Re: [PATCH v2 0/9] Various NFSv4 state error handling fixes

From: Trond Myklebust <trondmy@hammerspace.com>
To: "aglo@umich.edu" <aglo@umich.edu>
Cc: "linux-nfs@vger.kernel.org" <linux-nfs@vger.kernel.org>,
	"Anna.Schumaker@netapp.com" <Anna.Schumaker@netapp.com>
Subject: Re: [PATCH v2 0/9] Various NFSv4 state error handling fixes
Date: Thu, 19 Sep 2019 01:49:09 +0000	[thread overview]
Message-ID: <4c620de7a944ff38644eceb4fed863f5092f1a15.camel@hammerspace.com> (raw)
In-Reply-To: <CAN-5tyF27z=+3tbU1De_wR0aosiczn67dNanBmBe4icj=uAYwQ@mail.gmail.com>

Hi Olga

On Wed, 2019-09-18 at 15:38 -0400, Olga Kornievskaia wrote:
> Hi Trond,
> 
> These set of patches do not address the locking problem. It's
> actually
> not the locking patch (which I thought it was as I reverted it and
> still had the issue). Without the whole patch series the unlock works
> fine so something in these new patches. Something is up with the 2
> patches:
> NFSv4: Handle NFS4ERR_OLD_STATEID in CLOSE/OPEN_DOWNGRADE
> NFSv4: Handle NFS4ERR_OLD_STATEID in LOCKU
> 
> If I remove either one separately, unlock fails but if I remove both
> unlock works.

Can you describe how you are testing this, and perhaps provide
wireshark traces that show how we're triggering these problems?

Thanks!
  Trond

> On Mon, Sep 16, 2019 at 4:46 PM Trond Myklebust <trondmy@gmail.com>
> wrote:
> > Various NFSv4 fixes to ensure we handle state errors correctly. In
> > particular, we need to ensure that for COMPOUNDs like CLOSE and
> > DELEGRETURN, that may have an embedded LAYOUTRETURN, we handle the
> > layout state errors so that a retry of either the LAYOUTRETURN, or
> > the later CLOSE/DELEGRETURN does not corrupt the LAYOUTRETURN
> > reply.
> > 
> > Also ensure that if we get a NFS4ERR_OLD_STATEID, then we do our
> > best to still try to destroy the state on the server, in order to
> > avoid causing state leakage.
> > 
> > v2: Fix bug reports from Olga
> >  - Try to avoid sending old stateids on CLOSE/OPEN_DOWNGRADE when
> >    doing fully serialised NFSv4.0.
> >  - Ensure LOCKU initialises the stateid correctly.
> > 
> > Trond Myklebust (9):
> >   pNFS: Ensure we do clear the return-on-close layout stateid on
> > fatal
> >     errors
> >   NFSv4: Clean up pNFS return-on-close error handling
> >   NFSv4: Handle NFS4ERR_DELAY correctly in return-on-close
> >   NFSv4: Handle RPC level errors in LAYOUTRETURN
> >   NFSv4: Add a helper to increment stateid seqids
> >   pNFS: Handle NFS4ERR_OLD_STATEID on layoutreturn by bumping the
> > state
> >     seqid
> >   NFSv4: Fix OPEN_DOWNGRADE error handling
> >   NFSv4: Handle NFS4ERR_OLD_STATEID in CLOSE/OPEN_DOWNGRADE
> >   NFSv4: Handle NFS4ERR_OLD_STATEID in LOCKU
> > 
> >  fs/nfs/nfs4_fs.h   |  11 ++-
> >  fs/nfs/nfs4proc.c  | 204 ++++++++++++++++++++++++++++++-----------
> > ----
> >  fs/nfs/nfs4state.c |  16 ----
> >  fs/nfs/pnfs.c      |  71 ++++++++++++++--
> >  fs/nfs/pnfs.h      |  17 +++-
> >  5 files changed, 229 insertions(+), 90 deletions(-)
> > 
> > --
> > 2.21.0
> > 
-- 
Trond Myklebust
Linux NFS client maintainer, Hammerspace
trond.myklebust@hammerspace.com