Re: [PATCH] knfsd: nfsd: Handle ERESTARTSYS from syscalls.

[Jeff Layton <jlayton@redhat.com>]

On Mon, 16 Jun 2008 12:54:46 -0400
Trond Myklebust <trond.myklebust@fys.uio.no> wrote:

> On Mon, 2008-06-16 at 11:09 -0400, Chuck Lever wrote:
> 
> > I think an error reply is much better than no reply in nearly every
> > case.  NFS3ERR_JUKEBOX/NFS4ERR_DELAY is an interesting idea, but
> > something else again will probably be required for v4.1 with sessions.
> 
> NFS3ERR_JUKEBOX/NFS4ERR_DELAY may be inappropriate if the nfs daemon has
> already started handling the RPC call, since you may be interrupting a
> non-idempotent operation.
> 

But if you drop the reply, the client will probably still end up
retransmitting the request. It seems like a JUKEBOX/DELAY error
is at least a defined error to the client instead of leaving it
guessing. Either way, the client could still end up on the wrong
side of a non-idempotent op.

Also, am I right that this should really only be happening if nfsd
catches a SIGKILL? All other signals should be masked off when
doing the local file operation. Or do we have a potential race if
we catch a signal just after svc_recv returns but before the new
sigmask is set and svc_process is called?

> The only complete solution to this problem is NFSv4.1 with persistent
> sessions.
> 

That's probably the case, though we should probably try to do
best-effort for earlier versions.

-- 
Jeff Layton <jlayton@redhat.com>