On Mon, Aug 15, 2016 at 08:41:52AM +0200, Martin Pieuchot wrote:
> On 08/08/16(Mon) 11:42, Martin Pieuchot wrote:
> > On the train back from n2k16 I found the real cause of the hang reported
> > by Dimitris Papastamos [0] and exposed by our recent
> > changes to the routing table.
> >
> > When an interface is removed/detached the kernel delete all the
> > corresponding route entries. At this moment the interface is
> > DOWN and the corresponding route as well. So the priority check
> > should consider that.
> >
> > Without the diff attached or the workaround to stop iterating when
> > an error occurs, the kernel would loop forever since it can't remove
> > the RTF_BROADCAST entry.
> >
> > [0] https://marc.info/?l=openbsd-bugs&m=146909621511954&w=2
> >
> > ok?
>
> Anyone?
>
This is correct. OK claudio@
IIRC this is not the first time we go burned by RTP_MASK so I wonder if we
should use a macro to access prio that does the masking internally or
removing the DOWN prio but then then all checks need to also check for
RTF_UP...
> > Index: net/route.c
> > ===================================================================
> > RCS file: /cvs/src/sys/net/route.c,v
> > retrieving revision 1.313
> > diff -u -p -r1.313 route.c
> > --- net/route.c 22 Jul 2016 11:03:30 -0000 1.313
> > +++ net/route.c 8 Aug 2016 09:33:15 -0000
> > @@ -873,7 +873,7 @@ rtrequest_delete(struct rt_addrinfo *inf
> > * kernel.
> > */
> > if ((rt->rt_flags & (RTF_LOCAL|RTF_BROADCAST)) &&
> > - prio != RTP_LOCAL) {
> > + (prio & RTP_MASK) != RTP_LOCAL) {
> > rtfree(rt);
> > return (EINVAL);
> > }
>
--
:wq Claudio
