From: trik-news at gmx.de (Spiro Trikaliotis)
Subject: Win32 Device Drivers Communication Vulnerabilities + PoC for Symantec Norton AntiVirus '2002 (probably all versions)  Device Driver

Hello,

On Fri, Aug 02, 2002 at 10:39:44AM +0200, [SEC-LABS TEAM]: wrote:
> 
> The Sec-Labs security research group found a bug in Win32 Device Drivers Communication,  the white-paper for this vulnerability can
> be viewed at http://sec-labs.hack.pl  , the exploit code for Symantec Norton AntiVirus '2002 (probably all versions) Device Driver  is also stored at our homepage. 
> Enjoy.

from the posting at the Full-Disclosure Mailing list, I found the article on
http://sec-labs.hack.pl/papers/win32ddc.php.

In this article, you write:

   "As far as I know, Windows system is not checking if the memory location
   (agruments/memory pointers) is valid, and it isn't a windows vulerability?
    - give me a break!"

For this, I want to comment the following:

Each IRP on NT/2K/XP has a method argument, that is, the last 3 bits
indicate which access method is used.

The DDK defines the following types (see devioctl.h):

#define METHOD_BUFFERED                 0
#define METHOD_IN_DIRECT                1
#define METHOD_OUT_DIRECT               2
#define METHOD_NEITHER                  3

The IRP you used for this proof of concept (222A87h) obviously has 
METHOD_NEITHER. The DDK clearly states what these methods mean:
(Windows DDK/Kernel-Mode Driver Architecture/Reference/System-Defined I/O
Function Codes/IRP Function Codes and IOCTRL/Defining I/O Control Codes):

  "METHOD_BUFFERED, if the driver transfers small amounts of data for the
  request 
  With this method, IRPs containing the I/O control code will supply a pointer
  to the buffer into which or from which to transfer data at
  Irp->AssociatedIrp.SystemBuffer. Most I/O control codes for device and
  intermediate drivers use this TransferType value. 

  METHOD_IN_DIRECT, if the underlying device driver will read a large amount
  of data for the request using DMA or PIO and must transfer the data quickly 
  With this method, IRPs containing the I/O control code will supply a pointer
  to an MDL, describing the output buffer at Irp->MdlAddress. 

  METHOD_OUT_DIRECT, if the underlying device driver will write a large amount
  of data to the device for the request using DMA or PIO and must transfer the
  data quickly 
  With this method, IRPs containing the I/O control code will supply a pointer
  to an MDL, describing the data buffer, at Irp->MdlAddress. 

  METHOD_NEITHER, if the driver can be sent such a request only while it is
  running in the context of the thread that originates the I/O control request 
  Only a highest-level kernel-mode driver is guaranteed to meet this
  condition, so this value is seldom used for the I/O control codes passed to
  device drivers. With this method, the highest-level driver must determine
  whether to set up buffered or direct access to user data on receipt of the
  request, possibly must lock down the user buffer, and must wrap its access
  to the user buffer in a structured exception handler. Otherwise, the
  originating user-mode caller might change the buffered data out from under
  the driver or the caller could be swapped out just as the driver is
  accessing the user buffer."

There are even more explicit texts.

In fact, METHOD_NEITHER means literally what the name means: Windows does
not touch or check the buffer pointer or the buffer itself in any way, does 
not perform any checks nor anything else. All text books I've read so far 
even advice not to use METHOD_NEITHER unless one knows exactly what he is 
doing - something the driver writer obviously did not in your example.

The DDK even contains the following statement
(Windows DDK/Kernel-Mode Driver Architecture/Design-Guide/Driver Programming
Techniques/Driver Reliability Issues/Errors in Referencing User-Space
Addresses):

  Errors in Referencing User-Space Addresses
  A driver should validate any address in user space before trying to use it.
  The I/O Manager does not validate such addresses, nor does it validate
  pointers that are embedded in buffers passed to drivers.

  Failure to Validate Addresses Passed in Type3 (METHOD_NEITHER) IOCTLs and
  FSCTLs
  The I/O Manager does no validation whatsoever for METHOD_NEITHER IOCTLs and
  FSCTLs. To ensure that user-space addresses are valid, the driver must use
  the ProbeForRead and ProbeForWrite routines, enclosing all buffer references
  in try/except blocks. 


So, unless you find another IOCTL with another method, I think it is not
correct to call this a windows vulerability. The driver writer *explicitly*
chose to shut down any tests Windows makes, but did not handle this case
correctly. You can blame Microsoft for very much, but how can you blame 
Microsoft for this?

Curious,
   Spiro.

Powered by blists - more mailing lists