From: itojun@iijlab.net
Subject: vlan and ethernet "slightly jumbo" frames
Date: Mon, 03 Jun 2002 10:21:48 +0900

> 	at this moment if_vlan.c controls ethernet "slightly jumbo" frames
> 	(ETHERCAP_VLAN_MTU).  there are other customers which uses "slightly
> 	jumbo" frames, namely MPLS over ethernet (see www.ayame.org).
> 	i guess it better to move ETHERCAP_VLAN_MTU handling into
> 	if_ethersubr.c.  just a thought.

I completly agree.

I am a member of ayame project.  We are now implementing an MPLS
forwarding function on NetBSD network stack.

The MPLS labeled packet on ethernet is one of "slightly jumbo"
frames, as itojun said, because the label is represented by the
shim-header located between the ether header and the IP header.

Currently, we are using the "ETHERCAP_VLAN_MTU" flag to use
the shim-header.
But because this flag is only for VLAN use, some operation which
changes the vlan configuration causes inconsistensy
(if ec_nvlans=0, the mpls function becomes disabled, etc..).

If we have a generalized in-kernel API for enabling "slightly
jumbo" frame exists in sys/net/if_ethersubr.c 
(which reference-counts the number of users and toggle
"slightly jumbo" frame appropriately), it must be convenient 
for various similar uses.

						Yojiro UO