What to do about nologin(8)?

[John Baldwin]

On Monday 23 February 2004 03:39 pm, Tim Kientzle wrote:
> John Baldwin wrote:
> > On Monday 23 February 2004 02:58 pm, Doug Rabson wrote:
> >>On Mon, 2004-02-23 at 17:45, Colin Percival wrote:
> >>>   For security reasons, nologin(8) must be statically linked;
> >>>as a result, adding logging has increased the binary size ...
> >>
> >>How about:
> >>
> >>7: Use 'system("logger ...") to log the failed login?
> >
> > Wouldn't that be subject to the same LD_LIBRARY_PATH concerns since
> > logger is dynamically linked and you could trojan it's libc?
>
> Not if nologin clears the environment first.
>
> Related to this, I think I've found a solution to
> the underlying problem:  ignore login's "-p"
> option if the user shell isn't in /etc/shells.
>
> This blocks environment-poisoning attacks against
> nologin via /usr/bin/login.
>
> With this change, it might even be possible to go
> back to the shell script version of nologin.

My point (sigh) is that doing system("logger") has the same problem set as 
making nologin dynamic, so it's not a solution to the original problem.  
Also, personally, I would rather have nologin be static than fix the one 
known case of login -p and just hope no other cases pop up in the future.  
Call me paranoid. :)

-- 
John Baldwin <jhb at FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve"  =  http://www.FreeBSD.org