maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[asan/win] Use SRW locks to fix a race in BlockingMutex 

Summary:
Before my change, BlockingMutex used Windows critial sections. Critical
sections can only be initialized by calling InitializeCriticalSection,
dynamically.

The primary sanitizer allocator expects to be able to reinterpret zero
initialized memory as a BlockingMutex and immediately lock it.
RegionInfo contains a mutex, and it placement new is never called for
it. These objects are accessed via:
  RegionInfo *GetRegionInfo(uptr class_id) const {
    DCHECK_LT(class_id, kNumClasses);
    RegionInfo *regions = reinterpret_cast<RegionInfo *>(SpaceEnd());
    return &regions[class_id];
  }
The memory comes from the OS without any other initialization.

For various reasons described in the comments, BlockingMutex::Lock would
check if the object appeared to be zero-initialized, and it would lazily
call the LinkerInitialized constructor to initialize the critical
section. This pattern is obviously racy, and the code had a bunch of
FIXMEs about it.

The best fix here is to use slim reader writer locks, which can start
out zero-initialized. They are available starting in Windows Vista. I
think it's safe to go ahead and use them today.

Reviewers: kcc, vitalybuka

Subscribers: kubamracek, llvm-commits

Differential Revision: https://reviews.llvm.org/D49893

llvm-svn: 338331
This commit is contained in:
Reid Kleckner 2018-07-30 23:32:33 +00:00
parent a51403f5cc
commit a5ed43c1c9
2 changed files with 8 additions and 34 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
compiler-rt/lib/sanitizer_common/sanitizer_mutex.h
View File

@ -73,13 +73,8 @@ class SpinMutex : public StaticSpinMutex {
class BlockingMutex {
public:
#if SANITIZER_WINDOWS
// Windows does not currently support LinkerInitialized
explicit BlockingMutex(LinkerInitialized);
#else
explicit constexpr BlockingMutex(LinkerInitialized)
: opaque_storage_ {0, }, owner_(0) {}
#endif
: opaque_storage_ {0, }, owner_{0} {}
BlockingMutex();
void Lock();
void Unlock();

35
compiler-rt/lib/sanitizer_common/sanitizer_win.cc
View File

@ -767,43 +767,22 @@ void *internal_start_thread(void (*func)(void *arg), void *arg) { return 0; }
void internal_join_thread(void *th) { }
// ---------------------- BlockingMutex ---------------- {{{1
const uptr LOCK_UNINITIALIZED = 0;
const uptr LOCK_READY = (uptr)-1;
BlockingMutex::BlockingMutex(LinkerInitialized li) {
// FIXME: see comments in BlockingMutex::Lock() for the details.
CHECK(li == LINKER_INITIALIZED || owner_ == LOCK_UNINITIALIZED);
CHECK(sizeof(CRITICAL_SECTION) <= sizeof(opaque_storage_));
InitializeCriticalSection((LPCRITICAL_SECTION)opaque_storage_);
owner_ = LOCK_READY;
}
BlockingMutex::BlockingMutex() {
CHECK(sizeof(CRITICAL_SECTION) <= sizeof(opaque_storage_));
InitializeCriticalSection((LPCRITICAL_SECTION)opaque_storage_);
owner_ = LOCK_READY;
CHECK(sizeof(SRWLOCK) <= sizeof(opaque_storage_));
internal_memset(this, 0, sizeof(*this));
}
void BlockingMutex::Lock() {
if (owner_ == LOCK_UNINITIALIZED) {
// FIXME: hm, global BlockingMutex objects are not initialized?!?
// This might be a side effect of the clang+cl+link Frankenbuild...
new(this) BlockingMutex((LinkerInitialized)(LINKER_INITIALIZED + 1));
// FIXME: If it turns out the linker doesn't invoke our
// constructors, we should probably manually Lock/Unlock all the global
// locks while we're starting in one thread to avoid double-init races.
}
EnterCriticalSection((LPCRITICAL_SECTION)opaque_storage_);
CHECK_EQ(owner_, LOCK_READY);
AcquireSRWLockExclusive((PSRWLOCK)opaque_storage_);
CHECK_EQ(owner_, 0);
owner_ = GetThreadSelf();
}
void BlockingMutex::Unlock() {
CHECK_EQ(owner_, GetThreadSelf());
owner_ = LOCK_READY;
LeaveCriticalSection((LPCRITICAL_SECTION)opaque_storage_);
CheckLocked();
owner_ = 0;
ReleaseSRWLockExclusive((PSRWLOCK)opaque_storage_);
}
void BlockingMutex::CheckLocked() {