|
Hello,
On 10 Apr 2003, John Reuning wrote:
> My original wording didn't express my real question, though. I'm
> wondering why the network stack behaves as it does. Why are arp replies
> sent for all interfaces regardless of which interface receives the arp
> request?
Because Linux routing agrees that all these senders have access
to this IP, so we give them access to valid link layer address.
This behavior is usually observed on routers configured without
source address validation enabled. As this is the default behavior
specified in RFC1812 (rp_filter=0), Linux simply allows access to
this IP on any interface.
> arp is part of the transition from network layer to link layer, right?
> So why should an alias on lo, an interface that doesn't really generate
> network frames, trigger an arp reply. Do other unix tcp/ip
Note that these packets are not passed via the lo interface,
also, we do not send ARP replies via lo, why we should care about
the lo's NOARP flag?
> implementation work like this? I can't seem to make a solaris 7 system
> generate arp replies for an lo alias.
The different systems have different policy for IP addresses
configured on loopback device. Note that in Linux, this behavior
has nothing to do with the lo interface, you can configure IP
on eth1 and then again to see our ARP reply for it on eth0.
Regards
--
Julian Anastasov <ja@xxxxxx>
|
