2 * SPDX-License-Identifier: LGPL-2.1-only
4 * Copyright (C) 2011 David Goulet <david.goulet@polymtl.ca>
5 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
11 #include <sys/types.h>
12 #include <sys/socket.h>
15 #include <sys/types.h>
22 #include <semaphore.h>
27 #include <urcu/uatomic.h>
29 #include <urcu/compiler.h>
30 #include <lttng/urcu/urcu-ust.h>
32 #include <lttng/ust-events.h>
33 #include <lttng/ust-abi.h>
34 #include <lttng/ust.h>
35 #include <lttng/ust-error.h>
36 #include <lttng/ust-ctl.h>
37 #include <lttng/ust-libc-wrapper.h>
38 #include <urcu/tls-compat.h>
41 #include <usterr-signal-safe.h>
42 #include <ust-helper.h>
43 #include "tracepoint-internal.h"
44 #include "lttng-tracer-core.h"
46 #include "../libringbuffer/rb-init.h"
47 #include "lttng-ust-statedump.h"
49 #include "../libringbuffer/getcpu.h"
51 #include "ust-events-internal.h"
52 #include "context-internal.h"
53 #include "ust-compat.h"
55 /* Concatenate lttng ust shared library name with its major version number. */
56 #define LTTNG_UST_LIB_SO_NAME "liblttng-ust.so." __ust_stringify(CONFIG_LTTNG_UST_LIBRARY_VERSION_MAJOR)
59 * Has lttng ust comm constructor been called ?
61 static int initialized
;
64 * The ust_lock/ust_unlock lock is used as a communication thread mutex.
65 * Held when handling a command, also held by fork() to deal with
66 * removal of threads, and by exit path.
68 * The UST lock is the centralized mutex across UST tracing control and
71 * ust_exit_mutex must never nest in ust_mutex.
73 * ust_fork_mutex must never nest in ust_mutex.
75 * ust_mutex_nest is a per-thread nesting counter, allowing the perf
76 * counter lazy initialization called by events within the statedump,
77 * which traces while the ust_mutex is held.
79 * ust_lock nests within the dynamic loader lock (within glibc) because
80 * it is taken within the library constructor.
82 * The ust fd tracker lock nests within the ust_mutex.
84 static pthread_mutex_t ust_mutex
= PTHREAD_MUTEX_INITIALIZER
;
86 /* Allow nesting the ust_mutex within the same thread. */
87 static DEFINE_URCU_TLS(int, ust_mutex_nest
);
90 * ust_exit_mutex protects thread_active variable wrt thread exit. It
91 * cannot be done by ust_mutex because pthread_cancel(), which takes an
92 * internal libc lock, cannot nest within ust_mutex.
94 * It never nests within a ust_mutex.
96 static pthread_mutex_t ust_exit_mutex
= PTHREAD_MUTEX_INITIALIZER
;
99 * ust_fork_mutex protects base address statedump tracing against forks. It
100 * prevents the dynamic loader lock to be taken (by base address statedump
101 * tracing) while a fork is happening, thus preventing deadlock issues with
102 * the dynamic loader lock.
104 static pthread_mutex_t ust_fork_mutex
= PTHREAD_MUTEX_INITIALIZER
;
106 /* Should the ust comm thread quit ? */
107 static int lttng_ust_comm_should_quit
;
110 * This variable can be tested by applications to check whether
111 * lttng-ust is loaded. They simply have to define their own
112 * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the
113 * library constructor.
115 int lttng_ust_loaded
__attribute__((weak
));
118 * Return 0 on success, -1 if should quit.
119 * The lock is taken in both cases.
124 sigset_t sig_all_blocked
, orig_mask
;
127 ret
= pthread_setcancelstate(PTHREAD_CANCEL_DISABLE
, &oldstate
);
129 ERR("pthread_setcancelstate: %s", strerror(ret
));
131 if (oldstate
!= PTHREAD_CANCEL_ENABLE
) {
132 ERR("pthread_setcancelstate: unexpected oldstate");
134 sigfillset(&sig_all_blocked
);
135 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_mask
);
137 ERR("pthread_sigmask: %s", strerror(ret
));
139 if (!URCU_TLS(ust_mutex_nest
)++)
140 pthread_mutex_lock(&ust_mutex
);
141 ret
= pthread_sigmask(SIG_SETMASK
, &orig_mask
, NULL
);
143 ERR("pthread_sigmask: %s", strerror(ret
));
145 if (lttng_ust_comm_should_quit
) {
153 * ust_lock_nocheck() can be used in constructors/destructors, because
154 * they are already nested within the dynamic loader lock, and therefore
155 * have exclusive access against execution of liblttng-ust destructor.
158 void ust_lock_nocheck(void)
160 sigset_t sig_all_blocked
, orig_mask
;
163 ret
= pthread_setcancelstate(PTHREAD_CANCEL_DISABLE
, &oldstate
);
165 ERR("pthread_setcancelstate: %s", strerror(ret
));
167 if (oldstate
!= PTHREAD_CANCEL_ENABLE
) {
168 ERR("pthread_setcancelstate: unexpected oldstate");
170 sigfillset(&sig_all_blocked
);
171 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_mask
);
173 ERR("pthread_sigmask: %s", strerror(ret
));
175 if (!URCU_TLS(ust_mutex_nest
)++)
176 pthread_mutex_lock(&ust_mutex
);
177 ret
= pthread_sigmask(SIG_SETMASK
, &orig_mask
, NULL
);
179 ERR("pthread_sigmask: %s", strerror(ret
));
186 void ust_unlock(void)
188 sigset_t sig_all_blocked
, orig_mask
;
191 sigfillset(&sig_all_blocked
);
192 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_mask
);
194 ERR("pthread_sigmask: %s", strerror(ret
));
196 if (!--URCU_TLS(ust_mutex_nest
))
197 pthread_mutex_unlock(&ust_mutex
);
198 ret
= pthread_sigmask(SIG_SETMASK
, &orig_mask
, NULL
);
200 ERR("pthread_sigmask: %s", strerror(ret
));
202 ret
= pthread_setcancelstate(PTHREAD_CANCEL_ENABLE
, &oldstate
);
204 ERR("pthread_setcancelstate: %s", strerror(ret
));
206 if (oldstate
!= PTHREAD_CANCEL_DISABLE
) {
207 ERR("pthread_setcancelstate: unexpected oldstate");
212 * Wait for either of these before continuing to the main
214 * - the register_done message from sessiond daemon
215 * (will let the sessiond daemon enable sessions before main
217 * - sessiond daemon is not reachable.
218 * - timeout (ensuring applications are resilient to session
221 static sem_t constructor_wait
;
223 * Doing this for both the global and local sessiond.
226 sem_count_initial_value
= 4,
229 static int sem_count
= sem_count_initial_value
;
232 * Counting nesting within lttng-ust. Used to ensure that calling fork()
233 * from liblttng-ust does not execute the pre/post fork handlers.
235 static DEFINE_URCU_TLS(int, lttng_ust_nest_count
);
238 * Info about socket and associated listener thread.
242 pthread_t ust_listener
; /* listener thread */
244 int registration_done
;
249 char sock_path
[PATH_MAX
];
253 char wait_shm_path
[PATH_MAX
];
255 /* Keep track of lazy state dump not performed yet. */
256 int statedump_pending
;
257 int initial_statedump_done
;
258 /* Keep procname for statedump */
259 char procname
[LTTNG_UST_ABI_PROCNAME_LEN
];
262 /* Socket from app (connect) to session daemon (listen) for communication */
263 struct sock_info global_apps
= {
268 .registration_done
= 0,
272 .sock_path
= LTTNG_DEFAULT_RUNDIR
"/" LTTNG_UST_SOCK_FILENAME
,
276 .wait_shm_path
= "/" LTTNG_UST_WAIT_FILENAME
,
278 .statedump_pending
= 0,
279 .initial_statedump_done
= 0,
283 /* TODO: allow global_apps_sock_path override */
285 struct sock_info local_apps
= {
289 .registration_done
= 0,
290 .allowed
= 0, /* Check setuid bit first */
296 .statedump_pending
= 0,
297 .initial_statedump_done
= 0,
301 static int wait_poll_fallback
;
303 static const char *cmd_name_mapping
[] = {
304 [ LTTNG_UST_ABI_RELEASE
] = "Release",
305 [ LTTNG_UST_ABI_SESSION
] = "Create Session",
306 [ LTTNG_UST_ABI_TRACER_VERSION
] = "Get Tracer Version",
308 [ LTTNG_UST_ABI_TRACEPOINT_LIST
] = "Create Tracepoint List",
309 [ LTTNG_UST_ABI_WAIT_QUIESCENT
] = "Wait for Quiescent State",
310 [ LTTNG_UST_ABI_REGISTER_DONE
] = "Registration Done",
311 [ LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST
] = "Create Tracepoint Field List",
313 [ LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE
] = "Create event notifier group",
315 /* Session FD commands */
316 [ LTTNG_UST_ABI_CHANNEL
] = "Create Channel",
317 [ LTTNG_UST_ABI_SESSION_START
] = "Start Session",
318 [ LTTNG_UST_ABI_SESSION_STOP
] = "Stop Session",
320 /* Channel FD commands */
321 [ LTTNG_UST_ABI_STREAM
] = "Create Stream",
322 [ LTTNG_UST_ABI_EVENT
] = "Create Event",
324 /* Event and Channel FD commands */
325 [ LTTNG_UST_ABI_CONTEXT
] = "Create Context",
326 [ LTTNG_UST_ABI_FLUSH_BUFFER
] = "Flush Buffer",
328 /* Event, Channel and Session commands */
329 [ LTTNG_UST_ABI_ENABLE
] = "Enable",
330 [ LTTNG_UST_ABI_DISABLE
] = "Disable",
332 /* Tracepoint list commands */
333 [ LTTNG_UST_ABI_TRACEPOINT_LIST_GET
] = "List Next Tracepoint",
334 [ LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST_GET
] = "List Next Tracepoint Field",
336 /* Event FD commands */
337 [ LTTNG_UST_ABI_FILTER
] = "Create Filter",
338 [ LTTNG_UST_ABI_EXCLUSION
] = "Add exclusions to event",
340 /* Event notifier group commands */
341 [ LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE
] = "Create event notifier",
343 /* Session and event notifier group commands */
344 [ LTTNG_UST_ABI_COUNTER
] = "Create Counter",
346 /* Counter commands */
347 [ LTTNG_UST_ABI_COUNTER_GLOBAL
] = "Create Counter Global",
348 [ LTTNG_UST_ABI_COUNTER_CPU
] = "Create Counter CPU",
351 static const char *str_timeout
;
352 static int got_timeout_env
;
354 extern void lttng_ring_buffer_client_overwrite_init(void);
355 extern void lttng_ring_buffer_client_overwrite_rt_init(void);
356 extern void lttng_ring_buffer_client_discard_init(void);
357 extern void lttng_ring_buffer_client_discard_rt_init(void);
358 extern void lttng_ring_buffer_metadata_client_init(void);
359 extern void lttng_ring_buffer_client_overwrite_exit(void);
360 extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
361 extern void lttng_ring_buffer_client_discard_exit(void);
362 extern void lttng_ring_buffer_client_discard_rt_exit(void);
363 extern void lttng_ring_buffer_metadata_client_exit(void);
365 extern void lttng_counter_client_percpu_32_modular_init(void);
367 extern void lttng_counter_client_percpu_32_modular_exit(void);
369 extern void lttng_counter_client_percpu_64_modular_init(void);
371 extern void lttng_counter_client_percpu_64_modular_exit(void);
373 static char *get_map_shm(struct sock_info
*sock_info
);
375 ssize_t
lttng_ust_read(int fd
, void *buf
, size_t len
)
378 size_t copied
= 0, to_copy
= len
;
381 ret
= read(fd
, buf
+ copied
, to_copy
);
386 } while ((ret
> 0 && to_copy
> 0)
387 || (ret
< 0 && errno
== EINTR
));
394 * Returns the HOME directory path. Caller MUST NOT free(3) the returned
398 const char *get_lttng_home_dir(void)
402 val
= (const char *) lttng_ust_getenv("LTTNG_HOME");
406 return (const char *) lttng_ust_getenv("HOME");
410 * Force a read (imply TLS fixup for dlopen) of TLS variables.
413 void lttng_fixup_nest_count_tls(void)
415 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count
)));
419 void lttng_fixup_ust_mutex_nest_tls(void)
421 asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest
)));
425 * Fixup lttng-ust urcu TLS.
428 void lttng_fixup_lttng_ust_urcu_tls(void)
430 (void) lttng_ust_urcu_read_ongoing();
433 void lttng_ust_fixup_tls(void)
435 lttng_fixup_lttng_ust_urcu_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_time_ns_tls();
447 lttng_fixup_uts_ns_tls();
450 int lttng_get_notify_socket(void *owner
)
452 struct sock_info
*info
= owner
;
454 return info
->notify_socket
;
459 char* lttng_ust_sockinfo_get_procname(void *owner
)
461 struct sock_info
*info
= owner
;
463 return info
->procname
;
467 void print_cmd(int cmd
, int handle
)
469 const char *cmd_name
= "Unknown";
471 if (cmd
>= 0 && cmd
< LTTNG_ARRAY_SIZE(cmd_name_mapping
)
472 && cmd_name_mapping
[cmd
]) {
473 cmd_name
= cmd_name_mapping
[cmd
];
475 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
477 lttng_ust_obj_get_name(handle
), handle
);
481 int setup_global_apps(void)
484 assert(!global_apps
.wait_shm_mmap
);
486 global_apps
.wait_shm_mmap
= get_map_shm(&global_apps
);
487 if (!global_apps
.wait_shm_mmap
) {
488 WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing.");
489 global_apps
.allowed
= 0;
494 global_apps
.allowed
= 1;
495 lttng_pthread_getname_np(global_apps
.procname
, LTTNG_UST_ABI_PROCNAME_LEN
);
500 int setup_local_apps(void)
503 const char *home_dir
;
506 assert(!local_apps
.wait_shm_mmap
);
510 * Disallow per-user tracing for setuid binaries.
512 if (uid
!= geteuid()) {
513 assert(local_apps
.allowed
== 0);
517 home_dir
= get_lttng_home_dir();
519 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
520 assert(local_apps
.allowed
== 0);
524 local_apps
.allowed
= 1;
525 snprintf(local_apps
.sock_path
, PATH_MAX
, "%s/%s/%s",
527 LTTNG_DEFAULT_HOME_RUNDIR
,
528 LTTNG_UST_SOCK_FILENAME
);
529 snprintf(local_apps
.wait_shm_path
, PATH_MAX
, "/%s-%u",
530 LTTNG_UST_WAIT_FILENAME
,
533 local_apps
.wait_shm_mmap
= get_map_shm(&local_apps
);
534 if (!local_apps
.wait_shm_mmap
) {
535 WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing.");
536 local_apps
.allowed
= 0;
541 lttng_pthread_getname_np(local_apps
.procname
, LTTNG_UST_ABI_PROCNAME_LEN
);
547 * Get socket timeout, in ms.
548 * -1: wait forever. 0: don't wait. >0: timeout, in ms.
551 long get_timeout(void)
553 long constructor_delay_ms
= LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS
;
555 if (!got_timeout_env
) {
556 str_timeout
= lttng_ust_getenv("LTTNG_UST_REGISTER_TIMEOUT");
560 constructor_delay_ms
= strtol(str_timeout
, NULL
, 10);
561 /* All negative values are considered as "-1". */
562 if (constructor_delay_ms
< -1)
563 constructor_delay_ms
= -1;
564 return constructor_delay_ms
;
567 /* Timeout for notify socket send and recv. */
569 long get_notify_sock_timeout(void)
571 return get_timeout();
574 /* Timeout for connecting to cmd and notify sockets. */
576 long get_connect_sock_timeout(void)
578 return get_timeout();
582 * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
585 int get_constructor_timeout(struct timespec
*constructor_timeout
)
587 long constructor_delay_ms
;
590 constructor_delay_ms
= get_timeout();
592 switch (constructor_delay_ms
) {
593 case -1:/* fall-through */
595 return constructor_delay_ms
;
601 * If we are unable to find the current time, don't wait.
603 ret
= clock_gettime(CLOCK_REALTIME
, constructor_timeout
);
608 constructor_timeout
->tv_sec
+= constructor_delay_ms
/ 1000UL;
609 constructor_timeout
->tv_nsec
+=
610 (constructor_delay_ms
% 1000UL) * 1000000UL;
611 if (constructor_timeout
->tv_nsec
>= 1000000000UL) {
612 constructor_timeout
->tv_sec
++;
613 constructor_timeout
->tv_nsec
-= 1000000000UL;
615 /* Timeout wait (constructor_delay_ms). */
620 void get_allow_blocking(void)
622 const char *str_allow_blocking
=
623 lttng_ust_getenv("LTTNG_UST_ALLOW_BLOCKING");
625 if (str_allow_blocking
) {
626 DBG("%s environment variable is set",
627 "LTTNG_UST_ALLOW_BLOCKING");
628 lttng_ust_ringbuffer_set_allow_blocking();
633 int register_to_sessiond(int socket
, enum ustctl_socket_type type
)
635 return ustcomm_send_reg_msg(socket
,
638 lttng_alignof(uint8_t) * CHAR_BIT
,
639 lttng_alignof(uint16_t) * CHAR_BIT
,
640 lttng_alignof(uint32_t) * CHAR_BIT
,
641 lttng_alignof(uint64_t) * CHAR_BIT
,
642 lttng_alignof(unsigned long) * CHAR_BIT
);
646 int send_reply(int sock
, struct ustcomm_ust_reply
*lur
)
650 len
= ustcomm_send_unix_sock(sock
, lur
, sizeof(*lur
));
653 DBG("message successfully sent");
656 if (len
== -ECONNRESET
) {
657 DBG("remote end closed connection");
662 DBG("incorrect message size: %zd", len
);
668 void decrement_sem_count(unsigned int count
)
672 assert(uatomic_read(&sem_count
) >= count
);
674 if (uatomic_read(&sem_count
) <= 0) {
678 ret
= uatomic_add_return(&sem_count
, -count
);
680 ret
= sem_post(&constructor_wait
);
686 int handle_register_done(struct sock_info
*sock_info
)
688 if (sock_info
->registration_done
)
690 sock_info
->registration_done
= 1;
692 decrement_sem_count(1);
693 if (!sock_info
->statedump_pending
) {
694 sock_info
->initial_statedump_done
= 1;
695 decrement_sem_count(1);
702 int handle_register_failed(struct sock_info
*sock_info
)
704 if (sock_info
->registration_done
)
706 sock_info
->registration_done
= 1;
707 sock_info
->initial_statedump_done
= 1;
709 decrement_sem_count(2);
715 * Only execute pending statedump after the constructor semaphore has
716 * been posted by the current listener thread. This means statedump will
717 * only be performed after the "registration done" command is received
718 * from this thread's session daemon.
720 * This ensures we don't run into deadlock issues with the dynamic
721 * loader mutex, which is held while the constructor is called and
722 * waiting on the constructor semaphore. All operations requiring this
723 * dynamic loader lock need to be postponed using this mechanism.
725 * In a scenario with two session daemons connected to the application,
726 * it is possible that the first listener thread which receives the
727 * registration done command issues its statedump while the dynamic
728 * loader lock is still held by the application constructor waiting on
729 * the semaphore. It will however be allowed to proceed when the
730 * second session daemon sends the registration done command to the
731 * second listener thread. This situation therefore does not produce
735 void handle_pending_statedump(struct sock_info
*sock_info
)
737 if (sock_info
->registration_done
&& sock_info
->statedump_pending
) {
738 sock_info
->statedump_pending
= 0;
739 pthread_mutex_lock(&ust_fork_mutex
);
740 lttng_handle_pending_statedump(sock_info
);
741 pthread_mutex_unlock(&ust_fork_mutex
);
743 if (!sock_info
->initial_statedump_done
) {
744 sock_info
->initial_statedump_done
= 1;
745 decrement_sem_count(1);
751 const char *bytecode_type_str(uint32_t cmd
)
754 case LTTNG_UST_ABI_CAPTURE
:
756 case LTTNG_UST_ABI_FILTER
:
764 int handle_bytecode_recv(struct sock_info
*sock_info
,
765 int sock
, struct ustcomm_ust_msg
*lum
)
767 struct lttng_ust_bytecode_node
*bytecode
= NULL
;
768 enum lttng_ust_bytecode_node_type type
;
769 const struct lttng_ust_abi_objd_ops
*ops
;
770 uint32_t data_size
, data_size_max
, reloc_offset
;
776 case LTTNG_UST_ABI_FILTER
:
777 type
= LTTNG_UST_BYTECODE_NODE_TYPE_FILTER
;
778 data_size
= lum
->u
.filter
.data_size
;
779 data_size_max
= LTTNG_UST_ABI_FILTER_BYTECODE_MAX_LEN
;
780 reloc_offset
= lum
->u
.filter
.reloc_offset
;
781 seqnum
= lum
->u
.filter
.seqnum
;
783 case LTTNG_UST_ABI_CAPTURE
:
784 type
= LTTNG_UST_BYTECODE_NODE_TYPE_CAPTURE
;
785 data_size
= lum
->u
.capture
.data_size
;
786 data_size_max
= LTTNG_UST_ABI_CAPTURE_BYTECODE_MAX_LEN
;
787 reloc_offset
= lum
->u
.capture
.reloc_offset
;
788 seqnum
= lum
->u
.capture
.seqnum
;
794 if (data_size
> data_size_max
) {
795 ERR("Bytecode %s data size is too large: %u bytes",
796 bytecode_type_str(lum
->cmd
), data_size
);
801 if (reloc_offset
> data_size
) {
802 ERR("Bytecode %s reloc offset %u is not within data",
803 bytecode_type_str(lum
->cmd
), reloc_offset
);
808 /* Allocate the structure AND the `data[]` field. */
809 bytecode
= zmalloc(sizeof(*bytecode
) + data_size
);
815 bytecode
->bc
.len
= data_size
;
816 bytecode
->bc
.reloc_offset
= reloc_offset
;
817 bytecode
->bc
.seqnum
= seqnum
;
818 bytecode
->type
= type
;
820 len
= ustcomm_recv_unix_sock(sock
, bytecode
->bc
.data
, bytecode
->bc
.len
);
822 case 0: /* orderly shutdown */
826 if (len
== bytecode
->bc
.len
) {
827 DBG("Bytecode %s data received",
828 bytecode_type_str(lum
->cmd
));
830 } else if (len
< 0) {
831 DBG("Receive failed from lttng-sessiond with errno %d",
833 if (len
== -ECONNRESET
) {
834 ERR("%s remote end closed connection",
842 DBG("Incorrect %s bytecode data message size: %zd",
843 bytecode_type_str(lum
->cmd
), len
);
849 ops
= lttng_ust_abi_objd_ops(lum
->handle
);
856 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
857 (unsigned long) &bytecode
,
868 int handle_message(struct sock_info
*sock_info
,
869 int sock
, struct ustcomm_ust_msg
*lum
)
872 const struct lttng_ust_abi_objd_ops
*ops
;
873 struct ustcomm_ust_reply lur
;
874 union lttng_ust_abi_args args
;
875 char ctxstr
[LTTNG_UST_ABI_SYM_NAME_LEN
]; /* App context string. */
878 memset(&lur
, 0, sizeof(lur
));
881 ret
= -LTTNG_UST_ERR_EXITING
;
885 ops
= lttng_ust_abi_objd_ops(lum
->handle
);
892 case LTTNG_UST_ABI_REGISTER_DONE
:
893 if (lum
->handle
== LTTNG_UST_ABI_ROOT_HANDLE
)
894 ret
= handle_register_done(sock_info
);
898 case LTTNG_UST_ABI_RELEASE
:
899 if (lum
->handle
== LTTNG_UST_ABI_ROOT_HANDLE
)
902 ret
= lttng_ust_abi_objd_unref(lum
->handle
, 1);
904 case LTTNG_UST_ABI_CAPTURE
:
905 case LTTNG_UST_ABI_FILTER
:
906 ret
= handle_bytecode_recv(sock_info
, sock
, lum
);
910 case LTTNG_UST_ABI_EXCLUSION
:
912 /* Receive exclusion names */
913 struct lttng_ust_excluder_node
*node
;
916 count
= lum
->u
.exclusion
.count
;
918 /* There are no names to read */
922 node
= zmalloc(sizeof(*node
) +
923 count
* LTTNG_UST_ABI_SYM_NAME_LEN
);
928 node
->excluder
.count
= count
;
929 len
= ustcomm_recv_unix_sock(sock
, node
->excluder
.names
,
930 count
* LTTNG_UST_ABI_SYM_NAME_LEN
);
932 case 0: /* orderly shutdown */
937 if (len
== count
* LTTNG_UST_ABI_SYM_NAME_LEN
) {
938 DBG("Exclusion data received");
940 } else if (len
< 0) {
941 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
942 if (len
== -ECONNRESET
) {
943 ERR("%s remote end closed connection", sock_info
->name
);
952 DBG("Incorrect exclusion data message size: %zd", len
);
959 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
960 (unsigned long) &node
,
967 case LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE
:
969 int event_notifier_notif_fd
, close_ret
;
971 len
= ustcomm_recv_event_notifier_notif_fd_from_sessiond(sock
,
972 &event_notifier_notif_fd
);
974 case 0: /* orderly shutdown */
981 DBG("Receive failed from lttng-sessiond with errno %d",
983 if (len
== -ECONNRESET
) {
984 ERR("%s remote end closed connection",
992 DBG("Incorrect event notifier fd message size: %zd",
998 args
.event_notifier_handle
.event_notifier_notif_fd
=
999 event_notifier_notif_fd
;
1001 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1002 (unsigned long) &lum
->u
,
1006 if (args
.event_notifier_handle
.event_notifier_notif_fd
>= 0) {
1007 lttng_ust_lock_fd_tracker();
1008 close_ret
= close(args
.event_notifier_handle
.event_notifier_notif_fd
);
1009 lttng_ust_unlock_fd_tracker();
1015 case LTTNG_UST_ABI_CHANNEL
:
1020 len
= ustcomm_recv_channel_from_sessiond(sock
,
1021 &chan_data
, lum
->u
.channel
.len
,
1024 case 0: /* orderly shutdown */
1028 if (len
== lum
->u
.channel
.len
) {
1029 DBG("channel data received");
1031 } else if (len
< 0) {
1032 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1033 if (len
== -ECONNRESET
) {
1034 ERR("%s remote end closed connection", sock_info
->name
);
1041 DBG("incorrect channel data message size: %zd", len
);
1046 args
.channel
.chan_data
= chan_data
;
1047 args
.channel
.wakeup_fd
= wakeup_fd
;
1049 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1050 (unsigned long) &lum
->u
,
1054 if (args
.channel
.wakeup_fd
>= 0) {
1057 lttng_ust_lock_fd_tracker();
1058 close_ret
= close(args
.channel
.wakeup_fd
);
1059 lttng_ust_unlock_fd_tracker();
1060 args
.channel
.wakeup_fd
= -1;
1064 free(args
.channel
.chan_data
);
1067 case LTTNG_UST_ABI_STREAM
:
1071 /* Receive shm_fd, wakeup_fd */
1072 ret
= ustcomm_recv_stream_from_sessiond(sock
,
1074 &args
.stream
.shm_fd
,
1075 &args
.stream
.wakeup_fd
);
1081 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1082 (unsigned long) &lum
->u
,
1086 if (args
.stream
.shm_fd
>= 0) {
1087 lttng_ust_lock_fd_tracker();
1088 close_ret
= close(args
.stream
.shm_fd
);
1089 lttng_ust_unlock_fd_tracker();
1090 args
.stream
.shm_fd
= -1;
1094 if (args
.stream
.wakeup_fd
>= 0) {
1095 lttng_ust_lock_fd_tracker();
1096 close_ret
= close(args
.stream
.wakeup_fd
);
1097 lttng_ust_unlock_fd_tracker();
1098 args
.stream
.wakeup_fd
= -1;
1104 case LTTNG_UST_ABI_CONTEXT
:
1105 switch (lum
->u
.context
.ctx
) {
1106 case LTTNG_UST_ABI_CONTEXT_APP_CONTEXT
:
1109 size_t ctxlen
, recvlen
;
1111 ctxlen
= strlen("$app.") + lum
->u
.context
.u
.app_ctx
.provider_name_len
- 1
1112 + strlen(":") + lum
->u
.context
.u
.app_ctx
.ctx_name_len
;
1113 if (ctxlen
>= LTTNG_UST_ABI_SYM_NAME_LEN
) {
1114 ERR("Application context string length size is too large: %zu bytes",
1119 strcpy(ctxstr
, "$app.");
1120 p
= &ctxstr
[strlen("$app.")];
1121 recvlen
= ctxlen
- strlen("$app.");
1122 len
= ustcomm_recv_unix_sock(sock
, p
, recvlen
);
1124 case 0: /* orderly shutdown */
1128 if (len
== recvlen
) {
1129 DBG("app context data received");
1131 } else if (len
< 0) {
1132 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1133 if (len
== -ECONNRESET
) {
1134 ERR("%s remote end closed connection", sock_info
->name
);
1141 DBG("incorrect app context data message size: %zd", len
);
1146 /* Put : between provider and ctxname. */
1147 p
[lum
->u
.context
.u
.app_ctx
.provider_name_len
- 1] = ':';
1148 args
.app_context
.ctxname
= ctxstr
;
1155 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1156 (unsigned long) &lum
->u
,
1162 case LTTNG_UST_ABI_COUNTER
:
1166 len
= ustcomm_recv_counter_from_sessiond(sock
,
1167 &counter_data
, lum
->u
.counter
.len
);
1169 case 0: /* orderly shutdown */
1173 if (len
== lum
->u
.counter
.len
) {
1174 DBG("counter data received");
1176 } else if (len
< 0) {
1177 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1178 if (len
== -ECONNRESET
) {
1179 ERR("%s remote end closed connection", sock_info
->name
);
1186 DBG("incorrect counter data message size: %zd", len
);
1191 args
.counter
.counter_data
= counter_data
;
1193 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1194 (unsigned long) &lum
->u
,
1198 free(args
.counter
.counter_data
);
1201 case LTTNG_UST_ABI_COUNTER_GLOBAL
:
1203 /* Receive shm_fd */
1204 ret
= ustcomm_recv_counter_shm_from_sessiond(sock
,
1205 &args
.counter_shm
.shm_fd
);
1211 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1212 (unsigned long) &lum
->u
,
1216 if (args
.counter_shm
.shm_fd
>= 0) {
1219 lttng_ust_lock_fd_tracker();
1220 close_ret
= close(args
.counter_shm
.shm_fd
);
1221 lttng_ust_unlock_fd_tracker();
1222 args
.counter_shm
.shm_fd
= -1;
1228 case LTTNG_UST_ABI_COUNTER_CPU
:
1230 /* Receive shm_fd */
1231 ret
= ustcomm_recv_counter_shm_from_sessiond(sock
,
1232 &args
.counter_shm
.shm_fd
);
1238 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1239 (unsigned long) &lum
->u
,
1243 if (args
.counter_shm
.shm_fd
>= 0) {
1246 lttng_ust_lock_fd_tracker();
1247 close_ret
= close(args
.counter_shm
.shm_fd
);
1248 lttng_ust_unlock_fd_tracker();
1249 args
.counter_shm
.shm_fd
= -1;
1255 case LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE
:
1257 /* Receive struct lttng_ust_event_notifier */
1258 struct lttng_ust_abi_event_notifier event_notifier
;
1260 if (sizeof(event_notifier
) != lum
->u
.event_notifier
.len
) {
1261 DBG("incorrect event notifier data message size: %u", lum
->u
.event_notifier
.len
);
1265 len
= ustcomm_recv_unix_sock(sock
, &event_notifier
, sizeof(event_notifier
));
1267 case 0: /* orderly shutdown */
1271 if (len
== sizeof(event_notifier
)) {
1272 DBG("event notifier data received");
1274 } else if (len
< 0) {
1275 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1276 if (len
== -ECONNRESET
) {
1277 ERR("%s remote end closed connection", sock_info
->name
);
1284 DBG("incorrect event notifier data message size: %zd", len
);
1290 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1291 (unsigned long) &event_notifier
,
1300 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1301 (unsigned long) &lum
->u
,
1308 lur
.handle
= lum
->handle
;
1312 lur
.ret_code
= LTTNG_UST_OK
;
1315 * Use -LTTNG_UST_ERR as wildcard for UST internal
1316 * error that are not caused by the transport, except if
1317 * we already have a more precise error message to
1320 if (ret
> -LTTNG_UST_ERR
) {
1321 /* Translate code to UST error. */
1324 lur
.ret_code
= -LTTNG_UST_ERR_EXIST
;
1327 lur
.ret_code
= -LTTNG_UST_ERR_INVAL
;
1330 lur
.ret_code
= -LTTNG_UST_ERR_NOENT
;
1333 lur
.ret_code
= -LTTNG_UST_ERR_PERM
;
1336 lur
.ret_code
= -LTTNG_UST_ERR_NOSYS
;
1339 lur
.ret_code
= -LTTNG_UST_ERR
;
1348 case LTTNG_UST_ABI_TRACER_VERSION
:
1349 lur
.u
.version
= lum
->u
.version
;
1351 case LTTNG_UST_ABI_TRACEPOINT_LIST_GET
:
1352 memcpy(&lur
.u
.tracepoint
, &lum
->u
.tracepoint
, sizeof(lur
.u
.tracepoint
));
1356 DBG("Return value: %d", lur
.ret_val
);
1361 * Performed delayed statedump operations outside of the UST
1362 * lock. We need to take the dynamic loader lock before we take
1363 * the UST lock internally within handle_pending_statedump().
1365 handle_pending_statedump(sock_info
);
1368 ret
= -LTTNG_UST_ERR_EXITING
;
1372 ret
= send_reply(sock
, &lur
);
1374 DBG("error sending reply");
1379 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1382 if (lur
.ret_code
== LTTNG_UST_OK
) {
1384 case LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST_GET
:
1385 len
= ustcomm_send_unix_sock(sock
,
1386 &args
.field_list
.entry
,
1387 sizeof(args
.field_list
.entry
));
1392 if (len
!= sizeof(args
.field_list
.entry
)) {
1406 void cleanup_sock_info(struct sock_info
*sock_info
, int exiting
)
1410 if (sock_info
->root_handle
!= -1) {
1411 ret
= lttng_ust_abi_objd_unref(sock_info
->root_handle
, 1);
1413 ERR("Error unref root handle");
1415 sock_info
->root_handle
= -1;
1417 sock_info
->registration_done
= 0;
1418 sock_info
->initial_statedump_done
= 0;
1421 * wait_shm_mmap, socket and notify socket are used by listener
1422 * threads outside of the ust lock, so we cannot tear them down
1423 * ourselves, because we cannot join on these threads. Leave
1424 * responsibility of cleaning up these resources to the OS
1430 if (sock_info
->socket
!= -1) {
1431 ret
= ustcomm_close_unix_sock(sock_info
->socket
);
1433 ERR("Error closing ust cmd socket");
1435 sock_info
->socket
= -1;
1437 if (sock_info
->notify_socket
!= -1) {
1438 ret
= ustcomm_close_unix_sock(sock_info
->notify_socket
);
1440 ERR("Error closing ust notify socket");
1442 sock_info
->notify_socket
= -1;
1444 if (sock_info
->wait_shm_mmap
) {
1447 page_size
= LTTNG_UST_PAGE_SIZE
;
1448 if (page_size
<= 0) {
1452 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1454 ret
= munmap(sock_info
->wait_shm_mmap
, page_size
);
1456 ERR("Error unmapping wait shm");
1459 sock_info
->wait_shm_mmap
= NULL
;
1464 * Using fork to set umask in the child process (not multi-thread safe).
1465 * We deal with the shm_open vs ftruncate race (happening when the
1466 * sessiond owns the shm and does not let everybody modify it, to ensure
1467 * safety against shm_unlink) by simply letting the mmap fail and
1468 * retrying after a few seconds.
1469 * For global shm, everybody has rw access to it until the sessiond
1473 int get_wait_shm(struct sock_info
*sock_info
, size_t mmap_size
)
1475 int wait_shm_fd
, ret
;
1479 * Try to open read-only.
1481 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
, O_RDONLY
, 0);
1482 if (wait_shm_fd
>= 0) {
1485 size_t bytes_read
= 0;
1488 * Try to read the fd. If unable to do so, try opening
1492 len
= read(wait_shm_fd
,
1493 &((char *) &tmp_read
)[bytes_read
],
1494 sizeof(tmp_read
) - bytes_read
);
1498 } while ((len
< 0 && errno
== EINTR
)
1499 || (len
> 0 && bytes_read
< sizeof(tmp_read
)));
1500 if (bytes_read
!= sizeof(tmp_read
)) {
1501 ret
= close(wait_shm_fd
);
1503 ERR("close wait_shm_fd");
1508 } else if (wait_shm_fd
< 0 && errno
!= ENOENT
) {
1510 * Real-only open did not work, and it's not because the
1511 * entry was not present. It's a failure that prohibits
1514 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
1520 * If the open failed because the file did not exist, or because
1521 * the file was not truncated yet, try creating it ourself.
1523 URCU_TLS(lttng_ust_nest_count
)++;
1525 URCU_TLS(lttng_ust_nest_count
)--;
1530 * Parent: wait for child to return, in which case the
1531 * shared memory map will have been created.
1533 pid
= wait(&status
);
1534 if (pid
< 0 || !WIFEXITED(status
) || WEXITSTATUS(status
) != 0) {
1539 * Try to open read-only again after creation.
1541 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
, O_RDONLY
, 0);
1542 if (wait_shm_fd
< 0) {
1544 * Real-only open did not work. It's a failure
1545 * that prohibits using shm.
1547 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
1551 } else if (pid
== 0) {
1555 create_mode
= S_IRUSR
| S_IWUSR
| S_IRGRP
;
1556 if (sock_info
->global
)
1557 create_mode
|= S_IROTH
| S_IWGRP
| S_IWOTH
;
1559 * We're alone in a child process, so we can modify the
1560 * process-wide umask.
1562 umask(~create_mode
);
1564 * Try creating shm (or get rw access).
1565 * We don't do an exclusive open, because we allow other
1566 * processes to create+ftruncate it concurrently.
1568 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
,
1569 O_RDWR
| O_CREAT
, create_mode
);
1570 if (wait_shm_fd
>= 0) {
1571 ret
= ftruncate(wait_shm_fd
, mmap_size
);
1573 PERROR("ftruncate");
1574 _exit(EXIT_FAILURE
);
1576 _exit(EXIT_SUCCESS
);
1579 * For local shm, we need to have rw access to accept
1580 * opening it: this means the local sessiond will be
1581 * able to wake us up. For global shm, we open it even
1582 * if rw access is not granted, because the root.root
1583 * sessiond will be able to override all rights and wake
1586 if (!sock_info
->global
&& errno
!= EACCES
) {
1587 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
1588 _exit(EXIT_FAILURE
);
1591 * The shm exists, but we cannot open it RW. Report
1594 _exit(EXIT_SUCCESS
);
1599 if (wait_shm_fd
>= 0 && !sock_info
->global
) {
1600 struct stat statbuf
;
1603 * Ensure that our user is the owner of the shm file for
1604 * local shm. If we do not own the file, it means our
1605 * sessiond will not have access to wake us up (there is
1606 * probably a rogue process trying to fake our
1607 * sessiond). Fallback to polling method in this case.
1609 ret
= fstat(wait_shm_fd
, &statbuf
);
1614 if (statbuf
.st_uid
!= getuid())
1620 ret
= close(wait_shm_fd
);
1622 PERROR("Error closing fd");
1628 char *get_map_shm(struct sock_info
*sock_info
)
1631 int wait_shm_fd
, ret
;
1632 char *wait_shm_mmap
;
1634 page_size
= sysconf(_SC_PAGE_SIZE
);
1635 if (page_size
<= 0) {
1639 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1643 lttng_ust_lock_fd_tracker();
1644 wait_shm_fd
= get_wait_shm(sock_info
, page_size
);
1645 if (wait_shm_fd
< 0) {
1646 lttng_ust_unlock_fd_tracker();
1650 ret
= lttng_ust_add_fd_to_tracker(wait_shm_fd
);
1652 ret
= close(wait_shm_fd
);
1654 PERROR("Error closing fd");
1656 lttng_ust_unlock_fd_tracker();
1661 lttng_ust_unlock_fd_tracker();
1663 wait_shm_mmap
= mmap(NULL
, page_size
, PROT_READ
,
1664 MAP_SHARED
, wait_shm_fd
, 0);
1666 /* close shm fd immediately after taking the mmap reference */
1667 lttng_ust_lock_fd_tracker();
1668 ret
= close(wait_shm_fd
);
1670 lttng_ust_delete_fd_from_tracker(wait_shm_fd
);
1672 PERROR("Error closing fd");
1674 lttng_ust_unlock_fd_tracker();
1676 if (wait_shm_mmap
== MAP_FAILED
) {
1677 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1680 return wait_shm_mmap
;
1687 void wait_for_sessiond(struct sock_info
*sock_info
)
1689 /* Use ust_lock to check if we should quit. */
1693 if (wait_poll_fallback
) {
1698 assert(sock_info
->wait_shm_mmap
);
1700 DBG("Waiting for %s apps sessiond", sock_info
->name
);
1701 /* Wait for futex wakeup */
1702 if (uatomic_read((int32_t *) sock_info
->wait_shm_mmap
))
1705 while (lttng_ust_futex_async((int32_t *) sock_info
->wait_shm_mmap
,
1706 FUTEX_WAIT
, 0, NULL
, NULL
, 0)) {
1709 /* Value already changed. */
1712 /* Retry if interrupted by signal. */
1713 break; /* Get out of switch. */
1715 wait_poll_fallback
= 1;
1717 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1718 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1719 "Please upgrade your kernel "
1720 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1721 "mainline). LTTng-UST will use polling mode fallback.");
1722 if (ust_err_debug_enabled())
1740 * This thread does not allocate any resource, except within
1741 * handle_message, within mutex protection. This mutex protects against
1743 * The other moment it allocates resources is at socket connection, which
1744 * is also protected by the mutex.
1747 void *ust_listener_thread(void *arg
)
1749 struct sock_info
*sock_info
= arg
;
1750 int sock
, ret
, prev_connect_failed
= 0, has_waited
= 0, fd
;
1753 lttng_ust_fixup_tls();
1755 * If available, add '-ust' to the end of this thread's
1758 ret
= lttng_ust_setustprocname();
1760 ERR("Unable to set UST process name");
1763 /* Restart trying to connect to the session daemon */
1765 if (prev_connect_failed
) {
1766 /* Wait for sessiond availability with pipe */
1767 wait_for_sessiond(sock_info
);
1771 * Sleep for 5 seconds before retrying after a
1772 * sequence of failure / wait / failure. This
1773 * deals with a killed or broken session daemon.
1779 prev_connect_failed
= 0;
1786 if (sock_info
->socket
!= -1) {
1787 /* FD tracker is updated by ustcomm_close_unix_sock() */
1788 ret
= ustcomm_close_unix_sock(sock_info
->socket
);
1790 ERR("Error closing %s ust cmd socket",
1793 sock_info
->socket
= -1;
1795 if (sock_info
->notify_socket
!= -1) {
1796 /* FD tracker is updated by ustcomm_close_unix_sock() */
1797 ret
= ustcomm_close_unix_sock(sock_info
->notify_socket
);
1799 ERR("Error closing %s ust notify socket",
1802 sock_info
->notify_socket
= -1;
1807 * Register. We need to perform both connect and sending
1808 * registration message before doing the next connect otherwise
1809 * we may reach unix socket connect queue max limits and block
1810 * on the 2nd connect while the session daemon is awaiting the
1811 * first connect registration message.
1813 /* Connect cmd socket */
1814 lttng_ust_lock_fd_tracker();
1815 ret
= ustcomm_connect_unix_sock(sock_info
->sock_path
,
1816 get_connect_sock_timeout());
1818 lttng_ust_unlock_fd_tracker();
1819 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info
->name
);
1820 prev_connect_failed
= 1;
1823 * If we cannot find the sessiond daemon, don't delay
1824 * constructor execution.
1826 ret
= handle_register_failed(sock_info
);
1832 ret
= lttng_ust_add_fd_to_tracker(fd
);
1836 PERROR("close on sock_info->socket");
1839 lttng_ust_unlock_fd_tracker();
1844 sock_info
->socket
= ret
;
1845 lttng_ust_unlock_fd_tracker();
1849 * Unlock/relock ust lock because connect is blocking (with
1850 * timeout). Don't delay constructors on the ust lock for too
1858 * Create only one root handle per listener thread for the whole
1859 * process lifetime, so we ensure we get ID which is statically
1860 * assigned to the root handle.
1862 if (sock_info
->root_handle
== -1) {
1863 ret
= lttng_abi_create_root_handle();
1865 ERR("Error creating root handle");
1868 sock_info
->root_handle
= ret
;
1871 ret
= register_to_sessiond(sock_info
->socket
, USTCTL_SOCKET_CMD
);
1873 ERR("Error registering to %s ust cmd socket",
1875 prev_connect_failed
= 1;
1877 * If we cannot register to the sessiond daemon, don't
1878 * delay constructor execution.
1880 ret
= handle_register_failed(sock_info
);
1888 * Unlock/relock ust lock because connect is blocking (with
1889 * timeout). Don't delay constructors on the ust lock for too
1896 /* Connect notify socket */
1897 lttng_ust_lock_fd_tracker();
1898 ret
= ustcomm_connect_unix_sock(sock_info
->sock_path
,
1899 get_connect_sock_timeout());
1901 lttng_ust_unlock_fd_tracker();
1902 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info
->name
);
1903 prev_connect_failed
= 1;
1906 * If we cannot find the sessiond daemon, don't delay
1907 * constructor execution.
1909 ret
= handle_register_failed(sock_info
);
1916 ret
= lttng_ust_add_fd_to_tracker(fd
);
1920 PERROR("close on sock_info->notify_socket");
1923 lttng_ust_unlock_fd_tracker();
1928 sock_info
->notify_socket
= ret
;
1929 lttng_ust_unlock_fd_tracker();
1933 * Unlock/relock ust lock because connect is blocking (with
1934 * timeout). Don't delay constructors on the ust lock for too
1941 timeout
= get_notify_sock_timeout();
1944 * Give at least 10ms to sessiond to reply to
1949 ret
= ustcomm_setsockopt_rcv_timeout(sock_info
->notify_socket
,
1952 WARN("Error setting socket receive timeout");
1954 ret
= ustcomm_setsockopt_snd_timeout(sock_info
->notify_socket
,
1957 WARN("Error setting socket send timeout");
1959 } else if (timeout
< -1) {
1960 WARN("Unsupported timeout value %ld", timeout
);
1963 ret
= register_to_sessiond(sock_info
->notify_socket
,
1964 USTCTL_SOCKET_NOTIFY
);
1966 ERR("Error registering to %s ust notify socket",
1968 prev_connect_failed
= 1;
1970 * If we cannot register to the sessiond daemon, don't
1971 * delay constructor execution.
1973 ret
= handle_register_failed(sock_info
);
1978 sock
= sock_info
->socket
;
1984 struct ustcomm_ust_msg lum
;
1986 len
= ustcomm_recv_unix_sock(sock
, &lum
, sizeof(lum
));
1988 case 0: /* orderly shutdown */
1989 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info
->name
);
1994 * Either sessiond has shutdown or refused us by closing the socket.
1995 * In either case, we don't want to delay construction execution,
1996 * and we need to wait before retry.
1998 prev_connect_failed
= 1;
2000 * If we cannot register to the sessiond daemon, don't
2001 * delay constructor execution.
2003 ret
= handle_register_failed(sock_info
);
2008 print_cmd(lum
.cmd
, lum
.handle
);
2009 ret
= handle_message(sock_info
, sock
, &lum
);
2011 ERR("Error handling message for %s socket",
2014 * Close socket if protocol error is
2022 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
2024 DBG("incorrect message size (%s socket): %zd", sock_info
->name
, len
);
2026 if (len
== -ECONNRESET
) {
2027 DBG("%s remote end closed connection", sock_info
->name
);
2038 /* Cleanup socket handles before trying to reconnect */
2039 lttng_ust_abi_objd_table_owner_cleanup(sock_info
);
2041 goto restart
; /* try to reconnect */
2046 pthread_mutex_lock(&ust_exit_mutex
);
2047 sock_info
->thread_active
= 0;
2048 pthread_mutex_unlock(&ust_exit_mutex
);
2053 * Weak symbol to call when the ust malloc wrapper is not loaded.
2055 __attribute__((weak
))
2056 void lttng_ust_libc_wrapper_malloc_init(void)
2061 * sessiond monitoring thread: monitor presence of global and per-user
2062 * sessiond by polling the application common named pipe.
2064 void __attribute__((constructor
)) lttng_ust_init(void)
2066 struct timespec constructor_timeout
;
2067 sigset_t sig_all_blocked
, orig_parent_mask
;
2068 pthread_attr_t thread_attr
;
2073 if (uatomic_xchg(&initialized
, 1) == 1)
2077 * Fixup interdependency between TLS fixup mutex (which happens
2078 * to be the dynamic linker mutex) and ust_lock, taken within
2081 lttng_ust_fixup_tls();
2083 lttng_ust_loaded
= 1;
2086 * We need to ensure that the liblttng-ust library is not unloaded to avoid
2087 * the unloading of code used by the ust_listener_threads as we can not
2088 * reliably know when they exited. To do that, manually load
2089 * liblttng-ust.so to increment the dynamic loader's internal refcount for
2090 * this library so it never becomes zero, thus never gets unloaded from the
2091 * address space of the process. Since we are already running in the
2092 * constructor of the LTTNG_UST_LIB_SO_NAME library, calling dlopen will
2093 * simply increment the refcount and no additionnal work is needed by the
2094 * dynamic loader as the shared library is already loaded in the address
2095 * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
2096 * unloading of the UST library if its refcount becomes zero (which should
2097 * never happen). Do the return value check but discard the handle at the
2098 * end of the function as it's not needed.
2100 handle
= dlopen(LTTNG_UST_LIB_SO_NAME
, RTLD_LAZY
| RTLD_NODELETE
);
2102 ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SO_NAME
);
2106 * We want precise control over the order in which we construct
2107 * our sub-libraries vs starting to receive commands from
2108 * sessiond (otherwise leading to errors when trying to create
2109 * sessiond before the init functions are completed).
2112 lttng_ust_getenv_init(); /* Needs ust_err_init() to be completed. */
2113 lttng_ust_tp_init();
2114 lttng_ust_init_fd_tracker();
2115 lttng_ust_clock_init();
2116 lttng_ust_getcpu_init();
2117 lttng_ust_statedump_init();
2118 lttng_ring_buffer_metadata_client_init();
2119 lttng_ring_buffer_client_overwrite_init();
2120 lttng_ring_buffer_client_overwrite_rt_init();
2121 lttng_ring_buffer_client_discard_init();
2122 lttng_ring_buffer_client_discard_rt_init();
2123 lttng_counter_client_percpu_32_modular_init();
2124 lttng_counter_client_percpu_64_modular_init();
2125 lttng_perf_counter_init();
2127 * Invoke ust malloc wrapper init before starting other threads.
2129 lttng_ust_libc_wrapper_malloc_init();
2131 timeout_mode
= get_constructor_timeout(&constructor_timeout
);
2133 get_allow_blocking();
2135 ret
= sem_init(&constructor_wait
, 0, 0);
2140 ret
= setup_global_apps();
2142 assert(global_apps
.allowed
== 0);
2143 DBG("global apps setup returned %d", ret
);
2146 ret
= setup_local_apps();
2148 assert(local_apps
.allowed
== 0);
2149 DBG("local apps setup returned %d", ret
);
2152 /* A new thread created by pthread_create inherits the signal mask
2153 * from the parent. To avoid any signal being received by the
2154 * listener thread, we block all signals temporarily in the parent,
2155 * while we create the listener thread.
2157 sigfillset(&sig_all_blocked
);
2158 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_parent_mask
);
2160 ERR("pthread_sigmask: %s", strerror(ret
));
2163 ret
= pthread_attr_init(&thread_attr
);
2165 ERR("pthread_attr_init: %s", strerror(ret
));
2167 ret
= pthread_attr_setdetachstate(&thread_attr
, PTHREAD_CREATE_DETACHED
);
2169 ERR("pthread_attr_setdetachstate: %s", strerror(ret
));
2172 if (global_apps
.allowed
) {
2173 pthread_mutex_lock(&ust_exit_mutex
);
2174 ret
= pthread_create(&global_apps
.ust_listener
, &thread_attr
,
2175 ust_listener_thread
, &global_apps
);
2177 ERR("pthread_create global: %s", strerror(ret
));
2179 global_apps
.thread_active
= 1;
2180 pthread_mutex_unlock(&ust_exit_mutex
);
2182 handle_register_done(&global_apps
);
2185 if (local_apps
.allowed
) {
2186 pthread_mutex_lock(&ust_exit_mutex
);
2187 ret
= pthread_create(&local_apps
.ust_listener
, &thread_attr
,
2188 ust_listener_thread
, &local_apps
);
2190 ERR("pthread_create local: %s", strerror(ret
));
2192 local_apps
.thread_active
= 1;
2193 pthread_mutex_unlock(&ust_exit_mutex
);
2195 handle_register_done(&local_apps
);
2197 ret
= pthread_attr_destroy(&thread_attr
);
2199 ERR("pthread_attr_destroy: %s", strerror(ret
));
2202 /* Restore original signal mask in parent */
2203 ret
= pthread_sigmask(SIG_SETMASK
, &orig_parent_mask
, NULL
);
2205 ERR("pthread_sigmask: %s", strerror(ret
));
2208 switch (timeout_mode
) {
2209 case 1: /* timeout wait */
2211 ret
= sem_timedwait(&constructor_wait
,
2212 &constructor_timeout
);
2213 } while (ret
< 0 && errno
== EINTR
);
2217 ERR("Timed out waiting for lttng-sessiond");
2220 PERROR("sem_timedwait");
2223 ERR("Unexpected error \"%s\" returned by sem_timedwait",
2228 case -1:/* wait forever */
2230 ret
= sem_wait(&constructor_wait
);
2231 } while (ret
< 0 && errno
== EINTR
);
2238 ERR("Unexpected error \"%s\" returned by sem_wait",
2243 case 0: /* no timeout */
2249 void lttng_ust_cleanup(int exiting
)
2251 cleanup_sock_info(&global_apps
, exiting
);
2252 cleanup_sock_info(&local_apps
, exiting
);
2253 local_apps
.allowed
= 0;
2254 global_apps
.allowed
= 0;
2256 * The teardown in this function all affect data structures
2257 * accessed under the UST lock by the listener thread. This
2258 * lock, along with the lttng_ust_comm_should_quit flag, ensure
2259 * that none of these threads are accessing this data at this
2262 lttng_ust_abi_exit();
2263 lttng_ust_abi_events_exit();
2264 lttng_perf_counter_exit();
2265 lttng_ring_buffer_client_discard_rt_exit();
2266 lttng_ring_buffer_client_discard_exit();
2267 lttng_ring_buffer_client_overwrite_rt_exit();
2268 lttng_ring_buffer_client_overwrite_exit();
2269 lttng_ring_buffer_metadata_client_exit();
2270 lttng_counter_client_percpu_32_modular_exit();
2271 lttng_counter_client_percpu_64_modular_exit();
2272 lttng_ust_statedump_destroy();
2273 lttng_ust_tp_exit();
2275 /* Reinitialize values for fork */
2276 sem_count
= sem_count_initial_value
;
2277 lttng_ust_comm_should_quit
= 0;
2282 void __attribute__((destructor
)) lttng_ust_exit(void)
2287 * Using pthread_cancel here because:
2288 * A) we don't want to hang application teardown.
2289 * B) the thread is not allocating any resource.
2293 * Require the communication thread to quit. Synchronize with
2294 * mutexes to ensure it is not in a mutex critical section when
2295 * pthread_cancel is later called.
2298 lttng_ust_comm_should_quit
= 1;
2301 pthread_mutex_lock(&ust_exit_mutex
);
2302 /* cancel threads */
2303 if (global_apps
.thread_active
) {
2304 ret
= pthread_cancel(global_apps
.ust_listener
);
2306 ERR("Error cancelling global ust listener thread: %s",
2309 global_apps
.thread_active
= 0;
2312 if (local_apps
.thread_active
) {
2313 ret
= pthread_cancel(local_apps
.ust_listener
);
2315 ERR("Error cancelling local ust listener thread: %s",
2318 local_apps
.thread_active
= 0;
2321 pthread_mutex_unlock(&ust_exit_mutex
);
2324 * Do NOT join threads: use of sys_futex makes it impossible to
2325 * join the threads without using async-cancel, but async-cancel
2326 * is delivered by a signal, which could hit the target thread
2327 * anywhere in its code path, including while the ust_lock() is
2328 * held, causing a deadlock for the other thread. Let the OS
2329 * cleanup the threads if there are stalled in a syscall.
2331 lttng_ust_cleanup(1);
2335 void ust_context_ns_reset(void)
2337 lttng_context_pid_ns_reset();
2338 lttng_context_cgroup_ns_reset();
2339 lttng_context_ipc_ns_reset();
2340 lttng_context_mnt_ns_reset();
2341 lttng_context_net_ns_reset();
2342 lttng_context_user_ns_reset();
2343 lttng_context_time_ns_reset();
2344 lttng_context_uts_ns_reset();
2348 void ust_context_vuids_reset(void)
2350 lttng_context_vuid_reset();
2351 lttng_context_veuid_reset();
2352 lttng_context_vsuid_reset();
2356 void ust_context_vgids_reset(void)
2358 lttng_context_vgid_reset();
2359 lttng_context_vegid_reset();
2360 lttng_context_vsgid_reset();
2364 * We exclude the worker threads across fork and clone (except
2365 * CLONE_VM), because these system calls only keep the forking thread
2366 * running in the child. Therefore, we don't want to call fork or clone
2367 * in the middle of an tracepoint or ust tracing state modification.
2368 * Holding this mutex protects these structures across fork and clone.
2370 void ust_before_fork(sigset_t
*save_sigset
)
2373 * Disable signals. This is to avoid that the child intervenes
2374 * before it is properly setup for tracing. It is safer to
2375 * disable all signals, because then we know we are not breaking
2376 * anything by restoring the original mask.
2381 /* Fixup lttng-ust TLS. */
2382 lttng_ust_fixup_tls();
2384 if (URCU_TLS(lttng_ust_nest_count
))
2386 /* Disable signals */
2387 sigfillset(&all_sigs
);
2388 ret
= sigprocmask(SIG_BLOCK
, &all_sigs
, save_sigset
);
2390 PERROR("sigprocmask");
2393 pthread_mutex_lock(&ust_fork_mutex
);
2396 lttng_ust_urcu_before_fork();
2397 lttng_ust_lock_fd_tracker();
2401 static void ust_after_fork_common(sigset_t
*restore_sigset
)
2405 DBG("process %d", getpid());
2406 lttng_perf_unlock();
2407 lttng_ust_unlock_fd_tracker();
2410 pthread_mutex_unlock(&ust_fork_mutex
);
2412 /* Restore signals */
2413 ret
= sigprocmask(SIG_SETMASK
, restore_sigset
, NULL
);
2415 PERROR("sigprocmask");
2419 void ust_after_fork_parent(sigset_t
*restore_sigset
)
2421 if (URCU_TLS(lttng_ust_nest_count
))
2423 DBG("process %d", getpid());
2424 lttng_ust_urcu_after_fork_parent();
2425 /* Release mutexes and reenable signals */
2426 ust_after_fork_common(restore_sigset
);
2430 * After fork, in the child, we need to cleanup all the leftover state,
2431 * except the worker thread which already magically disappeared thanks
2432 * to the weird Linux fork semantics. After tyding up, we call
2433 * lttng_ust_init() again to start over as a new PID.
2435 * This is meant for forks() that have tracing in the child between the
2436 * fork and following exec call (if there is any).
2438 void ust_after_fork_child(sigset_t
*restore_sigset
)
2440 if (URCU_TLS(lttng_ust_nest_count
))
2442 lttng_context_vpid_reset();
2443 lttng_context_vtid_reset();
2444 lttng_context_procname_reset();
2445 ust_context_ns_reset();
2446 ust_context_vuids_reset();
2447 ust_context_vgids_reset();
2448 DBG("process %d", getpid());
2449 /* Release urcu mutexes */
2450 lttng_ust_urcu_after_fork_child();
2451 lttng_ust_cleanup(0);
2452 /* Release mutexes and reenable signals */
2453 ust_after_fork_common(restore_sigset
);
2457 void ust_after_setns(void)
2459 ust_context_ns_reset();
2460 ust_context_vuids_reset();
2461 ust_context_vgids_reset();
2464 void ust_after_unshare(void)
2466 ust_context_ns_reset();
2467 ust_context_vuids_reset();
2468 ust_context_vgids_reset();
2471 void ust_after_setuid(void)
2473 ust_context_vuids_reset();
2476 void ust_after_seteuid(void)
2478 ust_context_vuids_reset();
2481 void ust_after_setreuid(void)
2483 ust_context_vuids_reset();
2486 void ust_after_setresuid(void)
2488 ust_context_vuids_reset();
2491 void ust_after_setgid(void)
2493 ust_context_vgids_reset();
2496 void ust_after_setegid(void)
2498 ust_context_vgids_reset();
2501 void ust_after_setregid(void)
2503 ust_context_vgids_reset();
2506 void ust_after_setresgid(void)
2508 ust_context_vgids_reset();
2511 void lttng_ust_sockinfo_session_enabled(void *owner
)
2513 struct sock_info
*sock_info
= owner
;
2514 sock_info
->statedump_pending
= 1;