On Sat, Mar 28, 2015 at 9:30 AM, Juliusz Chroboczek <[email protected]> wrote: >> Say wlan0 vanishes. All the routes going out that interface are no >> longer valid, but from what I understood of this patch, it will loop for >> a while, then give up. > > If wlan0 vanishes, this will be recognised the next time check_interfaces > is run, and all neighbours visible through wlan0 will be flushed. > > The issue we're having is a race condition -- if wlan0 goes down and then > back up before we run check_interfaces, and the IP addresses don't change, > then check_interfaces will not notice the transition, and Babel will think > that its routes through wlan0 are still up -- end you end up with a FIB > that is not a subset of the RIB. Ouch.
There is a bunch of work towards unifying ipv4 and ipv6 behavior that might be worrisome OR helpful. See: https://www.netdev01.org/docs/prabhu-linux_ipv4_ipv6_inconsistencies_talk_slides.pdf A bunch of other presos from that conference were interesting https://www.netdev01.org/downloads > So it might be a good idea to run check_interfaces early when we get > EAGAIN, but I'm not sure what consequences it might have -- EAGAIN can > also happen when we're under load, and we'd rather not be repeatedly > scanning our interfaces in that case. > > It would be better to get async notifications from the kernel about > interfaces going down. Amplified to netdev. >> Not clear to me if this would happen for 4 hello intervals before the >> interface is recognised as gone? > > No, the hellos are used to notice vanishing neighbours, not vanishing > interfaces. That's a completely different mechanism. got it. Keeping calm and patching on! > > -- Juliusz -- Dave Täht Let's make wifi fast, less jittery and reliable again! https://plus.google.com/u/0/107942175615993706558/posts/TVX3o84jjmb _______________________________________________ Babel-users mailing list [email protected] http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/babel-users
