Bind 8 crashing on bad NAT

[Sybren Stuvel]

Hi!

I'm using an ADSL connection, provided by the Dutch company KPN. They (KPN) 
use some kind of weird method of Network Address Translation (NAT). Our IP 
address, from the Internet point of view, is 194.109.221.152. Our IP 
address, from our server's point of view, is 10.67.195.129. No, I'm not 
mistaken, this is really the IP address our server has:

ppp0      Link encap:Point-to-Point Protocol
           inet addr:10.67.195.129  P-t-P:195.190.229.51  Mask:255.255.255.255
           UP POINTOPOINT RUNNING NOARP MULTICAST  MTU:1500  Metric:1
           RX packets:26374 errors:0 dropped:0 overruns:0 frame:0
           TX packets:22454 errors:0 dropped:0 overruns:0 carrier:0
           collisions:0 txqueuelen:10

Because of this, named often crashes and complains about lame servers, 
which aren't lame at all.

I've tried both bind 8 and 9. Bind 8 complains a lot, but is ok again after 
a 'ndc restart'. Bind 9 doesn't crash, but has other problems. With bind 9, 
it often can't find a domain, unless I look it up using another server first:

dig kernel.org <- fails, using our bind 9 name server
dig kernel.org @ns.xs4all.nl <- ok, using our provider's name server
dig kernel.org <- ok, using our own bind 9 name server again.

Isn't that weird???

Does anyone have an answer to these problems? Especially the first (bind 8) 
problem is important to us.

Greetz,
MeltDown

Linux version 2.2.16-22
RedHat 7
Bind version 8.2.2_P5-25 and 9.0.0rc6