> For DNS it would be overkill I agree, but for other protocols (like Coda
> which does bulk transfer over an UDP(?) based RPC protocol in userspace)
> it could be useful. Also a system call is still much much cheaper than
> sending and receiving an ARP packet, which also introduces "bubbles"
> into long running data transfers.
Probably, probably. Do not forget only, that arps are sent only
every reachable_time+delay_time seconds, and only if no another
notifications were received (from TCP). User notifications
(or even checking for time to make them) will be made every transaction.
Certainly, it can be tested by direct experiments, but now
I am inclined to believe, that the feature would be not very useful.
> > Well, I think, when this element will appear (or recognized to be useless)
> > for IPv6, we will port it back to IPv4.
>
> Is this considered by the working groups?
This item is in semi-dead state, but at least the question is formulated
in Advandced API without any explicit proposal.
Certainly, nobody prohibited us to make it first 8)
> My thinking was just that everything that is possible for a kernel protocol
> should be possible for a user level protocol too - no unfair advantage
> for the kernel ;)
We can make it from user space, though it is not very direct.
RTM_NEWNEIGH, nud_state=NUD_REACHABLE, called for target
without NLM_F_CREATE will make this work.
ioctl() would be much more convenient (and faster), certainly.
Alexey
-
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/