> > Well i  sure have.  I've experienced  both NFS file corruption and DNS
> > request corruption at  the hands  of  this.  BTW, the  person who said
> > that COMPAT_42 turns off udpcksum in netbsd 1.1  was correct.  In fact
> > this seems  to  be the only   thing that  COMPAT_42  does  (not  to be
> > confused  with TCP_COMPAT_42).  Unless there is  some  good reason for
> > this I'd suggest getting rid of that option, there just isn't any good
> > reason to ever turn off UDP checksums.
>
> Is there a good test to determine when this lossage exists?

You mean whether it's possible to remotely probe if UDP checksum
checking and generation are off?  Yes, it is.  Send a datagram
with erroneous UDP checksum to a "random" high-numbered port
(should not be and probably isn't in use) on the target host and
see if you get an ICMP Port Unreachable back.  This technique is
similar to the one tcpdump uses, and someone (I think Benoit
Grange from NIC France) actually hacked tcpdump and produced
"ckudpcksum", which can be had from

	ftp://ftp.nic.fr/pub/programmes/DNS/ckudpcksum.tar.gz

Regards,

- Havard