Re: nm-platform suspend/resume routing problem

[Dan Williams <dcbw redhat com>]

On Tue, 2013-07-23 at 17:40 -0400, Pavel Simerda wrote:
> Hi,
>
> didn't catch you on IRC, IMO the routing problem will be more general and suspend/resume only triggers it. 
> In my opinion, setting interface down can trigger inconsistent cache state and we have to clear or maybe 
> even refill the address/route cache when link down event is received. To handle this properly, a new 
> automated test to check that would be needed. Also, more information about what protocol version the route 
> was and what was its source may be needed.

No problem, I posted a patch to the bug that seems to work around the
issue.  It appears that the kernel never sends RTM_DELROUTE events for
an interface's routes (including the default route) when the interface
goes down, but it does remove those routes from the routing table.
Thus, even if the platform was listening to the kernel, it wouldn't see
the default route removal events anyway.

Note that I don't think there's a problem with RTM_DELADDR, just routes.
I do see the RTM_DELADDR events when taking the interface down.

Perhaps the kernel just expects userspace to know that when an interface
goes !IFF_UP that all that interface's routes are now removed?  That
seems kinda odd, would be nice to have RTM_DELROUTE events whenever a
route is deleted instead of deducing magic.

Dan