On (07/23/07 12:20), Peter Memishian wrote: > > > Not sure I understand: what is the problem if we put the same address > > on both the meta- and the underlying interface? > > How would that work? A given IP address can only be IFF_UP once, unless > you're thinking of the whole IFF_UNNUMBERED mess. > > > Given that the meta interface is just a virtual interface, what would > > break if they shared the address? > > I'm not aware of any precedent for this in the kernel, so I think it would > be a painful exercise. That aside, if another interface joined the group, > the data address on the IPMP meta-interface would need to vanish, which > might surprise applications. (And there's the broader question of how > such an environment could be configured with e.g., DHCP -- would it > somehow know about IPMP singleton and maintain its lease information on > two different interfaces? Sounds icky.)
yes, I was thinking of UNNUMBERED interfaces, but I see your point about the ickiness introduced when we go from singleton -> non-singleton ipmp. > > Alternatively, in the re-architected model, with the new anonymous > > group proposal, can we add a meta-interface to the anonymous group? > > We could, but what does that buy us? I guess I was trying to ask "are we going to explicitly prohibit this, or is there some foreseeable use for it in the future?" --Sowmini
