Re: attempt to re-implement sillyrename for NFS in 2.1.*

Matthias Urlichs (smurf@lap.noris.de)
27 Aug 1997 04:47:10 +0200


Claus-Justus Heine <Heine@physik.rwth-aachen.de> writes:

> Broken NFS.
> My UGLY hack is included below :-((((
>
This just shows that it was a fundamental mistake to implement stateful
things (i.e., open files which can be renamed but which still stay open)
with a stateless protocol, i.e. NFS.

I hesitate to even speculate what this mess will do in the presence of more
than one client. FIXME. In other words, forget it, this won't work, it
can't work; we're going to make a reasonable effort here BUT the
reasonableness should stop well before the point where it would be easier
to go ahead and implement a new-and-improved (hell, _anything_ must be
better than this mess) stateful file sharing protocol from scratch.

I'd define "reasonable" as "'tail -f FOO & rm FOO' has a chance of working".
Think not about renames, nor hardlinks, it's not worth the bother. IMHO.

NB: _Is_ anybody working on a usable stateful shared file system for Linux?

-- 
Matthias Urlichs