Re: 8250 serial console fixes -- issue

[Pavel Machek <pavel@ucw.cz>]

On Po 06-02-06 20:55:00, Russell King wrote:
> On Mon, Feb 06, 2006 at 08:26:55PM +0000, Pavel Machek wrote:
> > Hi!
> > 
> > > > The serial console driver has a host of issues
> > > > 
> > > > [...]
> > > > 
> > > >  - [SECURITY] 'r' should require DCD to be asserted
> > > >    before outputing characters. Otherwise we talk to
> > > >    Hayes modem command mode.  This allows a non-root
> > > >    user to re-program the modem and is a major security
> > > >    issue is people configure calling line identification
> > > >    or encryption to restrict use of the serial console.
> > > 
> > > How is this possible?  A normal user can't produce arbitarily formatted
> > > kernel messages, and if they have access to /dev/ttyS they can do what
> > > ever they like with the port anyway.
> > 
> > Maybe not *arbitrary* messages, but any user probably can fake enough
> > to
> > confuse modem. Name your process \nATD609123456\n and cause it to eat
> > all memory, or something like that. OOM killer will print name...
> 
> As I say, it's a problem which needs fixing elsewhere.  What if
> the process was called:
> 
> \nSystem Halted\n
> 
> (which will fit in the kernel's process name.)  Or maybe some escape
> sequence which reprograms your terminal.
> 
> As I say, this is a problem which needs solving by other means.  Maybe
> flush_old_exec() should be a little more careful about what it copies,
> changing non-alphanumeric characters to '?' ?

I'm afraid that auditing kernel to never ever print \n from user will
be quite a long job. If I get

Killed process 1234
System Halted
due to OOM

I am going to figure it out no problem, but modems do not have that
kind of abilities...
								Pavel

-- 
Web maintainer for suspend.sf.net (www.sf.net/projects/suspend) wanted...