Re: [PATCH] nfsd4: fix seqid_mutating_error

["J. Bruce Fields" <bfields@fieldses.org>]

On Tue, Aug 23, 2011 at 03:58:05PM -0700, Myklebust, Trond wrote:
> > -----Original Message-----
> > From: J. Bruce Fields [mailto:bfields@fieldses.org]
> > diff --git a/include/linux/nfs4.h b/include/linux/nfs4.h
> > index 76f99e8..b875b03 100644
> > --- a/include/linux/nfs4.h
> > +++ b/include/linux/nfs4.h
> > @@ -373,6 +373,22 @@ enum nfsstat4 {
> >  	NFS4ERR_DELEG_REVOKED	= 10087,	/* deleg./layout revoked
> */
> >  };
> > 
> > +static inline bool seqid_mutating_err(u32 err)
> > +{
> > +	/* rfc 3530 section 8.1.5: */
> > +	switch (err) {
> > +	case NFS4ERR_STALE_CLIENTID:
> > +	case NFS4ERR_STALE_STATEID:
> > +	case NFS4ERR_BAD_STATEID:
> > +	case NFS4ERR_BAD_SEQID:
> > +	case NFS4ERR_BADXDR:
> > +	case NFS4ERR_RESOURCE:
> > +	case NFS4ERR_NOFILEHANDLE:
> > +		return false;
> > +	};
> > +	return true;
> > +}
> > +
> >  /*
> >   * Note: NF4BAD is not actually part of the protocol; it is just used
> >   * internally by nfsd.
> 
> Should we make it take a 's32' rather than a 'u32' argument? We do
> usually expect those errors to be signed.

On the server side it's unsigned--bigendian in fact.  The caller is:

> > +	if (seqid_mutating_err(ntohl(nfserr)) && stateowner) { 	\

I put this in nfs4.h which is mostly protocol constants, and I figure
u32 is the type closest to the relevant xdr, and we should convert as
necessary in callers.  I dunno.

--b.