16 May
2024
16 May
'24
8:49 p.m.
On Thu, May 16, 2024 at 10:44 AM John Cowan <cowan(a)ccil.org> wrote:
NFS v4 provides idmapping between client uids/gids and
their server equivalents using a config file.
NFS v4 came much later, of course. But even earlier NFS
implementations provided something like this, and UID 0 has been
mapped to "nobody" for many decades. But that's not terribly relevant;
the point is that, by default, the NFS protocol relies on UIDs having
meaning, as opposed to properly authenticated principles independent
of the underlying implementation of "identity", as in e.g. AFS. To be
fair, it's my understanding that NFSv4 _does_ improve on the situation
here.
- Dan C.
On Thu, May 16, 2024 at 9:46 AM Dan Cross
<crossd(a)gmail.com> wrote:
>
> On Thu, May 16, 2024 at 7:51 AM John Cowan <cowan(a)ccil.org> wrote:
> > [snip]
> > This appears to be a VHS vs. Betamax battle: NFS was not transparent, but Sun
had far more marketing clout. However, the Manchester Connection required a single uid
space (as far as I can tell), which may also have been a (perceived) institutional
barrier.
>
> So did NFS, for that matter.
>
> This is one of those areas where Unix appears creaky in comparison to
> Plan 9. `ssh` is all about remote access to resources, whereas plan 9
> was all about resource sharing: you'd set up a namespace with all of
> the resources (exposed as files from wherever they ultimately came
> from) you cared about, and then operate on those "locally"; the
> resources were shared with you and access was transparent, via a
> consistent, file-based interface. You want to `diff` two remote files?
> Import the filesystems they're both on, mount those somewhere, and
> `diff /n/host1/file /n/host2/file`.
>
> I think the `sshfs`+FUSE model that Doug mentioned is about the
> closest you're going to get these days.
>
> - Dan C.
>
>
> - Dan C.