jamal <[EMAIL PROTECTED]> writes:

> Ok, its a little more than grat arp (which i assumed it was). However,
> the cause-fatale given in section 4 for why you wanna broadcast does
> sound very lame. To quote: You do this in case "two disjoint network
> links are joined".

It isn't quite a gratuitous arp, as this only applies to arp
reply packets.  But it is certainly the same general principle.

> I translate this to mean that somehow someone is going to take two hubs
> or two switches of already established connections and join them
> together and therefore linklocal addresses in one broadcast domain will
> possibly conflict with linklocal in the other - it didnt seem practical
> at all. I could be wrong and there maybe other cases not mentioned in
> the RFC where you would need to do this. Hopefully, the original poster
> can explain what kind of problems they saw that prompted the patch.

I suspect a better scenario is if you have two switches connected
normally and someone doing maintenance in the switch closet temporarily
unplugs them, untangles the cables and replugs them.

Similar things can happen with an unmananged wireless network when
to machines are barely within range of each other.

I have seen a piece of this problem when two machines were assigned
the same ip address.  An arp request is sent out for some IP
address.  Two unicast arp replies are received and the kernel picks
one.  After an appropriate interval the arp cache is refreshed
and the kernel starts using the other arp reply and your tcp
connections break.  If the machines are configured sufficiently similarly
this can get very confusing.

Now I would make the suggestion that we only broadcast the arp
reply if the source of the arp request is not already in our arp cache,
when we receive the arp reply.

It is really only when we start communicating with a new machine that
we need to broadcast our replies.  And this allows the periodic arp
refresh between two machines to remain unicast.

Eric
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to