Subject: Re: dhcpd(8) _cannot_ be completely disabled on an interface
To: John Nemeth <jnemeth@victoria.tc.ca>
From: Dennis Ferguson <dennis@juniper.net>
List: tech-net
Date: 01/12/2002 16:12:01
> On May 30,  9:25am, Dennis Ferguson wrote:
> } 
> } I think DHCPOFFER messages are always sent addressed to 'yiaddr' address,
> } and this is never 255.255.255.255.
> 
>      Wrong again.  You are neglecting the cases of a relayed request or
> when the client sets the broadcast flag.  Perhaps you should go read
> RFC 2131 again.

I enjoyed how you entirely deleted the context, which was DCHPOFFER messages
addressed to the client's MAC address, as in:

>> Sometimes, the DHCPOFFER message needs to be sent broadcast back to
>> the client.  In other words, the dest IP address is 0xffffffff, and
>> the dest MAC address is the client's.  I don't think you can do this
>> with UDP.
>
> I'd challenge you to quote the text in RFC 2131 which requires this.
> This is such a huge layering and semantic violation that I don't think
> anyone would even dare to write it down, let alone have it remain in
> the document all the way to draft standard status, even if it could
> otherwise be widely implemented.
>
> I think DHCPOFFER messages are always sent addressed to 'yiaddr' address,
> and this is never 255.255.255.255.

DHCPOFFER messages which are sent to the client's destination MAC address
are always addressed to the `yiaddr' IP address.  DHCPOFFER messages which have
something other than the `yiaddr' address as their IP destination are not
sent to the client's MAC address.  The combination of IP and MAC destination
addressed mentioned in the quote above never occurs.

If I've missed something about either the broadcast flag or relayed
requests which requires you to send a packet to an IP destination of
255.255.255.255 with the MAC address set to the client's, perhaps you
could actually quote the text in RFC 2131 instead of just hinting that
it exists?

Thanks,
Dennis Ferguson