[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Still more mileage from the old grey mare... was: Trojan Horse

On Mon, 2 Sep 1996, Bill Stewart wrote:

> >The key to the success is that the application in question has to be 
> >compromised to respond to these codes, either by design or by hacking. 
> >Either way the individual responsible must modify the execution 
> >mechanism, not just the data itself.
> A well-written program is hard to exploit, but badly written programs
> can often be exploited in ways that allow execution of untrusted code.
> For instance, the fingerd bug exploited so spectacularly by Robert Morris
> handed a program more input that it was ready to accept, and the program
> stupidly kept writing the input into the array, past the end, and out
> into the stack, where it could be later interpreted as executable code.

That is one instance where it was obscure and workable. I am not familiar 
with the UNIX binary map, but in the Intel DOS world (includes Windows to 
a large degree), the stack generally lies at the top of DSEG, not CSEG  and 
flows down. Older incarnations of x86 allowed for such wonders as:

	push	ds
	push	cs
	pop	ds

	[ series of writes ]

	pop	ds

and viola, self modifying code by one method. Ugly, but it works. 
Variations on the same theme exist, now that Intel does not allow direct 
manipulation of the IP anymore. It's not that difficult to do. Some of the 
digital audio work I was involved in several years ago required modifying 
code sequencing on the fly to accomodate time delays and different types 
of processor performance.

Intel binaries are not difficult to modify directly - especially when 
they are DOS binaries (note the plethora of viruses in the medium). Even 
good code can be reworked without having source. One of the tricks is to 
grab the intial JMP off the binary header and point it down past the 
bottom of the existing binary towards the code you have added. This way 
on startup DOS does not truncate the new binary size via Function 31h. 
Now all you have to do is mudge the read section and look for your 
trigger before executing the additional code. What I have just described 
is a very simplistic viral mechanism.

> If a popular GIF or JPEG interpreter was written that badly, you could 
> possibly devise a GIF that lies about how big it is and encourages
> the program to scribble on its stack.  Now, there probably aren't any
> like that, and it'd probably have to be Netscape or MSIE or Lview
> to be widespread enough to make an attack like that worthwhile.
> (I'd bet on MSIE, of the three of them :-)  Does Microsoft have some sort
> of Really Cool Extension to JPEG, allowing Macros for Self-Modifying JPEGs,
> trying to out-do Netscape's animated GIFs?)

It wouldn't surprise me.

> >Let's see -current examples of computing items with this kind of a 
> >"feature"... magic cookies, macros, OLE, DDE, MS Objects, JAVA, and the 
> >list keeps growing.
> Back when Good Times came out, everyone denied that it was possible
> for there to be any risk from a text file (though, as I pointed out,
> escape-sequence hacks have been used occasionally for over 15 years),
> and not long after that, the MSWord Macro Viruses started appearing.
> Bad Code can't always be hacked usefully, but it can always be hacked...
> #			Thanks;  Bill
> # Bill Stewart, +1-415-442-2215 [email protected]
> # <A HREF="http://idiom.com/~wcs"> 	
> # You can get PGP software outside the US at ftp.ox.ac.uk/pub/crypto

Escape sequence hacking in DOS has been done since the day ANSI.SYS came 
out to play. IT was a quick and dirty trick to send escape sequence laden 
files to the unsuspecting and ask them to type them to the screen. Then 
the fun began.

When you stop and think about it though, any application that functions 
as a data engine of some type is susecptible at some level to this form 
of attack. The issue is really which method to employ that will give you