We have encountered a number of Windows kernel crashes in the win32k.sys driver while processing corrupted TTF font files. The most frequent one occurring for the bug reported here is as follows: --- PAGE_FAULT_IN_NONPAGED_AREA (50) Invalid system memory was referenced. This cannot be protected by try-except, it must be protected by a Probe. Typically the address is just plain bad or it is pointing at freed memory. Arguments: Arg1: 8273777f, memory referenced. Arg2: 00000000, value 0 = read operation, 1 = write operation. Arg3: 919c279f, If non-zero, the instruction address which referenced the bad memory address. Arg4: 00000000, (reserved) Debugging Details: ------------------ FAULTING_IP: win32k!bGeneratePath+60 919c279f 8b0f mov ecx,dword ptr [edi] MM_INTERNAL_CODE: 0 DEFAULT_BUCKET_ID: WIN7_DRIVER_FAULT BUGCHECK_STR: 0x50 PROCESS_NAME: csrss.exe CURRENT_IRQL: 0 ANALYSIS_VERSION: 6.3.9600.17237 (debuggers(dbg).140716-0327) x86fre LAST_CONTROL_TRANSFER: from 91a9b6af to 919c279f STACK_TEXT: 99ee4a14 91a9b6af 00000000 000003e0 00000010 win32k!bGeneratePath+0x60 99ee4a40 91a9a105 fbc62cf0 00000005 faebeda0 win32k!ttfdQueryTrueTypeOutline+0x79 99ee4a90 91a82fef 00000000 fbc62cf0 00000005 win32k!ttfdSemQueryTrueTypeOutline+0x45 99ee4ad8 91a65175 00000000 fbc62cf0 00000005 win32k!PDEVOBJ::QueryTrueTypeOutline+0x3e 99ee4b90 91a5cd60 fbc62cf0 fbc62cf0 00000003 win32k!GreGetGlyphOutlineInternal+0x4f5 99ee4c0c 8286c87a 2801007e 0000003b 00000003 win32k!NtGdiGetGlyphOutline+0x95 99ee4c0c 770570b4 2801007e 0000003b 00000003 nt!KiFastCallEntry+0x12a WARNING: Frame IP not in any known module. Following frames may be wrong. 002df760 00000000 00000000 00000000 00000000 0x770570b4 --- We have observed the invalid memory addresses accessed by the win32k!bGeneratePath function to be seemingly "wild", e.g. 0x8273777f, 0xe9849de5, 0xc7617bc7, 0xf2edc7eb etc. The above crash dump comes from an old version of Windows 7 32-bit, because symbols for win32k.sys on the latest build are currently unavailable on the Microsoft Symbol Server. Nevertheless, a crash summary from an up-to-date system is as follows: --- cut --- PAGE_FAULT_IN_NONPAGED_AREA (50) Invalid system memory was referenced. This cannot be protected by try-except, it must be protected by a Probe. Typically the address is just plain bad or it is pointing at freed memory. Arguments: Arg1: 8128f57d, memory referenced. Arg2: 00000000, value 0 = read operation, 1 = write operation. Arg3: 925375f6, If non-zero, the instruction address which referenced the bad memory address. Arg4: 00000000, (reserved) Debugging Details: ------------------ FAULTING_IP: win32k!PATHOBJ_bCloseFigure+76 925375f6 8b0f mov ecx,dword ptr [edi] MM_INTERNAL_CODE: 0 DEFAULT_BUCKET_ID: WIN7_DRIVER_FAULT BUGCHECK_STR: 0x50 PROCESS_NAME: csrss.exe CURRENT_IRQL: 0 ANALYSIS_VERSION: 6.3.9600.17237 (debuggers(dbg).140716-0327) x86fre LAST_CONTROL_TRANSFER: from 9261b9c8 to 925375f6 STACK_TEXT: WARNING: Stack unwind information not available. Following frames may be wrong. 89277a10 9261b9c8 00000000 00000150 00000010 win32k!PATHOBJ_bCloseFigure+0x76 89277a3c 9261a316 fbb26cf0 0000000c fba36f38 win32k!XLATEOBJ_hGetColorTransform+0x423bf 89277a8c 926019b4 00000000 fbb26cf0 0000000c win32k!XLATEOBJ_hGetColorTransform+0x40d0d 89277ad4 925e33e5 00000000 fbb26cf0 0000000c win32k!XLATEOBJ_hGetColorTransform+0x283ab 89277b90 925dafcc fbb26cf0 fbb26cf0 00000003 win32k!XLATEOBJ_hGetColorTransform+0x9ddc 89277c0c 82837986 2201061c 00000029 00000003 win32k!XLATEOBJ_hGetColorTransform+0x19c3 89277c0c 772b6c74 2201061c 00000029 00000003 nt!KiSystemServicePostCall 0019f608 00000000 00000000 00000000 00000000 0x772b6c74 --- cut --- While the above crashes are the most common ones, we have also encountered bugchecks (likely caused by the same issue) at the following other locations on old Windows 7 32-bit: --- win32k!vQsplineToPolyBezier+43 91522614 8b4608 mov eax,dword ptr [esi+8] --- win32k!vQsplineToPolyBezier+83 92292654 8941fc mov dword ptr [ecx-4],eax --- ... and on latest Windows 7 32-bit: --- win32k!EngDeleteRgn+3293 91e0747c 8b460c mov eax,dword ptr [esi+0Ch] --- The crash in win32k!vQsplineToPolyBezier+83 strongly suggests that the failures are caused or may lead to memory corruption, and consequently to arbitrary code execution. While we have not determined the specific root cause of the vulnerability, we have pinpointed the offending mutations to reside in the "fpgm" table. In case of the few samples we have examined, the problem seems to stem from changing one of the instructions in the FPGM program to "FLIPPT". The issue reproduces on Windows 7 (other platforms unchecked). It is easiest to reproduce with Special Pools enabled for win32k.sys (leading to an immediate crash when the bug is triggered), but it it also possible to observe a system crash on a default Windows installation. In order to reproduce the problem with the provided samples, it is necessary to use a custom program which calls the GetGlyphOutline() API with various parameters over all of the font's glyphs. Attached is an archive with several proof-of-concept mutated TTF files.