[Top] [All Lists]

routable interfaces WAS( Re: [PATCH] hashed device lookup (Does NOT mee

To: Gleb Natapov <gleb@xxxxxxxxxxx>
Subject: routable interfaces WAS( Re: [PATCH] hashed device lookup (Does NOT meet Linus' sumission policy!)
From: jamal <hadi@xxxxxxxxxx>
Date: Sun, 7 Jan 2001 13:02:24 -0500 (EST)
Cc: Chris Wedgwood <cw@xxxxxxxx>, Ben Greear <greearb@xxxxxxxxxxxxxxx>, linux-kernel <linux-kernel@xxxxxxxxxxxxxxx>, "netdev@xxxxxxxxxxx" <netdev@xxxxxxxxxxx>
In-reply-to: <>
Sender: owner-netdev@xxxxxxxxxxx

On Sun, 7 Jan 2001, Gleb Natapov wrote:

> > One could have the route daemon take charge of management of these
> > devices, a master device like "eth0" and a attached device like "vlan0".
> > They both share the same ifindex but different have labels.
> > Basically, i dont think there would be a problem.
> >
> Theoretically it seems to be possible but it's much harder to do in Zebra than
> in kernel. And "eth0" shouldn't share ifindex with "vlan0" I don't think SNMP
> will be happy about that.

A very good reason why you would want them to have separate ifindices.
Essentially, vlans have to be separate interfaces today. Other "virtual"
interfaces such as aliased devices are not going to work with route
daemons today since they dont meet this requirement.

Not to rain on Ben's parade but:
My thought was to have the vlan be attached on the interface ifa list and
just give it a different label since it is a "virtual interface" on top
of the "physical interface". Now that you mention the SNMP requirement,
maybe an idea of major:minor ifindex makes sense. Say make the ifindex
a u32 with major 16 bit and minor 16 bit. This way we can have upto 2^16
physical interfaces and upto 2^16 virtual interfaces on the physical
interface. The search will be broken into two 16 bits.



<Prev in Thread] Current Thread [Next in Thread>