[lug] physical/logical network interfaces

[Hugh Brown]

On Tue, 22 Mar 2005, Nick Golder wrote:

> On 2005-03-22 10:31 -0700, Hugh Brown wrote:
> > I always love responding to my own posts.  So there really are 5 ports,
> > the onboard NIC, and two Intel pro 100 dual port cards.
> >
> > So the question morphs.  The four ports in the dual port cards have no
> > cables in them.  Only the onboard NIC is connected.  So now my conundrum
> > is this:
> >
> > Why would the kernel allow me to configure eth1 as an interface (and
> > receive traffic for that interface) when there's no cable connected to it?
> >
> > Is it an artifact of all of the cards using the same driver (e100)?
> >
> > This strikes me as obscurely broken.
> >
>
> The kernel sees 5 NICs and enumerates them all.  The question is why
> shouldn't the kernel allow you configure a NIC that doesn't have a cable
> connected.  The kernel thinks that if you didn't want to be
> able to configure them, you probably wouldn't have them installed.
>
> It would strike me as "obscurely broken" if I couldn't configure a NIC
> that isn't pysically connected to a network.

Agreed.  Here's my thought process on what's "broken"

I have a disconnected nic named eth1.  I configure it.  Since there's no
cabled connected, I would expect that packets destined for it would fall
on the floor.

Instead, what I'm seeing is eth0 is picking up all packets destined to
eth1 and the IP assigned to eth1 is associated with the MAC address for
eth0.

So what I think is "broken" is that eth0 has pre-empted eth1, not that I
was able to configure eth1 in the first place.

Hugh