On 11/26/18 12:06 PM, Alexis Bauvin wrote: > Moreover, the issue of mixing default and non-default vrf needs to be > addressed. For now it is stale, as I don’t see any solution (except for > rewriting the whole thing as you suggested before) to address the > "Address already in use" made by a socket of the default vrf owning the > port across all vrfs. > I tested both Vyatta’s changes and SO_REUSEPORT, and neither of them seem > to work for this case.
That suggests to me the reopen should be done internally so that the socket failure can cause the enslavement to fail with a message passed back to the user via extack. ie., If changing the vrf association breaks vxlan, we should detect that and fail the change.
