[Dibbler] dibbler-server an debian-armel

[Tomasz Mrugalski]

On Fri, 15 Aug 2008 somebody known as Christian Andersen wrote:

> The first dump shows the attempt of Windows Vista to communicate with
> dibbler-server, but without success (no answer from dibbler). The second
> dump shows the successful communication of Windows Vista with
> wide-dhcpv6-server.
I'm not 100% sure (see below), but I believe I've tracked the rootcause of 
this issue.

Short answer: it's a Vista bug that confuses dibbler.

Long answer:
Vista sends FQDN option that is malformed. They encode domain name in a 
wrong manner. (That was confirmed by Xerox). They also had some problems 
with the FQDN option sent by Microsoft.

This malformed option sometimes confuses Dibbler. I've prepared snapshot 
with fixed FQDN parsing:

http://klub.com.pl/dhcpv6/snapshots/2008-08-15-possible-vista-bug-workaround/

And for disbelievers (and folks interested in details), here's the 
explanation:

According to RFC4704 (FQDN), section 4.2 domain names must be coded as 
specified in RFC3315, section 8. RFC3315 in section 8 states that domain 
encoding must be stored in a compressed form as explained in section 3.1 
of RFC1035. And here's the corresponding part of that RFC:

Each label is represented as a one octet length field followed by that
number of octets.

Microsoft just puts domain name and terminates it with 0.
That is *wrong*.

End of explanation.


Ok, now for the issue that you reported. I've said that I'm not 100% sure. 
Here are my doubts. The log you've provided states that dibbler-server 
received UDP packet with 79 bytes as payload:

2008.08.15 12:20:27 Server Debug     Received 79 bytes on interface eth0/2
(socket=5, addr=fe80::20d:56ff:fe7a:2baf.).

However, the dump that you provided contains SOLICIT messages that are 100 
bytes long (length in UDP is 108, but that includes UDP header, i.e. 8 
bytes). The difference is 21 bytes. That means that they are completely 
different packets.

Regards,
Tomek