|
|
25845f |
commit 8f9450a0b7a9e78267e8ae1ab1000ebca08e473e
|
|
|
25845f |
Author: Torvald Riegel <triegel@redhat.com>
|
|
|
25845f |
Date: Sat Dec 24 00:40:46 2016 +0100
|
|
|
25845f |
|
|
|
25845f |
Add compiler barriers around modifications of the robust mutex list.
|
|
|
25845f |
|
|
|
25845f |
Any changes to the per-thread list of robust mutexes currently acquired as
|
|
|
25845f |
well as the pending-operations entry are not simply sequential code but
|
|
|
25845f |
basically concurrent with any actions taken by the kernel when it tries
|
|
|
25845f |
to clean up after a crash. This is not quite like multi-thread concurrency
|
|
|
25845f |
but more like signal-handler concurrency.
|
|
|
25845f |
This patch fixes latent bugs by adding compiler barriers where necessary so
|
|
|
25845f |
that it is ensured that the kernel crash handling sees consistent data.
|
|
|
25845f |
|
|
|
25845f |
This is meant to be easy to backport, so we do not use C11-style signal
|
|
|
25845f |
fences yet.
|
|
|
25845f |
|
|
|
25845f |
* nptl/descr.h (ENQUEUE_MUTEX_BOTH, DEQUEUE_MUTEX): Add compiler
|
|
|
25845f |
barriers and comments.
|
|
|
25845f |
* nptl/pthread_mutex_lock.c (__pthread_mutex_lock_full): Likewise.
|
|
|
25845f |
* nptl/pthread_mutex_timedlock.c (pthread_mutex_timedlock): Likewise.
|
|
|
25845f |
* nptl/pthread_mutex_unlock.c (__pthread_mutex_unlock_full): Likewise.
|
|
|
25845f |
|
|
|
25845f |
Index: glibc-2.17-c758a686/nptl/descr.h
|
|
|
25845f |
===================================================================
|
|
|
25845f |
--- glibc-2.17-c758a686.orig/nptl/descr.h
|
|
|
25845f |
+++ glibc-2.17-c758a686/nptl/descr.h
|
|
|
25845f |
@@ -180,7 +180,16 @@ struct pthread
|
|
|
25845f |
but the pointer to the next/previous element of the list points
|
|
|
25845f |
in the middle of the object, the __next element. Whenever
|
|
|
25845f |
casting to __pthread_list_t we need to adjust the pointer
|
|
|
25845f |
- first. */
|
|
|
25845f |
+ first.
|
|
|
25845f |
+ These operations are effectively concurrent code in that the thread
|
|
|
25845f |
+ can get killed at any point in time and the kernel takes over. Thus,
|
|
|
25845f |
+ the __next elements are a kind of concurrent list and we need to
|
|
|
25845f |
+ enforce using compiler barriers that the individual operations happen
|
|
|
25845f |
+ in such a way that the kernel always sees a consistent list. The
|
|
|
25845f |
+ backward links (ie, the __prev elements) are not used by the kernel.
|
|
|
25845f |
+ FIXME We should use relaxed MO atomic operations here and signal fences
|
|
|
25845f |
+ because this kind of concurrency is similar to synchronizing with a
|
|
|
25845f |
+ signal handler. */
|
|
|
25845f |
# define QUEUE_PTR_ADJUST (offsetof (__pthread_list_t, __next))
|
|
|
25845f |
|
|
|
25845f |
# define ENQUEUE_MUTEX_BOTH(mutex, val) \
|
|
|
25845f |
@@ -192,6 +201,8 @@ struct pthread
|
|
|
25845f |
mutex->__data.__list.__next = THREAD_GETMEM (THREAD_SELF, \
|
|
|
25845f |
robust_head.list); \
|
|
|
25845f |
mutex->__data.__list.__prev = (void *) &THREAD_SELF->robust_head; \
|
|
|
25845f |
+ /* Ensure that the new list entry is ready before we insert it. */ \
|
|
|
25845f |
+ __asm ("" ::: "memory"); \
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list, \
|
|
|
25845f |
(void *) (((uintptr_t) &mutex->__data.__list.__next) \
|
|
|
25845f |
| val)); \
|
|
|
25845f |
@@ -206,6 +217,9 @@ struct pthread
|
|
|
25845f |
((char *) (((uintptr_t) mutex->__data.__list.__prev) & ~1ul) \
|
|
|
25845f |
- QUEUE_PTR_ADJUST); \
|
|
|
25845f |
prev->__next = mutex->__data.__list.__next; \
|
|
|
25845f |
+ /* Ensure that we remove the entry from the list before we change the \
|
|
|
25845f |
+ __next pointer of the entry, which is read by the kernel. */ \
|
|
|
25845f |
+ __asm ("" ::: "memory"); \
|
|
|
25845f |
mutex->__data.__list.__prev = NULL; \
|
|
|
25845f |
mutex->__data.__list.__next = NULL; \
|
|
|
25845f |
} while (0)
|
|
|
25845f |
@@ -220,6 +234,8 @@ struct pthread
|
|
|
25845f |
do { \
|
|
|
25845f |
mutex->__data.__list.__next \
|
|
|
25845f |
= THREAD_GETMEM (THREAD_SELF, robust_list.__next); \
|
|
|
25845f |
+ /* Ensure that the new list entry is ready before we insert it. */ \
|
|
|
25845f |
+ __asm ("" ::: "memory"); \
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_list.__next, \
|
|
|
25845f |
(void *) (((uintptr_t) &mutex->__data.__list) | val)); \
|
|
|
25845f |
} while (0)
|
|
|
25845f |
@@ -240,6 +256,9 @@ struct pthread
|
|
|
25845f |
} \
|
|
|
25845f |
\
|
|
|
25845f |
runp->__next = next->__next; \
|
|
|
25845f |
+ /* Ensure that we remove the entry from the list before we change the \
|
|
|
25845f |
+ __next pointer of the entry, which is read by the kernel. */ \
|
|
|
25845f |
+ __asm ("" ::: "memory"); \
|
|
|
25845f |
mutex->__data.__list.__next = NULL; \
|
|
|
25845f |
} \
|
|
|
25845f |
} while (0)
|
|
|
25845f |
Index: glibc-2.17-c758a686/nptl/pthread_mutex_lock.c
|
|
|
25845f |
===================================================================
|
|
|
25845f |
--- glibc-2.17-c758a686.orig/nptl/pthread_mutex_lock.c
|
|
|
25845f |
+++ glibc-2.17-c758a686/nptl/pthread_mutex_lock.c
|
|
|
25845f |
@@ -181,6 +181,9 @@ __pthread_mutex_lock_full (pthread_mutex
|
|
|
25845f |
case PTHREAD_MUTEX_ROBUST_ADAPTIVE_NP:
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending,
|
|
|
25845f |
&mutex->__data.__list.__next);
|
|
|
25845f |
+ /* We need to set op_pending before starting the operation. Also
|
|
|
25845f |
+ see comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
|
|
|
25845f |
oldval = mutex->__data.__lock;
|
|
|
25845f |
/* This is set to FUTEX_WAITERS iff we might have shared the
|
|
|
25845f |
@@ -228,7 +231,12 @@ __pthread_mutex_lock_full (pthread_mutex
|
|
|
25845f |
/* But it is inconsistent unless marked otherwise. */
|
|
|
25845f |
mutex->__data.__owner = PTHREAD_MUTEX_INCONSISTENT;
|
|
|
25845f |
|
|
|
25845f |
+ /* We must not enqueue the mutex before we have acquired it.
|
|
|
25845f |
+ Also see comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
ENQUEUE_MUTEX (mutex);
|
|
|
25845f |
+ /* We need to clear op_pending after we enqueue the mutex. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
|
|
|
25845f |
/* Note that we deliberately exit here. If we fall
|
|
|
25845f |
@@ -250,6 +258,8 @@ __pthread_mutex_lock_full (pthread_mutex
|
|
|
25845f |
int kind = PTHREAD_MUTEX_TYPE (mutex);
|
|
|
25845f |
if (kind == PTHREAD_MUTEX_ROBUST_ERRORCHECK_NP)
|
|
|
25845f |
{
|
|
|
25845f |
+ /* We do not need to ensure ordering wrt another memory
|
|
|
25845f |
+ access. Also see comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending,
|
|
|
25845f |
NULL);
|
|
|
25845f |
return EDEADLK;
|
|
|
25845f |
@@ -257,6 +267,8 @@ __pthread_mutex_lock_full (pthread_mutex
|
|
|
25845f |
|
|
|
25845f |
if (kind == PTHREAD_MUTEX_ROBUST_RECURSIVE_NP)
|
|
|
25845f |
{
|
|
|
25845f |
+ /* We do not need to ensure ordering wrt another memory
|
|
|
25845f |
+ access. */
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending,
|
|
|
25845f |
NULL);
|
|
|
25845f |
|
|
|
25845f |
@@ -309,12 +321,19 @@ __pthread_mutex_lock_full (pthread_mutex
|
|
|
25845f |
mutex->__data.__count = 0;
|
|
|
25845f |
int private = PTHREAD_ROBUST_MUTEX_PSHARED (mutex);
|
|
|
25845f |
lll_unlock (mutex->__data.__lock, private);
|
|
|
25845f |
+ /* FIXME This violates the mutex destruction requirements. See
|
|
|
25845f |
+ __pthread_mutex_unlock_full. */
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
return ENOTRECOVERABLE;
|
|
|
25845f |
}
|
|
|
25845f |
|
|
|
25845f |
mutex->__data.__count = 1;
|
|
|
25845f |
+ /* We must not enqueue the mutex before we have acquired it.
|
|
|
25845f |
+ Also see comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
ENQUEUE_MUTEX (mutex);
|
|
|
25845f |
+ /* We need to clear op_pending after we enqueue the mutex. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
break;
|
|
|
25845f |
|
|
|
25845f |
@@ -331,10 +350,15 @@ __pthread_mutex_lock_full (pthread_mutex
|
|
|
25845f |
int robust = mutex->__data.__kind & PTHREAD_MUTEX_ROBUST_NORMAL_NP;
|
|
|
25845f |
|
|
|
25845f |
if (robust)
|
|
|
25845f |
- /* Note: robust PI futexes are signaled by setting bit 0. */
|
|
|
25845f |
- THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending,
|
|
|
25845f |
- (void *) (((uintptr_t) &mutex->__data.__list.__next)
|
|
|
25845f |
- | 1));
|
|
|
25845f |
+ {
|
|
|
25845f |
+ /* Note: robust PI futexes are signaled by setting bit 0. */
|
|
|
25845f |
+ THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending,
|
|
|
25845f |
+ (void *) (((uintptr_t) &mutex->__data.__list.__next)
|
|
|
25845f |
+ | 1));
|
|
|
25845f |
+ /* We need to set op_pending before starting the operation. Also
|
|
|
25845f |
+ see comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
+ }
|
|
|
25845f |
|
|
|
25845f |
oldval = mutex->__data.__lock;
|
|
|
25845f |
|
|
|
25845f |
@@ -343,12 +367,16 @@ __pthread_mutex_lock_full (pthread_mutex
|
|
|
25845f |
{
|
|
|
25845f |
if (kind == PTHREAD_MUTEX_ERRORCHECK_NP)
|
|
|
25845f |
{
|
|
|
25845f |
+ /* We do not need to ensure ordering wrt another memory
|
|
|
25845f |
+ access. */
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
return EDEADLK;
|
|
|
25845f |
}
|
|
|
25845f |
|
|
|
25845f |
if (kind == PTHREAD_MUTEX_RECURSIVE_NP)
|
|
|
25845f |
{
|
|
|
25845f |
+ /* We do not need to ensure ordering wrt another memory
|
|
|
25845f |
+ access. */
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
|
|
|
25845f |
/* Just bump the counter. */
|
|
|
25845f |
@@ -411,7 +439,12 @@ __pthread_mutex_lock_full (pthread_mutex
|
|
|
25845f |
/* But it is inconsistent unless marked otherwise. */
|
|
|
25845f |
mutex->__data.__owner = PTHREAD_MUTEX_INCONSISTENT;
|
|
|
25845f |
|
|
|
25845f |
+ /* We must not enqueue the mutex before we have acquired it.
|
|
|
25845f |
+ Also see comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
ENQUEUE_MUTEX_PI (mutex);
|
|
|
25845f |
+ /* We need to clear op_pending after we enqueue the mutex. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
|
|
|
25845f |
/* Note that we deliberately exit here. If we fall
|
|
|
25845f |
@@ -439,6 +472,8 @@ __pthread_mutex_lock_full (pthread_mutex
|
|
|
25845f |
PTHREAD_ROBUST_MUTEX_PSHARED (mutex)),
|
|
|
25845f |
0, 0);
|
|
|
25845f |
|
|
|
25845f |
+ /* To the kernel, this will be visible after the kernel has
|
|
|
25845f |
+ acquired the mutex in the syscall. */
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
return ENOTRECOVERABLE;
|
|
|
25845f |
}
|
|
|
25845f |
@@ -446,7 +481,12 @@ __pthread_mutex_lock_full (pthread_mutex
|
|
|
25845f |
mutex->__data.__count = 1;
|
|
|
25845f |
if (robust)
|
|
|
25845f |
{
|
|
|
25845f |
+ /* We must not enqueue the mutex before we have acquired it.
|
|
|
25845f |
+ Also see comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
ENQUEUE_MUTEX_PI (mutex);
|
|
|
25845f |
+ /* We need to clear op_pending after we enqueue the mutex. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
}
|
|
|
25845f |
}
|
|
|
25845f |
Index: glibc-2.17-c758a686/nptl/pthread_mutex_timedlock.c
|
|
|
25845f |
===================================================================
|
|
|
25845f |
--- glibc-2.17-c758a686.orig/nptl/pthread_mutex_timedlock.c
|
|
|
25845f |
+++ glibc-2.17-c758a686/nptl/pthread_mutex_timedlock.c
|
|
|
25845f |
@@ -140,6 +140,9 @@ pthread_mutex_timedlock (pthread_mutex_t
|
|
|
25845f |
case PTHREAD_MUTEX_ROBUST_ADAPTIVE_NP:
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending,
|
|
|
25845f |
&mutex->__data.__list.__next);
|
|
|
25845f |
+ /* We need to set op_pending before starting the operation. Also
|
|
|
25845f |
+ see comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
|
|
|
25845f |
oldval = mutex->__data.__lock;
|
|
|
25845f |
/* This is set to FUTEX_WAITERS iff we might have shared the
|
|
|
25845f |
@@ -177,7 +180,12 @@ pthread_mutex_timedlock (pthread_mutex_t
|
|
|
25845f |
/* But it is inconsistent unless marked otherwise. */
|
|
|
25845f |
mutex->__data.__owner = PTHREAD_MUTEX_INCONSISTENT;
|
|
|
25845f |
|
|
|
25845f |
+ /* We must not enqueue the mutex before we have acquired it.
|
|
|
25845f |
+ Also see comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
ENQUEUE_MUTEX (mutex);
|
|
|
25845f |
+ /* We need to clear op_pending after we enqueue the mutex. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
|
|
|
25845f |
/* Note that we deliberately exit here. If we fall
|
|
|
25845f |
@@ -193,6 +201,8 @@ pthread_mutex_timedlock (pthread_mutex_t
|
|
|
25845f |
int kind = PTHREAD_MUTEX_TYPE (mutex);
|
|
|
25845f |
if (kind == PTHREAD_MUTEX_ROBUST_ERRORCHECK_NP)
|
|
|
25845f |
{
|
|
|
25845f |
+ /* We do not need to ensure ordering wrt another memory
|
|
|
25845f |
+ access. Also see comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending,
|
|
|
25845f |
NULL);
|
|
|
25845f |
return EDEADLK;
|
|
|
25845f |
@@ -200,6 +210,8 @@ pthread_mutex_timedlock (pthread_mutex_t
|
|
|
25845f |
|
|
|
25845f |
if (kind == PTHREAD_MUTEX_ROBUST_RECURSIVE_NP)
|
|
|
25845f |
{
|
|
|
25845f |
+ /* We do not need to ensure ordering wrt another memory
|
|
|
25845f |
+ access. */
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending,
|
|
|
25845f |
NULL);
|
|
|
25845f |
|
|
|
25845f |
@@ -294,12 +306,19 @@ pthread_mutex_timedlock (pthread_mutex_t
|
|
|
25845f |
mutex->__data.__count = 0;
|
|
|
25845f |
int private = PTHREAD_ROBUST_MUTEX_PSHARED (mutex);
|
|
|
25845f |
lll_unlock (mutex->__data.__lock, private);
|
|
|
25845f |
+ /* FIXME This violates the mutex destruction requirements. See
|
|
|
25845f |
+ __pthread_mutex_unlock_full. */
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
return ENOTRECOVERABLE;
|
|
|
25845f |
}
|
|
|
25845f |
|
|
|
25845f |
mutex->__data.__count = 1;
|
|
|
25845f |
+ /* We must not enqueue the mutex before we have acquired it.
|
|
|
25845f |
+ Also see comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
ENQUEUE_MUTEX (mutex);
|
|
|
25845f |
+ /* We need to clear op_pending after we enqueue the mutex. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
break;
|
|
|
25845f |
|
|
|
25845f |
@@ -316,10 +335,15 @@ pthread_mutex_timedlock (pthread_mutex_t
|
|
|
25845f |
int robust = mutex->__data.__kind & PTHREAD_MUTEX_ROBUST_NORMAL_NP;
|
|
|
25845f |
|
|
|
25845f |
if (robust)
|
|
|
25845f |
- /* Note: robust PI futexes are signaled by setting bit 0. */
|
|
|
25845f |
- THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending,
|
|
|
25845f |
- (void *) (((uintptr_t) &mutex->__data.__list.__next)
|
|
|
25845f |
- | 1));
|
|
|
25845f |
+ {
|
|
|
25845f |
+ /* Note: robust PI futexes are signaled by setting bit 0. */
|
|
|
25845f |
+ THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending,
|
|
|
25845f |
+ (void *) (((uintptr_t) &mutex->__data.__list.__next)
|
|
|
25845f |
+ | 1));
|
|
|
25845f |
+ /* We need to set op_pending before starting the operation. Also
|
|
|
25845f |
+ see comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
+ }
|
|
|
25845f |
|
|
|
25845f |
oldval = mutex->__data.__lock;
|
|
|
25845f |
|
|
|
25845f |
@@ -328,12 +352,16 @@ pthread_mutex_timedlock (pthread_mutex_t
|
|
|
25845f |
{
|
|
|
25845f |
if (kind == PTHREAD_MUTEX_ERRORCHECK_NP)
|
|
|
25845f |
{
|
|
|
25845f |
+ /* We do not need to ensure ordering wrt another memory
|
|
|
25845f |
+ access. */
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
return EDEADLK;
|
|
|
25845f |
}
|
|
|
25845f |
|
|
|
25845f |
if (kind == PTHREAD_MUTEX_RECURSIVE_NP)
|
|
|
25845f |
{
|
|
|
25845f |
+ /* We do not need to ensure ordering wrt another memory
|
|
|
25845f |
+ access. */
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
|
|
|
25845f |
/* Just bump the counter. */
|
|
|
25845f |
@@ -420,7 +448,12 @@ pthread_mutex_timedlock (pthread_mutex_t
|
|
|
25845f |
/* But it is inconsistent unless marked otherwise. */
|
|
|
25845f |
mutex->__data.__owner = PTHREAD_MUTEX_INCONSISTENT;
|
|
|
25845f |
|
|
|
25845f |
+ /* We must not enqueue the mutex before we have acquired it.
|
|
|
25845f |
+ Also see comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
ENQUEUE_MUTEX_PI (mutex);
|
|
|
25845f |
+ /* We need to clear op_pending after we enqueue the mutex. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
|
|
|
25845f |
/* Note that we deliberately exit here. If we fall
|
|
|
25845f |
@@ -443,6 +476,8 @@ pthread_mutex_timedlock (pthread_mutex_t
|
|
|
25845f |
PTHREAD_ROBUST_MUTEX_PSHARED (mutex)),
|
|
|
25845f |
0, 0);
|
|
|
25845f |
|
|
|
25845f |
+ /* To the kernel, this will be visible after the kernel has
|
|
|
25845f |
+ acquired the mutex in the syscall. */
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
return ENOTRECOVERABLE;
|
|
|
25845f |
}
|
|
|
25845f |
@@ -450,7 +485,12 @@ pthread_mutex_timedlock (pthread_mutex_t
|
|
|
25845f |
mutex->__data.__count = 1;
|
|
|
25845f |
if (robust)
|
|
|
25845f |
{
|
|
|
25845f |
+ /* We must not enqueue the mutex before we have acquired it.
|
|
|
25845f |
+ Also see comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
ENQUEUE_MUTEX_PI (mutex);
|
|
|
25845f |
+ /* We need to clear op_pending after we enqueue the mutex. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
}
|
|
|
25845f |
}
|
|
|
25845f |
Index: glibc-2.17-c758a686/nptl/pthread_mutex_unlock.c
|
|
|
25845f |
===================================================================
|
|
|
25845f |
--- glibc-2.17-c758a686.orig/nptl/pthread_mutex_unlock.c
|
|
|
25845f |
+++ glibc-2.17-c758a686/nptl/pthread_mutex_unlock.c
|
|
|
25845f |
@@ -143,6 +143,9 @@ __pthread_mutex_unlock_full (pthread_mut
|
|
|
25845f |
/* Remove mutex from the list. */
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending,
|
|
|
25845f |
&mutex->__data.__list.__next);
|
|
|
25845f |
+ /* We must set op_pending before we dequeue the mutex. Also see
|
|
|
25845f |
+ comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
DEQUEUE_MUTEX (mutex);
|
|
|
25845f |
|
|
|
25845f |
mutex->__data.__owner = newowner;
|
|
|
25845f |
@@ -159,6 +162,14 @@ __pthread_mutex_unlock_full (pthread_mut
|
|
|
25845f |
& FUTEX_WAITERS) != 0))
|
|
|
25845f |
lll_futex_wake (&mutex->__data.__lock, 1, private);
|
|
|
25845f |
|
|
|
25845f |
+ /* We must clear op_pending after we release the mutex.
|
|
|
25845f |
+ FIXME However, this violates the mutex destruction requirements
|
|
|
25845f |
+ because another thread could acquire the mutex, destroy it, and
|
|
|
25845f |
+ reuse the memory for something else; then, if this thread crashes,
|
|
|
25845f |
+ and the memory happens to have a value equal to the TID, the kernel
|
|
|
25845f |
+ will believe it is still related to the mutex (which has been
|
|
|
25845f |
+ destroyed already) and will modify some other random object. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
break;
|
|
|
25845f |
|
|
|
25845f |
@@ -223,6 +234,9 @@ __pthread_mutex_unlock_full (pthread_mut
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending,
|
|
|
25845f |
(void *) (((uintptr_t) &mutex->__data.__list.__next)
|
|
|
25845f |
| 1));
|
|
|
25845f |
+ /* We must set op_pending before we dequeue the mutex. Also see
|
|
|
25845f |
+ comments at ENQUEUE_MUTEX. */
|
|
|
25845f |
+ __asm ("" ::: "memory");
|
|
|
25845f |
DEQUEUE_MUTEX (mutex);
|
|
|
25845f |
}
|
|
|
25845f |
|
|
|
25845f |
@@ -247,6 +261,9 @@ __pthread_mutex_unlock_full (pthread_mut
|
|
|
25845f |
__lll_private_flag (FUTEX_UNLOCK_PI, private));
|
|
|
25845f |
}
|
|
|
25845f |
|
|
|
25845f |
+ /* This happens after the kernel releases the mutex but violates the
|
|
|
25845f |
+ mutex destruction requirements; see comments in the code handling
|
|
|
25845f |
+ PTHREAD_MUTEX_ROBUST_NORMAL_NP. */
|
|
|
25845f |
THREAD_SETMEM (THREAD_SELF, robust_head.list_op_pending, NULL);
|
|
|
25845f |
break;
|
|
|
25845f |
|