On Thu, 10 Sep 2026, Jeff Layton wrote: > Describe the interface that lets a caller take over rpcbind registration.
> --- a/Documentation/netlink/specs/nfsd.yaml > +++ b/Documentation/netlink/specs/nfsd.yaml > @@ -483,13 +527,22 @@ operations: > - > name: listener-set > - doc: set nfs running sockets > + doc: >- > + set nfs running sockets. A request that carries userspace-rpcbind > + is answered with a reply rather than a bare ack, and the addr list > + in that reply names only the listeners that have an rpcbind netid. "listeners that have an rpcbind netid" overstates the set. RDMA listeners have netids (rdma, rdma6) but are omitted from the reply. Following the way Solaris handles RDMA transport, NFSD never registered RDMA itself, so that is not a regression, but the doc should say what your filter actually selects. > attribute-set: server-sock > flags: [admin-perm] > do: > request: > attributes: > - addr > + - userspace-rpcbind > + reply: > + attributes: > + - addr > + - userspace-rpcbind > + - rpcbind The spec declares an unconditional reply, but patch 4 sends one only when the request carried userspace-rpcbind. ynl has no way to express a conditional reply, and no other in-tree spec does this. Either always send the reply (drop the userspace_rpcbind gate in patch 4, so a plain caller simply ignores it), or leave listener-set without a reply and put the registration data in a separate get operation. I lean toward always replying; that's is a little more "declarative". -- Chuck Lever (Come to NFS bake-a-thon! https://nfsv4bat.org)
