Dailydave mailing list archives

Re: A 3 a.m. Riddle


From: Nicolas Waisman <nicolas.waisman () immunitysec com>
Date: Wed, 30 May 2007 16:13:14 -0500

Sweet! You got it. Theoretically you have only one inc [edi], and the
idea was to make it:
  o generic (works on every server)
  o and non addy dependant (not based on any SP)
At least the used technique.

 So my approach was to point edi to the FreeListInUse Bitmask on the
Heap.
 The FreeListInUse bitmask is a 4 dword bitmask which every bit
correspond to an entry of the FreeList double linked list array. So if
the bit 4 is setted , that means that at that entry of the
FreeList[4] there would be at least, one Free Chunk.
 The trick then is to inc one of the bytes of the FreeListInUse which is
always at a fixed offset from the begging of the heap (So, if we use
the Default Heap, the begging of the bitmask would start at: 0x70158).
The bit we are overwriting need to be 0, so we can turn it on. 
 The next time we try to allocate with a size smaller than the one we
theoretically set on the bitmask, it will try to pop a chunk from the
double linked list (But since it doesn't have any Chunk, the FLINK/BLINK
should point to the FreeList Entry itself).
 When the magic is finally done, RtlAllocateHeap will end up returning
the address of the FreeList entry. And the exciment continues, at a
fixed offset from the returned address, there should be a the Heap
Function pointer CommitRoutine. voila!

 I bet there is probably better others ways to do it, but thats how i
did it back in the time. (It obviously sounds much easier than there is,
but the primitive works).

 Also, if you think about it (and we discuss it with Oded), there is
another trick with a similar primitive, when you Overwrite the Size
header of a "Lonely" Free Chunk. (Note, this technique theoretically
might work on XP SP2)

Nico




On Wed, May 30, 2007 at 10:30:05AM -0700, Matt Conover wrote:
Can you do any many "inc" as you want? I think there are a lot of options if
you use this against the heap. For example, change
Heap->Lookaside[x].ListHead.Flink to point into middle of existing chunk
(since heap base is reliable except for Vista), increment the ListHead
enough to point into the middle of chunk data, so that you can setup a fake
chunk and wait until it's allocated, then it will cause a 4-byte overwrite
without safe unlink check (lookaside has no safe unlink issues). You could
also use "inc" to change heap flags, that may also be interesting

On 5/30/07, Nicolas Waisman <nicolas.waisman () immunitysec com> wrote:

You can only do it one time.
Note: The riddle is taken from an old silenty patched bug on WINS.

Nico


On Wed, May 30, 2007 at 03:15:13PM +0100, Dave Korn wrote:
On 30 May 2007 07:13, Nicolas Waisman wrote:

Lets have a fun riddle to cheer up the spirit ( Mate at 11pm, its all
night insomnia.)

 The riddle: Let said you are trying to exploit a remote service on an
old Windows 2000 (whatever SP you want) and the primitive is the
following
    inc [edi]   // you control edi

 What would be the best option for edi?

  Depends what else you control apart from edi, and whether you can do
it more
than once.  If you can overwrite an SEH handler, point edi at an illegal
address to invoke your code.  If you can do it multiple times, perhaps
you can
point edi somewhere on the stack and increment a stored ebp to point at
data
you control.  Don't forget the possibility of pointing it at a
non-word-aligned address to e.g increment just the high byte of a stored
pointer.

    cheers,
      DaveK
--
Can't think of a witty .sigline today....

_______________________________________________
Dailydave mailing list
Dailydave () lists immunitysec com
http://lists.immunitysec.com/mailman/listinfo/dailydave


_______________________________________________
Dailydave mailing list
Dailydave () lists immunitysec com
http://lists.immunitysec.com/mailman/listinfo/dailydave

_______________________________________________
Dailydave mailing list
Dailydave () lists immunitysec com
http://lists.immunitysec.com/mailman/listinfo/dailydave


Current thread: