On Wed, 2024-01-24 at 13:55 +0000, Chuck Lever III wrote: > > > On Jan 24, 2024, at 6:24 AM, Jeff Layton <jlayton@xxxxxxxxxx> wrote: > > > > On Wed, 2024-01-24 at 10:52 +0100, Lorenzo Bianconi wrote: > > > [...] > > > > > > > > That's a great question. We do need to properly support the -H option to > > > > rpc.nfsd. What we do today is look up the hostname or address using > > > > getaddrinfo, and then open a listening socket for that address and then > > > > pass that fd down to the kernel, which I think then takes the socket and > > > > sticks it on sv_permsocks. > > > > > > > > All of that seems a bit klunky. Ideally, I'd say the best thing would be > > > > to allow userland to pass the sockaddr we look up directly via netlink, > > > > and then let the kernel open the socket. That will probably mean > > > > refactoring some of the svc_xprt_create machinery to take a sockaddr, > > > > but I don't think it looks too hard to do. > > > > > > Do we already have a specific use case for it? I think we can even add it > > > later when we have a defined use case for it on top of the current series. > > > > > > > Yes: > > > > rpc.nfsd -H makes nfsd listen on a particular address and port. By > > passing down the sockaddr instead of an already-opened socket > > descriptor, we can achieve the goal without having to open sockets in > > userland. > > Tearing down a listener that was created that way would be a > use case for: > > > Do we ever want/need to remove listening sockets? > > Normal practice when making any changes is to stop and restart where > > "stop" removes all sockets, unexports all filesystems, disables all > > versions. > Good point. Even if we don't want to allow it today, passing down a sockaddr over netlink gives us an interface to properly match a listening socket for shutting them down without taking down the server. It would be a little silly to make userland open a socket in order to close the one that nfsd is listening on. -- Jeff Layton <jlayton@xxxxxxxxxx>
