Subject: Re: Changing the PHY status reporting
To: None <tech-net@netbsd.org>
From: Pavel Cahyna <pavel.cahyna@st.mff.cuni.cz>
List: tech-net
Date: 02/19/2006 20:40:41
On Sun, Feb 19, 2006 at 07:06:34PM +0100, joerg@britannica.bec.de wrote:
> We have three options:
> (1) Leave out CARP. Screw all those who want to use and/or need to use
> it.

ucarp from pkgsrc does not work?

> (2) Use the same protocol number and just ignore IANA. As long as you
> don't have some weired configuration, both stacks should be able to drop
> the packets they are not interested in.
> (3) Use a different protocol number. This screws up interoperability
> with other CARP implementations. Not that a big deal, but should be
> mentioned. Doesn't answer the question which protocol number should be
> used instead. Any suggestions?

Let the user choose the number by editing /etc/protocols. 
253 or 254 - those are for "Use for experimentation and testing" [1] -
can be a reasonable default. Add an commented out entry for the OpenBSD
number in /etc/protocols. 

Unfortunately, the protocol number is only a part of the story. CARP also
"stole" a) the VRRP multicast address (224.0.0.18) [2] and b) the VRRP
virtual MAC address space (00-00-5E-00-01-x) [2].

a) is probably harmless if the protocol number will differ, conflict in b)
can be avoided by administrators choosing different IDs for carp and vrrp
groups (coordination is required anyway if multiple groups of the same
protocol are used in the same subnet.)

Still, it is not pretty at all.

[1] http://www.iana.org/assignments/protocol-numbers
[2] http://www.ietf.org/rfc/rfc3768.txt

Pavel