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
26 #include <sys/types.h>
27 #include <sys/socket.h>
30 #include <sys/types.h>
37 #include <semaphore.h>
42 #include <urcu/uatomic.h>
43 #include <urcu/futex.h>
44 #include <urcu/compiler.h>
46 #include <lttng/ust-events.h>
47 #include <lttng/ust-abi.h>
48 #include <lttng/ust.h>
49 #include <lttng/ust-error.h>
50 #include <lttng/ust-ctl.h>
51 #include <urcu/tls-compat.h>
54 #include <usterr-signal-safe.h>
56 #include "tracepoint-internal.h"
57 #include "lttng-tracer-core.h"
59 #include "../libringbuffer/rb-init.h"
60 #include "lttng-ust-statedump.h"
62 #include "../libringbuffer/getcpu.h"
65 /* Concatenate lttng ust shared library name with its major version number. */
66 #define LTTNG_UST_LIB_SO_NAME "liblttng-ust.so." __ust_stringify(CONFIG_LTTNG_UST_LIBRARY_VERSION_MAJOR)
69 * Has lttng ust comm constructor been called ?
71 static int initialized
;
74 * The ust_lock/ust_unlock lock is used as a communication thread mutex.
75 * Held when handling a command, also held by fork() to deal with
76 * removal of threads, and by exit path.
78 * The UST lock is the centralized mutex across UST tracing control and
81 * ust_exit_mutex must never nest in ust_mutex.
83 * ust_fork_mutex must never nest in ust_mutex.
85 * ust_mutex_nest is a per-thread nesting counter, allowing the perf
86 * counter lazy initialization called by events within the statedump,
87 * which traces while the ust_mutex is held.
89 * ust_lock nests within the dynamic loader lock (within glibc) because
90 * it is taken within the library constructor.
92 * The ust fd tracker lock nests within the ust_mutex.
94 static pthread_mutex_t ust_mutex
= PTHREAD_MUTEX_INITIALIZER
;
96 /* Allow nesting the ust_mutex within the same thread. */
97 static DEFINE_URCU_TLS(int, ust_mutex_nest
);
100 * ust_exit_mutex protects thread_active variable wrt thread exit. It
101 * cannot be done by ust_mutex because pthread_cancel(), which takes an
102 * internal libc lock, cannot nest within ust_mutex.
104 * It never nests within a ust_mutex.
106 static pthread_mutex_t ust_exit_mutex
= PTHREAD_MUTEX_INITIALIZER
;
109 * ust_fork_mutex protects base address statedump tracing against forks. It
110 * prevents the dynamic loader lock to be taken (by base address statedump
111 * tracing) while a fork is happening, thus preventing deadlock issues with
112 * the dynamic loader lock.
114 static pthread_mutex_t ust_fork_mutex
= PTHREAD_MUTEX_INITIALIZER
;
116 /* Should the ust comm thread quit ? */
117 static int lttng_ust_comm_should_quit
;
120 * This variable can be tested by applications to check whether
121 * lttng-ust is loaded. They simply have to define their own
122 * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the
123 * library constructor.
125 int lttng_ust_loaded
__attribute__((weak
));
128 * Return 0 on success, -1 if should quit.
129 * The lock is taken in both cases.
134 sigset_t sig_all_blocked
, orig_mask
;
137 ret
= pthread_setcancelstate(PTHREAD_CANCEL_DISABLE
, &oldstate
);
139 ERR("pthread_setcancelstate: %s", strerror(ret
));
141 if (oldstate
!= PTHREAD_CANCEL_ENABLE
) {
142 ERR("pthread_setcancelstate: unexpected oldstate");
144 sigfillset(&sig_all_blocked
);
145 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_mask
);
147 ERR("pthread_sigmask: %s", strerror(ret
));
149 if (!URCU_TLS(ust_mutex_nest
)++)
150 pthread_mutex_lock(&ust_mutex
);
151 ret
= pthread_sigmask(SIG_SETMASK
, &orig_mask
, NULL
);
153 ERR("pthread_sigmask: %s", strerror(ret
));
155 if (lttng_ust_comm_should_quit
) {
163 * ust_lock_nocheck() can be used in constructors/destructors, because
164 * they are already nested within the dynamic loader lock, and therefore
165 * have exclusive access against execution of liblttng-ust destructor.
168 void ust_lock_nocheck(void)
170 sigset_t sig_all_blocked
, orig_mask
;
173 ret
= pthread_setcancelstate(PTHREAD_CANCEL_DISABLE
, &oldstate
);
175 ERR("pthread_setcancelstate: %s", strerror(ret
));
177 if (oldstate
!= PTHREAD_CANCEL_ENABLE
) {
178 ERR("pthread_setcancelstate: unexpected oldstate");
180 sigfillset(&sig_all_blocked
);
181 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_mask
);
183 ERR("pthread_sigmask: %s", strerror(ret
));
185 if (!URCU_TLS(ust_mutex_nest
)++)
186 pthread_mutex_lock(&ust_mutex
);
187 ret
= pthread_sigmask(SIG_SETMASK
, &orig_mask
, NULL
);
189 ERR("pthread_sigmask: %s", strerror(ret
));
196 void ust_unlock(void)
198 sigset_t sig_all_blocked
, orig_mask
;
201 sigfillset(&sig_all_blocked
);
202 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_mask
);
204 ERR("pthread_sigmask: %s", strerror(ret
));
206 if (!--URCU_TLS(ust_mutex_nest
))
207 pthread_mutex_unlock(&ust_mutex
);
208 ret
= pthread_sigmask(SIG_SETMASK
, &orig_mask
, NULL
);
210 ERR("pthread_sigmask: %s", strerror(ret
));
212 ret
= pthread_setcancelstate(PTHREAD_CANCEL_ENABLE
, &oldstate
);
214 ERR("pthread_setcancelstate: %s", strerror(ret
));
216 if (oldstate
!= PTHREAD_CANCEL_DISABLE
) {
217 ERR("pthread_setcancelstate: unexpected oldstate");
222 * Wait for either of these before continuing to the main
224 * - the register_done message from sessiond daemon
225 * (will let the sessiond daemon enable sessions before main
227 * - sessiond daemon is not reachable.
228 * - timeout (ensuring applications are resilient to session
231 static sem_t constructor_wait
;
233 * Doing this for both the global and local sessiond.
236 sem_count_initial_value
= 4,
239 static int sem_count
= sem_count_initial_value
;
242 * Counting nesting within lttng-ust. Used to ensure that calling fork()
243 * from liblttng-ust does not execute the pre/post fork handlers.
245 static DEFINE_URCU_TLS(int, lttng_ust_nest_count
);
248 * Info about socket and associated listener thread.
252 pthread_t ust_listener
; /* listener thread */
254 int registration_done
;
259 char sock_path
[PATH_MAX
];
263 char wait_shm_path
[PATH_MAX
];
265 /* Keep track of lazy state dump not performed yet. */
266 int statedump_pending
;
267 int initial_statedump_done
;
268 /* Keep procname for statedump */
269 char procname
[LTTNG_UST_PROCNAME_LEN
];
272 /* Socket from app (connect) to session daemon (listen) for communication */
273 struct sock_info global_apps
= {
278 .registration_done
= 0,
282 .sock_path
= LTTNG_DEFAULT_RUNDIR
"/" LTTNG_UST_SOCK_FILENAME
,
286 .wait_shm_path
= "/" LTTNG_UST_WAIT_FILENAME
,
288 .statedump_pending
= 0,
289 .initial_statedump_done
= 0,
293 /* TODO: allow global_apps_sock_path override */
295 struct sock_info local_apps
= {
299 .registration_done
= 0,
300 .allowed
= 0, /* Check setuid bit first */
306 .statedump_pending
= 0,
307 .initial_statedump_done
= 0,
311 static int wait_poll_fallback
;
313 static const char *cmd_name_mapping
[] = {
314 [ LTTNG_UST_RELEASE
] = "Release",
315 [ LTTNG_UST_SESSION
] = "Create Session",
316 [ LTTNG_UST_TRACER_VERSION
] = "Get Tracer Version",
318 [ LTTNG_UST_TRACEPOINT_LIST
] = "Create Tracepoint List",
319 [ LTTNG_UST_WAIT_QUIESCENT
] = "Wait for Quiescent State",
320 [ LTTNG_UST_REGISTER_DONE
] = "Registration Done",
321 [ LTTNG_UST_TRACEPOINT_FIELD_LIST
] = "Create Tracepoint Field List",
323 [ LTTNG_UST_TRIGGER_GROUP_CREATE
] = "Create trigger group",
324 [ LTTNG_UST_TRIGGER_CREATE
] = "Create trigger",
326 /* Session FD commands */
327 [ LTTNG_UST_CHANNEL
] = "Create Channel",
328 [ LTTNG_UST_SESSION_START
] = "Start Session",
329 [ LTTNG_UST_SESSION_STOP
] = "Stop Session",
331 /* Channel FD commands */
332 [ LTTNG_UST_STREAM
] = "Create Stream",
333 [ LTTNG_UST_EVENT
] = "Create Event",
335 /* Event and Channel FD commands */
336 [ LTTNG_UST_CONTEXT
] = "Create Context",
337 [ LTTNG_UST_FLUSH_BUFFER
] = "Flush Buffer",
339 /* Event, Channel and Session commands */
340 [ LTTNG_UST_ENABLE
] = "Enable",
341 [ LTTNG_UST_DISABLE
] = "Disable",
343 /* Tracepoint list commands */
344 [ LTTNG_UST_TRACEPOINT_LIST_GET
] = "List Next Tracepoint",
345 [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET
] = "List Next Tracepoint Field",
347 /* Event FD commands */
348 [ LTTNG_UST_FILTER
] = "Create Filter",
349 [ LTTNG_UST_EXCLUSION
] = "Add exclusions to event",
352 static const char *str_timeout
;
353 static int got_timeout_env
;
355 extern void lttng_ring_buffer_client_overwrite_init(void);
356 extern void lttng_ring_buffer_client_overwrite_rt_init(void);
357 extern void lttng_ring_buffer_client_discard_init(void);
358 extern void lttng_ring_buffer_client_discard_rt_init(void);
359 extern void lttng_ring_buffer_metadata_client_init(void);
360 extern void lttng_ring_buffer_client_overwrite_exit(void);
361 extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
362 extern void lttng_ring_buffer_client_discard_exit(void);
363 extern void lttng_ring_buffer_client_discard_rt_exit(void);
364 extern void lttng_ring_buffer_metadata_client_exit(void);
366 static char *get_map_shm(struct sock_info
*sock_info
);
368 ssize_t
lttng_ust_read(int fd
, void *buf
, size_t len
)
371 size_t copied
= 0, to_copy
= len
;
374 ret
= read(fd
, buf
+ copied
, to_copy
);
379 } while ((ret
> 0 && to_copy
> 0)
380 || (ret
< 0 && errno
== EINTR
));
387 * Returns the HOME directory path. Caller MUST NOT free(3) the returned
391 const char *get_lttng_home_dir(void)
395 val
= (const char *) lttng_getenv("LTTNG_HOME");
399 return (const char *) lttng_getenv("HOME");
403 * Force a read (imply TLS fixup for dlopen) of TLS variables.
406 void lttng_fixup_nest_count_tls(void)
408 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count
)));
412 void lttng_fixup_ust_mutex_nest_tls(void)
414 asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest
)));
421 void lttng_fixup_urcu_bp_tls(void)
427 void lttng_ust_fixup_tls(void)
429 lttng_fixup_urcu_bp_tls();
430 lttng_fixup_ringbuffer_tls();
431 lttng_fixup_vtid_tls();
432 lttng_fixup_nest_count_tls();
433 lttng_fixup_procname_tls();
434 lttng_fixup_ust_mutex_nest_tls();
435 lttng_ust_fixup_perf_counter_tls();
436 lttng_ust_fixup_fd_tracker_tls();
437 lttng_fixup_cgroup_ns_tls();
438 lttng_fixup_ipc_ns_tls();
439 lttng_fixup_net_ns_tls();
440 lttng_fixup_uts_ns_tls();
443 int lttng_get_notify_socket(void *owner
)
445 struct sock_info
*info
= owner
;
447 return info
->notify_socket
;
452 char* lttng_ust_sockinfo_get_procname(void *owner
)
454 struct sock_info
*info
= owner
;
456 return info
->procname
;
460 void print_cmd(int cmd
, int handle
)
462 const char *cmd_name
= "Unknown";
464 if (cmd
>= 0 && cmd
< LTTNG_ARRAY_SIZE(cmd_name_mapping
)
465 && cmd_name_mapping
[cmd
]) {
466 cmd_name
= cmd_name_mapping
[cmd
];
468 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
470 lttng_ust_obj_get_name(handle
), handle
);
474 int setup_global_apps(void)
477 assert(!global_apps
.wait_shm_mmap
);
479 global_apps
.wait_shm_mmap
= get_map_shm(&global_apps
);
480 if (!global_apps
.wait_shm_mmap
) {
481 WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing.");
482 global_apps
.allowed
= 0;
487 global_apps
.allowed
= 1;
488 lttng_ust_getprocname(global_apps
.procname
);
493 int setup_local_apps(void)
496 const char *home_dir
;
499 assert(!local_apps
.wait_shm_mmap
);
503 * Disallow per-user tracing for setuid binaries.
505 if (uid
!= geteuid()) {
506 assert(local_apps
.allowed
== 0);
510 home_dir
= get_lttng_home_dir();
512 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
513 assert(local_apps
.allowed
== 0);
517 local_apps
.allowed
= 1;
518 snprintf(local_apps
.sock_path
, PATH_MAX
, "%s/%s/%s",
520 LTTNG_DEFAULT_HOME_RUNDIR
,
521 LTTNG_UST_SOCK_FILENAME
);
522 snprintf(local_apps
.wait_shm_path
, PATH_MAX
, "/%s-%u",
523 LTTNG_UST_WAIT_FILENAME
,
526 local_apps
.wait_shm_mmap
= get_map_shm(&local_apps
);
527 if (!local_apps
.wait_shm_mmap
) {
528 WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing.");
529 local_apps
.allowed
= 0;
534 lttng_ust_getprocname(local_apps
.procname
);
540 * Get socket timeout, in ms.
541 * -1: wait forever. 0: don't wait. >0: timeout, in ms.
544 long get_timeout(void)
546 long constructor_delay_ms
= LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS
;
548 if (!got_timeout_env
) {
549 str_timeout
= lttng_getenv("LTTNG_UST_REGISTER_TIMEOUT");
553 constructor_delay_ms
= strtol(str_timeout
, NULL
, 10);
554 /* All negative values are considered as "-1". */
555 if (constructor_delay_ms
< -1)
556 constructor_delay_ms
= -1;
557 return constructor_delay_ms
;
560 /* Timeout for notify socket send and recv. */
562 long get_notify_sock_timeout(void)
564 return get_timeout();
567 /* Timeout for connecting to cmd and notify sockets. */
569 long get_connect_sock_timeout(void)
571 return get_timeout();
575 * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
578 int get_constructor_timeout(struct timespec
*constructor_timeout
)
580 long constructor_delay_ms
;
583 constructor_delay_ms
= get_timeout();
585 switch (constructor_delay_ms
) {
586 case -1:/* fall-through */
588 return constructor_delay_ms
;
594 * If we are unable to find the current time, don't wait.
596 ret
= clock_gettime(CLOCK_REALTIME
, constructor_timeout
);
601 constructor_timeout
->tv_sec
+= constructor_delay_ms
/ 1000UL;
602 constructor_timeout
->tv_nsec
+=
603 (constructor_delay_ms
% 1000UL) * 1000000UL;
604 if (constructor_timeout
->tv_nsec
>= 1000000000UL) {
605 constructor_timeout
->tv_sec
++;
606 constructor_timeout
->tv_nsec
-= 1000000000UL;
608 /* Timeout wait (constructor_delay_ms). */
613 void get_allow_blocking(void)
615 const char *str_allow_blocking
=
616 lttng_getenv("LTTNG_UST_ALLOW_BLOCKING");
618 if (str_allow_blocking
) {
619 DBG("%s environment variable is set",
620 "LTTNG_UST_ALLOW_BLOCKING");
621 lttng_ust_ringbuffer_set_allow_blocking();
626 int register_to_sessiond(int socket
, enum ustctl_socket_type type
)
628 return ustcomm_send_reg_msg(socket
,
631 lttng_alignof(uint8_t) * CHAR_BIT
,
632 lttng_alignof(uint16_t) * CHAR_BIT
,
633 lttng_alignof(uint32_t) * CHAR_BIT
,
634 lttng_alignof(uint64_t) * CHAR_BIT
,
635 lttng_alignof(unsigned long) * CHAR_BIT
);
639 int send_reply(int sock
, struct ustcomm_ust_reply
*lur
)
643 len
= ustcomm_send_unix_sock(sock
, lur
, sizeof(*lur
));
646 DBG("message successfully sent");
649 if (len
== -ECONNRESET
) {
650 DBG("remote end closed connection");
655 DBG("incorrect message size: %zd", len
);
661 void decrement_sem_count(unsigned int count
)
665 assert(uatomic_read(&sem_count
) >= count
);
667 if (uatomic_read(&sem_count
) <= 0) {
671 ret
= uatomic_add_return(&sem_count
, -count
);
673 ret
= sem_post(&constructor_wait
);
679 int handle_register_done(struct sock_info
*sock_info
)
681 if (sock_info
->registration_done
)
683 sock_info
->registration_done
= 1;
685 decrement_sem_count(1);
686 if (!sock_info
->statedump_pending
) {
687 sock_info
->initial_statedump_done
= 1;
688 decrement_sem_count(1);
695 int handle_register_failed(struct sock_info
*sock_info
)
697 if (sock_info
->registration_done
)
699 sock_info
->registration_done
= 1;
700 sock_info
->initial_statedump_done
= 1;
702 decrement_sem_count(2);
708 * Only execute pending statedump after the constructor semaphore has
709 * been posted by the current listener thread. This means statedump will
710 * only be performed after the "registration done" command is received
711 * from this thread's session daemon.
713 * This ensures we don't run into deadlock issues with the dynamic
714 * loader mutex, which is held while the constructor is called and
715 * waiting on the constructor semaphore. All operations requiring this
716 * dynamic loader lock need to be postponed using this mechanism.
718 * In a scenario with two session daemons connected to the application,
719 * it is possible that the first listener thread which receives the
720 * registration done command issues its statedump while the dynamic
721 * loader lock is still held by the application constructor waiting on
722 * the semaphore. It will however be allowed to proceed when the
723 * second session daemon sends the registration done command to the
724 * second listener thread. This situation therefore does not produce
728 void handle_pending_statedump(struct sock_info
*sock_info
)
730 if (sock_info
->registration_done
&& sock_info
->statedump_pending
) {
731 sock_info
->statedump_pending
= 0;
732 pthread_mutex_lock(&ust_fork_mutex
);
733 lttng_handle_pending_statedump(sock_info
);
734 pthread_mutex_unlock(&ust_fork_mutex
);
736 if (!sock_info
->initial_statedump_done
) {
737 sock_info
->initial_statedump_done
= 1;
738 decrement_sem_count(1);
744 int handle_message(struct sock_info
*sock_info
,
745 int sock
, struct ustcomm_ust_msg
*lum
)
748 const struct lttng_ust_objd_ops
*ops
;
749 struct ustcomm_ust_reply lur
;
751 char ctxstr
[LTTNG_UST_SYM_NAME_LEN
]; /* App context string. */
754 memset(&lur
, 0, sizeof(lur
));
757 ret
= -LTTNG_UST_ERR_EXITING
;
761 ops
= objd_ops(lum
->handle
);
768 case LTTNG_UST_REGISTER_DONE
:
769 if (lum
->handle
== LTTNG_UST_ROOT_HANDLE
)
770 ret
= handle_register_done(sock_info
);
774 case LTTNG_UST_RELEASE
:
775 if (lum
->handle
== LTTNG_UST_ROOT_HANDLE
)
778 ret
= lttng_ust_objd_unref(lum
->handle
, 1);
780 case LTTNG_UST_FILTER
:
782 /* Receive filter data */
783 struct lttng_ust_filter_bytecode_node
*bytecode
;
785 if (lum
->u
.filter
.data_size
> FILTER_BYTECODE_MAX_LEN
) {
786 ERR("Filter data size is too large: %u bytes",
787 lum
->u
.filter
.data_size
);
792 if (lum
->u
.filter
.reloc_offset
> lum
->u
.filter
.data_size
) {
793 ERR("Filter reloc offset %u is not within data",
794 lum
->u
.filter
.reloc_offset
);
799 bytecode
= zmalloc(sizeof(*bytecode
) + lum
->u
.filter
.data_size
);
804 len
= ustcomm_recv_unix_sock(sock
, bytecode
->bc
.data
,
805 lum
->u
.filter
.data_size
);
807 case 0: /* orderly shutdown */
812 if (len
== lum
->u
.filter
.data_size
) {
813 DBG("filter data received");
815 } else if (len
< 0) {
816 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
817 if (len
== -ECONNRESET
) {
818 ERR("%s remote end closed connection", sock_info
->name
);
827 DBG("incorrect filter data message size: %zd", len
);
833 bytecode
->bc
.len
= lum
->u
.filter
.data_size
;
834 bytecode
->bc
.reloc_offset
= lum
->u
.filter
.reloc_offset
;
835 bytecode
->bc
.seqnum
= lum
->u
.filter
.seqnum
;
837 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
838 (unsigned long) bytecode
,
843 /* don't free bytecode if everything went fine. */
850 case LTTNG_UST_EXCLUSION
:
852 /* Receive exclusion names */
853 struct lttng_ust_excluder_node
*node
;
856 count
= lum
->u
.exclusion
.count
;
858 /* There are no names to read */
862 node
= zmalloc(sizeof(*node
) +
863 count
* LTTNG_UST_SYM_NAME_LEN
);
868 node
->excluder
.count
= count
;
869 len
= ustcomm_recv_unix_sock(sock
, node
->excluder
.names
,
870 count
* LTTNG_UST_SYM_NAME_LEN
);
872 case 0: /* orderly shutdown */
877 if (len
== count
* LTTNG_UST_SYM_NAME_LEN
) {
878 DBG("Exclusion data received");
880 } else if (len
< 0) {
881 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
882 if (len
== -ECONNRESET
) {
883 ERR("%s remote end closed connection", sock_info
->name
);
892 DBG("Incorrect exclusion data message size: %zd", len
);
899 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
900 (unsigned long) node
,
905 /* Don't free exclusion data if everything went fine. */
912 case LTTNG_UST_TRIGGER_GROUP_CREATE
:
914 int trigger_notif_fd
;
916 len
= ustcomm_recv_trigger_notif_fd_from_sessiond(sock
,
919 case 0: /* orderly shutdown */
926 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
927 if (len
== -ECONNRESET
) {
928 ERR("%s remote end closed connection", sock_info
->name
);
935 DBG("incorrect trigger fd message size: %zd", len
);
940 args
.trigger_handle
.trigger_notif_fd
= trigger_notif_fd
;
942 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
943 (unsigned long) &lum
->u
,
949 case LTTNG_UST_CHANNEL
:
954 len
= ustcomm_recv_channel_from_sessiond(sock
,
955 &chan_data
, lum
->u
.channel
.len
,
958 case 0: /* orderly shutdown */
962 if (len
== lum
->u
.channel
.len
) {
963 DBG("channel data received");
965 } else if (len
< 0) {
966 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
967 if (len
== -ECONNRESET
) {
968 ERR("%s remote end closed connection", sock_info
->name
);
975 DBG("incorrect channel data message size: %zd", len
);
980 args
.channel
.chan_data
= chan_data
;
981 args
.channel
.wakeup_fd
= wakeup_fd
;
983 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
984 (unsigned long) &lum
->u
,
990 case LTTNG_UST_STREAM
:
992 /* Receive shm_fd, wakeup_fd */
993 ret
= ustcomm_recv_stream_from_sessiond(sock
,
996 &args
.stream
.wakeup_fd
);
1002 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1003 (unsigned long) &lum
->u
,
1009 case LTTNG_UST_CONTEXT
:
1010 switch (lum
->u
.context
.ctx
) {
1011 case LTTNG_UST_CONTEXT_APP_CONTEXT
:
1014 size_t ctxlen
, recvlen
;
1016 ctxlen
= strlen("$app.") + lum
->u
.context
.u
.app_ctx
.provider_name_len
- 1
1017 + strlen(":") + lum
->u
.context
.u
.app_ctx
.ctx_name_len
;
1018 if (ctxlen
>= LTTNG_UST_SYM_NAME_LEN
) {
1019 ERR("Application context string length size is too large: %zu bytes",
1024 strcpy(ctxstr
, "$app.");
1025 p
= &ctxstr
[strlen("$app.")];
1026 recvlen
= ctxlen
- strlen("$app.");
1027 len
= ustcomm_recv_unix_sock(sock
, p
, recvlen
);
1029 case 0: /* orderly shutdown */
1033 if (len
== recvlen
) {
1034 DBG("app context data received");
1036 } else if (len
< 0) {
1037 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1038 if (len
== -ECONNRESET
) {
1039 ERR("%s remote end closed connection", sock_info
->name
);
1046 DBG("incorrect app context data message size: %zd", len
);
1051 /* Put : between provider and ctxname. */
1052 p
[lum
->u
.context
.u
.app_ctx
.provider_name_len
- 1] = ':';
1053 args
.app_context
.ctxname
= ctxstr
;
1060 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1061 (unsigned long) &lum
->u
,
1069 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1070 (unsigned long) &lum
->u
,
1077 lur
.handle
= lum
->handle
;
1081 lur
.ret_code
= LTTNG_UST_OK
;
1084 * Use -LTTNG_UST_ERR as wildcard for UST internal
1085 * error that are not caused by the transport, except if
1086 * we already have a more precise error message to
1089 if (ret
> -LTTNG_UST_ERR
) {
1090 /* Translate code to UST error. */
1093 lur
.ret_code
= -LTTNG_UST_ERR_EXIST
;
1096 lur
.ret_code
= -LTTNG_UST_ERR_INVAL
;
1099 lur
.ret_code
= -LTTNG_UST_ERR_NOENT
;
1102 lur
.ret_code
= -LTTNG_UST_ERR_PERM
;
1105 lur
.ret_code
= -LTTNG_UST_ERR_NOSYS
;
1108 lur
.ret_code
= -LTTNG_UST_ERR
;
1117 case LTTNG_UST_TRACER_VERSION
:
1118 lur
.u
.version
= lum
->u
.version
;
1120 case LTTNG_UST_TRACEPOINT_LIST_GET
:
1121 memcpy(&lur
.u
.tracepoint
, &lum
->u
.tracepoint
, sizeof(lur
.u
.tracepoint
));
1125 DBG("Return value: %d", lur
.ret_val
);
1130 * Performed delayed statedump operations outside of the UST
1131 * lock. We need to take the dynamic loader lock before we take
1132 * the UST lock internally within handle_pending_statedump().
1134 handle_pending_statedump(sock_info
);
1137 ret
= -LTTNG_UST_ERR_EXITING
;
1141 ret
= send_reply(sock
, &lur
);
1143 DBG("error sending reply");
1148 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1151 if (lur
.ret_code
== LTTNG_UST_OK
) {
1153 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET
:
1154 len
= ustcomm_send_unix_sock(sock
,
1155 &args
.field_list
.entry
,
1156 sizeof(args
.field_list
.entry
));
1161 if (len
!= sizeof(args
.field_list
.entry
)) {
1175 void cleanup_sock_info(struct sock_info
*sock_info
, int exiting
)
1179 if (sock_info
->root_handle
!= -1) {
1180 ret
= lttng_ust_objd_unref(sock_info
->root_handle
, 1);
1182 ERR("Error unref root handle");
1184 sock_info
->root_handle
= -1;
1186 sock_info
->registration_done
= 0;
1187 sock_info
->initial_statedump_done
= 0;
1190 * wait_shm_mmap, socket and notify socket are used by listener
1191 * threads outside of the ust lock, so we cannot tear them down
1192 * ourselves, because we cannot join on these threads. Leave
1193 * responsibility of cleaning up these resources to the OS
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 if (uatomic_read((int32_t *) sock_info
->wait_shm_mmap
))
1474 while (futex_async((int32_t *) sock_info
->wait_shm_mmap
,
1475 FUTEX_WAIT
, 0, NULL
, NULL
, 0)) {
1478 /* Value already changed. */
1481 /* Retry if interrupted by signal. */
1482 break; /* Get out of switch. */
1484 wait_poll_fallback
= 1;
1486 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1487 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1488 "Please upgrade your kernel "
1489 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1490 "mainline). LTTng-UST will use polling mode fallback.");
1509 * This thread does not allocate any resource, except within
1510 * handle_message, within mutex protection. This mutex protects against
1512 * The other moment it allocates resources is at socket connection, which
1513 * is also protected by the mutex.
1516 void *ust_listener_thread(void *arg
)
1518 struct sock_info
*sock_info
= arg
;
1519 int sock
, ret
, prev_connect_failed
= 0, has_waited
= 0, fd
;
1522 lttng_ust_fixup_tls();
1524 * If available, add '-ust' to the end of this thread's
1527 ret
= lttng_ust_setustprocname();
1529 ERR("Unable to set UST process name");
1532 /* Restart trying to connect to the session daemon */
1534 if (prev_connect_failed
) {
1535 /* Wait for sessiond availability with pipe */
1536 wait_for_sessiond(sock_info
);
1540 * Sleep for 5 seconds before retrying after a
1541 * sequence of failure / wait / failure. This
1542 * deals with a killed or broken session daemon.
1548 prev_connect_failed
= 0;
1555 if (sock_info
->socket
!= -1) {
1556 /* FD tracker is updated by ustcomm_close_unix_sock() */
1557 ret
= ustcomm_close_unix_sock(sock_info
->socket
);
1559 ERR("Error closing %s ust cmd socket",
1562 sock_info
->socket
= -1;
1564 if (sock_info
->notify_socket
!= -1) {
1565 /* FD tracker is updated by ustcomm_close_unix_sock() */
1566 ret
= ustcomm_close_unix_sock(sock_info
->notify_socket
);
1568 ERR("Error closing %s ust notify socket",
1571 sock_info
->notify_socket
= -1;
1576 * Register. We need to perform both connect and sending
1577 * registration message before doing the next connect otherwise
1578 * we may reach unix socket connect queue max limits and block
1579 * on the 2nd connect while the session daemon is awaiting the
1580 * first connect registration message.
1582 /* Connect cmd socket */
1583 lttng_ust_lock_fd_tracker();
1584 ret
= ustcomm_connect_unix_sock(sock_info
->sock_path
,
1585 get_connect_sock_timeout());
1587 lttng_ust_unlock_fd_tracker();
1588 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info
->name
);
1589 prev_connect_failed
= 1;
1592 * If we cannot find the sessiond daemon, don't delay
1593 * constructor execution.
1595 ret
= handle_register_failed(sock_info
);
1601 ret
= lttng_ust_add_fd_to_tracker(fd
);
1605 PERROR("close on sock_info->socket");
1608 lttng_ust_unlock_fd_tracker();
1613 sock_info
->socket
= ret
;
1614 lttng_ust_unlock_fd_tracker();
1618 * Unlock/relock ust lock because connect is blocking (with
1619 * timeout). Don't delay constructors on the ust lock for too
1627 * Create only one root handle per listener thread for the whole
1628 * process lifetime, so we ensure we get ID which is statically
1629 * assigned to the root handle.
1631 if (sock_info
->root_handle
== -1) {
1632 ret
= lttng_abi_create_root_handle();
1634 ERR("Error creating root handle");
1637 sock_info
->root_handle
= ret
;
1640 ret
= register_to_sessiond(sock_info
->socket
, USTCTL_SOCKET_CMD
);
1642 ERR("Error registering to %s ust cmd socket",
1644 prev_connect_failed
= 1;
1646 * If we cannot register to the sessiond daemon, don't
1647 * delay constructor execution.
1649 ret
= handle_register_failed(sock_info
);
1657 * Unlock/relock ust lock because connect is blocking (with
1658 * timeout). Don't delay constructors on the ust lock for too
1665 /* Connect notify socket */
1666 lttng_ust_lock_fd_tracker();
1667 ret
= ustcomm_connect_unix_sock(sock_info
->sock_path
,
1668 get_connect_sock_timeout());
1670 lttng_ust_unlock_fd_tracker();
1671 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info
->name
);
1672 prev_connect_failed
= 1;
1675 * If we cannot find the sessiond daemon, don't delay
1676 * constructor execution.
1678 ret
= handle_register_failed(sock_info
);
1685 ret
= lttng_ust_add_fd_to_tracker(fd
);
1689 PERROR("close on sock_info->notify_socket");
1692 lttng_ust_unlock_fd_tracker();
1697 sock_info
->notify_socket
= ret
;
1698 lttng_ust_unlock_fd_tracker();
1702 * Unlock/relock ust lock because connect is blocking (with
1703 * timeout). Don't delay constructors on the ust lock for too
1710 timeout
= get_notify_sock_timeout();
1713 * Give at least 10ms to sessiond to reply to
1718 ret
= ustcomm_setsockopt_rcv_timeout(sock_info
->notify_socket
,
1721 WARN("Error setting socket receive timeout");
1723 ret
= ustcomm_setsockopt_snd_timeout(sock_info
->notify_socket
,
1726 WARN("Error setting socket send timeout");
1728 } else if (timeout
< -1) {
1729 WARN("Unsupported timeout value %ld", timeout
);
1732 ret
= register_to_sessiond(sock_info
->notify_socket
,
1733 USTCTL_SOCKET_NOTIFY
);
1735 ERR("Error registering to %s ust notify socket",
1737 prev_connect_failed
= 1;
1739 * If we cannot register to the sessiond daemon, don't
1740 * delay constructor execution.
1742 ret
= handle_register_failed(sock_info
);
1747 sock
= sock_info
->socket
;
1753 struct ustcomm_ust_msg lum
;
1755 len
= ustcomm_recv_unix_sock(sock
, &lum
, sizeof(lum
));
1757 case 0: /* orderly shutdown */
1758 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info
->name
);
1763 * Either sessiond has shutdown or refused us by closing the socket.
1764 * In either case, we don't want to delay construction execution,
1765 * and we need to wait before retry.
1767 prev_connect_failed
= 1;
1769 * If we cannot register to the sessiond daemon, don't
1770 * delay constructor execution.
1772 ret
= handle_register_failed(sock_info
);
1777 print_cmd(lum
.cmd
, lum
.handle
);
1778 ret
= handle_message(sock_info
, sock
, &lum
);
1780 ERR("Error handling message for %s socket",
1783 * Close socket if protocol error is
1791 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1793 DBG("incorrect message size (%s socket): %zd", sock_info
->name
, len
);
1795 if (len
== -ECONNRESET
) {
1796 DBG("%s remote end closed connection", sock_info
->name
);
1807 /* Cleanup socket handles before trying to reconnect */
1808 lttng_ust_objd_table_owner_cleanup(sock_info
);
1810 goto restart
; /* try to reconnect */
1815 pthread_mutex_lock(&ust_exit_mutex
);
1816 sock_info
->thread_active
= 0;
1817 pthread_mutex_unlock(&ust_exit_mutex
);
1822 * Weak symbol to call when the ust malloc wrapper is not loaded.
1824 __attribute__((weak
))
1825 void lttng_ust_malloc_wrapper_init(void)
1830 * sessiond monitoring thread: monitor presence of global and per-user
1831 * sessiond by polling the application common named pipe.
1833 void __attribute__((constructor
)) lttng_ust_init(void)
1835 struct timespec constructor_timeout
;
1836 sigset_t sig_all_blocked
, orig_parent_mask
;
1837 pthread_attr_t thread_attr
;
1842 if (uatomic_xchg(&initialized
, 1) == 1)
1846 * Fixup interdependency between TLS fixup mutex (which happens
1847 * to be the dynamic linker mutex) and ust_lock, taken within
1850 lttng_ust_fixup_tls();
1852 lttng_ust_loaded
= 1;
1855 * We need to ensure that the liblttng-ust library is not unloaded to avoid
1856 * the unloading of code used by the ust_listener_threads as we can not
1857 * reliably know when they exited. To do that, manually load
1858 * liblttng-ust.so to increment the dynamic loader's internal refcount for
1859 * this library so it never becomes zero, thus never gets unloaded from the
1860 * address space of the process. Since we are already running in the
1861 * constructor of the LTTNG_UST_LIB_SO_NAME library, calling dlopen will
1862 * simply increment the refcount and no additionnal work is needed by the
1863 * dynamic loader as the shared library is already loaded in the address
1864 * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
1865 * unloading of the UST library if its refcount becomes zero (which should
1866 * never happen). Do the return value check but discard the handle at the
1867 * end of the function as it's not needed.
1869 handle
= dlopen(LTTNG_UST_LIB_SO_NAME
, RTLD_LAZY
| RTLD_NODELETE
);
1871 ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SO_NAME
);
1875 * We want precise control over the order in which we construct
1876 * our sub-libraries vs starting to receive commands from
1877 * sessiond (otherwise leading to errors when trying to create
1878 * sessiond before the init functions are completed).
1881 lttng_ust_getenv_init(); /* Needs init_usterr() to be completed. */
1883 lttng_ust_init_fd_tracker();
1884 lttng_ust_clock_init();
1885 lttng_ust_getcpu_init();
1886 lttng_ust_statedump_init();
1887 lttng_ring_buffer_metadata_client_init();
1888 lttng_ring_buffer_client_overwrite_init();
1889 lttng_ring_buffer_client_overwrite_rt_init();
1890 lttng_ring_buffer_client_discard_init();
1891 lttng_ring_buffer_client_discard_rt_init();
1892 lttng_perf_counter_init();
1894 * Invoke ust malloc wrapper init before starting other threads.
1896 lttng_ust_malloc_wrapper_init();
1898 timeout_mode
= get_constructor_timeout(&constructor_timeout
);
1900 get_allow_blocking();
1902 ret
= sem_init(&constructor_wait
, 0, 0);
1907 ret
= setup_global_apps();
1909 assert(global_apps
.allowed
== 0);
1910 DBG("global apps setup returned %d", ret
);
1913 ret
= setup_local_apps();
1915 assert(local_apps
.allowed
== 0);
1916 DBG("local apps setup returned %d", ret
);
1919 /* A new thread created by pthread_create inherits the signal mask
1920 * from the parent. To avoid any signal being received by the
1921 * listener thread, we block all signals temporarily in the parent,
1922 * while we create the listener thread.
1924 sigfillset(&sig_all_blocked
);
1925 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_parent_mask
);
1927 ERR("pthread_sigmask: %s", strerror(ret
));
1930 ret
= pthread_attr_init(&thread_attr
);
1932 ERR("pthread_attr_init: %s", strerror(ret
));
1934 ret
= pthread_attr_setdetachstate(&thread_attr
, PTHREAD_CREATE_DETACHED
);
1936 ERR("pthread_attr_setdetachstate: %s", strerror(ret
));
1939 if (global_apps
.allowed
) {
1940 pthread_mutex_lock(&ust_exit_mutex
);
1941 ret
= pthread_create(&global_apps
.ust_listener
, &thread_attr
,
1942 ust_listener_thread
, &global_apps
);
1944 ERR("pthread_create global: %s", strerror(ret
));
1946 global_apps
.thread_active
= 1;
1947 pthread_mutex_unlock(&ust_exit_mutex
);
1949 handle_register_done(&global_apps
);
1952 if (local_apps
.allowed
) {
1953 pthread_mutex_lock(&ust_exit_mutex
);
1954 ret
= pthread_create(&local_apps
.ust_listener
, &thread_attr
,
1955 ust_listener_thread
, &local_apps
);
1957 ERR("pthread_create local: %s", strerror(ret
));
1959 local_apps
.thread_active
= 1;
1960 pthread_mutex_unlock(&ust_exit_mutex
);
1962 handle_register_done(&local_apps
);
1964 ret
= pthread_attr_destroy(&thread_attr
);
1966 ERR("pthread_attr_destroy: %s", strerror(ret
));
1969 /* Restore original signal mask in parent */
1970 ret
= pthread_sigmask(SIG_SETMASK
, &orig_parent_mask
, NULL
);
1972 ERR("pthread_sigmask: %s", strerror(ret
));
1975 switch (timeout_mode
) {
1976 case 1: /* timeout wait */
1978 ret
= sem_timedwait(&constructor_wait
,
1979 &constructor_timeout
);
1980 } while (ret
< 0 && errno
== EINTR
);
1984 ERR("Timed out waiting for lttng-sessiond");
1987 PERROR("sem_timedwait");
1990 ERR("Unexpected error \"%s\" returned by sem_timedwait",
1995 case -1:/* wait forever */
1997 ret
= sem_wait(&constructor_wait
);
1998 } while (ret
< 0 && errno
== EINTR
);
2005 ERR("Unexpected error \"%s\" returned by sem_wait",
2010 case 0: /* no timeout */
2016 void lttng_ust_cleanup(int exiting
)
2018 cleanup_sock_info(&global_apps
, exiting
);
2019 cleanup_sock_info(&local_apps
, exiting
);
2020 local_apps
.allowed
= 0;
2021 global_apps
.allowed
= 0;
2023 * The teardown in this function all affect data structures
2024 * accessed under the UST lock by the listener thread. This
2025 * lock, along with the lttng_ust_comm_should_quit flag, ensure
2026 * that none of these threads are accessing this data at this
2029 lttng_ust_abi_exit();
2030 lttng_ust_events_exit();
2031 lttng_perf_counter_exit();
2032 lttng_ring_buffer_client_discard_rt_exit();
2033 lttng_ring_buffer_client_discard_exit();
2034 lttng_ring_buffer_client_overwrite_rt_exit();
2035 lttng_ring_buffer_client_overwrite_exit();
2036 lttng_ring_buffer_metadata_client_exit();
2037 lttng_ust_statedump_destroy();
2040 /* Reinitialize values for fork */
2041 sem_count
= sem_count_initial_value
;
2042 lttng_ust_comm_should_quit
= 0;
2047 void __attribute__((destructor
)) lttng_ust_exit(void)
2052 * Using pthread_cancel here because:
2053 * A) we don't want to hang application teardown.
2054 * B) the thread is not allocating any resource.
2058 * Require the communication thread to quit. Synchronize with
2059 * mutexes to ensure it is not in a mutex critical section when
2060 * pthread_cancel is later called.
2063 lttng_ust_comm_should_quit
= 1;
2066 pthread_mutex_lock(&ust_exit_mutex
);
2067 /* cancel threads */
2068 if (global_apps
.thread_active
) {
2069 ret
= pthread_cancel(global_apps
.ust_listener
);
2071 ERR("Error cancelling global ust listener thread: %s",
2074 global_apps
.thread_active
= 0;
2077 if (local_apps
.thread_active
) {
2078 ret
= pthread_cancel(local_apps
.ust_listener
);
2080 ERR("Error cancelling local ust listener thread: %s",
2083 local_apps
.thread_active
= 0;
2086 pthread_mutex_unlock(&ust_exit_mutex
);
2089 * Do NOT join threads: use of sys_futex makes it impossible to
2090 * join the threads without using async-cancel, but async-cancel
2091 * is delivered by a signal, which could hit the target thread
2092 * anywhere in its code path, including while the ust_lock() is
2093 * held, causing a deadlock for the other thread. Let the OS
2094 * cleanup the threads if there are stalled in a syscall.
2096 lttng_ust_cleanup(1);
2100 void ust_context_ns_reset(void)
2102 lttng_context_pid_ns_reset();
2103 lttng_context_cgroup_ns_reset();
2104 lttng_context_ipc_ns_reset();
2105 lttng_context_mnt_ns_reset();
2106 lttng_context_net_ns_reset();
2107 lttng_context_user_ns_reset();
2108 lttng_context_uts_ns_reset();
2112 void ust_context_vuids_reset(void)
2114 lttng_context_vuid_reset();
2115 lttng_context_veuid_reset();
2116 lttng_context_vsuid_reset();
2120 void ust_context_vgids_reset(void)
2122 lttng_context_vgid_reset();
2123 lttng_context_vegid_reset();
2124 lttng_context_vsgid_reset();
2128 * We exclude the worker threads across fork and clone (except
2129 * CLONE_VM), because these system calls only keep the forking thread
2130 * running in the child. Therefore, we don't want to call fork or clone
2131 * in the middle of an tracepoint or ust tracing state modification.
2132 * Holding this mutex protects these structures across fork and clone.
2134 void ust_before_fork(sigset_t
*save_sigset
)
2137 * Disable signals. This is to avoid that the child intervenes
2138 * before it is properly setup for tracing. It is safer to
2139 * disable all signals, because then we know we are not breaking
2140 * anything by restoring the original mask.
2145 /* Fixup lttng-ust TLS. */
2146 lttng_ust_fixup_tls();
2148 if (URCU_TLS(lttng_ust_nest_count
))
2150 /* Disable signals */
2151 sigfillset(&all_sigs
);
2152 ret
= sigprocmask(SIG_BLOCK
, &all_sigs
, save_sigset
);
2154 PERROR("sigprocmask");
2157 pthread_mutex_lock(&ust_fork_mutex
);
2160 urcu_bp_before_fork();
2161 lttng_ust_lock_fd_tracker();
2165 static void ust_after_fork_common(sigset_t
*restore_sigset
)
2169 DBG("process %d", getpid());
2170 lttng_perf_unlock();
2171 lttng_ust_unlock_fd_tracker();
2174 pthread_mutex_unlock(&ust_fork_mutex
);
2176 /* Restore signals */
2177 ret
= sigprocmask(SIG_SETMASK
, restore_sigset
, NULL
);
2179 PERROR("sigprocmask");
2183 void ust_after_fork_parent(sigset_t
*restore_sigset
)
2185 if (URCU_TLS(lttng_ust_nest_count
))
2187 DBG("process %d", getpid());
2188 urcu_bp_after_fork_parent();
2189 /* Release mutexes and reenable signals */
2190 ust_after_fork_common(restore_sigset
);
2194 * After fork, in the child, we need to cleanup all the leftover state,
2195 * except the worker thread which already magically disappeared thanks
2196 * to the weird Linux fork semantics. After tyding up, we call
2197 * lttng_ust_init() again to start over as a new PID.
2199 * This is meant for forks() that have tracing in the child between the
2200 * fork and following exec call (if there is any).
2202 void ust_after_fork_child(sigset_t
*restore_sigset
)
2204 if (URCU_TLS(lttng_ust_nest_count
))
2206 lttng_context_vpid_reset();
2207 lttng_context_vtid_reset();
2208 lttng_context_procname_reset();
2209 ust_context_ns_reset();
2210 ust_context_vuids_reset();
2211 ust_context_vgids_reset();
2212 DBG("process %d", getpid());
2213 /* Release urcu mutexes */
2214 urcu_bp_after_fork_child();
2215 lttng_ust_cleanup(0);
2216 /* Release mutexes and reenable signals */
2217 ust_after_fork_common(restore_sigset
);
2221 void ust_after_setns(void)
2223 ust_context_ns_reset();
2224 ust_context_vuids_reset();
2225 ust_context_vgids_reset();
2228 void ust_after_unshare(void)
2230 ust_context_ns_reset();
2231 ust_context_vuids_reset();
2232 ust_context_vgids_reset();
2235 void ust_after_setuid(void)
2237 ust_context_vuids_reset();
2240 void ust_after_seteuid(void)
2242 ust_context_vuids_reset();
2245 void ust_after_setreuid(void)
2247 ust_context_vuids_reset();
2250 void ust_after_setresuid(void)
2252 ust_context_vuids_reset();
2255 void ust_after_setgid(void)
2257 ust_context_vgids_reset();
2260 void ust_after_setegid(void)
2262 ust_context_vgids_reset();
2265 void ust_after_setregid(void)
2267 ust_context_vgids_reset();
2270 void ust_after_setresgid(void)
2272 ust_context_vgids_reset();
2275 void lttng_ust_sockinfo_session_enabled(void *owner
)
2277 struct sock_info
*sock_info
= owner
;
2278 sock_info
->statedump_pending
= 1;