[OpenAFS-devel] rx mtu

[Chaskiel M Grundman]

--On Friday, January 24, 2003 16:47:18 -0500 Jim Rees <rees@umich.edu>
wrote:
> Bug 1: I think "best" no 4 above is wrong.  It should be returning NULL.  

Perhaps. Perhaps not. One of the things the InitPeerParams code does is
potentially *reduce* the mtu to match the interface returned by
rxi_FindIfnet. I would think that returning the interface for the default
route would actually be useful for that purpose. (and note that the
IFF_POINTOPOINT detection would become useless in most cases)

maybe the first "if (ifn)" block should be getting NULL, but the second one
should not.

> Bug 2: The idea of "net" is some kind of leftover of the pre-CIDR days.
> It shouldn't be used at all.
I sort of disagree. If both of these change are made, then only machines on
directly attached subnets will actually get any mtu processing. Everyone
else will be stuck with RX_REMOTE_PACKET_SIZE, even if the packets will be
using a LAN that can pass larger packets around. using class-based netmasks
to make this decision is somewhat arbitrary and error prone, but I think
doing it is better than not doing it.