Re: [2.2] Network Interface aliasing

Andi Kleen (ak@muc.de)
Wed, 10 Nov 1999 12:50:40 +0100

On Tue, Nov 09, 1999 at 06:48:58PM +0100, Christopher E. Brown wrote:
> On 9 Nov 1999, Andi Kleen wrote:
>
> > cbrown@denalics.net (Christopher E. Brown) writes:
> >
> > > Example, machines A,B,C all have the same IP on a lookback or
> > > dummy interface. A load director/etc is currently routing all traffic
> > > to the cluster IP to machine A.
> > >
> > > Should machine Bs interface get into PROMISC mode all packets
> > > (even though they are sent to another MAC addr) will hit the IP layer,
> > > and should the IP be found on a local interface, the machine will
> > > respond.
> >
> > This is not true. When an interface in promisc mode receives a packet
> > not destined at its MAC address and not a {broad,multi}cast it sets its type
> > to PACKET_OTHERHOST. arp_rcv ignores such packets. IP does the same check.
> >
> > -Andi
>
>
> Hmm, I have not rechecked this in 2.2.x, however with <= 2.0.x
> this was not so.

Some early 2.0.x had a bug in this regard, but this was fixed years ago
in the 2.0 series.

>
> When an interface in promisc mode received a packet destined
> for another mac, but to an IP the local host owns it would, as you
> said, not arp. It *would* respond to the packets though.
>
>
> To restate, it was assumed that anything with a locally owned
> IP that made it to the IP layer was ours. Should one have an
> interface in promisc, where we receive a packet with an IP we have but
> destined to another MAC addr it would get passed to the IP layer, and
> dealt with exactly as if it was sent to our MAC addr.

Except when you have a buggy driver (that does not set PACKET_OTHERHOST
correctly -- normally all drivers do because the check is in the shared
eth_type_trans function) this should not happen.

-Andi

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/