Re: NFS bug (2.1.42..2.1.48)

Steven N. Hirsch (shirsch@ibm.net)
Thu, 7 Aug 1997 19:06:05 -0400 (EDT)


On 7 Aug 1997, Claus-Justus Heine wrote:

> hpa@transmeta.com (H. Peter Anvin) writes:
>
> >
> > It seems all versions of Linux from 2.1.42 on has had a problem that
> > it doesn't see a file that is created on an NFS system *by the same
> > system* immediately, but only some 5 seconds later. Now, I know the
> > network coherency issues are tricky, but it is a pretty severe bug for
> > the *same* system that creates the file not to see it immediately.
> > This, among other things, breaks configure (and has eaten my .newsrc
> > file far too many times :()
>
>
> Mmh. Are you really sure?
>
> I use 2.1.48 with my private patches to swap via NFS, root FS via NFS,
> everything else via NFS. When I do a
>
> touch /tmp/foo
>
> I really see /tmp/foo AT ONCE, no problem. At least your "doesn't see
> the created file at once" needs to be a bit more explained?
>

Verrryy interesting. I thought it was just me, but I had to give up on
the newer NFS client and server for similar reasons. One thing that
_really_ triggered it was running a large patch on the client against
kernel source living on the server. Dozens of ".nfsd....." files would
appear (and stay) throughout the hierarchy. After all was done, and with
no errors reported by patch, a build would fail miserably. When I took a
closer look, many files which were reported as being patched had their
original contents!

Strange. I gave up the 2.1.x series and went back to 2.0.30 with the
original NFS support. No problems at all (except it's slow).

Steve