misc things

Paul Rusty Russell Paul.Russell@rustcorp.com.au
Wed, 22 Sep 1999 03:06:12 +0930


In message <Pine.LNX.4.10.9909061840020.30977-100000@ontario.alcove-int> you wr
ite:
> Hi Rusty,
> 
> Hope you're doing well. Here it is :
> 
> #ifdef USE_POWERS_FOR_EVIL  /* yes, some actually read the source ! ;-) */

Hey, we have a winner!  Everyone else, pay attention...  (Bonus points
for figuring out what it does).

>  In packet-filter/extentions/ipt_unclean.c, it seems you forgot one 576
>  (maybe a macro would be good ?) :
>   limpk("First fragment size %u < 576\n", ntohs(iph->tot_len));

Thanks, fixed.

>  In ipnatctl.8 :
>  .\" Man page written by Rusty, based on iptables page by
>  .\" Herve Eychenne <eychenne@info.enserb.u-bordeaux.fr>, based on
>  .\" ipchains page by Paul ``Rusty'' Russell March 1997
>  .\" Based on the original ipfwadm man page by Jos Vos <jos@xos.nl>, who was
>  .\" probably the only one who actually grokked nroff.
>   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>  Thanks, I had a good laugh. :-))

I'd forgotten about that. 8-)

>  In iptables.c :
>  /* FIXME: Userspace can determine that there are no loops, and should
>     do so, elimintating loop check and simplifying this code. --RR */
> 
>  The point is that there should be *no* way to do any loop (thus hang the
>  kernel). Would the test be done in the lib or in iptables ?
>  In both cases, if in userspace, it would possible to bypass the tests.
>  But as long as rules can only be set as root, there would always be a way
>  to compromise kernel integrity...

There are precedents for allowing root to screw the machine up.  There
are also precedents the other way; it depends on how
optimization-obsessed the kernel author is, really.  The loop check
*could* be done statically in the kernel at table insertion time.
That means it has to be done without recursion...

<thinks>
Ick.
</thinks>

>  I guess you've already updated iptables man page (QUEUE, REJECT, state,
>  unclean, etc...), so I won't provide an already obsolete patch this
>  time. ;-)

Oh.  Yeah... thanks, will do.

Rusty.
--
Hacking time.