On Mon, 5 Sep 2016, Lou Berger wrote:
I'm testing a fix right now that inits (and frees ) AFI_ETHER tables...
Aha.
I think 'bgpd: fix leak introduced by peer nexthop tracking' which adds
a 'bgp_drop_peer_nexthop' to delete bncs, called via bgp_stop isn't
right. The bgp_info's can still have pointers to the bncs which can be
derefed later whne queueing processes the bncs.
The question is whether there's actually a leak? I do see entries
remaining in the nexthop cache after peers for those nexthops are gone
(and all their routes) - at least for connected nexthops. Those /do/
however get cleaned up if the connected link is downed. They do not get
cleaned up if the peers are deleted.
I think the right fix is to add an explicit ref-count on the BNCs,
instead of relying on the list-count on the number of paths - which
can't be shared with the bgp_start/bgp_stop path. ??
regards,
--
Paul Jakma | p...@jakma.org | @pjakma | Key ID: 0xD86BF79464A2FF6A
Fortune:
I hate trolls. Maybe I could metamorph it into something else -- like a
ravenous, two-headed, fire-breathing dragon.
-- Willow
_______________________________________________
Quagga-dev mailing list
Quagga-dev@lists.quagga.net
https://lists.quagga.net/mailman/listinfo/quagga-dev
