oss-sec mailing list archives

Re: Re: Offset2lib: bypassing full ASLR on 64bit Linux


From: Hector Marco <hecmargi () upv es>
Date: Thu, 11 Dec 2014 11:39:43 +0100

Hi Mike,


El 09/12/14 a las 02:04, Mike Hommey escribió:
Hi,

This is a disclosure of a weakness of the ASLR Linux implementation.
The problem appears when the executable is PIE compiled and it has an
address leak belonging to the executable. We named this weakness:
offset2lib.

In this scenario, an attacker is able to de-randomize all mmapped
areas (libraries, mapped files, etc.) by knowing only an address
belonging to the application and the offset2lib value.

We have built a PoC which bypasses on a 64 bit Linux system, the three
most widely adopted and effective protection techniques: No-eXecutable
bit (NX), address space layout randomization (ASLR) and stack smashing
protector (SSP). The exploit obtains a remote shell in less than one
second.

We have proposed the ASLRv3 which is a small Linux patch which removes
the offset2lib weakness.

Details of the weakness, steps to exploit the offset2lib weakness, a
working proof of concept exploit, recommendations and a demonstrative
video has been publish at:
http://cybersecurity.upv.es/attacks/offset2lib/offset2lib.html

If you have the base address of the executable and are able to read any
byte in the process address space, as my reading of the paper suggests,
then you don't even need offsetlib, and no amount of ASLR can save you.

Maybe you misunderstand our paper, we do not rely on been able to read
any byte of the process. Actually, our PoC works by writing bytes on
the stack to de-randomize the PIE executable. This is not new at all,
Packet Storm Security published a paper about how to do it:
http://www.intelligentexploit.com/articles/Smashing-the-stack,-an-example-from-2013.pdf (August 2013), One year later, almost the same
technique was published (as new?) in a paper called "Hacking Blind":
http://www.scs.stanford.edu/brop/bittau-brop.pdf (May 1014).

The conclusion of the Packet Storm paper was that even de-randomizing the process you can do nothing. This is because you can only redirect the control flow and that time they didn't know where to jump.

The Offset2lib is a step forward which enables to use (jump to)
the libraries without reading any other data from the executable
as "Hacking Blind" requires. By using the offset2lib to bypass the ASLR, the Hacking Blind will be deprecated until offset2lib is fixed.


Just read the ELF program headers that you can find at the base address
of the executable to find the PT_DYNAMIC segment. In that segment, find
the DT_DEBUG entry, which will give you a pointer to a r_debug struct,
which definition you can find in /usr/include/link.h.

That struct has a r_map member that gives you a linked list of libraries
the dynamic linker (ld.so) loaded, with the base address for each of
them.

Those data structures are used by gdb, so removing the DT_DEBUG pointer
would break debugging with gdb.

Mike


Hector Marco.


Current thread: