On Thu, Sep 10, 2020 at 05:25:19PM +0000, Trond Myklebust wrote:
> On Thu, 2020-09-10 at 13:02 -0400, J. Bruce Fields wrote:
> > On Tue, Sep 08, 2020 at 05:19:58PM -0400, trondmy@xxxxxxxxxx wrote:
> > > From: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
> > >
> > > If the mount point check in nfsd_export fails due to a transient
> > > error,
> > > then ignore it to avoid spurious NFSERR_STALE errors being returned
> > > by
> > > knfsd.
> >
> > What sort of transient errors?
> >
> > I guess this makes the upcall (and the original rpc) eventually time
> > out?
>
> The point here is that is_mountpoint() is making the assumption that
> _any_ error from stat() can be used to infer that there is nothing
> mounted. In reality, only a subset of errors allow it to make that
> assumption (specifically the errors ELOOP, ENAMETOOLONG, ENOENT, and
> ENOTDIR).
>
> In our case, we're seeing a problem when the underlying filesystem is a
> soft mounted NFSv4.2 client (i.e. we're running our Hammerspace use
> case of proxying NFSv4.2 to legacy NFSv3 clients) and that NFS4.2 mount
> times out due to a reboot of the underlying server, for instance.
>
> So, yes, in that case we want the upcall to time out instead of poking
> knfsd into declaring that the directory is stale.
Thanks for the explanation, I guess I agree that's the best you can do
in that situation.
--b.
> > > Signed-off-by: Trond Myklebust <trond.myklebust@xxxxxxxxxxxxxxx>
> > > ---
> > > utils/mountd/cache.c | 10 ++++++++--
> > > 1 file changed, 8 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/utils/mountd/cache.c b/utils/mountd/cache.c
> > > index 6cba2883026f..93e868341d15 100644
> > > --- a/utils/mountd/cache.c
> > > +++ b/utils/mountd/cache.c
> > > @@ -1411,7 +1411,10 @@ static void nfsd_export(int f)
> > >
> > > if (mp && !*mp)
> > > mp = found->m_export.e_path;
> > > - if (mp && !is_mountpoint(mp))
> > > + errno = 0;
> > > + if (mp && !is_mountpoint(mp)) {
> > > + if (errno != 0 && !path_lookup_error(errno))
> > > + goto out;
> > > /* Exportpoint is not mounted, so tell kernel
> > > it is
> > > * not available.
> > > * This will cause it not to appear in the V4
> > > Pseudo-root
> > > @@ -1420,9 +1423,12 @@ static void nfsd_export(int f)
> > > * And filehandle for this mountpoint from an
> > > earlier
> > > * mount will block in nfsd.fh lookup.
> > > */
> > > + xlog(L_WARNING,
> > > + "Cannot export path '%s': not a
> > > mountpoint",
> > > + path);
> > > dump_to_cache(f, buf, sizeof(buf), dom, path,
> > > NULL, 60);
> > > - else if (dump_to_cache(f, buf, sizeof(buf), dom, path,
> > > + } else if (dump_to_cache(f, buf, sizeof(buf), dom,
> > > path,
> > > &found->m_export, 0) < 0) {
> > > xlog(L_WARNING,
> > > "Cannot export %s, possibly unsupported
> > > filesystem"
> > > --
> > > 2.26.2
> --
> Trond Myklebust
> CTO, Hammerspace Inc
> 4984 El Camino Real, Suite 208
> Los Altos, CA 94022
> www.hammer.space
>
