[PATCH 29/29] nfsd: only allow filesystems that set EXPORT_OP_STABLE_HANDLES

Amir Goldstein amir73il at gmail.com
Sat Jan 17 01:46:50 AEDT 2026


On Fri, Jan 16, 2026 at 1:36 PM Jeff Layton <jlayton at kernel.org> wrote:
>
> On Thu, 2026-01-15 at 20:23 +0100, Amir Goldstein wrote:
> > On Thu, Jan 15, 2026 at 6:51 PM Jeff Layton <jlayton at kernel.org> wrote:
> > >
> > > Some filesystems have grown export operations in order to provide
> > > filehandles for local usage. Some of these filesystems are unsuitable
> > > for use with nfsd, since their filehandles are not persistent across
> > > reboots.
> > >
> > > In __fh_verify, check whether EXPORT_OP_STABLE_HANDLES is set
> > > and return nfserr_stale if it isn't.
> > >
> > > Signed-off-by: Jeff Layton <jlayton at kernel.org>
> > > ---
> > >  fs/nfsd/nfsfh.c | 4 ++++
> > >  1 file changed, 4 insertions(+)
> > >
> > > diff --git a/fs/nfsd/nfsfh.c b/fs/nfsd/nfsfh.c
> > > index ed85dd43da18e6d4c4667ff14dc035f2eacff1d6..da9d5fb2e6613c2707195da2e8678b3fcb3d444d 100644
> > > --- a/fs/nfsd/nfsfh.c
> > > +++ b/fs/nfsd/nfsfh.c
> > > @@ -334,6 +334,10 @@ __fh_verify(struct svc_rqst *rqstp,
> > >         dentry = fhp->fh_dentry;
> > >         exp = fhp->fh_export;
> > >
> > > +       error = nfserr_stale;
> > > +       if (!(dentry->d_sb->s_export_op->flags & EXPORT_OP_STABLE_HANDLES))
> > > +               goto out;
> > > +
> > >         trace_nfsd_fh_verify(rqstp, fhp, type, access);
> > >
> >
> > IDGI. Don't you want  to deny the export of those fs in check_export()?
> > By the same logic that check_export() checks for can_decode_fh()
> > not for can_encode_fh().
> >
>
> It certainly won't hurt to add a check for this to check_export(), and
> I've gone ahead and done so. To be clear, doing that won't prevent the
> filesystem from being exported, but you will get a warning like this
> when you try:
>
>     exportfs: /sys/fs/cgroup does not support NFS export
>
> That export will still show up in mountd though, so this is just a
> warning. Trying to mount it though will fail.
>

Oh, I did not know. What an odd user experience.
Anyway, better than no warning at all.

Thanks,
Amir.


More information about the Linux-erofs mailing list