157 case LTTNG_DOMAIN_LOG4J
:
3158 case LTTNG_DOMAIN_PYTHON
:
3159 case LTTNG_DOMAIN_UST
:
3161 if (!ust_app_supported()) {
3162 ret
= LTTNG_ERR_NO_UST
;
3165 /* Consumer is in an ERROR state. Report back to client */
3166 if (uatomic_read(&ust_consumerd_state
) == CONSUMER_ERROR
) {
3167 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3171 if (need_tracing_session
) {
3172 /* Create UST session if none exist. */
3173 if (cmd_ctx
->session
->ust_session
== NULL
) {
3174 ret
= create_ust_session(cmd_ctx
->session
,
3175 &cmd_ctx
->lsm
->domain
);
3176 if (ret
!= LTTNG_OK
) {
3181 /* Start the UST consumer daemons */
3183 pthread_mutex_lock(&ustconsumer64_data
.pid_mutex
);
3184 if (config
.consumerd64_bin_path
.value
&&
3185 ustconsumer64_data
.pid
== 0 &&
3186 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3187 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3188 ret
= start_consumerd(&ustconsumer64_data
);
3190 ret
= LTTNG_ERR_UST_CONSUMER64_FAIL
;
3191 uatomic_set(&ust_consumerd64_fd
, -EINVAL
);
3195 uatomic_set(&ust_consumerd64_fd
, ustconsumer64_data
.cmd_sock
);
3196 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3198 pthread_mutex_unlock(&ustconsumer64_data
.pid_mutex
);
3202 * Setup socket for consumer 64 bit. No need for atomic access
3203 * since it was set above and can ONLY be set in this thread.
3205 ret
= consumer_create_socket(&ustconsumer64_data
,
3206 cmd_ctx
->session
->ust_session
->consumer
);
3212 pthread_mutex_lock(&ustconsumer32_data
.pid_mutex
);
3213 if (config
.consumerd32_bin_path
.value
&&
3214 ustconsumer32_data
.pid
== 0 &&
3215 cmd_ctx
->lsm
->cmd_type
!= LTTNG_REGISTER_CONSUMER
) {
3216 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3217 ret
= start_consumerd(&ustconsumer32_data
);
3219 ret
= LTTNG_ERR_UST_CONSUMER32_FAIL
;
3220 uatomic_set(&ust_consumerd32_fd
, -EINVAL
);
3224 uatomic_set(&ust_consumerd32_fd
, ustconsumer32_data
.cmd_sock
);
3225 uatomic_set(&ust_consumerd_state
, CONSUMER_STARTED
);
3227 pthread_mutex_unlock(&ustconsumer32_data
.pid_mutex
);
3231 * Setup socket for consumer 64 bit. No need for atomic access
3232 * since it was set above and can ONLY be set in this thread.
3234 ret
= consumer_create_socket(&ustconsumer32_data
,
3235 cmd_ctx
->session
->ust_session
->consumer
);
3247 /* Validate consumer daemon state when start/stop trace command */
3248 if (cmd_ctx
->lsm
->cmd_type
== LTTNG_START_TRACE
||
3249 cmd_ctx
->lsm
->cmd_type
== LTTNG_STOP_TRACE
) {
3250 switch (cmd_ctx
->lsm
->domain
.type
) {
3251 case LTTNG_DOMAIN_NONE
:
3253 case LTTNG_DOMAIN_JUL
:
3254 case LTTNG_DOMAIN_LOG4J
:
3255 case LTTNG_DOMAIN_PYTHON
:
3256 case LTTNG_DOMAIN_UST
:
3257 if (uatomic_read(&ust_consumerd_state
) != CONSUMER_STARTED
) {
3258 ret
= LTTNG_ERR_NO_USTCONSUMERD
;
3262 case LTTNG_DOMAIN_KERNEL
:
3263 if (uatomic_read(&kernel_consumerd_state
) != CONSUMER_STARTED
) {
3264 ret
= LTTNG_ERR_NO_KERNCONSUMERD
;
3269 ret
= LTTNG_ERR_UNKNOWN_DOMAIN
;
3275 * Check that the UID or GID match that of the tracing session.
3276 * The root user can interact with all sessions.
3278 if (need_tracing_session
) {
3279 if (!session_access_ok(cmd_ctx
->session
,
3280 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3281 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
))) {
3282 ret
= LTTNG_ERR_EPERM
;
3288 * Send relayd information to consumer as soon as we have a domain and a
3291 if (cmd_ctx
->session
&& need_domain
) {
3293 * Setup relayd if not done yet. If the relayd information was already
3294 * sent to the consumer, this call will gracefully return.
3296 ret
= cmd_setup_relayd(cmd_ctx
->session
);
3297 if (ret
!= LTTNG_OK
) {
3302 /* Process by command type */
3303 switch (cmd_ctx
->lsm
->cmd_type
) {
3304 case LTTNG_ADD_CONTEXT
:
3307 * An LTTNG_ADD_CONTEXT command might have a supplementary
3308 * payload if the context being added is an application context.
3310 if (cmd_ctx
->lsm
->u
.context
.ctx
.ctx
==
3311 LTTNG_EVENT_CONTEXT_APP_CONTEXT
) {
3312 char *provider_name
= NULL
, *context_name
= NULL
;
3313 size_t provider_name_len
=
3314 cmd_ctx
->lsm
->u
.context
.provider_name_len
;
3315 size_t context_name_len
=
3316 cmd_ctx
->lsm
->u
.context
.context_name_len
;
3318 if (provider_name_len
== 0 || context_name_len
== 0) {
3320 * Application provider and context names MUST
3323 ret
= -LTTNG_ERR_INVALID
;
3327 provider_name
= zmalloc(provider_name_len
+ 1);
3328 if (!provider_name
) {
3329 ret
= -LTTNG_ERR_NOMEM
;
3332 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
=
3335 context_name
= zmalloc(context_name_len
+ 1);
3336 if (!context_name
) {
3337 ret
= -LTTNG_ERR_NOMEM
;
3338 goto error_add_context
;
3340 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
=
3343 ret
= lttcomm_recv_unix_sock(sock
, provider_name
,
3346 goto error_add_context
;
3349 ret
= lttcomm_recv_unix_sock(sock
, context_name
,
3352 goto error_add_context
;
3357 * cmd_add_context assumes ownership of the provider and context
3360 ret
= cmd_add_context(cmd_ctx
->session
,
3361 cmd_ctx
->lsm
->domain
.type
,
3362 cmd_ctx
->lsm
->u
.context
.channel_name
,
3363 &cmd_ctx
->lsm
->u
.context
.ctx
,
3364 kernel_poll_pipe
[1]);
3366 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
= NULL
;
3367 cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
= NULL
;
3369 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.provider_name
);
3370 free(cmd_ctx
->lsm
->u
.context
.ctx
.u
.app_ctx
.ctx_name
);
3376 case LTTNG_DISABLE_CHANNEL
:
3378 ret
= cmd_disable_channel(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3379 cmd_ctx
->lsm
->u
.disable
.channel_name
);
3382 case LTTNG_DISABLE_EVENT
:
3386 * FIXME: handle filter; for now we just receive the filter's
3387 * bytecode along with the filter expression which are sent by
3388 * liblttng-ctl and discard them.
3390 * This fixes an issue where the client may block while sending
3391 * the filter payload and encounter an error because the session
3392 * daemon closes the socket without ever handling this data.
3394 size_t count
= cmd_ctx
->lsm
->u
.disable
.expression_len
+
3395 cmd_ctx
->lsm
->u
.disable
.bytecode_len
;
3398 char data
[LTTNG_FILTER_MAX_LEN
];
3400 DBG("Discarding disable event command payload of size %zu", count
);
3402 ret
= lttcomm_recv_unix_sock(sock
, data
,
3403 count
> sizeof(data
) ? sizeof(data
) : count
);
3408 count
-= (size_t) ret
;
3411 /* FIXME: passing packed structure to non-packed pointer */
3412 ret
= cmd_disable_event(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3413 cmd_ctx
->lsm
->u
.disable
.channel_name
,
3414 &cmd_ctx
->lsm
->u
.disable
.event
);
3417 case LTTNG_ENABLE_CHANNEL
:
3419 cmd_ctx
->lsm
->u
.channel
.chan
.attr
.extended
.ptr
=
3420 (struct lttng_channel_extended
*) &cmd_ctx
->lsm
->u
.channel
.extended
;
3421 ret
= cmd_enable_channel(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3422 &cmd_ctx
->lsm
->u
.channel
.chan
,
3423 kernel_poll_pipe
[1]);
3426 case LTTNG_TRACK_PID
:
3428 ret
= cmd_track_pid(cmd_ctx
->session
,
3429 cmd_ctx
->lsm
->domain
.type
,
3430 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3433 case LTTNG_UNTRACK_PID
:
3435 ret
= cmd_untrack_pid(cmd_ctx
->session
,
3436 cmd_ctx
->lsm
->domain
.type
,
3437 cmd_ctx
->lsm
->u
.pid_tracker
.pid
);
3440 case LTTNG_ENABLE_EVENT
:
3442 struct lttng_event_exclusion
*exclusion
= NULL
;
3443 struct lttng_filter_bytecode
*bytecode
= NULL
;
3444 char *filter_expression
= NULL
;
3446 /* Handle exclusion events and receive it from the client. */
3447 if (cmd_ctx
->lsm
->u
.enable
.exclusion_count
> 0) {
3448 size_t count
= cmd_ctx
->lsm
->u
.enable
.exclusion_count
;
3450 exclusion
= zmalloc(sizeof(struct lttng_event_exclusion
) +
3451 (count
* LTTNG_SYMBOL_NAME_LEN
));
3453 ret
= LTTNG_ERR_EXCLUSION_NOMEM
;
3457 DBG("Receiving var len exclusion event list from client ...");
3458 exclusion
->count
= count
;
3459 ret
= lttcomm_recv_unix_sock(sock
, exclusion
->names
,
3460 count
* LTTNG_SYMBOL_NAME_LEN
);
3462 DBG("Nothing recv() from client var len data... continuing");
3465 ret
= LTTNG_ERR_EXCLUSION_INVAL
;
3470 /* Get filter expression from client. */
3471 if (cmd_ctx
->lsm
->u
.enable
.expression_len
> 0) {
3472 size_t expression_len
=
3473 cmd_ctx
->lsm
->u
.enable
.expression_len
;
3475 if (expression_len
> LTTNG_FILTER_MAX_LEN
) {
3476 ret
= LTTNG_ERR_FILTER_INVAL
;
3481 filter_expression
= zmalloc(expression_len
);
3482 if (!filter_expression
) {
3484 ret
= LTTNG_ERR_FILTER_NOMEM
;
3488 /* Receive var. len. data */
3489 DBG("Receiving var len filter's expression from client ...");
3490 ret
= lttcomm_recv_unix_sock(sock
, filter_expression
,
3493 DBG("Nothing recv() from client car len data... continuing");
3495 free(filter_expression
);
3497 ret
= LTTNG_ERR_FILTER_INVAL
;
3502 /* Handle filter and get bytecode from client. */
3503 if (cmd_ctx
->lsm
->u
.enable
.bytecode_len
> 0) {
3504 size_t bytecode_len
= cmd_ctx
->lsm
->u
.enable
.bytecode_len
;
3506 if (bytecode_len
> LTTNG_FILTER_MAX_LEN
) {
3507 ret
= LTTNG_ERR_FILTER_INVAL
;
3508 free(filter_expression
);
3513 bytecode
= zmalloc(bytecode_len
);
3515 free(filter_expression
);
3517 ret
= LTTNG_ERR_FILTER_NOMEM
;
3521 /* Receive var. len. data */
3522 DBG("Receiving var len filter's bytecode from client ...");
3523 ret
= lttcomm_recv_unix_sock(sock
, bytecode
, bytecode_len
);
3525 DBG("Nothing recv() from client car len data... continuing");
3527 free(filter_expression
);
3530 ret
= LTTNG_ERR_FILTER_INVAL
;
3534 if ((bytecode
->len
+ sizeof(*bytecode
)) != bytecode_len
) {
3535 free(filter_expression
);
3538 ret
= LTTNG_ERR_FILTER_INVAL
;
3543 ret
= cmd_enable_event(cmd_ctx
->session
, &cmd_ctx
->lsm
->domain
,
3544 cmd_ctx
->lsm
->u
.enable
.channel_name
,
3545 &cmd_ctx
->lsm
->u
.enable
.event
,
3546 filter_expression
, bytecode
, exclusion
,
3547 kernel_poll_pipe
[1]);
3550 case LTTNG_LIST_TRACEPOINTS
:
3552 struct lttng_event
*events
;
3555 session_lock_list();
3556 nb_events
= cmd_list_tracepoints(cmd_ctx
->lsm
->domain
.type
, &events
);
3557 session_unlock_list();
3558 if (nb_events
< 0) {
3559 /* Return value is a negative lttng_error_code. */
3565 * Setup lttng message with payload size set to the event list size in
3566 * bytes and then copy list into the llm payload.
3568 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3569 sizeof(struct lttng_event
) * nb_events
);
3579 case LTTNG_LIST_TRACEPOINT_FIELDS
:
3581 struct lttng_event_field
*fields
;
3584 session_lock_list();
3585 nb_fields
= cmd_list_tracepoint_fields(cmd_ctx
->lsm
->domain
.type
,
3587 session_unlock_list();
3588 if (nb_fields
< 0) {
3589 /* Return value is a negative lttng_error_code. */
3595 * Setup lttng message with payload size set to the event list size in
3596 * bytes and then copy list into the llm payload.
3598 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, fields
,
3599 sizeof(struct lttng_event_field
) * nb_fields
);
3609 case LTTNG_LIST_SYSCALLS
:
3611 struct lttng_event
*events
;
3614 nb_events
= cmd_list_syscalls(&events
);
3615 if (nb_events
< 0) {
3616 /* Return value is a negative lttng_error_code. */
3622 * Setup lttng message with payload size set to the event list size in
3623 * bytes and then copy list into the llm payload.
3625 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, events
,
3626 sizeof(struct lttng_event
) * nb_events
);
3636 case LTTNG_LIST_TRACKER_PIDS
:
3638 int32_t *pids
= NULL
;
3641 nr_pids
= cmd_list_tracker_pids(cmd_ctx
->session
,
3642 cmd_ctx
->lsm
->domain
.type
, &pids
);
3644 /* Return value is a negative lttng_error_code. */
3650 * Setup lttng message with payload size set to the event list size in
3651 * bytes and then copy list into the llm payload.
3653 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, pids
,
3654 sizeof(int32_t) * nr_pids
);
3664 case LTTNG_SET_CONSUMER_URI
:
3667 struct lttng_uri
*uris
;
3669 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3670 len
= nb_uri
* sizeof(struct lttng_uri
);
3673 ret
= LTTNG_ERR_INVALID
;
3677 uris
= zmalloc(len
);
3679 ret
= LTTNG_ERR_FATAL
;
3683 /* Receive variable len data */
3684 DBG("Receiving %zu URI(s) from client ...", nb_uri
);
3685 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3687 DBG("No URIs received from client... continuing");
3689 ret
= LTTNG_ERR_SESSION_FAIL
;
3694 ret
= cmd_set_consumer_uri(cmd_ctx
->session
, nb_uri
, uris
);
3696 if (ret
!= LTTNG_OK
) {
3703 case LTTNG_START_TRACE
:
3706 * On the first start, if we have a kernel session and we have
3707 * enabled time or size-based rotations, we have to make sure
3708 * the kernel tracer supports it.
3710 if (!cmd_ctx
->session
->has_been_started
&& \
3711 cmd_ctx
->session
->kernel_session
&& \
3712 (cmd_ctx
->session
->rotate_timer_period
|| \
3713 cmd_ctx
->session
->rotate_size
) && \
3714 !check_rotate_compatible()) {
3715 DBG("Kernel tracer version is not compatible with the rotation feature");
3716 ret
= LTTNG_ERR_ROTATION_WRONG_VERSION
;
3719 ret
= cmd_start_trace(cmd_ctx
->session
);
3722 case LTTNG_STOP_TRACE
:
3724 ret
= cmd_stop_trace(cmd_ctx
->session
);
3727 case LTTNG_CREATE_SESSION
:
3730 struct lttng_uri
*uris
= NULL
;
3732 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
3733 len
= nb_uri
* sizeof(struct lttng_uri
);
3736 uris
= zmalloc(len
);
3738 ret
= LTTNG_ERR_FATAL
;
3742 /* Receive variable len data */
3743 DBG("Waiting for %zu URIs from client ...", nb_uri
);
3744 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
3746 DBG("No URIs received from client... continuing");
3748 ret
= LTTNG_ERR_SESSION_FAIL
;
3753 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
3754 DBG("Creating session with ONE network URI is a bad call");
3755 ret
= LTTNG_ERR_SESSION_FAIL
;
3761 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
, nb_uri
,
3762 &cmd_ctx
->creds
, 0);
3768 case LTTNG_DESTROY_SESSION
:
3770 ret
= cmd_destroy_session(cmd_ctx
->session
, kernel_poll_pipe
[1]);
3772 /* Set session to NULL so we do not unlock it after free. */
3773 cmd_ctx
->session
= NULL
;
3776 case LTTNG_LIST_DOMAINS
:
3779 struct lttng_domain
*domains
= NULL
;
3781 nb_dom
= cmd_list_domains(cmd_ctx
->session
, &domains
);
3783 /* Return value is a negative lttng_error_code. */
3788 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, domains
,
3789 nb_dom
* sizeof(struct lttng_domain
));
3799 case LTTNG_LIST_CHANNELS
:
3801 ssize_t payload_size
;
3802 struct lttng_channel
*channels
= NULL
;
3804 payload_size
= cmd_list_channels(cmd_ctx
->lsm
->domain
.type
,
3805 cmd_ctx
->session
, &channels
);
3806 if (payload_size
< 0) {
3807 /* Return value is a negative lttng_error_code. */
3808 ret
= -payload_size
;
3812 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, channels
,
3823 case LTTNG_LIST_EVENTS
:
3826 struct lttng_event
*events
= NULL
;
3827 struct lttcomm_event_command_header cmd_header
;
3830 memset(&cmd_header
, 0, sizeof(cmd_header
));
3831 /* Extended infos are included at the end of events */
3832 nb_event
= cmd_list_events(cmd_ctx
->lsm
->domain
.type
,
3833 cmd_ctx
->session
, cmd_ctx
->lsm
->u
.list
.channel_name
,
3834 &events
, &total_size
);
3837 /* Return value is a negative lttng_error_code. */
3842 cmd_header
.nb_events
= nb_event
;
3843 ret
= setup_lttng_msg(cmd_ctx
, events
, total_size
,
3844 &cmd_header
, sizeof(cmd_header
));
3854 case LTTNG_LIST_SESSIONS
:
3856 unsigned int nr_sessions
;
3857 void *sessions_payload
;
3860 session_lock_list();
3861 nr_sessions
= lttng_sessions_count(
3862 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3863 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3864 payload_len
= sizeof(struct lttng_session
) * nr_sessions
;
3865 sessions_payload
= zmalloc(payload_len
);
3867 if (!sessions_payload
) {
3868 session_unlock_list();
3873 cmd_list_lttng_sessions(sessions_payload
,
3874 LTTNG_SOCK_GET_UID_CRED(&cmd_ctx
->creds
),
3875 LTTNG_SOCK_GET_GID_CRED(&cmd_ctx
->creds
));
3876 session_unlock_list();
3878 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, sessions_payload
,
3880 free(sessions_payload
);
3889 case LTTNG_REGISTER_CONSUMER
:
3891 struct consumer_data
*cdata
;
3893 switch (cmd_ctx
->lsm
->domain
.type
) {
3894 case LTTNG_DOMAIN_KERNEL
:
3895 cdata
= &kconsumer_data
;
3898 ret
= LTTNG_ERR_UND
;
3902 ret
= cmd_register_consumer(cmd_ctx
->session
, cmd_ctx
->lsm
->domain
.type
,
3903 cmd_ctx
->lsm
->u
.reg
.path
, cdata
);
3906 case LTTNG_DATA_PENDING
:
3909 uint8_t pending_ret_byte
;
3911 pending_ret
= cmd_data_pending(cmd_ctx
->session
);
3916 * This function may returns 0 or 1 to indicate whether or not
3917 * there is data pending. In case of error, it should return an
3918 * LTTNG_ERR code. However, some code paths may still return
3919 * a nondescript error code, which we handle by returning an
3922 if (pending_ret
== 0 || pending_ret
== 1) {
3924 * ret will be set to LTTNG_OK at the end of
3927 } else if (pending_ret
< 0) {
3928 ret
= LTTNG_ERR_UNK
;
3935 pending_ret_byte
= (uint8_t) pending_ret
;
3937 /* 1 byte to return whether or not data is pending */
3938 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
,
3939 &pending_ret_byte
, 1);
3948 case LTTNG_SNAPSHOT_ADD_OUTPUT
:
3950 struct lttcomm_lttng_output_id reply
;
3952 ret
= cmd_snapshot_add_output(cmd_ctx
->session
,
3953 &cmd_ctx
->lsm
->u
.snapshot_output
.output
, &reply
.id
);
3954 if (ret
!= LTTNG_OK
) {
3958 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &reply
,
3964 /* Copy output list into message payload */
3968 case LTTNG_SNAPSHOT_DEL_OUTPUT
:
3970 ret
= cmd_snapshot_del_output(cmd_ctx
->session
,
3971 &cmd_ctx
->lsm
->u
.snapshot_output
.output
);
3974 case LTTNG_SNAPSHOT_LIST_OUTPUT
:
3977 struct lttng_snapshot_output
*outputs
= NULL
;
3979 nb_output
= cmd_snapshot_list_outputs(cmd_ctx
->session
, &outputs
);
3980 if (nb_output
< 0) {
3985 assert((nb_output
> 0 && outputs
) || nb_output
== 0);
3986 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, outputs
,
3987 nb_output
* sizeof(struct lttng_snapshot_output
));
3997 case LTTNG_SNAPSHOT_RECORD
:
3999 ret
= cmd_snapshot_record(cmd_ctx
->session
,
4000 &cmd_ctx
->lsm
->u
.snapshot_record
.output
,
4001 cmd_ctx
->lsm
->u
.snapshot_record
.wait
);
4004 case LTTNG_CREATE_SESSION_SNAPSHOT
:
4007 struct lttng_uri
*uris
= NULL
;
4009 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4010 len
= nb_uri
* sizeof(struct lttng_uri
);
4013 uris
= zmalloc(len
);
4015 ret
= LTTNG_ERR_FATAL
;
4019 /* Receive variable len data */
4020 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4021 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4023 DBG("No URIs received from client... continuing");
4025 ret
= LTTNG_ERR_SESSION_FAIL
;
4030 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4031 DBG("Creating session with ONE network URI is a bad call");
4032 ret
= LTTNG_ERR_SESSION_FAIL
;
4038 ret
= cmd_create_session_snapshot(cmd_ctx
->lsm
->session
.name
, uris
,
4039 nb_uri
, &cmd_ctx
->creds
);
4043 case LTTNG_CREATE_SESSION_LIVE
:
4046 struct lttng_uri
*uris
= NULL
;
4048 nb_uri
= cmd_ctx
->lsm
->u
.uri
.size
;
4049 len
= nb_uri
* sizeof(struct lttng_uri
);
4052 uris
= zmalloc(len
);
4054 ret
= LTTNG_ERR_FATAL
;
4058 /* Receive variable len data */
4059 DBG("Waiting for %zu URIs from client ...", nb_uri
);
4060 ret
= lttcomm_recv_unix_sock(sock
, uris
, len
);
4062 DBG("No URIs received from client... continuing");
4064 ret
= LTTNG_ERR_SESSION_FAIL
;
4069 if (nb_uri
== 1 && uris
[0].dtype
!= LTTNG_DST_PATH
) {
4070 DBG("Creating session with ONE network URI is a bad call");
4071 ret
= LTTNG_ERR_SESSION_FAIL
;
4077 ret
= cmd_create_session_uri(cmd_ctx
->lsm
->session
.name
, uris
,
4078 nb_uri
, &cmd_ctx
->creds
, cmd_ctx
->lsm
->u
.session_live
.timer_interval
);
4082 case LTTNG_SAVE_SESSION
:
4084 ret
= cmd_save_sessions(&cmd_ctx
->lsm
->u
.save_session
.attr
,
4088 case LTTNG_SET_SESSION_SHM_PATH
:
4090 ret
= cmd_set_session_shm_path(cmd_ctx
->session
,
4091 cmd_ctx
->lsm
->u
.set_shm_path
.shm_path
);
4094 case LTTNG_REGENERATE_METADATA
:
4096 ret
= cmd_regenerate_metadata(cmd_ctx
->session
);
4099 case LTTNG_REGENERATE_STATEDUMP
:
4101 ret
= cmd_regenerate_statedump(cmd_ctx
->session
);
4104 case LTTNG_REGISTER_TRIGGER
:
4106 ret
= cmd_register_trigger(cmd_ctx
, sock
,
4107 notification_thread_handle
);
4110 case LTTNG_UNREGISTER_TRIGGER
:
4112 ret
= cmd_unregister_trigger(cmd_ctx
, sock
,
4113 notification_thread_handle
);
4116 case LTTNG_ROTATE_SESSION
:
4118 struct lttng_rotate_session_return rotate_return
;
4120 DBG("Client rotate session \"%s\"", cmd_ctx
->session
->name
);
4122 memset(&rotate_return
, 0, sizeof(rotate_return
));
4123 if (cmd_ctx
->session
->kernel_session
&& !check_rotate_compatible()) {
4124 DBG("Kernel tracer version is not compatible with the rotation feature");
4125 ret
= LTTNG_ERR_ROTATION_WRONG_VERSION
;
4129 ret
= cmd_rotate_session(cmd_ctx
->session
, &rotate_return
);
4135 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &rotate_return
,
4136 sizeof(rotate_return
));
4145 case LTTNG_ROTATION_GET_INFO
:
4147 struct lttng_rotation_get_info_return get_info_return
;
4149 memset(&get_info_return
, 0, sizeof(get_info_return
));
4150 ret
= cmd_rotate_get_info(cmd_ctx
->session
, &get_info_return
,
4151 cmd_ctx
->lsm
->u
.get_rotation_info
.rotation_id
);
4157 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &get_info_return
,
4158 sizeof(get_info_return
));
4167 case LTTNG_SESSION_GET_CURRENT_OUTPUT
:
4169 struct lttng_session_get_current_output_return output_return
;
4171 memset(&output_return
, 0, sizeof(output_return
));
4172 ret
= cmd_session_get_current_output(cmd_ctx
->session
,
4179 ret
= setup_lttng_msg_no_cmd_header(cmd_ctx
, &output_return
,
4180 sizeof(output_return
));
4189 case LTTNG_ROTATION_SET_SCHEDULE
:
4191 if (cmd_ctx
->session
->kernel_session
&& !check_rotate_compatible()) {
4192 DBG("Kernel tracer version does not support session rotations");
4193 ret
= LTTNG_ERR_ROTATION_WRONG_VERSION
;
4197 ret
= cmd_rotation_set_schedule(cmd_ctx
->session
,
4198 cmd_ctx
->lsm
->u
.rotate_setup
.timer_us
,
4199 cmd_ctx
->lsm
->u
.rotate_setup
.size
);
4209 ret
= LTTNG_ERR_UND
;
4214 if (cmd_ctx
->llm
== NULL
) {
4215 DBG("Missing llm structure. Allocating one.");
4216 if (setup_lttng_msg_no_cmd_header(cmd_ctx
, NULL
, 0) < 0) {
4220 /* Set return code */
4221 cmd_ctx
->llm
->ret_code
= ret
;
4223 if (cmd_ctx
->session
) {
4224 session_unlock(cmd_ctx
->session
);
4226 if (need_tracing_session
) {
4227 session_unlock_list();
4230 assert(!rcu_read_ongoing());
4235 * Thread managing health check socket.
4237 static void *thread_manage_health(void *data
)
4239 int sock
= -1, new_sock
= -1, ret
, i
, pollfd
, err
= -1;
4240 uint32_t revents
, nb_fd
;
4241 struct lttng_poll_event events
;
4242 struct health_comm_msg msg
;
4243 struct health_comm_reply reply
;
4245 DBG("[thread] Manage health check started");
4247 rcu_register_thread();
4249 /* We might hit an error path before this is created. */
4250 lttng_poll_init(&events
);
4252 /* Create unix socket */
4253 sock
= lttcomm_create_unix_sock(config
.health_unix_sock_path
.value
);
4255 ERR("Unable to create health check Unix socket");
4260 /* lttng health client socket path permissions */
4261 ret
= chown(config
.health_unix_sock_path
.value
, 0,
4262 utils_get_group_id(config
.tracing_group_name
.value
));
4264 ERR("Unable to set group on %s", config
.health_unix_sock_path
.value
);
4269 ret
= chmod(config
.health_unix_sock_path
.value
,
4270 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
4272 ERR("Unable to set permissions on %s", config
.health_unix_sock_path
.value
);
4279 * Set the CLOEXEC flag. Return code is useless because either way, the
4282 (void) utils_set_fd_cloexec(sock
);
4284 ret
= lttcomm_listen_unix_sock(sock
);
4290 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4291 * more will be added to this poll set.
4293 ret
= sessiond_set_thread_pollset(&events
, 2);
4298 /* Add the application registration socket */
4299 ret
= lttng_poll_add(&events
, sock
, LPOLLIN
| LPOLLPRI
);
4304 sessiond_notify_ready();
4307 DBG("Health check ready");
4309 /* Inifinite blocking call, waiting for transmission */
4311 ret
= lttng_poll_wait(&events
, -1);
4314 * Restart interrupted system call.
4316 if (errno
== EINTR
) {
4324 for (i
= 0; i
< nb_fd
; i
++) {
4325 /* Fetch once the poll data */
4326 revents
= LTTNG_POLL_GETEV(&events
, i
);
4327 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4330 /* No activity for this FD (poll implementation). */
4334 /* Thread quit pipe has been closed. Killing thread. */
4335 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4341 /* Event on the registration socket */
4342 if (pollfd
== sock
) {
4343 if (revents
& LPOLLIN
) {
4345 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4346 ERR("Health socket poll error");
4349 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4355 new_sock
= lttcomm_accept_unix_sock(sock
);
4361 * Set the CLOEXEC flag. Return code is useless because either way, the
4364 (void) utils_set_fd_cloexec(new_sock
);
4366 DBG("Receiving data from client for health...");
4367 ret
= lttcomm_recv_unix_sock(new_sock
, (void *)&msg
, sizeof(msg
));
4369 DBG("Nothing recv() from client... continuing");
4370 ret
= close(new_sock
);
4377 rcu_thread_online();
4379 memset(&reply
, 0, sizeof(reply
));
4380 for (i
= 0; i
< NR_HEALTH_SESSIOND_TYPES
; i
++) {
4382 * health_check_state returns 0 if health is
4385 if (!health_check_state(health_sessiond
, i
)) {
4386 reply
.ret_code
|= 1ULL << i
;
4390 DBG2("Health check return value %" PRIx64
, reply
.ret_code
);
4392 ret
= send_unix_sock(new_sock
, (void *) &reply
, sizeof(reply
));
4394 ERR("Failed to send health data back to client");
4397 /* End of transmission */
4398 ret
= close(new_sock
);
4407 ERR("Health error occurred in %s", __func__
);
4409 DBG("Health check thread dying");
4410 unlink(config
.health_unix_sock_path
.value
);
4418 lttng_poll_clean(&events
);
4420 rcu_unregister_thread();
4425 * This thread manage all clients request using the unix client socket for
4428 static void *thread_manage_clients(void *data
)
4430 int sock
= -1, ret
, i
, pollfd
, err
= -1;
4432 uint32_t revents
, nb_fd
;
4433 struct command_ctx
*cmd_ctx
= NULL
;
4434 struct lttng_poll_event events
;
4436 DBG("[thread] Manage client started");
4438 rcu_register_thread();
4440 health_register(health_sessiond
, HEALTH_SESSIOND_TYPE_CMD
);
4442 health_code_update();
4444 ret
= lttcomm_listen_unix_sock(client_sock
);
4450 * Pass 2 as size here for the thread quit pipe and client_sock. Nothing
4451 * more will be added to this poll set.
4453 ret
= sessiond_set_thread_pollset(&events
, 2);
4455 goto error_create_poll
;
4458 /* Add the application registration socket */
4459 ret
= lttng_poll_add(&events
, client_sock
, LPOLLIN
| LPOLLPRI
);
4464 sessiond_notify_ready();
4465 ret
= sem_post(&load_info
->message_thread_ready
);
4467 PERROR("sem_post message_thread_ready");
4471 /* This testpoint is after we signal readiness to the parent. */
4472 if (testpoint(sessiond_thread_manage_clients
)) {
4476 if (testpoint(sessiond_thread_manage_clients_before_loop
)) {
4480 health_code_update();
4483 DBG("Accepting client command ...");
4485 /* Inifinite blocking call, waiting for transmission */
4487 health_poll_entry();
4488 ret
= lttng_poll_wait(&events
, -1);
4492 * Restart interrupted system call.
4494 if (errno
== EINTR
) {
4502 for (i
= 0; i
< nb_fd
; i
++) {
4503 /* Fetch once the poll data */
4504 revents
= LTTNG_POLL_GETEV(&events
, i
);
4505 pollfd
= LTTNG_POLL_GETFD(&events
, i
);
4507 health_code_update();
4510 /* No activity for this FD (poll implementation). */
4514 /* Thread quit pipe has been closed. Killing thread. */
4515 ret
= sessiond_check_thread_quit_pipe(pollfd
, revents
);
4521 /* Event on the registration socket */
4522 if (pollfd
== client_sock
) {
4523 if (revents
& LPOLLIN
) {
4525 } else if (revents
& (LPOLLERR
| LPOLLHUP
| LPOLLRDHUP
)) {
4526 ERR("Client socket poll error");
4529 ERR("Unexpected poll events %u for sock %d", revents
, pollfd
);
4535 DBG("Wait for client response");
4537 health_code_update();
4539 sock
= lttcomm_accept_unix_sock(client_sock
);
4545 * Set the CLOEXEC flag. Return code is useless because either way, the
4548 (void) utils_set_fd_cloexec(sock
);
4550 /* Set socket option for credentials retrieval */
4551 ret
= lttcomm_setsockopt_creds_unix_sock(sock
);
4556 /* Allocate context command to process the client request */
4557 cmd_ctx
= zmalloc(sizeof(struct command_ctx
));
4558 if (cmd_ctx
== NULL
) {
4559 PERROR("zmalloc cmd_ctx");
4563 /* Allocate data buffer for reception */
4564 cmd_ctx
->lsm
= zmalloc(sizeof(struct lttcomm_session_msg
));
4565 if (cmd_ctx
->lsm
== NULL
) {
4566 PERROR("zmalloc cmd_ctx->lsm");
4570 cmd_ctx
->llm
= NULL
;
4571 cmd_ctx
->session
= NULL
;
4573 health_code_update();
4576 * Data is received from the lttng client. The struct
4577 * lttcomm_session_msg (lsm) contains the command and data request of
4580 DBG("Receiving data from client ...");
4581 ret
= lttcomm_recv_creds_unix_sock(sock
, cmd_ctx
->lsm
,
4582 sizeof(struct lttcomm_session_msg
), &cmd_ctx
->creds
);
4584 DBG("Nothing recv() from client... continuing");
4590 clean_command_ctx(&cmd_ctx
);
4594 health_code_update();
4596 // TODO: Validate cmd_ctx including sanity check for
4597 // security purpose.
4599 rcu_thread_online();
4601 * This function dispatch the work to the kernel or userspace tracer
4602 * libs and fill the lttcomm_lttng_msg data structure of all the needed
4603 * informations for the client. The command context struct contains
4604 * everything this function may needs.
4606 ret
= process_client_msg(cmd_ctx
, sock
, &sock_error
);
4607 rcu_thread_offline();
4615 * TODO: Inform client somehow of the fatal error. At
4616 * this point, ret < 0 means that a zmalloc failed
4617 * (ENOMEM). Error detected but still accept
4618 * command, unless a socket error has been
4621 clean_command_ctx(&cmd_ctx
);
4625 health_code_update();
4627 DBG("Sending response (size: %d, retcode: %s (%d))",
4628 cmd_ctx
->lttng_msg_size
,
4629 lttng_strerror(-cmd_ctx
->llm
->ret_code
),
4630 cmd_ctx
->llm
->ret_code
);
4631 ret
= send_unix_sock(sock
, cmd_ctx
->llm
, cmd_ctx
->lttng_msg_size
);
4633 ERR("Failed to send data back to client");
4636 /* End of transmission */
4643 clean_command_ctx(&cmd_ctx
);
4645 health_code_update();
4657 lttng_poll_clean(&events
);
4658 clean_command_ctx(&cmd_ctx
);
4662 unlink(config
.client_unix_sock_path
.value
);
4663 if (client_sock
>= 0) {
4664 ret
= close(client_sock
);
4672 ERR("Health error occurred in %s", __func__
);
4675 health_unregister(health_sessiond
);
4677 DBG("Client thread dying");
4679 rcu_unregister_thread();
4682 * Since we are creating the consumer threads, we own them, so we need
4683 * to join them before our thread exits.
4685 ret
= join_consumer_thread(&kconsumer_data
);
4688 PERROR("join_consumer");
4691 ret
= join_consumer_thread(&ustconsumer32_data
);
4694 PERROR("join_consumer ust32");
4697 ret
= join_consumer_thread(&ustconsumer64_data
);
4700 PERROR("join_consumer ust64");
4705 static int string_match(const char *str1
, const char *str2
)
4707 return (str1
&& str2
) && !strcmp(str1
, str2
);
4711 * Take an option from the getopt output and set it in the right variable to be
4714 * Return 0 on success else a negative value.
4716 static int set_option(int opt
, const char *arg
, const char *optname
)
4720 if (string_match(optname
, "client-sock") || opt
== 'c') {
4721 if (!arg
|| *arg
== '\0') {
4725 if (lttng_is_setuid_setgid()) {
4726 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4727 "-c, --client-sock");
4729 config_string_set(&config
.client_unix_sock_path
,
4731 if (!config
.client_unix_sock_path
.value
) {
4736 } else if (string_match(optname
, "apps-sock") || opt
== 'a') {
4737 if (!arg
|| *arg
== '\0') {
4741 if (lttng_is_setuid_setgid()) {
4742 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4745 config_string_set(&config
.apps_unix_sock_path
,
4747 if (!config
.apps_unix_sock_path
.value
) {
4752 } else if (string_match(optname
, "daemonize") || opt
== 'd') {
4753 config
.daemonize
= true;
4754 } else if (string_match(optname
, "background") || opt
== 'b') {
4755 config
.background
= true;
4756 } else if (string_match(optname
, "group") || opt
== 'g') {
4757 if (!arg
|| *arg
== '\0') {
4761 if (lttng_is_setuid_setgid()) {
4762 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4765 config_string_set(&config
.tracing_group_name
,
4767 if (!config
.tracing_group_name
.value
) {
4772 } else if (string_match(optname
, "help") || opt
== 'h') {
4773 ret
= utils_show_help(8, "lttng-sessiond", help_msg
);
4775 ERR("Cannot show --help for `lttng-sessiond`");
4778 exit(ret
? EXIT_FAILURE
: EXIT_SUCCESS
);
4779 } else if (string_match(optname
, "version") || opt
== 'V') {
4780 fprintf(stdout
, "%s\n", VERSION
);
4782 } else if (string_match(optname
, "sig-parent") || opt
== 'S') {
4783 config
.sig_parent
= true;
4784 } else if (string_match(optname
, "kconsumerd-err-sock")) {
4785 if (!arg
|| *arg
== '\0') {
4789 if (lttng_is_setuid_setgid()) {
4790 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4791 "--kconsumerd-err-sock");
4793 config_string_set(&config
.kconsumerd_err_unix_sock_path
,
4795 if (!config
.kconsumerd_err_unix_sock_path
.value
) {
4800 } else if (string_match(optname
, "kconsumerd-cmd-sock")) {
4801 if (!arg
|| *arg
== '\0') {
4805 if (lttng_is_setuid_setgid()) {
4806 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4807 "--kconsumerd-cmd-sock");
4809 config_string_set(&config
.kconsumerd_cmd_unix_sock_path
,
4811 if (!config
.kconsumerd_cmd_unix_sock_path
.value
) {
4816 } else if (string_match(optname
, "ustconsumerd64-err-sock")) {
4817 if (!arg
|| *arg
== '\0') {
4821 if (lttng_is_setuid_setgid()) {
4822 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4823 "--ustconsumerd64-err-sock");
4825 config_string_set(&config
.consumerd64_err_unix_sock_path
,
4827 if (!config
.consumerd64_err_unix_sock_path
.value
) {
4832 } else if (string_match(optname
, "ustconsumerd64-cmd-sock")) {
4833 if (!arg
|| *arg
== '\0') {
4837 if (lttng_is_setuid_setgid()) {
4838 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4839 "--ustconsumerd64-cmd-sock");
4841 config_string_set(&config
.consumerd64_cmd_unix_sock_path
,
4843 if (!config
.consumerd64_cmd_unix_sock_path
.value
) {
4848 } else if (string_match(optname
, "ustconsumerd32-err-sock")) {
4849 if (!arg
|| *arg
== '\0') {
4853 if (lttng_is_setuid_setgid()) {
4854 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4855 "--ustconsumerd32-err-sock");
4857 config_string_set(&config
.consumerd32_err_unix_sock_path
,
4859 if (!config
.consumerd32_err_unix_sock_path
.value
) {
4864 } else if (string_match(optname
, "ustconsumerd32-cmd-sock")) {
4865 if (!arg
|| *arg
== '\0') {
4869 if (lttng_is_setuid_setgid()) {
4870 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4871 "--ustconsumerd32-cmd-sock");
4873 config_string_set(&config
.consumerd32_cmd_unix_sock_path
,
4875 if (!config
.consumerd32_cmd_unix_sock_path
.value
) {
4880 } else if (string_match(optname
, "no-kernel")) {
4881 config
.no_kernel
= true;
4882 } else if (string_match(optname
, "quiet") || opt
== 'q') {
4883 lttng_opt_quiet
= true;
4884 } else if (string_match(optname
, "verbose") || opt
== 'v') {
4885 /* Verbose level can increase using multiple -v */
4887 /* Value obtained from config file */
4888 config
.verbose
= config_parse_value(arg
);
4890 /* -v used on command line */
4893 /* Clamp value to [0, 3] */
4894 config
.verbose
= config
.verbose
< 0 ? 0 :
4895 (config
.verbose
<= 3 ? config
.verbose
: 3);
4896 } else if (string_match(optname
, "verbose-consumer")) {
4898 config
.verbose_consumer
= config_parse_value(arg
);
4900 config
.verbose_consumer
++;
4902 } else if (string_match(optname
, "consumerd32-path")) {
4903 if (!arg
|| *arg
== '\0') {
4907 if (lttng_is_setuid_setgid()) {
4908 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4909 "--consumerd32-path");
4911 config_string_set(&config
.consumerd32_bin_path
,
4913 if (!config
.consumerd32_bin_path
.value
) {
4918 } else if (string_match(optname
, "consumerd32-libdir")) {
4919 if (!arg
|| *arg
== '\0') {
4923 if (lttng_is_setuid_setgid()) {
4924 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4925 "--consumerd32-libdir");
4927 config_string_set(&config
.consumerd32_lib_dir
,
4929 if (!config
.consumerd32_lib_dir
.value
) {
4934 } else if (string_match(optname
, "consumerd64-path")) {
4935 if (!arg
|| *arg
== '\0') {
4939 if (lttng_is_setuid_setgid()) {
4940 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4941 "--consumerd64-path");
4943 config_string_set(&config
.consumerd64_bin_path
,
4945 if (!config
.consumerd64_bin_path
.value
) {
4950 } else if (string_match(optname
, "consumerd64-libdir")) {
4951 if (!arg
|| *arg
== '\0') {
4955 if (lttng_is_setuid_setgid()) {
4956 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4957 "--consumerd64-libdir");
4959 config_string_set(&config
.consumerd64_lib_dir
,
4961 if (!config
.consumerd64_lib_dir
.value
) {
4966 } else if (string_match(optname
, "pidfile") || opt
== 'p') {
4967 if (!arg
|| *arg
== '\0') {
4971 if (lttng_is_setuid_setgid()) {
4972 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4975 config_string_set(&config
.pid_file_path
, strdup(arg
));
4976 if (!config
.pid_file_path
.value
) {
4981 } else if (string_match(optname
, "agent-tcp-port")) {
4982 if (!arg
|| *arg
== '\0') {
4986 if (lttng_is_setuid_setgid()) {
4987 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
4988 "--agent-tcp-port");
4993 v
= strtoul(arg
, NULL
, 0);
4994 if (errno
!= 0 || !isdigit(arg
[0])) {
4995 ERR("Wrong value in --agent-tcp-port parameter: %s", arg
);
4998 if (v
== 0 || v
>= 65535) {
4999 ERR("Port overflow in --agent-tcp-port parameter: %s", arg
);
5002 config
.agent_tcp_port
= (uint32_t) v
;
5003 DBG3("Agent TCP port set to non default: %u", config
.agent_tcp_port
);
5005 } else if (string_match(optname
, "load") || opt
== 'l') {
5006 if (!arg
|| *arg
== '\0') {
5010 if (lttng_is_setuid_setgid()) {
5011 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5014 config_string_set(&config
.load_session_path
, strdup(arg
));
5015 if (!config
.load_session_path
.value
) {
5020 } else if (string_match(optname
, "kmod-probes")) {
5021 if (!arg
|| *arg
== '\0') {
5025 if (lttng_is_setuid_setgid()) {
5026 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5029 config_string_set(&config
.kmod_probes_list
, strdup(arg
));
5030 if (!config
.kmod_probes_list
.value
) {
5035 } else if (string_match(optname
, "extra-kmod-probes")) {
5036 if (!arg
|| *arg
== '\0') {
5040 if (lttng_is_setuid_setgid()) {
5041 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5042 "--extra-kmod-probes");
5044 config_string_set(&config
.kmod_extra_probes_list
,
5046 if (!config
.kmod_extra_probes_list
.value
) {
5051 } else if (string_match(optname
, "config") || opt
== 'f') {
5052 /* This is handled in set_options() thus silent skip. */
5055 /* Unknown option or other error.
5056 * Error is printed by getopt, just return */
5061 if (ret
== -EINVAL
) {
5062 const char *opt_name
= "unknown";
5065 for (i
= 0; i
< sizeof(long_options
) / sizeof(struct option
);
5067 if (opt
== long_options
[i
].val
) {
5068 opt_name
= long_options
[i
].name
;
5073 WARN("Invalid argument provided for option \"%s\", using default value.",
5081 * config_entry_handler_cb used to handle options read from a config file.
5082 * See config_entry_handler_cb comment in common/config/session-config.h for the
5083 * return value conventions.
5085 static int config_entry_handler(const struct config_entry
*entry
, void *unused
)
5089 if (!entry
|| !entry
->name
|| !entry
->value
) {
5094 /* Check if the option is to be ignored */
5095 for (i
= 0; i
< sizeof(config_ignore_options
) / sizeof(char *); i
++) {
5096 if (!strcmp(entry
->name
, config_ignore_options
[i
])) {
5101 for (i
= 0; i
< (sizeof(long_options
) / sizeof(struct option
)) - 1;
5104 /* Ignore if not fully matched. */
5105 if (strcmp(entry
->name
, long_options
[i
].name
)) {
5110 * If the option takes no argument on the command line, we have to
5111 * check if the value is "true". We support non-zero numeric values,
5114 if (!long_options
[i
].has_arg
) {
5115 ret
= config_parse_value(entry
->value
);
5118 WARN("Invalid configuration value \"%s\" for option %s",
5119 entry
->value
, entry
->name
);
5121 /* False, skip boolean config option. */
5126 ret
= set_option(long_options
[i
].val
, entry
->value
, entry
->name
);
5130 WARN("Unrecognized option \"%s\" in daemon configuration file.", entry
->name
);
5137 * daemon configuration loading and argument parsing
5139 static int set_options(int argc
, char **argv
)
5141 int ret
= 0, c
= 0, option_index
= 0;
5142 int orig_optopt
= optopt
, orig_optind
= optind
;
5144 const char *config_path
= NULL
;
5146 optstring
= utils_generate_optstring(long_options
,
5147 sizeof(long_options
) / sizeof(struct option
));
5153 /* Check for the --config option */
5154 while ((c
= getopt_long(argc
, argv
, optstring
, long_options
,
5155 &option_index
)) != -1) {
5159 } else if (c
!= 'f') {
5160 /* if not equal to --config option. */
5164 if (lttng_is_setuid_setgid()) {
5165 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
5168 config_path
= utils_expand_path(optarg
);
5170 ERR("Failed to resolve path: %s", optarg
);
5175 ret
= config_get_section_entries(config_path
, config_section_name
,
5176 config_entry_handler
, NULL
);
5179 ERR("Invalid configuration option at line %i", ret
);
5185 /* Reset getopt's global state */
5186 optopt
= orig_optopt
;
5187 optind
= orig_optind
;
5191 * getopt_long() will not set option_index if it encounters a
5194 c
= getopt_long(argc
, argv
, optstring
, long_options
,
5201 * Pass NULL as the long option name if popt left the index
5204 ret
= set_option(c
, optarg
,
5205 option_index
< 0 ? NULL
:
5206 long_options
[option_index
].name
);
5218 * Creates the two needed socket by the daemon.
5219 * apps_sock - The communication socket for all UST apps.
5220 * client_sock - The communication of the cli tool (lttng).
5222 static int init_daemon_socket(void)
5227 old_umask
= umask(0);
5229 /* Create client tool unix socket */
5230 client_sock
= lttcomm_create_unix_sock(config
.client_unix_sock_path
.value
);
5231 if (client_sock
< 0) {
5232 ERR("Create unix sock failed: %s", config
.client_unix_sock_path
.value
);
5237 /* Set the cloexec flag */
5238 ret
= utils_set_fd_cloexec(client_sock
);
5240 ERR("Unable to set CLOEXEC flag to the client Unix socket (fd: %d). "
5241 "Continuing but note that the consumer daemon will have a "
5242 "reference to this socket on exec()", client_sock
);
5245 /* File permission MUST be 660 */
5246 ret
= chmod(config
.client_unix_sock_path
.value
, S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5248 ERR("Set file permissions failed: %s", config
.client_unix_sock_path
.value
);
5253 /* Create the application unix socket */
5254 apps_sock
= lttcomm_create_unix_sock(config
.apps_unix_sock_path
.value
);
5255 if (apps_sock
< 0) {
5256 ERR("Create unix sock failed: %s", config
.apps_unix_sock_path
.value
);
5261 /* Set the cloexec flag */
5262 ret
= utils_set_fd_cloexec(apps_sock
);
5264 ERR("Unable to set CLOEXEC flag to the app Unix socket (fd: %d). "
5265 "Continuing but note that the consumer daemon will have a "
5266 "reference to this socket on exec()", apps_sock
);
5269 /* File permission MUST be 666 */
5270 ret
= chmod(config
.apps_unix_sock_path
.value
,
5271 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
| S_IROTH
| S_IWOTH
);
5273 ERR("Set file permissions failed: %s", config
.apps_unix_sock_path
.value
);
5278 DBG3("Session daemon client socket %d and application socket %d created",
5279 client_sock
, apps_sock
);
5287 * Check if the global socket is available, and if a daemon is answering at the
5288 * other side. If yes, error is returned.
5290 static int check_existing_daemon(void)
5292 /* Is there anybody out there ? */
5293 if (lttng_session_daemon_alive()) {
5301 * Set the tracing group gid onto the client socket.
5303 * Race window between mkdir and chown is OK because we are going from more
5304 * permissive (root.root) to less permissive (root.tracing).
5306 static int set_permissions(char *rundir
)
5311 gid
= utils_get_group_id(config
.tracing_group_name
.value
);
5313 /* Set lttng run dir */
5314 ret
= chown(rundir
, 0, gid
);
5316 ERR("Unable to set group on %s", rundir
);
5321 * Ensure all applications and tracing group can search the run
5322 * dir. Allow everyone to read the directory, since it does not
5323 * buy us anything to hide its content.
5325 ret
= chmod(rundir
, S_IRWXU
| S_IRGRP
| S_IXGRP
| S_IROTH
| S_IXOTH
);
5327 ERR("Unable to set permissions on %s", rundir
);
5331 /* lttng client socket path */
5332 ret
= chown(config
.client_unix_sock_path
.value
, 0, gid
);
5334 ERR("Unable to set group on %s", config
.client_unix_sock_path
.value
);
5338 /* kconsumer error socket path */
5339 ret
= chown(kconsumer_data
.err_unix_sock_path
, 0, 0);
5341 ERR("Unable to set group on %s", kconsumer_data
.err_unix_sock_path
);
5345 /* 64-bit ustconsumer error socket path */
5346 ret
= chown(ustconsumer64_data
.err_unix_sock_path
, 0, 0);
5348 ERR("Unable to set group on %s", ustconsumer64_data
.err_unix_sock_path
);
5352 /* 32-bit ustconsumer compat32 error socket path */
5353 ret
= chown(ustconsumer32_data
.err_unix_sock_path
, 0, 0);
5355 ERR("Unable to set group on %s", ustconsumer32_data
.err_unix_sock_path
);
5359 DBG("All permissions are set");
5365 * Create the lttng run directory needed for all global sockets and pipe.
5367 static int create_lttng_rundir(void)
5371 DBG3("Creating LTTng run directory: %s", config
.rundir
.value
);
5373 ret
= mkdir(config
.rundir
.value
, S_IRWXU
);
5375 if (errno
!= EEXIST
) {
5376 ERR("Unable to create %s", config
.rundir
.value
);
5388 * Setup sockets and directory needed by the consumerds' communication with the
5391 static int set_consumer_sockets(struct consumer_data
*consumer_data
)
5396 switch (consumer_data
->type
) {
5397 case LTTNG_CONSUMER_KERNEL
:
5398 path
= config
.kconsumerd_path
.value
;
5400 case LTTNG_CONSUMER64_UST
:
5401 path
= config
.consumerd64_path
.value
;
5403 case LTTNG_CONSUMER32_UST
:
5404 path
= config
.consumerd32_path
.value
;
5407 ERR("Consumer type unknown");
5413 DBG2("Creating consumer directory: %s", path
);
5415 ret
= mkdir(path
, S_IRWXU
| S_IRGRP
| S_IXGRP
);
5416 if (ret
< 0 && errno
!= EEXIST
) {
5418 ERR("Failed to create %s", path
);
5422 ret
= chown(path
, 0, utils_get_group_id(config
.tracing_group_name
.value
));
5424 ERR("Unable to set group on %s", path
);
5430 /* Create the consumerd error unix socket */
5431 consumer_data
->err_sock
=
5432 lttcomm_create_unix_sock(consumer_data
->err_unix_sock_path
);
5433 if (consumer_data
->err_sock
< 0) {
5434 ERR("Create unix sock failed: %s", consumer_data
->err_unix_sock_path
);
5440 * Set the CLOEXEC flag. Return code is useless because either way, the
5443 ret
= utils_set_fd_cloexec(consumer_data
->err_sock
);
5445 PERROR("utils_set_fd_cloexec");
5446 /* continue anyway */
5449 /* File permission MUST be 660 */
5450 ret
= chmod(consumer_data
->err_unix_sock_path
,
5451 S_IRUSR
| S_IWUSR
| S_IRGRP
| S_IWGRP
);
5453 ERR("Set file permissions failed: %s", consumer_data
->err_unix_sock_path
);
5463 * Signal handler for the daemon
5465 * Simply stop all worker threads, leaving main() return gracefully after
5466 * joining all threads and calling cleanup().
5468 static void sighandler(int sig
)
5472 DBG("SIGINT caught");
5476 DBG("SIGTERM caught");
5480 CMM_STORE_SHARED(recv_child_signal
, 1);
5488 * Setup signal handler for :
5489 * SIGINT, SIGTERM, SIGPIPE
5491 static int set_signal_handler(void)
5494 struct sigaction sa
;
5497 if ((ret
= sigemptyset(&sigset
)) < 0) {
5498 PERROR("sigemptyset");
5502 sa
.sa_mask
= sigset
;
5505 sa
.sa_handler
= sighandler
;
5506 if ((ret
= sigaction(SIGTERM
, &sa
, NULL
)) < 0) {
5507 PERROR("sigaction");
5511 if ((ret
= sigaction(SIGINT
, &sa
, NULL
)) < 0) {
5512 PERROR("sigaction");
5516 if ((ret
= sigaction(SIGUSR1
, &sa
, NULL
)) < 0) {
5517 PERROR("sigaction");
5521 sa
.sa_handler
= SIG_IGN
;
5522 if ((ret
= sigaction(SIGPIPE
, &sa
, NULL
)) < 0) {
5523 PERROR("sigaction");
5527 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
5533 * Set open files limit to unlimited. This daemon can open a large number of
5534 * file descriptors in order to consume multiple kernel traces.
5536 static void set_ulimit(void)
5541 /* The kernel does not allow an infinite limit for open files */
5542 lim
.rlim_cur
= 65535;
5543 lim
.rlim_max
= 65535;
5545 ret
= setrlimit(RLIMIT_NOFILE
, &lim
);
5547 PERROR("failed to set open files limit");
5551 static int write_pidfile(void)
5553 return utils_create_pid_file(getpid(), config
.pid_file_path
.value
);
5557 * Create lockfile using the rundir and return its fd.
5559 static int create_lockfile(void)
5561 return utils_create_lock_file(config
.lock_file_path
.value
);
5565 * Write agent TCP port using the rundir.
5567 static int write_agent_port(void)
5569 return utils_create_pid_file(config
.agent_tcp_port
,
5570 config
.agent_port_file_path
.value
);
5573 static int set_clock_plugin_env(void)
5576 char *env_value
= NULL
;
5578 if (!config
.lttng_ust_clock_plugin
.value
) {
5582 ret
= asprintf(&env_value
, "LTTNG_UST_CLOCK_PLUGIN=%s",
5583 config
.lttng_ust_clock_plugin
.value
);
5589 ret
= putenv(env_value
);
5592 PERROR("putenv of LTTNG_UST_CLOCK_PLUGIN");
5596 DBG("Updated LTTNG_UST_CLOCK_PLUGIN environment variable to \"%s\"",
5597 config
.lttng_ust_clock_plugin
.value
);
5603 struct rotation_thread_timer_queue
*create_rotate_timer_queue(void)
5605 struct rotation_thread_timer_queue
*queue
= NULL
;
5607 queue
= zmalloc(sizeof(struct rotation_thread_timer_queue
));
5609 PERROR("Failed to allocate timer rotate queue");
5613 queue
->event_pipe
= lttng_pipe_open(FD_CLOEXEC
| O_NONBLOCK
);
5614 CDS_INIT_LIST_HEAD(&queue
->list
);
5615 pthread_mutex_init(&queue
->lock
, NULL
);
5622 void destroy_rotate_timer_queue(struct rotation_thread_timer_queue
*queue
)
5624 struct sessiond_rotation_timer
*node
, *tmp_node
;
5630 lttng_pipe_destroy(queue
->event_pipe
);
5632 pthread_mutex_lock(&queue
->lock
);
5633 /* Empty wait queue. */
5634 cds_list_for_each_entry_safe(node
, tmp_node
, &queue
->list
, head
) {
5635 cds_list_del(&node
->head
);
5638 pthread_mutex_unlock(&queue
->lock
);
5640 pthread_mutex_destroy(&queue
->lock
);
5647 int main(int argc
, char **argv
)
5649 int ret
= 0, retval
= 0;
5651 const char *env_app_timeout
;
5652 struct lttng_pipe
*ust32_channel_monitor_pipe
= NULL
,
5653 *ust64_channel_monitor_pipe
= NULL
,
5654 *kernel_channel_monitor_pipe
= NULL
;
5655 bool notification_thread_running
= false;
5656 bool rotation_thread_running
= false;
5657 bool timer_thread_running
= false;
5658 struct lttng_pipe
*ust32_channel_rotate_pipe
= NULL
,
5659 *ust64_channel_rotate_pipe
= NULL
,
5660 *kernel_channel_rotate_pipe
= NULL
;
5661 struct timer_thread_parameters timer_thread_ctx
;
5662 /* Queue of rotation jobs populated by the sessiond-timer. */
5663 struct rotation_thread_timer_queue
*rotation_timer_queue
= NULL
;
5665 init_kernel_workarounds();
5667 rcu_register_thread();
5669 if (set_signal_handler()) {
5671 goto exit_set_signal_handler
;
5674 if (sessiond_timer_signal_init()) {
5676 goto exit_set_signal_handler
;
5679 page_size
= sysconf(_SC_PAGESIZE
);
5680 if (page_size
< 0) {
5681 PERROR("sysconf _SC_PAGESIZE");
5682 page_size
= LONG_MAX
;
5683 WARN("Fallback page size to %ld", page_size
);
5686 ret
= sessiond_config_init(&config
);
5689 goto exit_set_signal_handler
;
5693 * Parse arguments and load the daemon configuration file.
5695 * We have an exit_options exit path to free memory reserved by
5696 * set_options. This is needed because the rest of sessiond_cleanup()
5697 * depends on ht_cleanup_thread, which depends on lttng_daemonize, which
5698 * depends on set_options.
5701 if (set_options(argc
, argv
)) {
5706 /* Init config from environment variables. */
5707 sessiond_config_apply_env_config(&config
);
5710 * Resolve all paths received as arguments, configuration option, or
5711 * through environment variable as absolute paths. This is necessary
5712 * since daemonizing causes the sessiond's current working directory
5715 ret
= sessiond_config_resolve_paths(&config
);
5721 lttng_opt_verbose
= config
.verbose
;
5722 lttng_opt_quiet
= config
.quiet
;
5723 kconsumer_data
.err_unix_sock_path
=
5724 config
.kconsumerd_err_unix_sock_path
.value
;
5725 kconsumer_data
.cmd_unix_sock_path
=
5726 config
.kconsumerd_cmd_unix_sock_path
.value
;
5727 ustconsumer32_data
.err_unix_sock_path
=
5728 config
.consumerd32_err_unix_sock_path
.value
;
5729 ustconsumer32_data
.cmd_unix_sock_path
=
5730 config
.consumerd32_cmd_unix_sock_path
.value
;
5731 ustconsumer64_data
.err_unix_sock_path
=
5732 config
.consumerd64_err_unix_sock_path
.value
;
5733 ustconsumer64_data
.cmd_unix_sock_path
=
5734 config
.consumerd64_cmd_unix_sock_path
.value
;
5735 set_clock_plugin_env();
5737 sessiond_config_log(&config
);
5740 if (config
.daemonize
|| config
.background
) {
5743 ret
= lttng_daemonize(&child_ppid
, &recv_child_signal
,
5744 !config
.background
);
5751 * We are in the child. Make sure all other file descriptors are
5752 * closed, in case we are called with more opened file
5753 * descriptors than the standard ones.
5755 for (i
= 3; i
< sysconf(_SC_OPEN_MAX
); i
++) {
5760 if (run_as_create_worker(argv
[0]) < 0) {
5761 goto exit_create_run_as_worker_cleanup
;
5765 * Starting from here, we can create threads. This needs to be after
5766 * lttng_daemonize due to RCU.
5770 * Initialize the health check subsystem. This call should set the
5771 * appropriate time values.
5773 health_sessiond
= health_app_create(NR_HEALTH_SESSIOND_TYPES
);
5774 if (!health_sessiond
) {
5775 PERROR("health_app_create error");
5777 goto exit_health_sessiond_cleanup
;
5780 /* Create thread to clean up RCU hash tables */
5781 if (init_ht_cleanup_thread(&ht_cleanup_thread
)) {
5783 goto exit_ht_cleanup
;
5786 /* Create thread quit pipe */
5787 if (init_thread_quit_pipe()) {
5789 goto exit_init_data
;
5792 /* Check if daemon is UID = 0 */
5793 is_root
= !getuid();
5795 if (create_lttng_rundir()) {
5797 goto exit_init_data
;
5801 /* Create global run dir with root access */
5803 kernel_channel_monitor_pipe
= lttng_pipe_open(0);
5804 if (!kernel_channel_monitor_pipe
) {
5805 ERR("Failed to create kernel consumer channel monitor pipe");
5807 goto exit_init_data
;
5809 kconsumer_data
.channel_monitor_pipe
=
5810 lttng_pipe_release_writefd(
5811 kernel_channel_monitor_pipe
);
5812 if (kconsumer_data
.channel_monitor_pipe
< 0) {
5814 goto exit_init_data
;
5816 kernel_channel_rotate_pipe
= lttng_pipe_open(0);
5817 if (!kernel_channel_rotate_pipe
) {
5818 ERR("Failed to create kernel consumer channel rotate pipe");
5820 goto exit_init_data
;
5822 kconsumer_data
.channel_rotate_pipe
=
5823 lttng_pipe_release_writefd(
5824 kernel_channel_rotate_pipe
);
5825 if (kconsumer_data
.channel_rotate_pipe
< 0) {
5827 goto exit_init_data
;
5831 lockfile_fd
= create_lockfile();
5832 if (lockfile_fd
< 0) {
5834 goto exit_init_data
;
5837 /* Set consumer initial state */
5838 kernel_consumerd_state
= CONSUMER_STOPPED
;
5839 ust_consumerd_state
= CONSUMER_STOPPED
;
5841 ust32_channel_monitor_pipe
= lttng_pipe_open(0);
5842 if (!ust32_channel_monitor_pipe
) {
5843 ERR("Failed to create 32-bit user space consumer channel monitor pipe");
5845 goto exit_init_data
;
5847 ustconsumer32_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
5848 ust32_channel_monitor_pipe
);
5849 if (ustconsumer32_data
.channel_monitor_pipe
< 0) {
5851 goto exit_init_data
;
5853 ust32_channel_rotate_pipe
= lttng_pipe_open(0);
5854 if (!ust32_channel_rotate_pipe
) {
5855 ERR("Failed to create 32-bit user space consumer channel rotate pipe");
5857 goto exit_init_data
;
5859 ustconsumer32_data
.channel_rotate_pipe
= lttng_pipe_release_writefd(
5860 ust32_channel_rotate_pipe
);
5861 if (ustconsumer32_data
.channel_rotate_pipe
< 0) {
5863 goto exit_init_data
;
5867 * The rotation_timer_queue structure is shared between the sessiond timer
5868 * thread and the rotation thread. The main() keeps the ownership and
5869 * destroys it when both threads have quit.
5871 rotation_timer_queue
= create_rotate_timer_queue();
5872 if (!rotation_timer_queue
) {
5874 goto exit_init_data
;
5876 timer_thread_ctx
.rotation_timer_queue
= rotation_timer_queue
;
5878 ust64_channel_monitor_pipe
= lttng_pipe_open(0);
5879 if (!ust64_channel_monitor_pipe
) {
5880 ERR("Failed to create 64-bit user space consumer channel monitor pipe");
5882 goto exit_init_data
;
5884 ustconsumer64_data
.channel_monitor_pipe
= lttng_pipe_release_writefd(
5885 ust64_channel_monitor_pipe
);
5886 if (ustconsumer64_data
.channel_monitor_pipe
< 0) {
5888 goto exit_init_data
;
5890 ust64_channel_rotate_pipe
= lttng_pipe_open(0);
5891 if (!ust64_channel_rotate_pipe
) {
5892 ERR("Failed to create 64-bit user space consumer channel rotate pipe");
5894 goto exit_init_data
;
5896 ustconsumer64_data
.channel_rotate_pipe
= lttng_pipe_release_writefd(
5897 ust64_channel_rotate_pipe
);
5898 if (ustconsumer64_data
.channel_rotate_pipe
< 0) {
5900 goto exit_init_data
;
5904 * See if daemon already exist.
5906 if (check_existing_daemon()) {
5907 ERR("Already running daemon.\n");
5909 * We do not goto exit because we must not cleanup()
5910 * because a daemon is already running.
5913 goto exit_init_data
;
5917 * Init UST app hash table. Alloc hash table before this point since
5918 * cleanup() can get called after that point.
5920 if (ust_app_ht_alloc()) {
5921 ERR("Failed to allocate UST app hash table");
5923 goto exit_init_data
;
5927 * Initialize agent app hash table. We allocate the hash table here
5928 * since cleanup() can get called after this point.
5930 if (agent_app_ht_alloc()) {
5931 ERR("Failed to allocate Agent app hash table");
5933 goto exit_init_data
;
5937 * These actions must be executed as root. We do that *after* setting up
5938 * the sockets path because we MUST make the check for another daemon using
5939 * those paths *before* trying to set the kernel consumer sockets and init
5943 if (set_consumer_sockets(&kconsumer_data
)) {
5945 goto exit_init_data
;
5948 /* Setup kernel tracer */
5949 if (!config
.no_kernel
) {
5950 init_kernel_tracer();
5951 if (kernel_tracer_fd
>= 0) {
5952 ret
= syscall_init_table();
5954 ERR("Unable to populate syscall table. "
5955 "Syscall tracing won't work "
5956 "for this session daemon.");
5961 /* Set ulimit for open files */
5964 /* init lttng_fd tracking must be done after set_ulimit. */
5967 if (set_consumer_sockets(&ustconsumer64_data
)) {
5969 goto exit_init_data
;
5972 if (set_consumer_sockets(&ustconsumer32_data
)) {
5974 goto exit_init_data
;
5977 /* Setup the needed unix socket */
5978 if (init_daemon_socket()) {
5980 goto exit_init_data
;
5983 /* Set credentials to socket */
5984 if (is_root
&& set_permissions(config
.rundir
.value
)) {
5986 goto exit_init_data
;
5989 /* Get parent pid if -S, --sig-parent is specified. */
5990 if (config
.sig_parent
) {
5994 /* Setup the kernel pipe for waking up the kernel thread */
5995 if (is_root
&& !config
.no_kernel
) {
5996 if (utils_create_pipe_cloexec(kernel_poll_pipe
)) {
5998 goto exit_init_data
;
6002 /* Setup the thread apps communication pipe. */
6003 if (utils_create_pipe_cloexec(apps_cmd_pipe
)) {
6005 goto exit_init_data
;
6008 /* Setup the thread apps notify communication pipe. */
6009 if (utils_create_pipe_cloexec(apps_cmd_notify_pipe
)) {
6011 goto exit_init_data
;
6014 /* Initialize global buffer per UID and PID registry. */
6015 buffer_reg_init_uid_registry();
6016 buffer_reg_init_pid_registry();
6018 /* Init UST command queue. */
6019 cds_wfcq_init(&ust_cmd_queue
.head
, &ust_cmd_queue
.tail
);
6022 * Get session list pointer. This pointer MUST NOT be free'd. This list
6023 * is statically declared in session.c
6025 session_list_ptr
= session_get_list();
6029 /* Check for the application socket timeout env variable. */
6030 env_app_timeout
= getenv(DEFAULT_APP_SOCKET_TIMEOUT_ENV
);
6031 if (env_app_timeout
) {
6032 config
.app_socket_timeout
= atoi(env_app_timeout
);
6034 config
.app_socket_timeout
= DEFAULT_APP_SOCKET_RW_TIMEOUT
;
6037 ret
= write_pidfile();
6039 ERR("Error in write_pidfile");
6041 goto exit_init_data
;
6043 ret
= write_agent_port();
6045 ERR("Error in write_agent_port");
6047 goto exit_init_data
;
6050 /* Initialize communication library */
6052 /* Initialize TCP timeout values */
6053 lttcomm_inet_init();
6055 if (load_session_init_data(&load_info
) < 0) {
6057 goto exit_init_data
;
6059 load_info
->path
= config
.load_session_path
.value
;
6061 /* Create health-check thread. */
6062 ret
= pthread_create(&health_thread
, default_pthread_attr(),
6063 thread_manage_health
, (void *) NULL
);
6066 PERROR("pthread_create health");
6071 /* notification_thread_data acquires the pipes' read side. */
6072 notification_thread_handle
= notification_thread_handle_create(
6073 ust32_channel_monitor_pipe
,
6074 ust64_channel_monitor_pipe
,
6075 kernel_channel_monitor_pipe
);
6076 if (!notification_thread_handle
) {
6078 ERR("Failed to create notification thread shared data");
6080 goto exit_notification
;
6083 /* Create notification thread. */
6084 ret
= pthread_create(¬ification_thread
, default_pthread_attr(),
6085 thread_notification
, notification_thread_handle
);
6088 PERROR("pthread_create notification");
6091 goto exit_notification
;
6093 notification_thread_running
= true;
6095 /* Create timer thread. */
6096 ret
= pthread_create(&timer_thread
, default_pthread_attr(),
6097 sessiond_timer_thread
, &timer_thread_ctx
);
6100 PERROR("pthread_create timer");
6103 goto exit_notification
;
6105 timer_thread_running
= true;
6107 /* rotation_thread_data acquires the pipes' read side. */
6108 rotation_thread_handle
= rotation_thread_handle_create(
6109 ust32_channel_rotate_pipe
,
6110 ust64_channel_rotate_pipe
,
6111 kernel_channel_rotate_pipe
,
6112 thread_quit_pipe
[0],
6113 rotation_timer_queue
);
6114 if (!rotation_thread_handle
) {
6116 ERR("Failed to create rotation thread shared data");
6121 /* Create rotation thread. */
6122 ret
= pthread_create(&rotation_thread
, default_pthread_attr(),
6123 thread_rotation
, rotation_thread_handle
);
6126 PERROR("pthread_create rotation");
6131 rotation_thread_running
= true;
6133 /* Create thread to manage the client socket */
6134 ret
= pthread_create(&client_thread
, default_pthread_attr(),
6135 thread_manage_clients
, (void *) NULL
);
6138 PERROR("pthread_create clients");
6144 /* Create thread to dispatch registration */
6145 ret
= pthread_create(&dispatch_thread
, default_pthread_attr(),
6146 thread_dispatch_ust_registration
, (void *) NULL
);
6149 PERROR("pthread_create dispatch");
6155 /* Create thread to manage application registration. */
6156 ret
= pthread_create(®_apps_thread
, default_pthread_attr(),
6157 thread_registration_apps
, (void *) NULL
);
6160 PERROR("pthread_create registration");
6166 /* Create thread to manage application socket */
6167 ret
= pthread_create(&apps_thread
, default_pthread_attr(),
6168 thread_manage_apps
, (void *) NULL
);
6171 PERROR("pthread_create apps");
6177 /* Create thread to manage application notify socket */
6178 ret
= pthread_create(&apps_notify_thread
, default_pthread_attr(),
6179 ust_thread_manage_notify
, (void *) NULL
);
6182 PERROR("pthread_create notify");
6185 goto exit_apps_notify
;
6188 /* Create agent registration thread. */
6189 ret
= pthread_create(&agent_reg_thread
, default_pthread_attr(),
6190 agent_thread_manage_registration
, (void *) NULL
);
6193 PERROR("pthread_create agent");
6196 goto exit_agent_reg
;
6199 /* Don't start this thread if kernel tracing is not requested nor root */
6200 if (is_root
&& !config
.no_kernel
) {
6201 /* Create kernel thread to manage kernel event */
6202 ret
= pthread_create(&kernel_thread
, default_pthread_attr(),
6203 thread_manage_kernel
, (void *) NULL
);
6206 PERROR("pthread_create kernel");
6213 /* Create session loading thread. */
6214 ret
= pthread_create(&load_session_thread
, default_pthread_attr(),
6215 thread_load_session
, load_info
);
6218 PERROR("pthread_create load_session_thread");
6221 goto exit_load_session
;
6225 * This is where we start awaiting program completion (e.g. through
6226 * signal that asks threads to teardown).
6229 ret
= pthread_join(load_session_thread
, &status
);
6232 PERROR("pthread_join load_session_thread");
6237 if (is_root
&& !config
.no_kernel
) {
6238 ret
= pthread_join(kernel_thread
, &status
);
6241 PERROR("pthread_join");
6247 ret
= pthread_join(agent_reg_thread
, &status
);
6250 PERROR("pthread_join agent");
6255 ret
= pthread_join(apps_notify_thread
, &status
);
6258 PERROR("pthread_join apps notify");
6263 ret
= pthread_join(apps_thread
, &status
);
6266 PERROR("pthread_join apps");
6271 ret
= pthread_join(reg_apps_thread
, &status
);
6274 PERROR("pthread_join");
6280 * Join dispatch thread after joining reg_apps_thread to ensure
6281 * we don't leak applications in the queue.
6283 ret
= pthread_join(dispatch_thread
, &status
);
6286 PERROR("pthread_join");
6291 ret
= pthread_join(client_thread
, &status
);
6294 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 if (notification_thread_running
) {
6337 notification_thread_command_quit(
6338 notification_thread_handle
);
6339 ret
= pthread_join(notification_thread
, &status
);
6342 PERROR("pthread_join notification thread");
6346 notification_thread_handle_destroy(notification_thread_handle
);
6349 if (rotation_thread_handle
) {
6350 if (rotation_thread_running
) {
6351 ret
= pthread_join(rotation_thread
, &status
);
6354 PERROR("pthread_join rotation thread");
6358 rotation_thread_handle_destroy(rotation_thread_handle
);
6361 if (timer_thread_running
) {
6362 kill(getpid(), LTTNG_SESSIOND_SIG_EXIT
);
6363 ret
= pthread_join(timer_thread
, &status
);
6366 PERROR("pthread_join timer thread");
6372 * After the rotation and timer thread have quit, we can safely destroy
6373 * the rotation_timer_queue.
6375 destroy_rotate_timer_queue(rotation_timer_queue
);
6377 rcu_thread_offline();
6378 rcu_unregister_thread();
6380 ret
= fini_ht_cleanup_thread(&ht_cleanup_thread
);
6384 lttng_pipe_destroy(ust32_channel_monitor_pipe
);
6385 lttng_pipe_destroy(ust64_channel_monitor_pipe
);
6386 lttng_pipe_destroy(kernel_channel_monitor_pipe
);
6387 lttng_pipe_destroy(ust32_channel_rotate_pipe
);
6388 lttng_pipe_destroy(ust64_channel_rotate_pipe
);
6389 lttng_pipe_destroy(kernel_channel_rotate_pipe
);
6392 health_app_destroy(health_sessiond
);
6393 exit_health_sessiond_cleanup
:
6394 exit_create_run_as_worker_cleanup
:
6397 sessiond_cleanup_options();
6399 exit_set_signal_handler
: