Re: [PATCH 0/3] Enable the setting of a kernel module parameter from nfs.conf

[Trond Myklebust <trondmy@hammerspace.com>]

On Tue, 2021-04-20 at 13:40 -0400, bfields@fieldses.org wrote:
> On Tue, Apr 20, 2021 at 05:28:08PM +0000, Trond Myklebust wrote:
> > On Tue, 2021-04-20 at 13:18 -0400, J. Bruce Fields wrote:
> > > On Tue, Apr 20, 2021 at 02:31:58PM +0000, Trond Myklebust wrote:
> > > > I think the important thing is, as Chuck said, that the setting
> > > > of
> > > > the
> > > > uniquifier has to be automated. There are too many instances
> > > > out
> > > > there
> > > > of people who get confused because they are using a default
> > > > hostname,
> > > > such as 'localhost.localdomain' and are setting no uniquifier.
> > > > 
> > > > So the point is that it needs to be persisted by an automated
> > > > script if
> > > > unset.
> > > > 
> > > > While that script could use nfsconf to get/set the persisted
> > > > uniquifier, the worry is that such an automated change might be
> > > > made
> > > > while the user is performing some other edit of nfs.conf. What
> > > > happens
> > > > then?
> > > 
> > > The one thing I'm a little uneasy about is ignoring /etc/machine-
> > > id.
> > > Seems like distros *should* be creating it for us.  And it would
> > > be
> > > convenient to have one source of machine identity rather than
> > > separate
> > > ones for different subsystems.
> > > 
> > > Maybe we could use that if it exists, and fall back on generating
> > > our
> > > own only if it doesn't?
> > > 
> > > (Well, where "use it" actually means take a hash of it, as
> > > explained
> > > in
> > > machine-id(5).)
> > > 
> > 
> > Maybe, but that ties the nfs-utils package irrevocably to systemd.
> 
> Well, like I say, we could have a fallback.  Or even provide
> alternative
> scripts in nfs-utils and let the distro decide which to install
> depending on whether they use systemd.
> 
> But, whatever, those two alternatives (machine-id or vs. nfs
> generating
> its own uuid) are basically the same on some level.

Not quite. They cause the behaviour to differ depending on whether or
not systemd is installed. So if you imagine a system that gets updated
from "traditional init" to systemd, then that could cause the NFS
identity of the machine to change.

It would be better to be able to specify the identity in a form that is
independent of the platform.

So if the machine-id exists, then maybe we could indeed generate the
identity using the uuid in that file (although the question remains as
to why you'd want that?). However the generated value should then be
persisted separately so that it can be platform independent.

> 
> I agree with the basic idea that this should be automated rather than
> living in a configuration file that humans might have to deal with.
> 
> --b.

-- 
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@hammerspace.com