Re: Can anyone tell me why...

[Mike Connor <mikec@silug.org>]

Guess I'll watch for the answer then.  I've never noticed that the MMU 
could get state from the CPU regarding what exactly the fetch was for 
(loaded as a result of PC or just a plain ol' fetch).  

Mike


On Thu, 2 Jan 2003 fiaid@quasi-sane.com wrote:

> Trust me, I am no expert on this, but it was an interesting discussion I 
> was having with those guys, hopefully they will be on later.
> 
> > From what little I've messed with process stacks ( so be nice if I'm 
> > ignorant on the subject ), you'd have to make every push/pop from the 
> > stack handled by the kernel (obviously).  You'd have to enforce it with 
> > the MMU (which you'd be reconfiguring twice as often).
> 
> In the creation of the stack though, doesn't the VMM flag the MMU as
> execute/non-excute?  At this point the stack is flaged non-execute and it
> only takes that one step.  So at the beginning of the process running or
> at the fork when the new MMU's are created, why doesn't the underlying
> subsystems just say "all stacks are non-executable, i don't care what you
> think"?
> 
> > If you did something like one of the above and someone managed to set
> > the program counter to a stack location, it would hang the process or
> > kill it.  Which I guess would be a good thing.
> 
> Not all the time though, if you use a noop sled to catch the pointer into 
> the stack and have it fall down to your machine code it still hasn't 
> caught a signal yet.  This really depends on how smart the buffer overflow 
> writer was, if he created a large enough sled then they would be able 
> direct the pointer into a large enough location that it would be able to 
> noop slide down the code and execute it.  
> 
> > So at a minimum, the CPU would probably have an extra 100 instructions 
> > every time the process did a function call.
> 
> I am not sure about this, there might be additional instructions, but they 
> would be at a machine code level and wouldn't really take up all that much 
> CPU power.
> 
> Like I said earlier, I really know very little about this but I am 
> curious.  I know that we got some real developers here on the list, can 
> they shed some light?
> 
> Tighe
> 
> 


-
To unsubscribe, send email to majordomo@silug.org with
"unsubscribe silug-discuss" in the body.