> Alan Cox (alan@cymru.NET) wrote:
> : > Perhaps checking for own ip address should be done before routing is
> : > performed, Alan?
>
> : Probably. I think I'll avoid fiddling with that until post 2.0 just in case
>
> Moreover, RFC1812 teaches us to check for directly attached
> interfaces BEFORE route table lookup. (One consequence of this requirement
> is that kernel should correctly route packets even if no routing
> tables created!) We might follow this recommendation,
This would prevent some ugly scenario I've seen in HP-UX 9.0: Running
gated, the local interfaces finally timed out and the routes were
removed. From that moment on there were no more sends and receives.
You have to walk to the console and add the routes back manually.
Double-Ugly!
> but it is so ugly... When remaking routing code I decided
> that this recommendation was artifact of before-CIDR epoch and
> ignored it (RFC1812 is just draft, anyway).
What about auto-magically presetting the routing table when a
interface is configured? I don't know if there are useful hooks for
that.
>
> Now all the scheme is nice and simple, routing tables are
> the only source of routing information. If someone forget to create
> route table entry for his own address, it is his fault.
>
> Alexey Kuznetsov.
Ulrich Windl