On Mon, 2021-12-20 at 15:51 +0000, Chuck Lever III wrote:
>
>
> > On Dec 19, 2021, at 3:49 PM, Trond Myklebust
> > <trondmy@xxxxxxxxxxxxxxx> wrote:
> >
> > On Sun, 2021-12-19 at 18:15 +0000, Chuck Lever III wrote:
> > >
> > > > On Dec 18, 2021, at 8:38 PM, trondmy@xxxxxxxxxx wrote:
> > > >
> > > > From: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
> > > >
> > > > NFSv4 doesn't need rpcbind, so let's not refuse to start up
> > > > just
> > > > because
> > > > the rpcbind registration failed.
> > >
> > > Commit 7e55b59b2f32 ("SUNRPC/NFSD: Support a new option for
> > > ignoring
> > > the result of svc_register") added vs_rpcb_optnl, which is
> > > already
> > > set for nfsd4_version4. Is that not adequate?
> > >
> >
> > The other issue is that under certain circumstances, you may also
> > want
> > to run NFSv3 without rpcbind support. For instance, when you have a
> > knfsd server instance running as a data server, there is typically
> > no
> > need to run rpcbind.
>
> So what you are saying is that you'd like this to be a run-time
> setting
> instead of a compile-time setting. Got it.
>
> Note that this patch adds a non-container-aware administrative API.
> For
> the same reasons I NAK'd 9/10, I'm going to NAK this one and ask that
> you provide a version that is container-aware (ideally: not a module
> parameter).
>
> The new implementation should remove vs_rpcb_optnl, as a clean up.
>
>
This is not something that turns off the registration with rpcbind. It
is something that turns off the decision to abort knfsd if that
registration fails. That's not something that needs to be
containerised: it's just common sense and really wants to be the
default behaviour everywhere.
The only reason for the module parameter is to enable legacy behaviour.
--
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@xxxxxxxxxxxxxxx
