2 * SPDX-License-Identifier: LGPL-2.1-only
4 * Copyright (C) 2011 EfficiOS Inc.
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>
28 #include <urcu/compiler.h>
29 #include <lttng/urcu/urcu-ust.h>
31 #include <lttng/ust-utils.h>
32 #include <lttng/ust-events.h>
33 #include <lttng/ust-abi.h>
34 #include <lttng/ust-fork.h>
35 #include <lttng/ust-error.h>
36 #include <lttng/ust-ctl.h>
37 #include <lttng/ust-libc-wrapper.h>
38 #include <lttng/ust-thread.h>
39 #include <lttng/ust-tracer.h>
40 #include <lttng/ust-common.h>
41 #include <lttng/ust-cancelstate.h>
42 #include <urcu/tls-compat.h>
43 #include "lib/lttng-ust/futex.h"
44 #include "common/ustcomm.h"
45 #include "common/ust-fd.h"
46 #include "common/logging.h"
47 #include "common/macros.h"
48 #include "common/tracepoint.h"
49 #include "lttng-tracer-core.h"
50 #include "common/compat/pthread.h"
51 #include "common/procname.h"
52 #include "common/ringbuffer/rb-init.h"
53 #include "lttng-ust-statedump.h"
54 #include "common/clock.h"
55 #include "common/getenv.h"
56 #include "lib/lttng-ust/events.h"
57 #include "context-internal.h"
58 #include "common/align.h"
59 #include "common/counter-clients/clients.h"
60 #include "common/ringbuffer-clients/clients.h"
63 * Has lttng ust comm constructor been called ?
65 static int initialized
;
68 * The ust_lock/ust_unlock lock is used as a communication thread mutex.
69 * Held when handling a command, also held by fork() to deal with
70 * removal of threads, and by exit path.
72 * The UST lock is the centralized mutex across UST tracing control and
75 * ust_exit_mutex must never nest in ust_mutex.
77 * ust_fork_mutex must never nest in ust_mutex.
79 * ust_mutex_nest is a per-thread nesting counter, allowing the perf
80 * counter lazy initialization called by events within the statedump,
81 * which traces while the ust_mutex is held.
83 * ust_lock nests within the dynamic loader lock (within glibc) because
84 * it is taken within the library constructor.
86 * The ust fd tracker lock nests within the ust_mutex.
88 static pthread_mutex_t ust_mutex
= PTHREAD_MUTEX_INITIALIZER
;
90 /* Allow nesting the ust_mutex within the same thread. */
91 static DEFINE_URCU_TLS(int, ust_mutex_nest
);
94 * ust_exit_mutex protects thread_active variable wrt thread exit. It
95 * cannot be done by ust_mutex because pthread_cancel(), which takes an
96 * internal libc lock, cannot nest within ust_mutex.
98 * It never nests within a ust_mutex.
100 static pthread_mutex_t ust_exit_mutex
= PTHREAD_MUTEX_INITIALIZER
;
103 * ust_fork_mutex protects base address statedump tracing against forks. It
104 * prevents the dynamic loader lock to be taken (by base address statedump
105 * tracing) while a fork is happening, thus preventing deadlock issues with
106 * the dynamic loader lock.
108 static pthread_mutex_t ust_fork_mutex
= PTHREAD_MUTEX_INITIALIZER
;
110 /* Should the ust comm thread quit ? */
111 static int lttng_ust_comm_should_quit
;
114 * This variable can be tested by applications to check whether
115 * lttng-ust is loaded. They simply have to define their own
116 * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the
117 * library constructor.
119 int lttng_ust_loaded
__attribute__((weak
));
122 * Notes on async-signal-safety of ust lock: a few libc functions are used
123 * which are not strictly async-signal-safe:
125 * - pthread_setcancelstate
126 * - pthread_mutex_lock
127 * - pthread_mutex_unlock
129 * As of glibc 2.35, the implementation of pthread_setcancelstate only
130 * touches TLS data, and it appears to be safe to use from signal
131 * handlers. If the libc implementation changes, this will need to be
132 * revisited, and we may ask glibc to provide an async-signal-safe
133 * pthread_setcancelstate.
135 * As of glibc 2.35, the implementation of pthread_mutex_lock/unlock
136 * for fast mutexes only relies on the pthread_mutex_t structure.
137 * Disabling signals around all uses of this mutex ensures
138 * signal-safety. If the libc implementation changes and eventually uses
139 * other global resources, this will need to be revisited and we may
140 * need to implement our own mutex.
144 * Return 0 on success, -1 if should quit.
145 * The lock is taken in both cases.
150 sigset_t sig_all_blocked
, orig_mask
;
153 if (lttng_ust_cancelstate_disable_push()) {
154 ERR("lttng_ust_cancelstate_disable_push");
156 sigfillset(&sig_all_blocked
);
157 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_mask
);
159 ERR("pthread_sigmask: ret=%d", ret
);
161 if (!URCU_TLS(ust_mutex_nest
)++)
162 pthread_mutex_lock(&ust_mutex
);
163 ret
= pthread_sigmask(SIG_SETMASK
, &orig_mask
, NULL
);
165 ERR("pthread_sigmask: ret=%d", ret
);
167 if (lttng_ust_comm_should_quit
) {
175 * ust_lock_nocheck() can be used in constructors/destructors, because
176 * they are already nested within the dynamic loader lock, and therefore
177 * have exclusive access against execution of liblttng-ust destructor.
180 void ust_lock_nocheck(void)
182 sigset_t sig_all_blocked
, orig_mask
;
185 if (lttng_ust_cancelstate_disable_push()) {
186 ERR("lttng_ust_cancelstate_disable_push");
188 sigfillset(&sig_all_blocked
);
189 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_mask
);
191 ERR("pthread_sigmask: ret=%d", ret
);
193 if (!URCU_TLS(ust_mutex_nest
)++)
194 pthread_mutex_lock(&ust_mutex
);
195 ret
= pthread_sigmask(SIG_SETMASK
, &orig_mask
, NULL
);
197 ERR("pthread_sigmask: ret=%d", ret
);
204 void ust_unlock(void)
206 sigset_t sig_all_blocked
, orig_mask
;
209 sigfillset(&sig_all_blocked
);
210 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_mask
);
212 ERR("pthread_sigmask: ret=%d", ret
);
214 if (!--URCU_TLS(ust_mutex_nest
))
215 pthread_mutex_unlock(&ust_mutex
);
216 ret
= pthread_sigmask(SIG_SETMASK
, &orig_mask
, NULL
);
218 ERR("pthread_sigmask: ret=%d", ret
);
220 if (lttng_ust_cancelstate_disable_pop()) {
221 ERR("lttng_ust_cancelstate_disable_pop");
226 * Wait for either of these before continuing to the main
228 * - the register_done message from sessiond daemon
229 * (will let the sessiond daemon enable sessions before main
231 * - sessiond daemon is not reachable.
232 * - timeout (ensuring applications are resilient to session
235 static sem_t constructor_wait
;
237 * Doing this for both the global and local sessiond.
240 sem_count_initial_value
= 4,
243 static int sem_count
= sem_count_initial_value
;
246 * Counting nesting within lttng-ust. Used to ensure that calling fork()
247 * from liblttng-ust does not execute the pre/post fork handlers.
249 static DEFINE_URCU_TLS(int, lttng_ust_nest_count
);
252 * Info about socket and associated listener thread.
256 pthread_t ust_listener
; /* listener thread */
258 int registration_done
;
263 char sock_path
[PATH_MAX
];
267 char wait_shm_path
[PATH_MAX
];
269 /* Keep track of lazy state dump not performed yet. */
270 int statedump_pending
;
271 int initial_statedump_done
;
272 /* Keep procname for statedump */
273 char procname
[LTTNG_UST_CONTEXT_PROCNAME_LEN
];
276 /* Socket from app (connect) to session daemon (listen) for communication */
277 static struct sock_info global_apps
= {
282 .registration_done
= 0,
286 .sock_path
= LTTNG_DEFAULT_RUNDIR
"/" LTTNG_UST_SOCK_FILENAME
,
290 .wait_shm_path
= "/" LTTNG_UST_WAIT_FILENAME
,
292 .statedump_pending
= 0,
293 .initial_statedump_done
= 0,
297 /* TODO: allow global_apps_sock_path override */
299 static struct sock_info local_apps
= {
303 .registration_done
= 0,
304 .allowed
= 0, /* Check setuid bit first */
310 .statedump_pending
= 0,
311 .initial_statedump_done
= 0,
315 static int wait_poll_fallback
;
317 static const char *cmd_name_mapping
[] = {
318 [ LTTNG_UST_ABI_RELEASE
] = "Release",
319 [ LTTNG_UST_ABI_SESSION
] = "Create Session",
320 [ LTTNG_UST_ABI_TRACER_VERSION
] = "Get Tracer Version",
322 [ LTTNG_UST_ABI_TRACEPOINT_LIST
] = "Create Tracepoint List",
323 [ LTTNG_UST_ABI_WAIT_QUIESCENT
] = "Wait for Quiescent State",
324 [ LTTNG_UST_ABI_REGISTER_DONE
] = "Registration Done",
325 [ LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST
] = "Create Tracepoint Field List",
327 [ LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE
] = "Create event notifier group",
329 /* Session FD commands */
330 [ LTTNG_UST_ABI_CHANNEL
] = "Create Channel",
331 [ LTTNG_UST_ABI_SESSION_START
] = "Start Session",
332 [ LTTNG_UST_ABI_SESSION_STOP
] = "Stop Session",
334 /* Channel FD commands */
335 [ LTTNG_UST_ABI_STREAM
] = "Create Stream",
336 [ LTTNG_UST_ABI_EVENT
] = "Create Event",
338 /* Event and Channel FD commands */
339 [ LTTNG_UST_ABI_CONTEXT
] = "Create Context",
340 [ LTTNG_UST_ABI_FLUSH_BUFFER
] = "Flush Buffer",
342 /* Event, Channel and Session commands */
343 [ LTTNG_UST_ABI_ENABLE
] = "Enable",
344 [ LTTNG_UST_ABI_DISABLE
] = "Disable",
346 /* Tracepoint list commands */
347 [ LTTNG_UST_ABI_TRACEPOINT_LIST_GET
] = "List Next Tracepoint",
348 [ LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST_GET
] = "List Next Tracepoint Field",
350 /* Event FD commands */
351 [ LTTNG_UST_ABI_FILTER
] = "Create Filter",
352 [ LTTNG_UST_ABI_EXCLUSION
] = "Add exclusions to event",
354 /* Event notifier group commands */
355 [ LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE
] = "Create event notifier",
357 /* Session and event notifier group commands */
358 [ LTTNG_UST_ABI_COUNTER
] = "Create Counter",
360 /* Counter commands */
361 [ LTTNG_UST_ABI_COUNTER_GLOBAL
] = "Create Counter Global",
362 [ LTTNG_UST_ABI_COUNTER_CPU
] = "Create Counter CPU",
365 static const char *str_timeout
;
366 static int got_timeout_env
;
368 static char *get_map_shm(struct sock_info
*sock_info
);
371 * Returns the HOME directory path. Caller MUST NOT free(3) the returned
375 const char *get_lttng_home_dir(void)
379 val
= (const char *) lttng_ust_getenv("LTTNG_HOME");
383 return (const char *) lttng_ust_getenv("HOME");
387 * Force a read (imply TLS allocation for dlopen) of TLS variables.
390 void lttng_nest_count_alloc_tls(void)
392 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count
)));
396 void lttng_ust_mutex_nest_alloc_tls(void)
398 asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest
)));
402 * Allocate lttng-ust urcu TLS.
405 void lttng_lttng_ust_urcu_alloc_tls(void)
407 (void) lttng_ust_urcu_read_ongoing();
410 void lttng_ust_alloc_tls(void)
412 lttng_lttng_ust_urcu_alloc_tls();
413 lttng_ringbuffer_alloc_tls();
414 lttng_vtid_alloc_tls();
415 lttng_nest_count_alloc_tls();
416 lttng_procname_alloc_tls();
417 lttng_ust_mutex_nest_alloc_tls();
418 lttng_ust_perf_counter_alloc_tls();
419 lttng_ust_common_alloc_tls();
420 lttng_cgroup_ns_alloc_tls();
421 lttng_ipc_ns_alloc_tls();
422 lttng_net_ns_alloc_tls();
423 lttng_time_ns_alloc_tls();
424 lttng_uts_ns_alloc_tls();
425 lttng_ust_ring_buffer_client_discard_alloc_tls();
426 lttng_ust_ring_buffer_client_discard_rt_alloc_tls();
427 lttng_ust_ring_buffer_client_overwrite_alloc_tls();
428 lttng_ust_ring_buffer_client_overwrite_rt_alloc_tls();
432 * LTTng-UST uses Global Dynamic model TLS variables rather than IE
433 * model because many versions of glibc don't preallocate a pool large
434 * enough for TLS variables IE model defined in other shared libraries,
435 * and causes issues when using LTTng-UST for Java tracing.
437 * Because of this use of Global Dynamic TLS variables, users wishing to
438 * trace from signal handlers need to explicitly trigger the lazy
439 * allocation of those variables for each thread before using them.
440 * This can be triggered by calling lttng_ust_init_thread().
442 void lttng_ust_init_thread(void)
445 * Because those TLS variables are global dynamic, we need to
446 * ensure those are initialized before a signal handler nesting over
447 * this thread attempts to use them.
449 lttng_ust_alloc_tls();
452 int lttng_get_notify_socket(void *owner
)
454 struct sock_info
*info
= owner
;
456 return info
->notify_socket
;
460 char* lttng_ust_sockinfo_get_procname(void *owner
)
462 struct sock_info
*info
= owner
;
464 return info
->procname
;
468 void print_cmd(int cmd
, int handle
)
470 const char *cmd_name
= "Unknown";
472 if (cmd
>= 0 && cmd
< LTTNG_ARRAY_SIZE(cmd_name_mapping
)
473 && cmd_name_mapping
[cmd
]) {
474 cmd_name
= cmd_name_mapping
[cmd
];
476 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
478 lttng_ust_obj_get_name(handle
), handle
);
482 int setup_global_apps(void)
485 assert(!global_apps
.wait_shm_mmap
);
487 global_apps
.wait_shm_mmap
= get_map_shm(&global_apps
);
488 if (!global_apps
.wait_shm_mmap
) {
489 WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing.");
490 global_apps
.allowed
= 0;
495 global_apps
.allowed
= 1;
496 lttng_pthread_getname_np(global_apps
.procname
, LTTNG_UST_CONTEXT_PROCNAME_LEN
);
501 int setup_local_apps(void)
504 const char *home_dir
;
507 assert(!local_apps
.wait_shm_mmap
);
511 * Disallow per-user tracing for setuid binaries.
513 if (uid
!= geteuid()) {
514 assert(local_apps
.allowed
== 0);
518 home_dir
= get_lttng_home_dir();
520 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
521 assert(local_apps
.allowed
== 0);
525 local_apps
.allowed
= 1;
526 snprintf(local_apps
.sock_path
, PATH_MAX
, "%s/%s/%s",
528 LTTNG_DEFAULT_HOME_RUNDIR
,
529 LTTNG_UST_SOCK_FILENAME
);
530 snprintf(local_apps
.wait_shm_path
, PATH_MAX
, "/%s-%u",
531 LTTNG_UST_WAIT_FILENAME
,
534 local_apps
.wait_shm_mmap
= get_map_shm(&local_apps
);
535 if (!local_apps
.wait_shm_mmap
) {
536 WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing.");
537 local_apps
.allowed
= 0;
542 lttng_pthread_getname_np(local_apps
.procname
, LTTNG_UST_CONTEXT_PROCNAME_LEN
);
548 * Get socket timeout, in ms.
549 * -1: wait forever. 0: don't wait. >0: timeout, in ms.
552 long get_timeout(void)
554 long constructor_delay_ms
= LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS
;
556 if (!got_timeout_env
) {
557 str_timeout
= lttng_ust_getenv("LTTNG_UST_REGISTER_TIMEOUT");
561 constructor_delay_ms
= strtol(str_timeout
, NULL
, 10);
562 /* All negative values are considered as "-1". */
563 if (constructor_delay_ms
< -1)
564 constructor_delay_ms
= -1;
565 return constructor_delay_ms
;
568 /* Timeout for notify socket send and recv. */
570 long get_notify_sock_timeout(void)
572 return get_timeout();
575 /* Timeout for connecting to cmd and notify sockets. */
577 long get_connect_sock_timeout(void)
579 return get_timeout();
583 * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
586 int get_constructor_timeout(struct timespec
*constructor_timeout
)
588 long constructor_delay_ms
;
591 constructor_delay_ms
= get_timeout();
593 switch (constructor_delay_ms
) {
594 case -1:/* fall-through */
596 return constructor_delay_ms
;
602 * If we are unable to find the current time, don't wait.
604 ret
= clock_gettime(CLOCK_REALTIME
, constructor_timeout
);
609 constructor_timeout
->tv_sec
+= constructor_delay_ms
/ 1000UL;
610 constructor_timeout
->tv_nsec
+=
611 (constructor_delay_ms
% 1000UL) * 1000000UL;
612 if (constructor_timeout
->tv_nsec
>= 1000000000UL) {
613 constructor_timeout
->tv_sec
++;
614 constructor_timeout
->tv_nsec
-= 1000000000UL;
616 /* Timeout wait (constructor_delay_ms). */
621 void get_allow_blocking(void)
623 const char *str_allow_blocking
=
624 lttng_ust_getenv("LTTNG_UST_ALLOW_BLOCKING");
626 if (str_allow_blocking
) {
627 DBG("%s environment variable is set",
628 "LTTNG_UST_ALLOW_BLOCKING");
629 lttng_ust_ringbuffer_set_allow_blocking();
634 int register_to_sessiond(int socket
, enum lttng_ust_ctl_socket_type type
,
635 const char *procname
)
637 return ustcomm_send_reg_msg(socket
,
640 lttng_ust_rb_alignof(uint8_t) * CHAR_BIT
,
641 lttng_ust_rb_alignof(uint16_t) * CHAR_BIT
,
642 lttng_ust_rb_alignof(uint32_t) * CHAR_BIT
,
643 lttng_ust_rb_alignof(uint64_t) * CHAR_BIT
,
644 lttng_ust_rb_alignof(unsigned long) * CHAR_BIT
,
649 int send_reply(int sock
, struct ustcomm_ust_reply
*lur
)
653 len
= ustcomm_send_unix_sock(sock
, lur
, sizeof(*lur
));
656 DBG("message successfully sent");
659 if (len
== -ECONNRESET
) {
660 DBG("remote end closed connection");
665 DBG("incorrect message size: %zd", len
);
671 void decrement_sem_count(unsigned int count
)
675 assert(uatomic_read(&sem_count
) >= count
);
677 if (uatomic_read(&sem_count
) <= 0) {
681 ret
= uatomic_add_return(&sem_count
, -count
);
683 ret
= sem_post(&constructor_wait
);
689 int handle_register_done(struct sock_info
*sock_info
)
691 if (sock_info
->registration_done
)
693 sock_info
->registration_done
= 1;
695 decrement_sem_count(1);
696 if (!sock_info
->statedump_pending
) {
697 sock_info
->initial_statedump_done
= 1;
698 decrement_sem_count(1);
705 int handle_register_failed(struct sock_info
*sock_info
)
707 if (sock_info
->registration_done
)
709 sock_info
->registration_done
= 1;
710 sock_info
->initial_statedump_done
= 1;
712 decrement_sem_count(2);
718 * Only execute pending statedump after the constructor semaphore has
719 * been posted by the current listener thread. This means statedump will
720 * only be performed after the "registration done" command is received
721 * from this thread's session daemon.
723 * This ensures we don't run into deadlock issues with the dynamic
724 * loader mutex, which is held while the constructor is called and
725 * waiting on the constructor semaphore. All operations requiring this
726 * dynamic loader lock need to be postponed using this mechanism.
728 * In a scenario with two session daemons connected to the application,
729 * it is possible that the first listener thread which receives the
730 * registration done command issues its statedump while the dynamic
731 * loader lock is still held by the application constructor waiting on
732 * the semaphore. It will however be allowed to proceed when the
733 * second session daemon sends the registration done command to the
734 * second listener thread. This situation therefore does not produce
738 void handle_pending_statedump(struct sock_info
*sock_info
)
740 if (sock_info
->registration_done
&& sock_info
->statedump_pending
) {
741 sock_info
->statedump_pending
= 0;
742 pthread_mutex_lock(&ust_fork_mutex
);
743 lttng_handle_pending_statedump(sock_info
);
744 pthread_mutex_unlock(&ust_fork_mutex
);
746 if (!sock_info
->initial_statedump_done
) {
747 sock_info
->initial_statedump_done
= 1;
748 decrement_sem_count(1);
754 const char *bytecode_type_str(uint32_t cmd
)
757 case LTTNG_UST_ABI_CAPTURE
:
759 case LTTNG_UST_ABI_FILTER
:
767 int handle_bytecode_recv(struct sock_info
*sock_info
,
768 int sock
, struct ustcomm_ust_msg
*lum
)
770 struct lttng_ust_bytecode_node
*bytecode
= NULL
;
771 enum lttng_ust_bytecode_type type
;
772 const struct lttng_ust_abi_objd_ops
*ops
;
773 uint32_t data_size
, data_size_max
, reloc_offset
;
779 case LTTNG_UST_ABI_FILTER
:
780 type
= LTTNG_UST_BYTECODE_TYPE_FILTER
;
781 data_size
= lum
->u
.filter
.data_size
;
782 data_size_max
= LTTNG_UST_ABI_FILTER_BYTECODE_MAX_LEN
;
783 reloc_offset
= lum
->u
.filter
.reloc_offset
;
784 seqnum
= lum
->u
.filter
.seqnum
;
786 case LTTNG_UST_ABI_CAPTURE
:
787 type
= LTTNG_UST_BYTECODE_TYPE_CAPTURE
;
788 data_size
= lum
->u
.capture
.data_size
;
789 data_size_max
= LTTNG_UST_ABI_CAPTURE_BYTECODE_MAX_LEN
;
790 reloc_offset
= lum
->u
.capture
.reloc_offset
;
791 seqnum
= lum
->u
.capture
.seqnum
;
797 if (data_size
> data_size_max
) {
798 ERR("Bytecode %s data size is too large: %u bytes",
799 bytecode_type_str(lum
->cmd
), data_size
);
804 if (reloc_offset
> data_size
) {
805 ERR("Bytecode %s reloc offset %u is not within data",
806 bytecode_type_str(lum
->cmd
), reloc_offset
);
811 /* Allocate the structure AND the `data[]` field. */
812 bytecode
= zmalloc(sizeof(*bytecode
) + data_size
);
818 bytecode
->bc
.len
= data_size
;
819 bytecode
->bc
.reloc_offset
= reloc_offset
;
820 bytecode
->bc
.seqnum
= seqnum
;
821 bytecode
->type
= type
;
823 len
= ustcomm_recv_unix_sock(sock
, bytecode
->bc
.data
, bytecode
->bc
.len
);
825 case 0: /* orderly shutdown */
829 if (len
== bytecode
->bc
.len
) {
830 DBG("Bytecode %s data received",
831 bytecode_type_str(lum
->cmd
));
833 } else if (len
< 0) {
834 DBG("Receive failed from lttng-sessiond with errno %d",
836 if (len
== -ECONNRESET
) {
837 ERR("%s remote end closed connection",
845 DBG("Incorrect %s bytecode data message size: %zd",
846 bytecode_type_str(lum
->cmd
), len
);
852 ops
= lttng_ust_abi_objd_ops(lum
->handle
);
859 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
860 (unsigned long) &bytecode
,
871 void prepare_cmd_reply(struct ustcomm_ust_reply
*lur
, uint32_t handle
, uint32_t cmd
, int ret
)
873 lur
->handle
= handle
;
877 lur
->ret_code
= LTTNG_UST_OK
;
880 * Use -LTTNG_UST_ERR as wildcard for UST internal
881 * error that are not caused by the transport, except if
882 * we already have a more precise error message to
885 if (ret
> -LTTNG_UST_ERR
) {
886 /* Translate code to UST error. */
889 lur
->ret_code
= -LTTNG_UST_ERR_EXIST
;
892 lur
->ret_code
= -LTTNG_UST_ERR_INVAL
;
895 lur
->ret_code
= -LTTNG_UST_ERR_NOENT
;
898 lur
->ret_code
= -LTTNG_UST_ERR_PERM
;
901 lur
->ret_code
= -LTTNG_UST_ERR_NOSYS
;
904 lur
->ret_code
= -LTTNG_UST_ERR
;
914 int handle_message(struct sock_info
*sock_info
,
915 int sock
, struct ustcomm_ust_msg
*lum
)
918 const struct lttng_ust_abi_objd_ops
*ops
;
919 struct ustcomm_ust_reply lur
;
920 union lttng_ust_abi_args args
;
921 char ctxstr
[LTTNG_UST_ABI_SYM_NAME_LEN
]; /* App context string. */
924 memset(&lur
, 0, sizeof(lur
));
927 ret
= -LTTNG_UST_ERR_EXITING
;
931 ops
= lttng_ust_abi_objd_ops(lum
->handle
);
938 case LTTNG_UST_ABI_FILTER
:
939 case LTTNG_UST_ABI_EXCLUSION
:
940 case LTTNG_UST_ABI_CHANNEL
:
941 case LTTNG_UST_ABI_STREAM
:
942 case LTTNG_UST_ABI_CONTEXT
:
944 * Those commands send additional payload after struct
945 * ustcomm_ust_msg, which makes it pretty much impossible to
946 * deal with "unknown command" errors without leaving the
947 * communication pipe in a out-of-sync state. This is part of
948 * the ABI between liblttng-ust-ctl and liblttng-ust, and
949 * should be fixed on the next breaking
950 * LTTNG_UST_ABI_MAJOR_VERSION protocol bump by indicating the
951 * total command message length as part of a message header so
952 * that the protocol can recover from invalid command errors.
956 case LTTNG_UST_ABI_CAPTURE
:
957 case LTTNG_UST_ABI_COUNTER
:
958 case LTTNG_UST_ABI_COUNTER_GLOBAL
:
959 case LTTNG_UST_ABI_COUNTER_CPU
:
960 case LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE
:
961 case LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE
:
963 * Those commands expect a reply to the struct ustcomm_ust_msg
964 * before sending additional payload.
966 prepare_cmd_reply(&lur
, lum
->handle
, lum
->cmd
, 0);
968 ret
= send_reply(sock
, &lur
);
970 DBG("error sending reply");
977 * Other commands either don't send additional payload, or are
984 case LTTNG_UST_ABI_REGISTER_DONE
:
985 if (lum
->handle
== LTTNG_UST_ABI_ROOT_HANDLE
)
986 ret
= handle_register_done(sock_info
);
990 case LTTNG_UST_ABI_RELEASE
:
991 if (lum
->handle
== LTTNG_UST_ABI_ROOT_HANDLE
)
994 ret
= lttng_ust_abi_objd_unref(lum
->handle
, 1);
996 case LTTNG_UST_ABI_CAPTURE
:
997 case LTTNG_UST_ABI_FILTER
:
998 ret
= handle_bytecode_recv(sock_info
, sock
, lum
);
1002 case LTTNG_UST_ABI_EXCLUSION
:
1004 /* Receive exclusion names */
1005 struct lttng_ust_excluder_node
*node
;
1008 count
= lum
->u
.exclusion
.count
;
1010 /* There are no names to read */
1014 node
= zmalloc(sizeof(*node
) +
1015 count
* LTTNG_UST_ABI_SYM_NAME_LEN
);
1020 node
->excluder
.count
= count
;
1021 len
= ustcomm_recv_unix_sock(sock
, node
->excluder
.names
,
1022 count
* LTTNG_UST_ABI_SYM_NAME_LEN
);
1024 case 0: /* orderly shutdown */
1029 if (len
== count
* LTTNG_UST_ABI_SYM_NAME_LEN
) {
1030 DBG("Exclusion data received");
1032 } else if (len
< 0) {
1033 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1034 if (len
== -ECONNRESET
) {
1035 ERR("%s remote end closed connection", sock_info
->name
);
1044 DBG("Incorrect exclusion data message size: %zd", len
);
1051 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1052 (unsigned long) &node
,
1059 case LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE
:
1061 int event_notifier_notif_fd
, close_ret
;
1063 len
= ustcomm_recv_event_notifier_notif_fd_from_sessiond(sock
,
1064 &event_notifier_notif_fd
);
1066 case 0: /* orderly shutdown */
1073 DBG("Receive failed from lttng-sessiond with errno %d",
1075 if (len
== -ECONNRESET
) {
1076 ERR("%s remote end closed connection",
1084 DBG("Incorrect event notifier fd message size: %zd",
1090 args
.event_notifier_handle
.event_notifier_notif_fd
=
1091 event_notifier_notif_fd
;
1093 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1094 (unsigned long) &lum
->u
,
1098 if (args
.event_notifier_handle
.event_notifier_notif_fd
>= 0) {
1099 lttng_ust_lock_fd_tracker();
1100 close_ret
= close(args
.event_notifier_handle
.event_notifier_notif_fd
);
1101 lttng_ust_unlock_fd_tracker();
1107 case LTTNG_UST_ABI_CHANNEL
:
1112 len
= ustcomm_recv_channel_from_sessiond(sock
,
1113 &chan_data
, lum
->u
.channel
.len
,
1116 case 0: /* orderly shutdown */
1120 if (len
== lum
->u
.channel
.len
) {
1121 DBG("channel data received");
1123 } else if (len
< 0) {
1124 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1125 if (len
== -ECONNRESET
) {
1126 ERR("%s remote end closed connection", sock_info
->name
);
1133 DBG("incorrect channel data message size: %zd", len
);
1138 args
.channel
.chan_data
= chan_data
;
1139 args
.channel
.wakeup_fd
= wakeup_fd
;
1141 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1142 (unsigned long) &lum
->u
,
1146 if (args
.channel
.wakeup_fd
>= 0) {
1149 lttng_ust_lock_fd_tracker();
1150 close_ret
= close(args
.channel
.wakeup_fd
);
1151 lttng_ust_unlock_fd_tracker();
1152 args
.channel
.wakeup_fd
= -1;
1156 free(args
.channel
.chan_data
);
1159 case LTTNG_UST_ABI_STREAM
:
1163 /* Receive shm_fd, wakeup_fd */
1164 ret
= ustcomm_recv_stream_from_sessiond(sock
,
1166 &args
.stream
.shm_fd
,
1167 &args
.stream
.wakeup_fd
);
1173 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1174 (unsigned long) &lum
->u
,
1178 if (args
.stream
.shm_fd
>= 0) {
1179 lttng_ust_lock_fd_tracker();
1180 close_ret
= close(args
.stream
.shm_fd
);
1181 lttng_ust_unlock_fd_tracker();
1182 args
.stream
.shm_fd
= -1;
1186 if (args
.stream
.wakeup_fd
>= 0) {
1187 lttng_ust_lock_fd_tracker();
1188 close_ret
= close(args
.stream
.wakeup_fd
);
1189 lttng_ust_unlock_fd_tracker();
1190 args
.stream
.wakeup_fd
= -1;
1196 case LTTNG_UST_ABI_CONTEXT
:
1197 switch (lum
->u
.context
.ctx
) {
1198 case LTTNG_UST_ABI_CONTEXT_APP_CONTEXT
:
1201 size_t ctxlen
, recvlen
;
1203 ctxlen
= strlen("$app.") + lum
->u
.context
.u
.app_ctx
.provider_name_len
- 1
1204 + strlen(":") + lum
->u
.context
.u
.app_ctx
.ctx_name_len
;
1205 if (ctxlen
>= LTTNG_UST_ABI_SYM_NAME_LEN
) {
1206 ERR("Application context string length size is too large: %zu bytes",
1211 strcpy(ctxstr
, "$app.");
1212 p
= &ctxstr
[strlen("$app.")];
1213 recvlen
= ctxlen
- strlen("$app.");
1214 len
= ustcomm_recv_unix_sock(sock
, p
, recvlen
);
1216 case 0: /* orderly shutdown */
1220 if (len
== recvlen
) {
1221 DBG("app context data received");
1223 } else if (len
< 0) {
1224 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1225 if (len
== -ECONNRESET
) {
1226 ERR("%s remote end closed connection", sock_info
->name
);
1233 DBG("incorrect app context data message size: %zd", len
);
1238 /* Put : between provider and ctxname. */
1239 p
[lum
->u
.context
.u
.app_ctx
.provider_name_len
- 1] = ':';
1240 args
.app_context
.ctxname
= ctxstr
;
1247 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1248 (unsigned long) &lum
->u
,
1254 case LTTNG_UST_ABI_COUNTER
:
1258 len
= ustcomm_recv_counter_from_sessiond(sock
,
1259 &counter_data
, lum
->u
.counter
.len
);
1261 case 0: /* orderly shutdown */
1265 if (len
== lum
->u
.counter
.len
) {
1266 DBG("counter data received");
1268 } else if (len
< 0) {
1269 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1270 if (len
== -ECONNRESET
) {
1271 ERR("%s remote end closed connection", sock_info
->name
);
1278 DBG("incorrect counter data message size: %zd", len
);
1283 args
.counter
.counter_data
= counter_data
;
1285 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1286 (unsigned long) &lum
->u
,
1290 free(args
.counter
.counter_data
);
1293 case LTTNG_UST_ABI_COUNTER_GLOBAL
:
1295 /* Receive shm_fd */
1296 ret
= ustcomm_recv_counter_shm_from_sessiond(sock
,
1297 &args
.counter_shm
.shm_fd
);
1303 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1304 (unsigned long) &lum
->u
,
1308 if (args
.counter_shm
.shm_fd
>= 0) {
1311 lttng_ust_lock_fd_tracker();
1312 close_ret
= close(args
.counter_shm
.shm_fd
);
1313 lttng_ust_unlock_fd_tracker();
1314 args
.counter_shm
.shm_fd
= -1;
1320 case LTTNG_UST_ABI_COUNTER_CPU
:
1322 /* Receive shm_fd */
1323 ret
= ustcomm_recv_counter_shm_from_sessiond(sock
,
1324 &args
.counter_shm
.shm_fd
);
1330 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1331 (unsigned long) &lum
->u
,
1335 if (args
.counter_shm
.shm_fd
>= 0) {
1338 lttng_ust_lock_fd_tracker();
1339 close_ret
= close(args
.counter_shm
.shm_fd
);
1340 lttng_ust_unlock_fd_tracker();
1341 args
.counter_shm
.shm_fd
= -1;
1347 case LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE
:
1349 /* Receive struct lttng_ust_event_notifier */
1350 struct lttng_ust_abi_event_notifier event_notifier
;
1352 if (sizeof(event_notifier
) != lum
->u
.event_notifier
.len
) {
1353 DBG("incorrect event notifier data message size: %u", lum
->u
.event_notifier
.len
);
1357 len
= ustcomm_recv_unix_sock(sock
, &event_notifier
, sizeof(event_notifier
));
1359 case 0: /* orderly shutdown */
1363 if (len
== sizeof(event_notifier
)) {
1364 DBG("event notifier data received");
1366 } else if (len
< 0) {
1367 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
1368 if (len
== -ECONNRESET
) {
1369 ERR("%s remote end closed connection", sock_info
->name
);
1376 DBG("incorrect event notifier data message size: %zd", len
);
1382 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1383 (unsigned long) &event_notifier
,
1392 ret
= ops
->cmd(lum
->handle
, lum
->cmd
,
1393 (unsigned long) &lum
->u
,
1400 prepare_cmd_reply(&lur
, lum
->handle
, lum
->cmd
, ret
);
1404 case LTTNG_UST_ABI_TRACER_VERSION
:
1405 lur
.u
.version
= lum
->u
.version
;
1407 case LTTNG_UST_ABI_TRACEPOINT_LIST_GET
:
1408 memcpy(&lur
.u
.tracepoint
, &lum
->u
.tracepoint
, sizeof(lur
.u
.tracepoint
));
1412 DBG("Return value: %d", lur
.ret_val
);
1417 * Performed delayed statedump operations outside of the UST
1418 * lock. We need to take the dynamic loader lock before we take
1419 * the UST lock internally within handle_pending_statedump().
1421 handle_pending_statedump(sock_info
);
1424 ret
= -LTTNG_UST_ERR_EXITING
;
1428 ret
= send_reply(sock
, &lur
);
1430 DBG("error sending reply");
1435 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1438 if (lur
.ret_code
== LTTNG_UST_OK
) {
1440 case LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST_GET
:
1441 len
= ustcomm_send_unix_sock(sock
,
1442 &args
.field_list
.entry
,
1443 sizeof(args
.field_list
.entry
));
1448 if (len
!= sizeof(args
.field_list
.entry
)) {
1462 void cleanup_sock_info(struct sock_info
*sock_info
, int exiting
)
1466 if (sock_info
->root_handle
!= -1) {
1467 ret
= lttng_ust_abi_objd_unref(sock_info
->root_handle
, 1);
1469 ERR("Error unref root handle");
1471 sock_info
->root_handle
= -1;
1476 * wait_shm_mmap, socket and notify socket are used by listener
1477 * threads outside of the ust lock, so we cannot tear them down
1478 * ourselves, because we cannot join on these threads. Leave
1479 * responsibility of cleaning up these resources to the OS
1485 sock_info
->registration_done
= 0;
1486 sock_info
->initial_statedump_done
= 0;
1488 if (sock_info
->socket
!= -1) {
1489 ret
= ustcomm_close_unix_sock(sock_info
->socket
);
1491 ERR("Error closing ust cmd socket");
1493 sock_info
->socket
= -1;
1495 if (sock_info
->notify_socket
!= -1) {
1496 ret
= ustcomm_close_unix_sock(sock_info
->notify_socket
);
1498 ERR("Error closing ust notify socket");
1500 sock_info
->notify_socket
= -1;
1502 if (sock_info
->wait_shm_mmap
) {
1505 page_size
= LTTNG_UST_PAGE_SIZE
;
1506 if (page_size
<= 0) {
1510 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1512 ret
= munmap(sock_info
->wait_shm_mmap
, page_size
);
1514 ERR("Error unmapping wait shm");
1517 sock_info
->wait_shm_mmap
= NULL
;
1522 * Using fork to set umask in the child process (not multi-thread safe).
1523 * We deal with the shm_open vs ftruncate race (happening when the
1524 * sessiond owns the shm and does not let everybody modify it, to ensure
1525 * safety against shm_unlink) by simply letting the mmap fail and
1526 * retrying after a few seconds.
1527 * For global shm, everybody has rw access to it until the sessiond
1531 int get_wait_shm(struct sock_info
*sock_info
, size_t mmap_size
)
1533 int wait_shm_fd
, ret
;
1537 * Try to open read-only.
1539 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
, O_RDONLY
, 0);
1540 if (wait_shm_fd
>= 0) {
1543 size_t bytes_read
= 0;
1546 * Try to read the fd. If unable to do so, try opening
1550 len
= read(wait_shm_fd
,
1551 &((char *) &tmp_read
)[bytes_read
],
1552 sizeof(tmp_read
) - bytes_read
);
1556 } while ((len
< 0 && errno
== EINTR
)
1557 || (len
> 0 && bytes_read
< sizeof(tmp_read
)));
1558 if (bytes_read
!= sizeof(tmp_read
)) {
1559 ret
= close(wait_shm_fd
);
1561 ERR("close wait_shm_fd");
1566 } else if (wait_shm_fd
< 0 && errno
!= ENOENT
) {
1568 * Real-only open did not work, and it's not because the
1569 * entry was not present. It's a failure that prohibits
1572 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
1578 * If the open failed because the file did not exist, or because
1579 * the file was not truncated yet, try creating it ourself.
1581 URCU_TLS(lttng_ust_nest_count
)++;
1583 URCU_TLS(lttng_ust_nest_count
)--;
1588 * Parent: wait for child to return, in which case the
1589 * shared memory map will have been created.
1591 pid
= wait(&status
);
1592 if (pid
< 0 || !WIFEXITED(status
) || WEXITSTATUS(status
) != 0) {
1597 * Try to open read-only again after creation.
1599 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
, O_RDONLY
, 0);
1600 if (wait_shm_fd
< 0) {
1602 * Real-only open did not work. It's a failure
1603 * that prohibits using shm.
1605 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
1609 } else if (pid
== 0) {
1613 create_mode
= S_IRUSR
| S_IWUSR
| S_IRGRP
;
1614 if (sock_info
->global
)
1615 create_mode
|= S_IROTH
| S_IWGRP
| S_IWOTH
;
1617 * We're alone in a child process, so we can modify the
1618 * process-wide umask.
1620 umask(~create_mode
);
1622 * Try creating shm (or get rw access).
1623 * We don't do an exclusive open, because we allow other
1624 * processes to create+ftruncate it concurrently.
1626 wait_shm_fd
= shm_open(sock_info
->wait_shm_path
,
1627 O_RDWR
| O_CREAT
, create_mode
);
1628 if (wait_shm_fd
>= 0) {
1629 ret
= ftruncate(wait_shm_fd
, mmap_size
);
1631 PERROR("ftruncate");
1632 _exit(EXIT_FAILURE
);
1634 _exit(EXIT_SUCCESS
);
1637 * For local shm, we need to have rw access to accept
1638 * opening it: this means the local sessiond will be
1639 * able to wake us up. For global shm, we open it even
1640 * if rw access is not granted, because the root.root
1641 * sessiond will be able to override all rights and wake
1644 if (!sock_info
->global
&& errno
!= EACCES
) {
1645 ERR("Error opening shm %s", sock_info
->wait_shm_path
);
1646 _exit(EXIT_FAILURE
);
1649 * The shm exists, but we cannot open it RW. Report
1652 _exit(EXIT_SUCCESS
);
1657 if (wait_shm_fd
>= 0 && !sock_info
->global
) {
1658 struct stat statbuf
;
1661 * Ensure that our user is the owner of the shm file for
1662 * local shm. If we do not own the file, it means our
1663 * sessiond will not have access to wake us up (there is
1664 * probably a rogue process trying to fake our
1665 * sessiond). Fallback to polling method in this case.
1667 ret
= fstat(wait_shm_fd
, &statbuf
);
1672 if (statbuf
.st_uid
!= getuid())
1678 ret
= close(wait_shm_fd
);
1680 PERROR("Error closing fd");
1686 char *get_map_shm(struct sock_info
*sock_info
)
1689 int wait_shm_fd
, ret
;
1690 char *wait_shm_mmap
;
1692 page_size
= sysconf(_SC_PAGE_SIZE
);
1693 if (page_size
<= 0) {
1697 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1701 lttng_ust_lock_fd_tracker();
1702 wait_shm_fd
= get_wait_shm(sock_info
, page_size
);
1703 if (wait_shm_fd
< 0) {
1704 lttng_ust_unlock_fd_tracker();
1708 ret
= lttng_ust_add_fd_to_tracker(wait_shm_fd
);
1710 ret
= close(wait_shm_fd
);
1712 PERROR("Error closing fd");
1714 lttng_ust_unlock_fd_tracker();
1719 lttng_ust_unlock_fd_tracker();
1721 wait_shm_mmap
= mmap(NULL
, page_size
, PROT_READ
,
1722 MAP_SHARED
, wait_shm_fd
, 0);
1724 /* close shm fd immediately after taking the mmap reference */
1725 lttng_ust_lock_fd_tracker();
1726 ret
= close(wait_shm_fd
);
1728 lttng_ust_delete_fd_from_tracker(wait_shm_fd
);
1730 PERROR("Error closing fd");
1732 lttng_ust_unlock_fd_tracker();
1734 if (wait_shm_mmap
== MAP_FAILED
) {
1735 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1738 return wait_shm_mmap
;
1745 void wait_for_sessiond(struct sock_info
*sock_info
)
1747 /* Use ust_lock to check if we should quit. */
1751 if (wait_poll_fallback
) {
1756 assert(sock_info
->wait_shm_mmap
);
1758 DBG("Waiting for %s apps sessiond", sock_info
->name
);
1759 /* Wait for futex wakeup */
1760 if (uatomic_read((int32_t *) sock_info
->wait_shm_mmap
))
1763 while (lttng_ust_futex_async((int32_t *) sock_info
->wait_shm_mmap
,
1764 FUTEX_WAIT
, 0, NULL
, NULL
, 0)) {
1767 /* Value already changed. */
1770 /* Retry if interrupted by signal. */
1771 break; /* Get out of switch. */
1773 wait_poll_fallback
= 1;
1775 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1776 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1777 "Please upgrade your kernel "
1778 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1779 "mainline). LTTng-UST will use polling mode fallback.");
1780 if (lttng_ust_logging_debug_enabled())
1798 * This thread does not allocate any resource, except within
1799 * handle_message, within mutex protection. This mutex protects against
1801 * The other moment it allocates resources is at socket connection, which
1802 * is also protected by the mutex.
1805 void *ust_listener_thread(void *arg
)
1807 struct sock_info
*sock_info
= arg
;
1808 int sock
, ret
, prev_connect_failed
= 0, has_waited
= 0, fd
;
1811 lttng_ust_alloc_tls();
1813 * If available, add '-ust' to the end of this thread's
1816 ret
= lttng_ust_setustprocname();
1818 ERR("Unable to set UST process name");
1821 /* Restart trying to connect to the session daemon */
1823 if (prev_connect_failed
) {
1824 /* Wait for sessiond availability with pipe */
1825 wait_for_sessiond(sock_info
);
1829 * Sleep for 5 seconds before retrying after a
1830 * sequence of failure / wait / failure. This
1831 * deals with a killed or broken session daemon.
1837 prev_connect_failed
= 0;
1844 if (sock_info
->socket
!= -1) {
1845 /* FD tracker is updated by ustcomm_close_unix_sock() */
1846 ret
= ustcomm_close_unix_sock(sock_info
->socket
);
1848 ERR("Error closing %s ust cmd socket",
1851 sock_info
->socket
= -1;
1853 if (sock_info
->notify_socket
!= -1) {
1854 /* FD tracker is updated by ustcomm_close_unix_sock() */
1855 ret
= ustcomm_close_unix_sock(sock_info
->notify_socket
);
1857 ERR("Error closing %s ust notify socket",
1860 sock_info
->notify_socket
= -1;
1865 * Register. We need to perform both connect and sending
1866 * registration message before doing the next connect otherwise
1867 * we may reach unix socket connect queue max limits and block
1868 * on the 2nd connect while the session daemon is awaiting the
1869 * first connect registration message.
1871 /* Connect cmd socket */
1872 lttng_ust_lock_fd_tracker();
1873 ret
= ustcomm_connect_unix_sock(sock_info
->sock_path
,
1874 get_connect_sock_timeout());
1876 lttng_ust_unlock_fd_tracker();
1877 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info
->name
);
1878 prev_connect_failed
= 1;
1881 * If we cannot find the sessiond daemon, don't delay
1882 * constructor execution.
1884 ret
= handle_register_failed(sock_info
);
1890 ret
= lttng_ust_add_fd_to_tracker(fd
);
1894 PERROR("close on sock_info->socket");
1897 lttng_ust_unlock_fd_tracker();
1902 sock_info
->socket
= ret
;
1903 lttng_ust_unlock_fd_tracker();
1907 * Unlock/relock ust lock because connect is blocking (with
1908 * timeout). Don't delay constructors on the ust lock for too
1916 * Create only one root handle per listener thread for the whole
1917 * process lifetime, so we ensure we get ID which is statically
1918 * assigned to the root handle.
1920 if (sock_info
->root_handle
== -1) {
1921 ret
= lttng_abi_create_root_handle();
1923 ERR("Error creating root handle");
1926 sock_info
->root_handle
= ret
;
1929 ret
= register_to_sessiond(sock_info
->socket
, LTTNG_UST_CTL_SOCKET_CMD
,
1930 sock_info
->procname
);
1932 ERR("Error registering to %s ust cmd socket",
1934 prev_connect_failed
= 1;
1936 * If we cannot register to the sessiond daemon, don't
1937 * delay constructor execution.
1939 ret
= handle_register_failed(sock_info
);
1947 * Unlock/relock ust lock because connect is blocking (with
1948 * timeout). Don't delay constructors on the ust lock for too
1955 /* Connect notify socket */
1956 lttng_ust_lock_fd_tracker();
1957 ret
= ustcomm_connect_unix_sock(sock_info
->sock_path
,
1958 get_connect_sock_timeout());
1960 lttng_ust_unlock_fd_tracker();
1961 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info
->name
);
1962 prev_connect_failed
= 1;
1965 * If we cannot find the sessiond daemon, don't delay
1966 * constructor execution.
1968 ret
= handle_register_failed(sock_info
);
1975 ret
= lttng_ust_add_fd_to_tracker(fd
);
1979 PERROR("close on sock_info->notify_socket");
1982 lttng_ust_unlock_fd_tracker();
1987 sock_info
->notify_socket
= ret
;
1988 lttng_ust_unlock_fd_tracker();
1992 * Unlock/relock ust lock because connect is blocking (with
1993 * timeout). Don't delay constructors on the ust lock for too
2000 timeout
= get_notify_sock_timeout();
2003 * Give at least 10ms to sessiond to reply to
2008 ret
= ustcomm_setsockopt_rcv_timeout(sock_info
->notify_socket
,
2011 WARN("Error setting socket receive timeout");
2013 ret
= ustcomm_setsockopt_snd_timeout(sock_info
->notify_socket
,
2016 WARN("Error setting socket send timeout");
2018 } else if (timeout
< -1) {
2019 WARN("Unsupported timeout value %ld", timeout
);
2022 ret
= register_to_sessiond(sock_info
->notify_socket
,
2023 LTTNG_UST_CTL_SOCKET_NOTIFY
, sock_info
->procname
);
2025 ERR("Error registering to %s ust notify socket",
2027 prev_connect_failed
= 1;
2029 * If we cannot register to the sessiond daemon, don't
2030 * delay constructor execution.
2032 ret
= handle_register_failed(sock_info
);
2037 sock
= sock_info
->socket
;
2043 struct ustcomm_ust_msg lum
;
2045 len
= ustcomm_recv_unix_sock(sock
, &lum
, sizeof(lum
));
2047 case 0: /* orderly shutdown */
2048 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info
->name
);
2053 * Either sessiond has shutdown or refused us by closing the socket.
2054 * In either case, we don't want to delay construction execution,
2055 * and we need to wait before retry.
2057 prev_connect_failed
= 1;
2059 * If we cannot register to the sessiond daemon, don't
2060 * delay constructor execution.
2062 ret
= handle_register_failed(sock_info
);
2067 print_cmd(lum
.cmd
, lum
.handle
);
2068 ret
= handle_message(sock_info
, sock
, &lum
);
2070 ERR("Error handling message for %s socket",
2073 * Close socket if protocol error is
2081 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len
);
2083 DBG("incorrect message size (%s socket): %zd", sock_info
->name
, len
);
2085 if (len
== -ECONNRESET
) {
2086 DBG("%s remote end closed connection", sock_info
->name
);
2097 /* Cleanup socket handles before trying to reconnect */
2098 lttng_ust_abi_objd_table_owner_cleanup(sock_info
);
2100 goto restart
; /* try to reconnect */
2105 pthread_mutex_lock(&ust_exit_mutex
);
2106 sock_info
->thread_active
= 0;
2107 pthread_mutex_unlock(&ust_exit_mutex
);
2112 * Weak symbol to call when the ust malloc wrapper is not loaded.
2114 __attribute__((weak
))
2115 void lttng_ust_libc_wrapper_malloc_ctor(void)
2120 * Use a symbol of the previous ABI to detect if liblttng-ust.so.0 is loaded in
2121 * the current process.
2123 #define LTTNG_UST_SONAME_0_SYM "ltt_probe_register"
2126 void lttng_ust_check_soname_0(void)
2128 if (!dlsym(RTLD_DEFAULT
, LTTNG_UST_SONAME_0_SYM
))
2131 CRIT("Incompatible library ABIs detected within the same process. "
2132 "The process is likely linked against different major soname of LTTng-UST which is unsupported. "
2133 "The detection was triggered by lookup of ABI 0 symbol \"%s\" in the Global Symbol Table\n",
2134 LTTNG_UST_SONAME_0_SYM
);
2138 * Expose a canary symbol of the previous ABI to ensure we catch uses of a
2139 * liblttng-ust.so.0 dlopen'd after .so.1 has been loaded. Use a different
2140 * symbol than the detection code to ensure we don't detect ourself.
2142 * This scheme will only work on systems where the global symbol table has
2143 * priority when resolving the symbols of a dlopened shared object, which is
2144 * the case on Linux but not on FreeBSD.
2146 void init_usterr(void);
2147 void init_usterr(void)
2149 CRIT("Incompatible library ABIs detected within the same process. "
2150 "The process is likely linked against different major soname of LTTng-UST which is unsupported. "
2151 "The detection was triggered by canary symbol \"%s\"\n", __func__
);
2155 * sessiond monitoring thread: monitor presence of global and per-user
2156 * sessiond by polling the application common named pipe.
2159 void lttng_ust_ctor(void)
2160 __attribute__((constructor
));
2162 void lttng_ust_ctor(void)
2164 struct timespec constructor_timeout
;
2165 sigset_t sig_all_blocked
, orig_parent_mask
;
2166 pthread_attr_t thread_attr
;
2171 if (uatomic_xchg(&initialized
, 1) == 1)
2175 * Fixup interdependency between TLS allocation mutex (which happens
2176 * to be the dynamic linker mutex) and ust_lock, taken within
2179 lttng_ust_alloc_tls();
2181 lttng_ust_loaded
= 1;
2184 * Check if we find a symbol of the previous ABI in the current process
2185 * as different ABIs of liblttng-ust can't co-exist in a process. If we
2186 * do so, emit a critical log message which will also abort if the
2187 * LTTNG_UST_ABORT_ON_CRITICAL environment variable is set.
2189 lttng_ust_check_soname_0();
2192 * We need to ensure that the liblttng-ust library is not unloaded to avoid
2193 * the unloading of code used by the ust_listener_threads as we can not
2194 * reliably know when they exited. To do that, manually load
2195 * liblttng-ust.so to increment the dynamic loader's internal refcount for
2196 * this library so it never becomes zero, thus never gets unloaded from the
2197 * address space of the process. Since we are already running in the
2198 * constructor of the LTTNG_UST_LIB_SONAME library, calling dlopen will
2199 * simply increment the refcount and no additional work is needed by the
2200 * dynamic loader as the shared library is already loaded in the address
2201 * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
2202 * unloading of the UST library if its refcount becomes zero (which should
2203 * never happen). Do the return value check but discard the handle at the
2204 * end of the function as it's not needed.
2206 handle
= dlopen(LTTNG_UST_LIB_SONAME
, RTLD_LAZY
| RTLD_NODELETE
);
2208 ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SONAME
);
2210 DBG("dlopened liblttng-ust shared library (%s).", LTTNG_UST_LIB_SONAME
);
2214 * We want precise control over the order in which we construct
2215 * our sub-libraries vs starting to receive commands from
2216 * sessiond (otherwise leading to errors when trying to create
2217 * sessiond before the init functions are completed).
2221 * Both the logging and getenv lazy-initialization uses getenv()
2222 * internally and thus needs to be explicitly initialized in
2223 * liblttng-ust before we start any threads as an unsuspecting normally
2224 * single threaded application using liblttng-ust could be using
2225 * setenv() which is not thread-safe.
2227 lttng_ust_logging_init();
2228 lttng_ust_getenv_init();
2230 /* Call the liblttng-ust-common constructor. */
2231 lttng_ust_common_ctor();
2233 lttng_ust_tp_init();
2234 lttng_ust_statedump_init();
2235 lttng_ust_ring_buffer_clients_init();
2236 lttng_ust_counter_clients_init();
2237 lttng_perf_counter_init();
2239 * Invoke ust malloc wrapper init before starting other threads.
2241 lttng_ust_libc_wrapper_malloc_ctor();
2243 timeout_mode
= get_constructor_timeout(&constructor_timeout
);
2245 get_allow_blocking();
2247 ret
= sem_init(&constructor_wait
, 0, 0);
2252 ret
= setup_global_apps();
2254 assert(global_apps
.allowed
== 0);
2255 DBG("global apps setup returned %d", ret
);
2258 ret
= setup_local_apps();
2260 assert(local_apps
.allowed
== 0);
2261 DBG("local apps setup returned %d", ret
);
2264 /* A new thread created by pthread_create inherits the signal mask
2265 * from the parent. To avoid any signal being received by the
2266 * listener thread, we block all signals temporarily in the parent,
2267 * while we create the listener thread.
2269 sigfillset(&sig_all_blocked
);
2270 ret
= pthread_sigmask(SIG_SETMASK
, &sig_all_blocked
, &orig_parent_mask
);
2272 ERR("pthread_sigmask: %s", strerror(ret
));
2275 ret
= pthread_attr_init(&thread_attr
);
2277 ERR("pthread_attr_init: %s", strerror(ret
));
2279 ret
= pthread_attr_setdetachstate(&thread_attr
, PTHREAD_CREATE_DETACHED
);
2281 ERR("pthread_attr_setdetachstate: %s", strerror(ret
));
2284 if (global_apps
.allowed
) {
2285 pthread_mutex_lock(&ust_exit_mutex
);
2286 ret
= pthread_create(&global_apps
.ust_listener
, &thread_attr
,
2287 ust_listener_thread
, &global_apps
);
2289 ERR("pthread_create global: %s", strerror(ret
));
2291 global_apps
.thread_active
= 1;
2292 pthread_mutex_unlock(&ust_exit_mutex
);
2294 handle_register_done(&global_apps
);
2297 if (local_apps
.allowed
) {
2298 pthread_mutex_lock(&ust_exit_mutex
);
2299 ret
= pthread_create(&local_apps
.ust_listener
, &thread_attr
,
2300 ust_listener_thread
, &local_apps
);
2302 ERR("pthread_create local: %s", strerror(ret
));
2304 local_apps
.thread_active
= 1;
2305 pthread_mutex_unlock(&ust_exit_mutex
);
2307 handle_register_done(&local_apps
);
2309 ret
= pthread_attr_destroy(&thread_attr
);
2311 ERR("pthread_attr_destroy: %s", strerror(ret
));
2314 /* Restore original signal mask in parent */
2315 ret
= pthread_sigmask(SIG_SETMASK
, &orig_parent_mask
, NULL
);
2317 ERR("pthread_sigmask: %s", strerror(ret
));
2320 switch (timeout_mode
) {
2321 case 1: /* timeout wait */
2323 ret
= sem_timedwait(&constructor_wait
,
2324 &constructor_timeout
);
2325 } while (ret
< 0 && errno
== EINTR
);
2329 ERR("Timed out waiting for lttng-sessiond");
2332 PERROR("sem_timedwait");
2335 ERR("Unexpected error \"%s\" returned by sem_timedwait",
2340 case -1:/* wait forever */
2342 ret
= sem_wait(&constructor_wait
);
2343 } while (ret
< 0 && errno
== EINTR
);
2350 ERR("Unexpected error \"%s\" returned by sem_wait",
2355 case 0: /* no timeout */
2361 void lttng_ust_cleanup(int exiting
)
2363 cleanup_sock_info(&global_apps
, exiting
);
2364 cleanup_sock_info(&local_apps
, exiting
);
2365 local_apps
.allowed
= 0;
2366 global_apps
.allowed
= 0;
2368 * The teardown in this function all affect data structures
2369 * accessed under the UST lock by the listener thread. This
2370 * lock, along with the lttng_ust_comm_should_quit flag, ensure
2371 * that none of these threads are accessing this data at this
2374 lttng_ust_abi_exit();
2375 lttng_ust_abi_events_exit();
2376 lttng_perf_counter_exit();
2377 lttng_ust_ring_buffer_clients_exit();
2378 lttng_ust_counter_clients_exit();
2379 lttng_ust_statedump_destroy();
2380 lttng_ust_tp_exit();
2382 /* Reinitialize values for fork */
2383 sem_count
= sem_count_initial_value
;
2384 lttng_ust_comm_should_quit
= 0;
2390 void lttng_ust_exit(void)
2391 __attribute__((destructor
));
2393 void lttng_ust_exit(void)
2398 * Using pthread_cancel here because:
2399 * A) we don't want to hang application teardown.
2400 * B) the thread is not allocating any resource.
2404 * Require the communication thread to quit. Synchronize with
2405 * mutexes to ensure it is not in a mutex critical section when
2406 * pthread_cancel is later called.
2409 lttng_ust_comm_should_quit
= 1;
2412 pthread_mutex_lock(&ust_exit_mutex
);
2413 /* cancel threads */
2414 if (global_apps
.thread_active
) {
2415 ret
= pthread_cancel(global_apps
.ust_listener
);
2417 ERR("Error cancelling global ust listener thread: %s",
2420 global_apps
.thread_active
= 0;
2423 if (local_apps
.thread_active
) {
2424 ret
= pthread_cancel(local_apps
.ust_listener
);
2426 ERR("Error cancelling local ust listener thread: %s",
2429 local_apps
.thread_active
= 0;
2432 pthread_mutex_unlock(&ust_exit_mutex
);
2435 * Do NOT join threads: use of sys_futex makes it impossible to
2436 * join the threads without using async-cancel, but async-cancel
2437 * is delivered by a signal, which could hit the target thread
2438 * anywhere in its code path, including while the ust_lock() is
2439 * held, causing a deadlock for the other thread. Let the OS
2440 * cleanup the threads if there are stalled in a syscall.
2442 lttng_ust_cleanup(1);
2446 void ust_context_ns_reset(void)
2448 lttng_context_pid_ns_reset();
2449 lttng_context_cgroup_ns_reset();
2450 lttng_context_ipc_ns_reset();
2451 lttng_context_mnt_ns_reset();
2452 lttng_context_net_ns_reset();
2453 lttng_context_user_ns_reset();
2454 lttng_context_time_ns_reset();
2455 lttng_context_uts_ns_reset();
2459 void ust_context_vuids_reset(void)
2461 lttng_context_vuid_reset();
2462 lttng_context_veuid_reset();
2463 lttng_context_vsuid_reset();
2467 void ust_context_vgids_reset(void)
2469 lttng_context_vgid_reset();
2470 lttng_context_vegid_reset();
2471 lttng_context_vsgid_reset();
2475 * We exclude the worker threads across fork and clone (except
2476 * CLONE_VM), because these system calls only keep the forking thread
2477 * running in the child. Therefore, we don't want to call fork or clone
2478 * in the middle of an tracepoint or ust tracing state modification.
2479 * Holding this mutex protects these structures across fork and clone.
2481 void lttng_ust_before_fork(sigset_t
*save_sigset
)
2484 * Disable signals. This is to avoid that the child intervenes
2485 * before it is properly setup for tracing. It is safer to
2486 * disable all signals, because then we know we are not breaking
2487 * anything by restoring the original mask.
2492 /* Allocate lttng-ust TLS. */
2493 lttng_ust_alloc_tls();
2495 if (URCU_TLS(lttng_ust_nest_count
))
2497 /* Disable signals */
2498 sigfillset(&all_sigs
);
2499 ret
= sigprocmask(SIG_BLOCK
, &all_sigs
, save_sigset
);
2501 PERROR("sigprocmask");
2504 pthread_mutex_lock(&ust_fork_mutex
);
2507 lttng_ust_urcu_before_fork();
2508 lttng_ust_lock_fd_tracker();
2512 static void ust_after_fork_common(sigset_t
*restore_sigset
)
2516 DBG("process %d", getpid());
2517 lttng_perf_unlock();
2518 lttng_ust_unlock_fd_tracker();
2521 pthread_mutex_unlock(&ust_fork_mutex
);
2523 /* Restore signals */
2524 ret
= sigprocmask(SIG_SETMASK
, restore_sigset
, NULL
);
2526 PERROR("sigprocmask");
2530 void lttng_ust_after_fork_parent(sigset_t
*restore_sigset
)
2532 if (URCU_TLS(lttng_ust_nest_count
))
2534 DBG("process %d", getpid());
2535 lttng_ust_urcu_after_fork_parent();
2536 /* Release mutexes and re-enable signals */
2537 ust_after_fork_common(restore_sigset
);
2541 * After fork, in the child, we need to cleanup all the leftover state,
2542 * except the worker thread which already magically disappeared thanks
2543 * to the weird Linux fork semantics. After tyding up, we call
2544 * lttng_ust_ctor() again to start over as a new PID.
2546 * This is meant for forks() that have tracing in the child between the
2547 * fork and following exec call (if there is any).
2549 void lttng_ust_after_fork_child(sigset_t
*restore_sigset
)
2551 if (URCU_TLS(lttng_ust_nest_count
))
2553 lttng_context_vpid_reset();
2554 lttng_context_vtid_reset();
2555 lttng_ust_context_procname_reset();
2556 ust_context_ns_reset();
2557 ust_context_vuids_reset();
2558 ust_context_vgids_reset();
2559 DBG("process %d", getpid());
2560 /* Release urcu mutexes */
2561 lttng_ust_urcu_after_fork_child();
2562 lttng_ust_cleanup(0);
2563 /* Release mutexes and re-enable signals */
2564 ust_after_fork_common(restore_sigset
);
2568 void lttng_ust_after_setns(void)
2570 ust_context_ns_reset();
2571 ust_context_vuids_reset();
2572 ust_context_vgids_reset();
2575 void lttng_ust_after_unshare(void)
2577 ust_context_ns_reset();
2578 ust_context_vuids_reset();
2579 ust_context_vgids_reset();
2582 void lttng_ust_after_setuid(void)
2584 ust_context_vuids_reset();
2587 void lttng_ust_after_seteuid(void)
2589 ust_context_vuids_reset();
2592 void lttng_ust_after_setreuid(void)
2594 ust_context_vuids_reset();
2597 void lttng_ust_after_setresuid(void)
2599 ust_context_vuids_reset();
2602 void lttng_ust_after_setgid(void)
2604 ust_context_vgids_reset();
2607 void lttng_ust_after_setegid(void)
2609 ust_context_vgids_reset();
2612 void lttng_ust_after_setregid(void)
2614 ust_context_vgids_reset();
2617 void lttng_ust_after_setresgid(void)
2619 ust_context_vgids_reset();
2622 void lttng_ust_sockinfo_session_enabled(void *owner
)
2624 struct sock_info
*sock_info
= owner
;
2625 sock_info
->statedump_pending
= 1;