[PLUG] How to umount NFS hung mouts

[Steven A. Adams]

On Thu, 2003-08-14 at 13:52, Felix Lee wrote:
> "Steven A. Adams" <stevea at nwtechops.com>:
> > Implicating the filesystem protocol because an application is not
> > properly using that protocol? You have to admit, that's really
> > stretching things.
> 
> The programmers were properly using the "unix filesystem
> protocol" (open, read, write, etc).  The issue is that adding
> "nfs behavior" to the "unix filesystem protocol" changes the
> protocol in ways that programmers may not have originally
> considered.  nfs implementations go to a _lot_ of trouble to try
> to conform to the unix protocol, except for some unavoidable
> problems like close(), but these guarantees apply to hard mounts,
> not soft mounts.
> 
> Look, programming is sometimes like mathematics, and statements
> like "this program works correctly" or "this program is secure"
> are similar to mathematical proofs.  The proofs become invalid if
> you change any of the axioms.  Using soft mounts instead of hard
> mounts is like saying "ok, I'm changing '+' so that 2 + 2 = 0
> when everyone in the room shouts 'fire!, but that's so rare that
> everything should still work."  Logic systems just fall apart
> when you make changes like that.  Every previously proven
> statement becomes questionable.  You have to do new proofs, and I
> don't know that anyone's done that for soft nfs mounts.  Someone
> should go look.
> --

I understand and what you are saying and it makes perfect sense.
However, none of this changes the fact that soft mounts are part of NFS.
Ignoring the additional code in an application that addresses
filesystems, including NFS, is not the fault of NFS. If you are going to
design and interface to RPC for an application you would not ignore the
portmapper service and expect the application to function properly.

None of this is really relevant to the question of when to and when not
to use soft mounts though. It looks like we have digressed again.
Excellent argument though :)