4 * Copyright (C) 2011 David Goulet <david.goulet@polymtl.ca>
5 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7 * This library is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; only
10 * version 2.1 of the License.
12 * This library is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with this library; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 #include <sys/types.h>
25 #include <sys/socket.h>
28 #include <sys/types.h>
35 #include <semaphore.h>
40 #include <urcu/uatomic.h>
41 #include <urcu/futex.h>
42 #include <urcu/compiler.h>
44 #include <lttng/ust-events.h>
45 #include <lttng/ust-abi.h>
46 #include <lttng/ust.h>
47 #include <lttng/ust-error.h>
48 #include <lttng/ust-ctl.h>
49 #include <lttng/ust-cancelstate.h>
50 #include <urcu/tls-compat.h>
53 #include <usterr-signal-safe.h>
55 #include "tracepoint-internal.h"
56 #include "lttng-tracer-core.h"
58 #include "../libringbuffer/rb-init.h"
59 #include "lttng-ust-statedump.h"
61 #include "../libringbuffer/getcpu.h"
64 /* Concatenate lttng ust shared library name with its major version number. */
65 #define LTTNG_UST_LIB_SO_NAME "liblttng-ust.so." __ust_stringify(CONFIG_LTTNG_UST_LIBRARY_VERSION_MAJOR)
68 * Has lttng ust comm constructor been called ?
70 static int initialized
;
73 * The ust_lock/ust_unlock lock is used as a communication thread mutex.
74 * Held when handling a command, also held by fork() to deal with
75 * removal of threads, and by exit path.
77 * The UST lock is the centralized mutex across UST tracing control and
80 * ust_exit_mutex must never nest in ust_mutex.
82 * ust_fork_mutex must never nest in ust_mutex.
84 * ust_mutex_nest is a per-thread nesting counter, allowing the perf
85 * counter lazy initialization called by events within the statedump,
86 * which traces while the ust_mutex is held.
88 * ust_lock nests within the dynamic loader lock (within glibc) because
89 * it is taken within the library constructor.
91 * The ust fd tracker lock nests within the ust_mutex.
93 static pthread_mutex_t ust_mutex
= PTHREAD_MUTEX_INITIALIZER
;
95 /* Allow nesting the ust_mutex within the same thread. */
96 static DEFINE_URCU_TLS(int, ust_mutex_nest
);
99 * ust_exit_mutex protects thread_active variable wrt thread exit. It
100 * cannot be done by ust_mutex because pthread_cancel(), which takes an
101 * internal libc lock, cannot nest within ust_mutex.
103 * It never nests within a ust_mutex.
105 static pthread_mutex_t ust_exit_mutex
= PTHREAD_MUTEX_INITIALIZER
;
108 * ust_fork_mutex protects base address statedump tracing against forks. It
109 * prevents the dynamic loader lock to be taken (by base address statedump
110 * tracing) while a fork is happening, thus preventing deadlock issues with
111 * the dynamic loader lock.
113 static pthread_mutex_t ust_fork_mutex
= PTHREAD_MUTEX_INITIALIZER
;
115 /* Should the ust comm thread quit ? */
116 static int lttng_ust_comm_should_quit
;
119 * This variable can be tested by applications to check whether
120 * lttng-ust is loaded. They simply have to define their own
121 * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the
122 * library constructor.
124 int lttng_ust_loaded
__attribute__((weak
));
127 * Notes on async-signal-safety of ust lock: a few libc functions are used
128 * which are not strictly async-signal-safe:
130 * - pthread_setcancelstate
131 * - pthread_mutex_lock
132 * - pthread_mutex_unlock
134 * As of glibc 2.35, the implementation of pthread_setcancelstate only
135 * touches TLS data, and it appears to be safe to use from signal
136 * handlers. If the libc implementation changes, this will need to be
137 * revisited, and we may ask glibc to provide an async-signal-safe
138 * pthread_setcancelstate.
140 * As of glibc 2.35, the implementation of pthread_mutex_lock/unlock
141 * for fast mutexes only relies on the pthread_mutex_t structure.
142 * Disabling signals around all uses of this mutex ensures
143 * signal-safety. If the libc implementation changes and eventually uses
144 * other global resources, this will need to be revisited and we may
145 * need to implement our own mutex.
149 * Return 0 on success, -1 if should quit.
150 * The lock is taken in both cases.
155 sigset_t sig_all_blocked
, orig_mask
;
158 if (lttng_ust_cancelstate_disable_push()) {
159 ERR("lttng_ust_cancelstate_disable_push");
161 sigfillset(&sig_all_blocked
);
162 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_mask
);
164 ERR("pthread_sigmask: ret=%d", ret
);
166 if (!URCU_TLS(ust_mutex_nest
)++)
167 pthread_mutex_lock(&ust_mutex
);
168 ret
= pthread_sigmask(SIG_SETMASK
, &orig_mask
, NULL
);
170 ERR("pthread_sigmask: ret=%d", ret
);
172 if (lttng_ust_comm_should_quit
) {
180 * ust_lock_nocheck() can be used in constructors/destructors, because
181 * they are already nested within the dynamic loader lock, and therefore
182 * have exclusive access against execution of liblttng-ust destructor.
185 void ust_lock_nocheck(void)
187 sigset_t sig_all_blocked
, orig_mask
;
190 if (lttng_ust_cancelstate_disable_push()) {
191 ERR("lttng_ust_cancelstate_disable_push");
193 sigfillset(&sig_all_blocked
);
194 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_mask
);
196 ERR("pthread_sigmask: ret=%d", ret
);
198 if (!URCU_TLS(ust_mutex_nest
)++)
199 pthread_mutex_lock(&ust_mutex
);
200 ret
= pthread_sigmask(SIG_SETMASK
, &orig_mask
, NULL
);
202 ERR("pthread_sigmask: ret=%d", ret
);
209 void ust_unlock(void)
211 sigset_t sig_all_blocked
, orig_mask
;
214 sigfillset(&sig_all_blocked
);
215 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_mask
);
217 ERR("pthread_sigmask: ret=%d", ret
);
219 if (!--URCU_TLS(ust_mutex_nest
))
220 pthread_mutex_unlock(&ust_mutex
);
221 ret
= pthread_sigmask(SIG_SETMASK
, &orig_mask
, NULL
);
223 ERR("pthread_sigmask: ret=%d", ret
);
225 if (lttng_ust_cancelstate_disable_pop()) {
226 ERR("lttng_ust_cancelstate_disable_pop");
231 * Wait for either of these before continuing to the main
233 * - the register_done message from sessiond daemon
234 * (will let the sessiond daemon enable sessions before main
236 * - sessiond daemon is not reachable.
237 * - timeout (ensuring applications are resilient to session
240 static sem_t constructor_wait
;
242 * Doing this for both the global and local sessiond.
245 sem_count_initial_value
= 4,
248 static int sem_count
= sem_count_initial_value
;
251 * Counting nesting within lttng-ust. Used to ensure that calling fork()
252 * from liblttng-ust does not execute the pre/post fork handlers.
254 static DEFINE_URCU_TLS(int, lttng_ust_nest_count
);
257 * Info about socket and associated listener thread.
261 pthread_t ust_listener
; /* listener thread */
263 int registration_done
;
268 char sock_path
[PATH_MAX
];
272 char wait_shm_path
[PATH_MAX
];
274 /* Keep track of lazy state dump not performed yet. */
275 int statedump_pending
;
276 int initial_statedump_done
;
277 /* Keep procname for statedump */
278 char procname
[LTTNG_UST_PROCNAME_LEN
];
281 /* Socket from app (connect) to session daemon (listen) for communication */
282 struct sock_info global_apps
= {
287 .registration_done
= 0,
291 .sock_path
= LTTNG_DEFAULT_RUNDIR
"/" LTTNG_UST_SOCK_FILENAME
,
295 .wait_shm_path
= "/" LTTNG_UST_WAIT_FILENAME
,
297 .statedump_pending
= 0,
298 .initial_statedump_done
= 0,
302 /* TODO: allow global_apps_sock_path override */
304 struct sock_info local_apps
= {
308 .registration_done
= 0,
309 .allowed
= 0, /* Check setuid bit first */
315 .statedump_pending
= 0,
316 .initial_statedump_done
= 0,
320 static int wait_poll_fallback
;
322 static const char *cmd_name_mapping
[] = {
323 [ LTTNG_UST_RELEASE
] = "Release",
324 [ LTTNG_UST_SESSION
] = "Create Session",
325 [ LTTNG_UST_TRACER_VERSION
] = "Get Tracer Version",
327 [ LTTNG_UST_TRACEPOINT_LIST
] = "Create Tracepoint List",
328 [ LTTNG_UST_WAIT_QUIESCENT
] = "Wait for Quiescent State",
329 [ LTTNG_UST_REGISTER_DONE
] = "Registration Done",
330 [ LTTNG_UST_TRACEPOINT_FIELD_LIST
] = "Create Tracepoint Field List",
332 /* Session FD commands */
333 [ LTTNG_UST_CHANNEL
] = "Create Channel",
334 [ LTTNG_UST_SESSION_START
] = "Start Session",
335 [ LTTNG_UST_SESSION_STOP
] = "Stop Session",
337 /* Channel FD commands */
338 [ LTTNG_UST_STREAM
] = "Create Stream",
339 [ LTTNG_UST_EVENT
] = "Create Event",
341 /* Event and Channel FD commands */
342 [ LTTNG_UST_CONTEXT
] = "Create Context",
343 [ LTTNG_UST_FLUSH_BUFFER
] = "Flush Buffer",
345 /* Event, Channel and Session commands */
346 [ LTTNG_UST_ENABLE
] = "Enable",
347 [ LTTNG_UST_DISABLE
] = "Disable",
349 /* Tracepoint list commands */
350 [ LTTNG_UST_TRACEPOINT_LIST_GET
] = "List Next Tracepoint",
351 [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET
] = "List Next Tracepoint Field",
353 /* Event FD commands */
354 [ LTTNG_UST_FILTER
] = "Create Filter",
355 [ LTTNG_UST_EXCLUSION
] = "Add exclusions to event",
358 static const char *str_timeout
;
359 static int got_timeout_env
;
361 extern void lttng_ring_buffer_client_overwrite_init(void);
362 extern void lttng_ring_buffer_client_overwrite_rt_init(void);
363 extern void lttng_ring_buffer_client_discard_init(void);
364 extern void lttng_ring_buffer_client_discard_rt_init(void);
365 extern void lttng_ring_buffer_metadata_client_init(void);
366 extern void lttng_ring_buffer_client_overwrite_exit(void);
367 extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
368 extern void lttng_ring_buffer_client_discard_exit(void);
369 extern void lttng_ring_buffer_client_discard_rt_exit(void);
370 extern void lttng_ring_buffer_metadata_client_exit(void);
372 static char *get_map_shm(struct sock_info
*sock_info
);
374 ssize_t
lttng_ust_read(int fd
, void *buf
, size_t len
)
377 size_t copied
= 0, to_copy
= len
;
380 ret
= read(fd
, buf
+ copied
, to_copy
);
385 } while ((ret
> 0 && to_copy
> 0)
386 || (ret
< 0 && errno
== EINTR
));
393 * Returns the HOME directory path. Caller MUST NOT free(3) the returned
397 const char *get_lttng_home_dir(void)
401 val
= (const char *) lttng_getenv("LTTNG_HOME");
405 return (const char *) lttng_getenv("HOME");
409 * Force a read (imply TLS fixup for dlopen) of TLS variables.
412 void lttng_fixup_nest_count_tls(void)
414 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count
)));
418 void lttng_fixup_ust_mutex_nest_tls(void)
420 asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest
)));
427 void lttng_fixup_urcu_bp_tls(void)
433 void lttng_ust_fixup_tls(void)
435 lttng_fixup_urcu_bp_tls();
436 lttng_fixup_ringbuffer_tls();
437 lttng_fixup_vtid_tls();
438 lttng_fixup_nest_count_tls();
439 lttng_fixup_procname_tls();
440 lttng_fixup_ust_mutex_nest_tls();
441 lttng_ust_fixup_perf_counter_tls();
442 lttng_ust_fixup_fd_tracker_tls();
443 lttng_fixup_cgroup_ns_tls();
444 lttng_fixup_ipc_ns_tls();
445 lttng_fixup_net_ns_tls();
446 lttng_fixup_uts_ns_tls();
449 int lttng_get_notify_socket(void *owner
)
451 struct sock_info
*info
= owner
;
453 return info
->notify_socket
;
458 char* lttng_ust_sockinfo_get_procname(void *owner
)
460 struct sock_info
*info
= owner
;
462 return info
->procname
;
466 void print_cmd(int cmd
, int handle
)
468 const char *cmd_name
= "Unknown";
470 if (cmd
>= 0 && cmd
< LTTNG_ARRAY_SIZE(cmd_name_mapping
)
471 && cmd_name_mapping
[cmd
]) {
472 cmd_name
= cmd_name_mapping
[cmd
];
474 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
476 lttng_ust_obj_get_name(handle
), handle
);
480 int setup_global_apps(void)
483 assert(!global_apps
.wait_shm_mmap
);
485 global_apps
.wait_shm_mmap
= get_map_shm(&global_apps
);
486 if (!global_apps
.wait_shm_mmap
) {
487 WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing.");
488 global_apps
.allowed
= 0;
493 global_apps
.allowed
= 1;
494 lttng_ust_getprocname(global_apps
.procname
);
499 int setup_local_apps(void)
502 const char *home_dir
;
505 assert(!local_apps
.wait_shm_mmap
);
509 * Disallow per-user tracing for setuid binaries.
511 if (uid
!= geteuid()) {
512 assert(local_apps
.allowed
== 0);
516 home_dir
= get_lttng_home_dir();
518 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
519 assert(local_apps
.allowed
== 0);
523 local_apps
.allowed
= 1;
524 snprintf(local_apps
.sock_path
, PATH_MAX
, "%s/%s/%s",
526 LTTNG_DEFAULT_HOME_RUNDIR
,
527 LTTNG_UST_SOCK_FILENAME
);
528 snprintf(local_apps
.wait_shm_path
, PATH_MAX
, "/%s-%u",
529 LTTNG_UST_WAIT_FILENAME
,
532 local_apps
.wait_shm_mmap
= get_map_shm(&local_apps
);
533 if (!local_apps
.wait_shm_mmap
) {
534 WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing.");
535 local_apps
.allowed
= 0;
540 lttng_ust_getprocname(local_apps
.procname
);
546 * Get socket timeout, in ms.
547 * -1: wait forever. 0: don't wait. >0: timeout, in ms.
550 long get_timeout(void)
552 long constructor_delay_ms
= LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS
;
554 if (!got_timeout_env
) {
555 str_timeout
= lttng_getenv("LTTNG_UST_REGISTER_TIMEOUT");
559 constructor_delay_ms
= strtol(str_timeout
, NULL
, 10);
560 /* All negative values are considered as "-1". */
561 if (constructor_delay_ms
< -1)
562 constructor_delay_ms
= -1;
563 return constructor_delay_ms
;
566 /* Timeout for notify socket send and recv. */
568 long get_notify_sock_timeout(void)
570 return get_timeout();
573 /* Timeout for connecting to cmd and notify sockets. */
575 long get_connect_sock_timeout(void)
577 return get_timeout();
581 * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
584 int get_constructor_timeout(struct timespec
*constructor_timeout
)
586 long constructor_delay_ms
;
589 constructor_delay_ms
= get_timeout();
591 switch (constructor_delay_ms
) {
592 case -1:/* fall-through */
594 return constructor_delay_ms
;
600 * If we are unable to find the current time, don't wait.
602 ret
= clock_gettime(CLOCK_REALTIME
, constructor_timeout
);
607 constructor_timeout
->tv_sec
+= constructor_delay_ms
/ 1000UL;
608 constructor_timeout
->tv_nsec
+=
609 (constructor_delay_ms
% 1000UL) * 1000000UL;
610 if (constructor_timeout
->tv_nsec
>= 1000000000UL) {
611 constructor_timeout
->tv_sec
++;
612 constructor_timeout
->tv_nsec
-= 1000000000UL;
614 /* Timeout wait (constructor_delay_ms). */
619 void get_allow_blocking(void)
621 const char *str_allow_blocking
=
622 lttng_getenv("LTTNG_UST_ALLOW_BLOCKING");
624 if (str_allow_blocking
) {
625 DBG("%s environment variable is set",
626 "LTTNG_UST_ALLOW_BLOCKING");
627 lttng_ust_ringbuffer_set_allow_blocking();
632 int register_to_sessiond(int socket
, enum ustctl_socket_type type
)
634 return ustcomm_send_reg_msg(socket
,
637 lttng_alignof(uint8_t) * CHAR_BIT
,
638 lttng_alignof(uint16_t) * CHAR_BIT
,
639 lttng_alignof(uint32_t) * CHAR_BIT
,
640 lttng_alignof(uint64_t) * CHAR_BIT
,
641 lttng_alignof(unsigned long) * CHAR_BIT
);
645 int send_reply(int sock
, struct ustcomm_ust_reply
*lur
)
649 len
= ustcomm_send_unix_sock(sock
, lur
, sizeof(*lur
));
652 DBG("message successfully sent");
655 if (len
== -ECONNRESET
) {
656 DBG("remote end closed connection");
661 DBG("incorrect message size: %zd", len
);
667 void decrement_sem_count(unsigned int count
)
671 assert(uatomic_read(&sem_count
) >= count
);
673 if (uatomic_read(&sem_count
) <= 0) {
677 ret
= uatomic_add_return(&sem_count
, -count
);
679 ret
= sem_post(&constructor_wait
);
685 int handle_register_done(struct sock_info
*sock_info
)
687 if (sock_info
->registration_done
)
689 sock_info
->registration_done
= 1;
691 decrement_sem_count(1);
692 if (!sock_info
->statedump_pending
) {
693 sock_info
->initial_statedump_done
= 1;
694 decrement_sem_count(1);
701 int handle_register_failed(struct sock_info
*sock_info
)
703 if (sock_info
->registration_done
)
705 sock_info
->registration_done
= 1;
706 sock_info
->initial_statedump_done
= 1;
708 decrement_sem_count(2);
714 * Only execute pending statedump after the constructor semaphore has
715 * been posted by the current listener thread. This means statedump will
716 * only be performed after the "registration done" command is received
717 * from this thread's session daemon.
719 * This ensures we don't run into deadlock issues with the dynamic
720 * loader mutex, which is held while the constructor is called and
721 * waiting on the constructor semaphore. All operations requiring this
722 * dynamic loader lock need to be postponed using this mechanism.
724 * In a scenario with two session daemons connected to the application,
725 * it is possible that the first listener thread which receives the
726 * registration done command issues its statedump while the dynamic
727 * loader lock is still held by the application constructor waiting on
728 * the semaphore. It will however be allowed to proceed when the
729 * second session daemon sends the registration done command to the
730 * second listener thread. This situation therefore does not produce
734 void handle_pending_statedump(struct sock_info
*sock_info
)
736 if (sock_info
->registration_done
&& sock_info
->statedump_pending
) {
737 sock_info
->statedump_pending
= 0;
738 pthread_mutex_lock(&ust_fork_mutex
);
739 lttng_handle_pending_statedump(sock_info
);
740 pthread_mutex_unlock(&ust_fork_mutex
);
742 if (!sock_info
->initial_statedump_done
) {
743 sock_info
->initial_statedump_done
= 1;
744 decrement_sem_count(1);
750 int handle_message(struct sock_info
*sock_info
,
751 int sock
, struct ustcomm_ust_msg
*lum
)
754 const struct lttng_ust_objd_ops
*ops
;
755 struct ustcomm_ust_reply lur
;
757 char ctxstr
[LTTNG_UST_SYM_NAME_LEN
]; /* App context string. */
760 memset(&lur
, 0, sizeof(lur
));
763 ret
= -LTTNG_UST_ERR_EXITING
;
767 ops
= objd_ops(lum
->handle
);
774 case LTTNG_UST_REGISTER_DONE
:
775 if (lum
->handle
== LTTNG_UST_ROOT_HANDLE
)
776 ret
= handle_register_done(sock_info
);
780 case LTTNG_UST_RELEASE
:
781 if (lum
->handle
== LTTNG_UST_ROOT_HANDLE
)
784 ret
= lttng_ust_objd_unref(lum
->handle
, 1);
786 case LTTNG_UST_FILTER
:
788 /* Receive filter data */
789 struct lttng_ust_filter_bytecode_node
*bytecode
;
791 if (lum
->u
.filter
.data_size
> FILTER_BYTECODE_MAX_LEN
) {
792 ERR("Filter data size is too large: %u bytes",
793 lum
->u
.filter
.data_size
);
798 if (lum
->u
.filter
.reloc_offset
> lum
->u
.filter
.data_size
) {
799 ERR("Filter reloc offset %u is not within data",
800 lum
->u
.filter
.reloc_offset
);
805 bytecode
= zmalloc(sizeof(*bytecode
) + lum
->u
.filter
.data_size
);
810 len
= ustcomm_recv_unix_sock(sock
, bytecode
->bc
.data
,
811 lum
->u
.filter
.data_size
);
813 case 0: /* orderly shutdown */
818 if (len
== lum
->u
.filter
.data_size
) {
819 DBG("filter data received");
821 } else if (len
< 0) {
822 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
823 if (len
== -ECONNRESET
) {
824 ERR("%s remote end closed connection", sock_info
->name
);
833 DBG("incorrect filter data message size: %zd", len
);
839 bytecode
->bc
.len
= lum
->u
.filter
.data_size
;
840 bytecode
->bc
.reloc_offset
= lum
->u
.filter
.reloc_offset
;
841 bytecode
->bc
.seqnum
= lum
->u
.filter
.seqnum
;
843 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
844 (unsigned long) bytecode
,
849 /* don't free bytecode if everything went fine. */
856 case LTTNG_UST_EXCLUSION
:
858 /* Receive exclusion names */
859 struct lttng_ust_excluder_node
*node
;
862 count
= lum
->u
.exclusion
.count
;
864 /* There are no names to read */
868 node
= zmalloc(sizeof(*node
) +
869 count
* LTTNG_UST_SYM_NAME_LEN
);
874 node
->excluder
.count
= count
;
875 len
= ustcomm_recv_unix_sock(sock
, node
->excluder
.names
,
876 count
* LTTNG_UST_SYM_NAME_LEN
);
878 case 0: /* orderly shutdown */
883 if (len
== count
* LTTNG_UST_SYM_NAME_LEN
) {
884 DBG("Exclusion data received");
886 } else if (len
< 0) {
887 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
888 if (len
== -ECONNRESET
) {
889 ERR("%s remote end closed connection", sock_info
->name
);
898 DBG("Incorrect exclusion data message size: %zd", len
);
905 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
906 (unsigned long) node
,
911 /* Don't free exclusion data if everything went fine. */
918 case LTTNG_UST_CHANNEL
:
923 len
= ustcomm_recv_channel_from_sessiond(sock
,
924 &chan_data
, lum
->u
.channel
.len
,
927 case 0: /* orderly shutdown */
931 if (len
== lum
->u
.channel
.len
) {
932 DBG("channel data received");
934 } else if (len
< 0) {
935 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
936 if (len
== -ECONNRESET
) {
937 ERR("%s remote end closed connection", sock_info
->name
);
944 DBG("incorrect channel data message size: %zd", len
);
949 args
.channel
.chan_data
= chan_data
;
950 args
.channel
.wakeup_fd
= wakeup_fd
;
952 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
953 (unsigned long) &lum
->u
,
957 if (args
.channel
.wakeup_fd
>= 0) {
960 lttng_ust_lock_fd_tracker();
961 close_ret
= close(args
.channel
.wakeup_fd
);
962 lttng_ust_unlock_fd_tracker();
963 args
.channel
.wakeup_fd
= -1;
967 free(args
.channel
.chan_data
);
970 case LTTNG_UST_STREAM
:
974 /* Receive shm_fd, wakeup_fd */
975 ret
= ustcomm_recv_stream_from_sessiond(sock
,
978 &args
.stream
.wakeup_fd
);
984 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
985 (unsigned long) &lum
->u
,
989 if (args
.stream
.shm_fd
>= 0) {
990 lttng_ust_lock_fd_tracker();
991 close_ret
= close(args
.stream
.shm_fd
);
992 lttng_ust_unlock_fd_tracker();
993 args
.stream
.shm_fd
= -1;
997 if (args
.stream
.wakeup_fd
>= 0) {
998 lttng_ust_lock_fd_tracker();
999 close_ret
= close(args
.stream
.wakeup_fd
);
1000 lttng_ust_unlock_fd_tracker();
1001 args
.stream
.wakeup_fd
= -1;
1007 case LTTNG_UST_CONTEXT
:
1008 switch (lum
->u
.context
.ctx
) {
1009 case LTTNG_UST_CONTEXT_APP_CONTEXT
:
1012 size_t ctxlen
, recvlen
;
1014 ctxlen
= strlen("$app.") + lum
->u
.context
.u
.app_ctx
.provider_name_len
- 1
1015 + strlen(":") + lum
->u
.context
.u
.app_ctx
.ctx_name_len
;
1016 if (ctxlen
>= LTTNG_UST_SYM_NAME_LEN
) {
1017 ERR("Application context string length size is too large: %zu bytes",
1022 strcpy(ctxstr
, "$app.");
1023 p
= &ctxstr
[strlen("$app.")];
1024 recvlen
= ctxlen
- strlen("$app.");
1025 len
= ustcomm_recv_unix_sock(sock
, p
, recvlen
);
1027 case 0: /* orderly shutdown */
1031 if (len
== recvlen
) {
1032 DBG("app context data received");
1034 } else if (len
< 0) {
1035 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1036 if (len
== -ECONNRESET
) {
1037 ERR("%s remote end closed connection", sock_info
->name
);
1044 DBG("incorrect app context data message size: %zd", len
);
1049 /* Put : between provider and ctxname. */
1050 p
[lum
->u
.context
.u
.app_ctx
.provider_name_len
- 1] = ':';
1051 args
.app_context
.ctxname
= ctxstr
;
1058 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1059 (unsigned long) &lum
->u
,
1067 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1068 (unsigned long) &lum
->u
,
1075 lur
.handle
= lum
->handle
;
1079 lur
.ret_code
= LTTNG_UST_OK
;
1082 * Use -LTTNG_UST_ERR as wildcard for UST internal
1083 * error that are not caused by the transport, except if
1084 * we already have a more precise error message to
1087 if (ret
> -LTTNG_UST_ERR
) {
1088 /* Translate code to UST error. */
1091 lur
.ret_code
= -LTTNG_UST_ERR_EXIST
;
1094 lur
.ret_code
= -LTTNG_UST_ERR_INVAL
;
1097 lur
.ret_code
= -LTTNG_UST_ERR_NOENT
;
1100 lur
.ret_code
= -LTTNG_UST_ERR_PERM
;
1103 lur
.ret_code
= -LTTNG_UST_ERR_NOSYS
;
1106 lur
.ret_code
= -LTTNG_UST_ERR
;
1115 case LTTNG_UST_TRACER_VERSION
:
1116 lur
.u
.version
= lum
->u
.version
;
1118 case LTTNG_UST_TRACEPOINT_LIST_GET
:
1119 memcpy(&lur
.u
.tracepoint
, &lum
->u
.tracepoint
, sizeof(lur
.u
.tracepoint
));
1123 DBG("Return value: %d", lur
.ret_val
);
1128 * Performed delayed statedump operations outside of the UST
1129 * lock. We need to take the dynamic loader lock before we take
1130 * the UST lock internally within handle_pending_statedump().
1132 handle_pending_statedump(sock_info
);
1135 ret
= -LTTNG_UST_ERR_EXITING
;
1139 ret
= send_reply(sock
, &lur
);
1141 DBG("error sending reply");
1146 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1149 if (lur
.ret_code
== LTTNG_UST_OK
) {
1151 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET
:
1152 len
= ustcomm_send_unix_sock(sock
,
1153 &args
.field_list
.entry
,
1154 sizeof(args
.field_list
.entry
));
1159 if (len
!= sizeof(args
.field_list
.entry
)) {
1173 void cleanup_sock_info(struct sock_info
*sock_info
, int exiting
)
1177 if (sock_info
->root_handle
!= -1) {
1178 ret
= lttng_ust_objd_unref(sock_info
->root_handle
, 1);
1180 ERR("Error unref root handle");
1182 sock_info
->root_handle
= -1;
1187 * wait_shm_mmap, socket and notify socket are used by listener
1188 * threads outside of the ust lock, so we cannot tear them down
1189 * ourselves, because we cannot join on these threads. Leave
1190 * responsibility of cleaning up these resources to the OS
1196 sock_info
->registration_done
= 0;
1197 sock_info
->initial_statedump_done
= 0;
1199 if (sock_info
->socket
!= -1) {
1200 ret
= ustcomm_close_unix_sock(sock_info
->socket
);
1202 ERR("Error closing ust cmd socket");
1204 sock_info
->socket
= -1;
1206 if (sock_info
->notify_socket
!= -1) {
1207 ret
= ustcomm_close_unix_sock(sock_info
->notify_socket
);
1209 ERR("Error closing ust notify socket");
1211 sock_info
->notify_socket
= -1;
1213 if (sock_info
->wait_shm_mmap
) {
1216 page_size
= sysconf(_SC_PAGE_SIZE
);
1217 if (page_size
<= 0) {
1221 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1223 ret
= munmap(sock_info
->wait_shm_mmap
, page_size
);
1225 ERR("Error unmapping wait shm");
1228 sock_info
->wait_shm_mmap
= NULL
;
1233 * Using fork to set umask in the child process (not multi-thread safe).
1234 * We deal with the shm_open vs ftruncate race (happening when the
1235 * sessiond owns the shm and does not let everybody modify it, to ensure
1236 * safety against shm_unlink) by simply letting the mmap fail and
1237 * retrying after a few seconds.
1238 * For global shm, everybody has rw access to it until the sessiond
1242 int get_wait_shm(struct sock_info
*sock_info
, size_t mmap_size
)
1244 int wait_shm_fd
, ret
;
1248 * Try to open read-only.
1250 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
, O_RDONLY
, 0);
1251 if (wait_shm_fd
>= 0) {
1254 size_t bytes_read
= 0;
1257 * Try to read the fd. If unable to do so, try opening
1261 len
= read(wait_shm_fd
,
1262 &((char *) &tmp_read
)[bytes_read
],
1263 sizeof(tmp_read
) - bytes_read
);
1267 } while ((len
< 0 && errno
== EINTR
)
1268 || (len
> 0 && bytes_read
< sizeof(tmp_read
)));
1269 if (bytes_read
!= sizeof(tmp_read
)) {
1270 ret
= close(wait_shm_fd
);
1272 ERR("close wait_shm_fd");
1277 } else if (wait_shm_fd
< 0 && errno
!= ENOENT
) {
1279 * Real-only open did not work, and it's not because the
1280 * entry was not present. It's a failure that prohibits
1283 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
1289 * If the open failed because the file did not exist, or because
1290 * the file was not truncated yet, try creating it ourself.
1292 URCU_TLS(lttng_ust_nest_count
)++;
1294 URCU_TLS(lttng_ust_nest_count
)--;
1299 * Parent: wait for child to return, in which case the
1300 * shared memory map will have been created.
1302 pid
= wait(&status
);
1303 if (pid
< 0 || !WIFEXITED(status
) || WEXITSTATUS(status
) != 0) {
1308 * Try to open read-only again after creation.
1310 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
, O_RDONLY
, 0);
1311 if (wait_shm_fd
< 0) {
1313 * Real-only open did not work. It's a failure
1314 * that prohibits using shm.
1316 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
1320 } else if (pid
== 0) {
1324 create_mode
= S_IRUSR
| S_IWUSR
| S_IRGRP
;
1325 if (sock_info
->global
)
1326 create_mode
|= S_IROTH
| S_IWGRP
| S_IWOTH
;
1328 * We're alone in a child process, so we can modify the
1329 * process-wide umask.
1331 umask(~create_mode
);
1333 * Try creating shm (or get rw access).
1334 * We don't do an exclusive open, because we allow other
1335 * processes to create+ftruncate it concurrently.
1337 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
,
1338 O_RDWR
| O_CREAT
, create_mode
);
1339 if (wait_shm_fd
>= 0) {
1340 ret
= ftruncate(wait_shm_fd
, mmap_size
);
1342 PERROR("ftruncate");
1343 _exit(EXIT_FAILURE
);
1345 _exit(EXIT_SUCCESS
);
1348 * For local shm, we need to have rw access to accept
1349 * opening it: this means the local sessiond will be
1350 * able to wake us up. For global shm, we open it even
1351 * if rw access is not granted, because the root.root
1352 * sessiond will be able to override all rights and wake
1355 if (!sock_info
->global
&& errno
!= EACCES
) {
1356 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
1357 _exit(EXIT_FAILURE
);
1360 * The shm exists, but we cannot open it RW. Report
1363 _exit(EXIT_SUCCESS
);
1368 if (wait_shm_fd
>= 0 && !sock_info
->global
) {
1369 struct stat statbuf
;
1372 * Ensure that our user is the owner of the shm file for
1373 * local shm. If we do not own the file, it means our
1374 * sessiond will not have access to wake us up (there is
1375 * probably a rogue process trying to fake our
1376 * sessiond). Fallback to polling method in this case.
1378 ret
= fstat(wait_shm_fd
, &statbuf
);
1383 if (statbuf
.st_uid
!= getuid())
1389 ret
= close(wait_shm_fd
);
1391 PERROR("Error closing fd");
1397 char *get_map_shm(struct sock_info
*sock_info
)
1400 int wait_shm_fd
, ret
;
1401 char *wait_shm_mmap
;
1403 page_size
= sysconf(_SC_PAGE_SIZE
);
1404 if (page_size
<= 0) {
1408 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1412 lttng_ust_lock_fd_tracker();
1413 wait_shm_fd
= get_wait_shm(sock_info
, page_size
);
1414 if (wait_shm_fd
< 0) {
1415 lttng_ust_unlock_fd_tracker();
1419 ret
= lttng_ust_add_fd_to_tracker(wait_shm_fd
);
1421 ret
= close(wait_shm_fd
);
1423 PERROR("Error closing fd");
1425 lttng_ust_unlock_fd_tracker();
1430 lttng_ust_unlock_fd_tracker();
1432 wait_shm_mmap
= mmap(NULL
, page_size
, PROT_READ
,
1433 MAP_SHARED
, wait_shm_fd
, 0);
1435 /* close shm fd immediately after taking the mmap reference */
1436 lttng_ust_lock_fd_tracker();
1437 ret
= close(wait_shm_fd
);
1439 lttng_ust_delete_fd_from_tracker(wait_shm_fd
);
1441 PERROR("Error closing fd");
1443 lttng_ust_unlock_fd_tracker();
1445 if (wait_shm_mmap
== MAP_FAILED
) {
1446 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1449 return wait_shm_mmap
;
1456 void wait_for_sessiond(struct sock_info
*sock_info
)
1458 /* Use ust_lock to check if we should quit. */
1462 if (wait_poll_fallback
) {
1467 assert(sock_info
->wait_shm_mmap
);
1469 DBG("Waiting for %s apps sessiond", sock_info
->name
);
1470 /* Wait for futex wakeup */
1471 while (!uatomic_read((int32_t *) sock_info
->wait_shm_mmap
)) {
1472 if (!futex_async((int32_t *) sock_info
->wait_shm_mmap
, FUTEX_WAIT
, 0, NULL
, NULL
, 0)) {
1474 * Prior queued wakeups queued by unrelated code
1475 * using the same address can cause futex wait to
1476 * return 0 even through the futex value is still
1477 * 0 (spurious wakeups). Check the value again
1478 * in user-space to validate whether it really
1485 /* Value already changed. */
1488 /* Retry if interrupted by signal. */
1489 break; /* Get out of switch. Check again. */
1491 wait_poll_fallback
= 1;
1493 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1494 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1495 "Please upgrade your kernel "
1496 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1497 "mainline). LTTng-UST will use polling mode fallback.");
1516 * This thread does not allocate any resource, except within
1517 * handle_message, within mutex protection. This mutex protects against
1519 * The other moment it allocates resources is at socket connection, which
1520 * is also protected by the mutex.
1523 void *ust_listener_thread(void *arg
)
1525 struct sock_info
*sock_info
= arg
;
1526 int sock
, ret
, prev_connect_failed
= 0, has_waited
= 0, fd
;
1529 lttng_ust_fixup_tls();
1531 * If available, add '-ust' to the end of this thread's
1534 ret
= lttng_ust_setustprocname();
1536 ERR("Unable to set UST process name");
1539 /* Restart trying to connect to the session daemon */
1541 if (prev_connect_failed
) {
1542 /* Wait for sessiond availability with pipe */
1543 wait_for_sessiond(sock_info
);
1547 * Sleep for 5 seconds before retrying after a
1548 * sequence of failure / wait / failure. This
1549 * deals with a killed or broken session daemon.
1555 prev_connect_failed
= 0;
1562 if (sock_info
->socket
!= -1) {
1563 /* FD tracker is updated by ustcomm_close_unix_sock() */
1564 ret
= ustcomm_close_unix_sock(sock_info
->socket
);
1566 ERR("Error closing %s ust cmd socket",
1569 sock_info
->socket
= -1;
1571 if (sock_info
->notify_socket
!= -1) {
1572 /* FD tracker is updated by ustcomm_close_unix_sock() */
1573 ret
= ustcomm_close_unix_sock(sock_info
->notify_socket
);
1575 ERR("Error closing %s ust notify socket",
1578 sock_info
->notify_socket
= -1;
1583 * Register. We need to perform both connect and sending
1584 * registration message before doing the next connect otherwise
1585 * we may reach unix socket connect queue max limits and block
1586 * on the 2nd connect while the session daemon is awaiting the
1587 * first connect registration message.
1589 /* Connect cmd socket */
1590 lttng_ust_lock_fd_tracker();
1591 ret
= ustcomm_connect_unix_sock(sock_info
->sock_path
,
1592 get_connect_sock_timeout());
1594 lttng_ust_unlock_fd_tracker();
1595 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info
->name
);
1596 prev_connect_failed
= 1;
1599 * If we cannot find the sessiond daemon, don't delay
1600 * constructor execution.
1602 ret
= handle_register_failed(sock_info
);
1608 ret
= lttng_ust_add_fd_to_tracker(fd
);
1612 PERROR("close on sock_info->socket");
1615 lttng_ust_unlock_fd_tracker();
1620 sock_info
->socket
= ret
;
1621 lttng_ust_unlock_fd_tracker();
1625 * Unlock/relock ust lock because connect is blocking (with
1626 * timeout). Don't delay constructors on the ust lock for too
1634 * Create only one root handle per listener thread for the whole
1635 * process lifetime, so we ensure we get ID which is statically
1636 * assigned to the root handle.
1638 if (sock_info
->root_handle
== -1) {
1639 ret
= lttng_abi_create_root_handle();
1641 ERR("Error creating root handle");
1644 sock_info
->root_handle
= ret
;
1647 ret
= register_to_sessiond(sock_info
->socket
, USTCTL_SOCKET_CMD
);
1649 ERR("Error registering to %s ust cmd socket",
1651 prev_connect_failed
= 1;
1653 * If we cannot register to the sessiond daemon, don't
1654 * delay constructor execution.
1656 ret
= handle_register_failed(sock_info
);
1664 * Unlock/relock ust lock because connect is blocking (with
1665 * timeout). Don't delay constructors on the ust lock for too
1672 /* Connect notify socket */
1673 lttng_ust_lock_fd_tracker();
1674 ret
= ustcomm_connect_unix_sock(sock_info
->sock_path
,
1675 get_connect_sock_timeout());
1677 lttng_ust_unlock_fd_tracker();
1678 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info
->name
);
1679 prev_connect_failed
= 1;
1682 * If we cannot find the sessiond daemon, don't delay
1683 * constructor execution.
1685 ret
= handle_register_failed(sock_info
);
1692 ret
= lttng_ust_add_fd_to_tracker(fd
);
1696 PERROR("close on sock_info->notify_socket");
1699 lttng_ust_unlock_fd_tracker();
1704 sock_info
->notify_socket
= ret
;
1705 lttng_ust_unlock_fd_tracker();
1709 * Unlock/relock ust lock because connect is blocking (with
1710 * timeout). Don't delay constructors on the ust lock for too
1717 timeout
= get_notify_sock_timeout();
1720 * Give at least 10ms to sessiond to reply to
1725 ret
= ustcomm_setsockopt_rcv_timeout(sock_info
->notify_socket
,
1728 WARN("Error setting socket receive timeout");
1730 ret
= ustcomm_setsockopt_snd_timeout(sock_info
->notify_socket
,
1733 WARN("Error setting socket send timeout");
1735 } else if (timeout
< -1) {
1736 WARN("Unsupported timeout value %ld", timeout
);
1739 ret
= register_to_sessiond(sock_info
->notify_socket
,
1740 USTCTL_SOCKET_NOTIFY
);
1742 ERR("Error registering to %s ust notify socket",
1744 prev_connect_failed
= 1;
1746 * If we cannot register to the sessiond daemon, don't
1747 * delay constructor execution.
1749 ret
= handle_register_failed(sock_info
);
1754 sock
= sock_info
->socket
;
1760 struct ustcomm_ust_msg lum
;
1762 len
= ustcomm_recv_unix_sock(sock
, &lum
, sizeof(lum
));
1764 case 0: /* orderly shutdown */
1765 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info
->name
);
1770 * Either sessiond has shutdown or refused us by closing the socket.
1771 * In either case, we don't want to delay construction execution,
1772 * and we need to wait before retry.
1774 prev_connect_failed
= 1;
1776 * If we cannot register to the sessiond daemon, don't
1777 * delay constructor execution.
1779 ret
= handle_register_failed(sock_info
);
1784 print_cmd(lum
.cmd
, lum
.handle
);
1785 ret
= handle_message(sock_info
, sock
, &lum
);
1787 ERR("Error handling message for %s socket",
1790 * Close socket if protocol error is
1798 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1800 DBG("incorrect message size (%s socket): %zd", sock_info
->name
, len
);
1802 if (len
== -ECONNRESET
) {
1803 DBG("%s remote end closed connection", sock_info
->name
);
1814 /* Cleanup socket handles before trying to reconnect */
1815 lttng_ust_objd_table_owner_cleanup(sock_info
);
1817 goto restart
; /* try to reconnect */
1822 pthread_mutex_lock(&ust_exit_mutex
);
1823 sock_info
->thread_active
= 0;
1824 pthread_mutex_unlock(&ust_exit_mutex
);
1829 * Weak symbol to call when the ust malloc wrapper is not loaded.
1831 __attribute__((weak
))
1832 void lttng_ust_malloc_wrapper_init(void)
1837 * sessiond monitoring thread: monitor presence of global and per-user
1838 * sessiond by polling the application common named pipe.
1840 void __attribute__((constructor
)) lttng_ust_init(void)
1842 struct timespec constructor_timeout
;
1843 sigset_t sig_all_blocked
, orig_parent_mask
;
1844 pthread_attr_t thread_attr
;
1849 if (uatomic_xchg(&initialized
, 1) == 1)
1853 * Fixup interdependency between TLS fixup mutex (which happens
1854 * to be the dynamic linker mutex) and ust_lock, taken within
1857 lttng_ust_fixup_tls();
1859 lttng_ust_loaded
= 1;
1862 * We need to ensure that the liblttng-ust library is not unloaded to avoid
1863 * the unloading of code used by the ust_listener_threads as we can not
1864 * reliably know when they exited. To do that, manually load
1865 * liblttng-ust.so to increment the dynamic loader's internal refcount for
1866 * this library so it never becomes zero, thus never gets unloaded from the
1867 * address space of the process. Since we are already running in the
1868 * constructor of the LTTNG_UST_LIB_SO_NAME library, calling dlopen will
1869 * simply increment the refcount and no additionnal work is needed by the
1870 * dynamic loader as the shared library is already loaded in the address
1871 * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
1872 * unloading of the UST library if its refcount becomes zero (which should
1873 * never happen). Do the return value check but discard the handle at the
1874 * end of the function as it's not needed.
1876 handle
= dlopen(LTTNG_UST_LIB_SO_NAME
, RTLD_LAZY
| RTLD_NODELETE
);
1878 ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SO_NAME
);
1882 * We want precise control over the order in which we construct
1883 * our sub-libraries vs starting to receive commands from
1884 * sessiond (otherwise leading to errors when trying to create
1885 * sessiond before the init functions are completed).
1888 lttng_ust_getenv_init(); /* Needs init_usterr() to be completed. */
1890 lttng_ust_init_fd_tracker();
1891 lttng_ust_clock_init();
1892 lttng_ust_getcpu_init();
1893 lttng_ust_statedump_init();
1894 lttng_ring_buffer_metadata_client_init();
1895 lttng_ring_buffer_client_overwrite_init();
1896 lttng_ring_buffer_client_overwrite_rt_init();
1897 lttng_ring_buffer_client_discard_init();
1898 lttng_ring_buffer_client_discard_rt_init();
1899 lttng_perf_counter_init();
1901 * Invoke ust malloc wrapper init before starting other threads.
1903 lttng_ust_malloc_wrapper_init();
1905 timeout_mode
= get_constructor_timeout(&constructor_timeout
);
1907 get_allow_blocking();
1909 ret
= sem_init(&constructor_wait
, 0, 0);
1914 ret
= setup_global_apps();
1916 assert(global_apps
.allowed
== 0);
1917 DBG("global apps setup returned %d", ret
);
1920 ret
= setup_local_apps();
1922 assert(local_apps
.allowed
== 0);
1923 DBG("local apps setup returned %d", ret
);
1926 /* A new thread created by pthread_create inherits the signal mask
1927 * from the parent. To avoid any signal being received by the
1928 * listener thread, we block all signals temporarily in the parent,
1929 * while we create the listener thread.
1931 sigfillset(&sig_all_blocked
);
1932 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_parent_mask
);
1934 ERR("pthread_sigmask: %s", strerror(ret
));
1937 ret
= pthread_attr_init(&thread_attr
);
1939 ERR("pthread_attr_init: %s", strerror(ret
));
1941 ret
= pthread_attr_setdetachstate(&thread_attr
, PTHREAD_CREATE_DETACHED
);
1943 ERR("pthread_attr_setdetachstate: %s", strerror(ret
));
1946 if (global_apps
.allowed
) {
1947 pthread_mutex_lock(&ust_exit_mutex
);
1948 ret
= pthread_create(&global_apps
.ust_listener
, &thread_attr
,
1949 ust_listener_thread
, &global_apps
);
1951 ERR("pthread_create global: %s", strerror(ret
));
1953 global_apps
.thread_active
= 1;
1954 pthread_mutex_unlock(&ust_exit_mutex
);
1956 handle_register_done(&global_apps
);
1959 if (local_apps
.allowed
) {
1960 pthread_mutex_lock(&ust_exit_mutex
);
1961 ret
= pthread_create(&local_apps
.ust_listener
, &thread_attr
,
1962 ust_listener_thread
, &local_apps
);
1964 ERR("pthread_create local: %s", strerror(ret
));
1966 local_apps
.thread_active
= 1;
1967 pthread_mutex_unlock(&ust_exit_mutex
);
1969 handle_register_done(&local_apps
);
1971 ret
= pthread_attr_destroy(&thread_attr
);
1973 ERR("pthread_attr_destroy: %s", strerror(ret
));
1976 /* Restore original signal mask in parent */
1977 ret
= pthread_sigmask(SIG_SETMASK
, &orig_parent_mask
, NULL
);
1979 ERR("pthread_sigmask: %s", strerror(ret
));
1982 switch (timeout_mode
) {
1983 case 1: /* timeout wait */
1985 ret
= sem_timedwait(&constructor_wait
,
1986 &constructor_timeout
);
1987 } while (ret
< 0 && errno
== EINTR
);
1991 ERR("Timed out waiting for lttng-sessiond");
1994 PERROR("sem_timedwait");
1997 ERR("Unexpected error \"%s\" returned by sem_timedwait",
2002 case -1:/* wait forever */
2004 ret
= sem_wait(&constructor_wait
);
2005 } while (ret
< 0 && errno
== EINTR
);
2012 ERR("Unexpected error \"%s\" returned by sem_wait",
2017 case 0: /* no timeout */
2023 void lttng_ust_cleanup(int exiting
)
2025 cleanup_sock_info(&global_apps
, exiting
);
2026 cleanup_sock_info(&local_apps
, exiting
);
2027 local_apps
.allowed
= 0;
2028 global_apps
.allowed
= 0;
2030 * The teardown in this function all affect data structures
2031 * accessed under the UST lock by the listener thread. This
2032 * lock, along with the lttng_ust_comm_should_quit flag, ensure
2033 * that none of these threads are accessing this data at this
2036 lttng_ust_abi_exit();
2037 lttng_ust_events_exit();
2038 lttng_perf_counter_exit();
2039 lttng_ring_buffer_client_discard_rt_exit();
2040 lttng_ring_buffer_client_discard_exit();
2041 lttng_ring_buffer_client_overwrite_rt_exit();
2042 lttng_ring_buffer_client_overwrite_exit();
2043 lttng_ring_buffer_metadata_client_exit();
2044 lttng_ust_statedump_destroy();
2047 /* Reinitialize values for fork */
2048 sem_count
= sem_count_initial_value
;
2049 lttng_ust_comm_should_quit
= 0;
2054 void __attribute__((destructor
)) lttng_ust_exit(void)
2059 * Using pthread_cancel here because:
2060 * A) we don't want to hang application teardown.
2061 * B) the thread is not allocating any resource.
2065 * Require the communication thread to quit. Synchronize with
2066 * mutexes to ensure it is not in a mutex critical section when
2067 * pthread_cancel is later called.
2070 lttng_ust_comm_should_quit
= 1;
2073 pthread_mutex_lock(&ust_exit_mutex
);
2074 /* cancel threads */
2075 if (global_apps
.thread_active
) {
2076 ret
= pthread_cancel(global_apps
.ust_listener
);
2078 ERR("Error cancelling global ust listener thread: %s",
2081 global_apps
.thread_active
= 0;
2084 if (local_apps
.thread_active
) {
2085 ret
= pthread_cancel(local_apps
.ust_listener
);
2087 ERR("Error cancelling local ust listener thread: %s",
2090 local_apps
.thread_active
= 0;
2093 pthread_mutex_unlock(&ust_exit_mutex
);
2096 * Do NOT join threads: use of sys_futex makes it impossible to
2097 * join the threads without using async-cancel, but async-cancel
2098 * is delivered by a signal, which could hit the target thread
2099 * anywhere in its code path, including while the ust_lock() is
2100 * held, causing a deadlock for the other thread. Let the OS
2101 * cleanup the threads if there are stalled in a syscall.
2103 lttng_ust_cleanup(1);
2107 void ust_context_ns_reset(void)
2109 lttng_context_pid_ns_reset();
2110 lttng_context_cgroup_ns_reset();
2111 lttng_context_ipc_ns_reset();
2112 lttng_context_mnt_ns_reset();
2113 lttng_context_net_ns_reset();
2114 lttng_context_user_ns_reset();
2115 lttng_context_uts_ns_reset();
2119 void ust_context_vuids_reset(void)
2121 lttng_context_vuid_reset();
2122 lttng_context_veuid_reset();
2123 lttng_context_vsuid_reset();
2127 void ust_context_vgids_reset(void)
2129 lttng_context_vgid_reset();
2130 lttng_context_vegid_reset();
2131 lttng_context_vsgid_reset();
2135 * We exclude the worker threads across fork and clone (except
2136 * CLONE_VM), because these system calls only keep the forking thread
2137 * running in the child. Therefore, we don't want to call fork or clone
2138 * in the middle of an tracepoint or ust tracing state modification.
2139 * Holding this mutex protects these structures across fork and clone.
2141 void ust_before_fork(sigset_t
*save_sigset
)
2144 * Disable signals. This is to avoid that the child intervenes
2145 * before it is properly setup for tracing. It is safer to
2146 * disable all signals, because then we know we are not breaking
2147 * anything by restoring the original mask.
2152 /* Fixup lttng-ust TLS. */
2153 lttng_ust_fixup_tls();
2155 if (URCU_TLS(lttng_ust_nest_count
))
2157 /* Disable signals */
2158 sigfillset(&all_sigs
);
2159 ret
= sigprocmask(SIG_BLOCK
, &all_sigs
, save_sigset
);
2161 PERROR("sigprocmask");
2164 pthread_mutex_lock(&ust_fork_mutex
);
2167 urcu_bp_before_fork();
2168 lttng_ust_lock_fd_tracker();
2172 static void ust_after_fork_common(sigset_t
*restore_sigset
)
2176 DBG("process %d", getpid());
2177 lttng_perf_unlock();
2178 lttng_ust_unlock_fd_tracker();
2181 pthread_mutex_unlock(&ust_fork_mutex
);
2183 /* Restore signals */
2184 ret
= sigprocmask(SIG_SETMASK
, restore_sigset
, NULL
);
2186 PERROR("sigprocmask");
2190 void ust_after_fork_parent(sigset_t
*restore_sigset
)
2192 if (URCU_TLS(lttng_ust_nest_count
))
2194 DBG("process %d", getpid());
2195 urcu_bp_after_fork_parent();
2196 /* Release mutexes and reenable signals */
2197 ust_after_fork_common(restore_sigset
);
2201 * After fork, in the child, we need to cleanup all the leftover state,
2202 * except the worker thread which already magically disappeared thanks
2203 * to the weird Linux fork semantics. After tyding up, we call
2204 * lttng_ust_init() again to start over as a new PID.
2206 * This is meant for forks() that have tracing in the child between the
2207 * fork and following exec call (if there is any).
2209 void ust_after_fork_child(sigset_t
*restore_sigset
)
2211 if (URCU_TLS(lttng_ust_nest_count
))
2213 lttng_context_vpid_reset();
2214 lttng_context_vtid_reset();
2215 lttng_context_procname_reset();
2216 ust_context_ns_reset();
2217 ust_context_vuids_reset();
2218 ust_context_vgids_reset();
2219 DBG("process %d", getpid());
2220 /* Release urcu mutexes */
2221 urcu_bp_after_fork_child();
2222 lttng_ust_cleanup(0);
2223 /* Release mutexes and reenable signals */
2224 ust_after_fork_common(restore_sigset
);
2228 void ust_after_setns(void)
2230 ust_context_ns_reset();
2231 ust_context_vuids_reset();
2232 ust_context_vgids_reset();
2235 void ust_after_unshare(void)
2237 ust_context_ns_reset();
2238 ust_context_vuids_reset();
2239 ust_context_vgids_reset();
2242 void ust_after_setuid(void)
2244 ust_context_vuids_reset();
2247 void ust_after_seteuid(void)
2249 ust_context_vuids_reset();
2252 void ust_after_setreuid(void)
2254 ust_context_vuids_reset();
2257 void ust_after_setresuid(void)
2259 ust_context_vuids_reset();
2262 void ust_after_setgid(void)
2264 ust_context_vgids_reset();
2267 void ust_after_setegid(void)
2269 ust_context_vgids_reset();
2272 void ust_after_setregid(void)
2274 ust_context_vgids_reset();
2277 void ust_after_setresgid(void)
2279 ust_context_vgids_reset();
2282 void lttng_ust_sockinfo_session_enabled(void *owner
)
2284 struct sock_info
*sock_info
= owner
;
2285 sock_info
->statedump_pending
= 1;