On Thu, May 26, 2016 at 02:01:05PM -0700, Jesse Gross wrote: > On Thu, May 26, 2016 at 1:34 PM, Thadeu Lima de Souza Cascardo > <casca...@redhat.com> wrote: > > One thing that may be seen as a regression is the MTU of those devices. It > > won't > > be IP_MAX_MTU as in latest upstream, but supporting older kernels restrict > > us to > > use 1500 for VXLAN and GENEVE. > > Is it possible to detect this and continue using the compat code on > those kernels? (I think it's a pretty narrow range of kernels that > support metadata mode on tunnels but not MTU adjustment.) As you say, > it's otherwise a regression from where we currently are as far as > jumbo frames go and also results in different behavior between tunnel > types and kernel versions.
We could try setting a larger MTU and, if that fails, revert to compat code. There still is going to be a range of kernel versions where this problem happens (between 4.3 and 4.5, I guess). The other issue is that the code introduced by David Wrang adds a non-strict option that only OVS compatibility code can use. That means we cannot use IP_MAX_MTU but a slightly smaller value. I don't think that would be a problem, though. I would still like to see any other comments on the patchset before I send a v2 fixing that. Thanks, Jesse. Cascardo. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
