View source for CVE-2014-1773 Internet Explorer內存損壞漏洞

==INFO==
<pre>
CClipStack::PushClipRect() Invalid Array Indexing

First, starting with the callstack from the crash with pageheap enabled:

ChildEBP RetAddr 
0919aac8 6b58459f MSHTML!CWorldTransform::IsAxisAligned
0919ab40 6ba228e9 MSHTML!CDispSurface::CClipStack::PushClipRect+0x1a2
0919aba4 6be25b58 MSHTML!CDispSurface::PushClipRectInternal+0x2f
0919abc0 6bf1118f MSHTML!CDispSurface::PushClipRectUser+0x26
0919ac24 6bf1157e MSHTML!CCanvasCompositor::ClearRightAndBelowRenderedRegion+0xcd
0919acd8 6b28108c MSHTML!CCanvasCompositor::ExecuteCompositionEffects+0x34d
0919ace0 6b2802f4 MSHTML!CCanvasCompositor::Flush+0x3d
0919ace8 6bf0c867 MSHTML!CCanvasCompositor::~CCanvasCompositor+0x10
0919ae28 6bf0b668 MSHTML!CCanvasRenderingContext2D::StrokeRectInternal+0x1b3
0919ae70 6bf0e032 MSHTML!CCanvasRenderingContext2D::ExecuteStrokeRect+0x1c5
0919aebc 6bd74b6f MSHTML!CCanvasRenderingContext2D::Var_strokeRect+0xac
0919aee0 6ae7056e MSHTML!CFastDOM::CCanvasRenderingContext2D::Trampoline_strokeRect+0x3b
0919af50 6ae6cdda jscript9!Js::JavascriptExternalFunction::ExternalFunctionThunk+0x165
0919b328 6ae6dc86 jscript9!Js::InterpreterStackFrame::Process+0x1e74
0919b474 09560fd9 jscript9!Js::InterpreterStackFrame::InterpreterThunk<1>+0x1e7

If we step back one function, this guy sets the object pointer that is used by the crash:

.text:63CDE575 ; public: long __thiscall CDispSurface::CClipStack::PushClipRect(class CRectF const &, class CWorldTransform const *, bool, bool)
.text:63CDE575 mov edi, edi
.text:63CDE577 push ebp
.text:63CDE578 mov ebp, esp
.text:63CDE57A sub esp, 64h
.text:63CDE57D and [ebp+var_8], 0
.text:63CDE581 mov edx, ecx
.text:63CDE583 push ebx
.text:63CDE584 push esi
.text:63CDE585 mov esi, [ebp+arg_0]
.text:63CDE588 push edi
.text:63CDE589 lea edi, [ebp+var_28]
.text:63CDE58C mov [ebp+var_4], edx
.text:63CDE58F movsd
.text:63CDE590 movsd
.text:63CDE591 movsd
.text:63CDE592 movsd
.text:63CDE593 mov esi, [ebp+arg_4]
.text:63CDE596 test esi, esi
.text:63CDE598 jnz loc_63B1A8DF
.text:63CDE59E
.text:63CDE59E loc_63CDE59E:
.text:63CDE59E imul ecx, [edx+4], 18h
.text:63CDE5A2 xor bl, bl
.text:63CDE5A4 mov eax, [edx+8] 
.text:63CDE5A7 add eax, 0FFFFFFE8h
.text:63CDE5AA add eax, ecx
.text:63CDE5AC mov [ebp+arg_0], eax
.text:63CDE5AF mov edi, [eax+14h]

The outer object, CDispSurface, contains a CClipStack object inside of it. The CClipStack
object starts at offset 0x64 into the CDispSurface object. CClipStack is basically just
a subclass of CImplAry it seems, which is a generic array class that Internet Explorer
uses all over the place for arrays. In the above function, CDispSurface::CClipStack::PushClipRect,
the 'this' pointer is the subobject CClipStack inside of the outer CDispSurface.

The CClipStack object looks something liek this:

DWORD dwMaxElems;
DWORD dwCurElems;
VOID *pElems;

So, looking at the above code, you can guess what is going on. What the above code is doing is
extracting the last element from the CClipStack array. Each element in the array is 0x18 bytes in size,
hence:

imul ecx, [edx+4], 18h ; dwCurElems * 0x18
mov eax, [edx+8] ;pElems

But then you can see the following:

add eax, 0FFFFFFE8h ; subtract 0x18 from pElems
add eax, ecx ; pElems += (dwCurElems * 0x18)

However, if the array is empty, dwCurElems * 0x18 '==' 0, and so pElems
will actually point BEHIND the start of the array, into the adjacent heap chunk
below it. The array itself contains objects of the type CWorldTransform, which you can
infer from the code that crashes shortly after extract the object from the array. ecx
is the object pointer taken from the 'mov edi, [eax+14h]' operation:

.text:63B1A940 ; public: bool __thiscall CWorldTransform::IsAxisAligned(void)const

.text:63B1A940 test dword ptr [ecx+8], 80000000h ;ecx == BAD

So, where is the array allocated:

MSHTML!CDispSurface::CClipStack::PushClipRect+0x32:
6b97e5a7 83c0e8 add eax,0FFFFFFE8h
0:013> !heap -p -a eax
address 0c4b0fa0 found in
_DPH_HEAP_ROOT @ 211000
in busy allocation ( DPH_HEAP_BLOCK: UserAddr UserSize - VirtAddr VirtSize)
c45123c: c4b0fa0 60 - c4b0000 2000
739f8e89 verifier!AVrfDebugPageHeapAllocate+0x00000229
77a95e7a ntdll!RtlDebugAllocateHeap+0x00000030
77a5a3ba ntdll!RtlpAllocateHeap+0x000000c4
77a25a70 ntdll!RtlAllocateHeap+0x0000023a
6b55c592 MSHTML!CImplAry::EnsureSizeWorker+0x00000061
6b584513 MSHTML!CDispSurface::BeginDraw+0x00000122
6b2847a0 MSHTML!CCanvasRenderingContext2D::BeginDraw+0x00000041
6b286155 MSHTML!CCanvasContextBase::OpenBitmapRenderTarget+0x00000014
6b283b77 MSHTML!CCanvasCompositor::Initialize+0x0000102f
6b285cf7 MSHTML!CCanvasRenderingContext2D::StrokeGeometry+0x00000131
6b285090 MSHTML!CCanvasRenderingContext2D::ExecuteStroke+0x000002c4
6b284dae MSHTML!CFastDOM::CCanvasRenderingContext2D::Trampoline_stroke+0x00000035
6ae7056e jscript9!Js::JavascriptExternalFunction::ExternalFunctionThunk+0x00000165
6ae6cdda jscript9!Js::InterpreterStackFrame::Process+0x00001e74
6ae6dc86 jscript9!Js::InterpreterStackFrame::InterpreterThunk<1>+0x000001e7

It is in the 0x60 LFH bin. We can control the chunk behind the array by doing some heap crafting.
But the bad part of that code above is that the calculation for extracting the object used is
this:

(BYTE *)pArrayStart - 0x18 + 0x14 => (BYTE *)pArrayStart - 0x4

That results in using the flags/index field of the LFH chunk header as an object pointer.
That's not really good as the bytes in that field are not super controllable (I think).



For seeing how to possibly control that second dword of the LFH header:

http://illmatics.com/Understanding_the_LFH.pdf


Collaboration with Sean Larsson
</pre>