On Wed, Jan 30, 2008 at 02:30:44PM -0500, James Carlson wrote:
> Glenn Faden writes:
> > If we're going to remove the route when the zone is halted, then we have
> > to prevent the another zone from using the same route. However, I think
> > that adding this complexity will not improve the customer experience.
> > For example, I'm afraid that there may be race conditions between a zone
> > that is booting and one that is shutting down with the same default
> > route. The simpler solution seemed adequate to me.
>
> I don't think it is. Try this:
>
> - set up a default route to 10.0.0.1 for a zone.
> - boot it.
> - shut it down.
> - change that default route to 10.0.0.2.
> - boot it again.
> - examine the kernel's forwarding (routing) table.
>
> You should end up with two default routes; one to the new destination,
> and another to the old destination. The system caches these things.
> ("-nostatic" might do the trick, but I'm not positive.)
>
this is really bad. i really think that a zone should cleanup any
routes it added to the system when it's shutdown. we don't want zones
leaving configuration turds lying around the system after zones
have been shutdown.
if a zone adds a route to the system successfully it needs to
record this fact in the zone state, a zone attribute would be
perfect for this. then at zone shutdown the framework can check
to see if there are any routes it needs to tear down.
ed
