Re: [PATCH v1 00/42] Update NFSD XDR functions

[Trond Myklebust <trondmy@hammerspace.com>]

On Fri, 2021-01-08 at 11:01 -0500, Bruce Fields wrote:
> On Fri, Jan 08, 2021 at 10:56:14AM -0500, Chuck Lever wrote:
> > 
> > 
> > > On Jan 8, 2021, at 10:52 AM, Bruce Fields <bfields@fieldses.org>
> > > wrote:
> > > 
> > > On Fri, Jan 08, 2021 at 10:50:09AM -0500, Chuck Lever wrote:
> > > > 
> > > > 
> > > > > On Jan 7, 2021, at 10:18 PM, bfields@fieldses.org wrote:
> > > > > 
> > > > > I haven't had a chance to review these, but thought I should
> > > > > mention I'm
> > > > > seeing a failure in xfstests generic/465 that I don't *think*
> > > > > is
> > > > > reproduceable before this series.  Unfortunately it's
> > > > > intermittent,
> > > > > though, so I'm not certain yet.
> > > > 
> > > > Confirming: does that failure occur with NFSv3?
> > > 
> > > I've only tried it over 4.2.
> > 
> > Interesting. This series shouldn't have any impact on NFSv4
> > direct I/O functionality:
> > 
> > fs/nfs_common/nfsacl.c          |  52 +++
> > fs/nfsd/nfs2acl.c               |  62 ++--
> > fs/nfsd/nfs3acl.c               |  42 ++-
> > fs/nfsd/nfs3proc.c              |  71 +++--
> > fs/nfsd/nfs3xdr.c               | 538 ++++++++++++++++++-----------
> > ---
> > fs/nfsd/nfsproc.c               |  74 +++--
> > fs/nfsd/nfssvc.c                |  34 --
> > fs/nfsd/nfsxdr.c                | 350 ++++++++++-----------
> > fs/nfsd/xdr.h                   |  12 +-
> > fs/nfsd/xdr3.h                  |  20 +-
> > include/linux/nfsacl.h          |   3 +
> > include/linux/sunrpc/msg_prot.h |   3 -
> > include/linux/sunrpc/xdr.h      |  13 +-
> > include/trace/events/sunrpc.h   |  15 +-
> > include/uapi/linux/nfs3.h       |   6 +
> > 15 files changed, 680 insertions(+), 615 deletions(-)
> > 
> > Can you try to nail it down a little?
> 
> I took a look back through my testing history and realized I've seen
> it
> fail previously.  So it was just coincidence that I saw it fail a
> couple
> times after applying the series but not before yesterday.  Sorry for
> the
> noise!
> 
> --b.

Just ignore generic/465. As far as NFS is concerned, the test has
utterly borked assumptions about O_DIRECT ordering.

-- 
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@hammerspace.com