MIME-Version: 1.0
References: <20211219013803.324724-1-trondmy@kernel.org> <20211219013803.324724-2-trondmy@kernel.org>
 <20211219013803.324724-3-trondmy@kernel.org> <20211219013803.324724-4-trondmy@kernel.org>
 <20211219013803.324724-5-trondmy@kernel.org> <20211219013803.324724-6-trondmy@kernel.org>
In-Reply-To: <20211219013803.324724-6-trondmy@kernel.org>
From:   Daire Byrne <daire@dneg.com>
Date:   Wed, 5 Jan 2022 16:10:36 +0000
Message-ID: <CAPt2mGPO1z3+Cr1FmQfwsKCa4hdT=2eczTqDxCjryt-y5Rm6yw@mail.gmail.com>
Subject: Re: [PATCH v2 05/10] nfs: Add export support for weak cache
 consistency attributes
To:     trondmy@kernel.org
Cc:     Chuck Lever <chuck.lever@oracle.com>,
        "J. Bruce Fields" <bfields@redhat.com>, linux-nfs@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi,

I was interested in testing these patches with our re-export
workloads. However, it looks like this specific patch and the previous
one (nfsd: Distinguish between required and optional NFSv3 post-op
attributes) are breaking re-exports in such a way as to cause
applications to randomly crash out with sigbus errors.

At a guess, loading applications and memory mapping data files are
particularly good at triggering this. I can see no errors logged on
either the re-export server or the eventual client (e.g. stale
filehandles). We mostly re-export NFSv4.2 servers to NFSv3 clients but
we do also have a few NFSv3 servers re-exported as NFSv3  too
(Netapps).

This only happens with patches 4 + 5 and vanilla 5.16-rc6 does not
have any issues. I haven't had a chance to dig much deeper yet, but
thought I'd flag it anyway.

Daire

On Sun, 19 Dec 2021 at 01:44, <trondmy@kernel.org> wrote:
>
> From: Trond Myklebust <trond.myklebust@primarydata.com>
>
> Allow knfsd to request weak cache consistency attributes on files that
> have delegations and/or have up to date attribute caches by propagating
> the information to NFS that the attributes being requested are optional.