The following crash was observed in Microsoft Office 2007 Excel with Microsoft
Office File Validation Add-In disabled and Application Verifier enabled for
testing and reproduction. This bug did not reproduce in Office 2010 or 2013.
Attached files:
Original File: 3013413838_orig.xls
Crashing File: 3013413838_crash.xls
Minimized Crashing File: 3013413838_min.xls
The minimized crashing file shows a one bit delta from the original file at
offset 0x139F. OffVis did not reveal anything unique about this offset in the
minimized file.
File Versions:
Excel.exe: 12.0.6718.5000
OGL.dll: 12.0.6719.5000
oart.dll: 12.0.6683.5002
GDI32.dll: 5.2.3790.5563
Observed Crash:
This crashing eip was observed 4 times in fuzzing results with various invalid
memory address being dereferenced.
eax=8a94e1a1 ebx=00000000 ecx=10a80598 edx=8a94e1a0 esi=0013d478 edi=0013d42c
eip=3bd18f75 esp=0013d3dc ebp=0013d3e0 iopl=0 nv up ei pl nz na pe nc
cs=001b ss=0023 ds=0023 es=0023 fs=003b gs=0000 efl=00010206
OGL!ScanOperation::Convert_24_sRGB:
3bd18f68 55 push ebp
3bd18f69 8bec mov ebp,esp
3bd18f6b 837d0800 cmp dword ptr [ebp+8],0
3bd18f6f 7431 je OGL!ScanOperation::Convert_24_sRGB+0x3a
(3bd18fa2)
3bd18f71 8d4201 lea eax,[edx+1]
3bd18f74 56 push esi
=> 3bd18f75 0fb65001 movzx edx,byte ptr [eax+1]
ds:0023:8a94e1a2=??
0:000> kb L8
ChildEBP RetAddr Args to Child
0013d3e0 3be703b3 0000666f 0013d42c 00000000
OGL!ScanOperation::Convert_24_sRGB+0xd
0013d3fc 3be18f32 00000000 8a94e1a0 0000666f OGL!EpAlphaBlender::Blend+0x55
0013d568 3bd9f6c1 0013d894 00000000 0013d58c OGL!ConvertBitmapData+0x61
0013d5a4 3bde4137 00000000 00000001 000e200b
OGL!GpMemoryBitmap::InternalLockBits+0x105
0013d5d0 3bdfa09b 05492fa8 0013d5f8 00000001 OGL!GpMemoryBitmap::LockBits+0xba
0013d608 3bdfac0c 0013d7bc 0013d894 0013d62c
OGL!CopyOnWriteBitmap::PipeLockBitsFromMemory+0xb8
0013d6e8 3bd2b7e7 0013d7bc 0013d894 0013d7d0
OGL!CopyOnWriteBitmap::PipeLockBits+0x553
0013d700 3be4cc56 0013d7bc 0013d894 00000001 OGL!GpBitmap::PipeLockBits+0x4e
The function OGL!ScanOperation::Convert_24_sRGB was called with edx pointing to
an invalid memory location: 0x8a94e1a0. Tracing back we can find that the heap
address where edx came from was allocated with the following call stack:
3be70fe2 OGL!GpMalloc+0x00000014
3bd58669 OGL!CopyOnWriteBitmap::CopyOnWriteBitmap+0x00000049
3be0517e OGL!CopyOnWriteBitmap::Create+0x00000021
3be0514d OGL!GpBitmap::GpBitmap+0x00000030
The edx value was copied in from the stack at the following location
OGL!GpMemoryBitmap::InitMemoryBitmap():
3bd4f6f0 8b45fc mov eax,dword ptr [ebp-4]
3bd4f6f3 6a06 push 6
3bd4f6f5 59 pop ecx
3bd4f6f6 8bf3 mov esi,ebx
=>3bd4f6f8 f3a5 rep movs dword ptr es:[edi],dword ptr [esi]
The stack location was set with the invalid value here in
OGL!ValidateBitmapInfo():
3bda46ed 8b4d08 mov ecx,dword ptr [ebp+8]
3bda46f0 895804 mov dword ptr [eax+4],ebx
3bda46f3 895008 mov dword ptr [eax+8],edx
3bda46f6 89480c mov dword ptr [eax+0Ch],ecx
=> 3bda46f9 897810 mov dword ptr [eax+10h],edi
Edi is set earlier as the result of an imul instruction that is then added to a
base heap pointer:
.text:3BDA46CB lea edi, [ebx-1]
.text:3BDA46CE imul edi, edx
.text:3BDA46D1 add edi, [ebp+arg_4] ; bad value here
With this PoC edi=0x0000666e and edx=0x00013350. The edx value is calculated
earlier in the same function. If 0xf9ef540 is the base pointer (arg_4) we end
up setting this value to be 0x666e*0x13350+0xf9ef540 or 0x8a94e1a0 as we saw in
the initial bad memory access. The heap chunk referenced at 0xf9ef540 has an
original allocation size of 15156 and we've set our pointer far out of bounds
of this allocation range.
There is a distinct lack of overflow checks and bounds checking in the
OGL!ValidateBitmapInfo function that may lead to memory corruption when doing
bitmap conversion later on in the code. For example, if the 0x13350 value is
able to grow to 0x27fd0 we can set the edi value to be 0xffffcb60 (0x666e *
0x27fd0 = 0xffffcb60) which leads to an out of bound write instead of an out of
bound read later in the code.
This bug is subject to a 90 day disclosure deadline. If 90 days elapse without
a broadly available patch, then the bug report will automatically become
visible to the public.
Original issue reported on code.google.com by scvi...@google.com on 30 Jun 2015 at 4:31
Original issue reported on code.google.com by
scvi...@google.com
on 30 Jun 2015 at 4:31Attachments: