Received-SPF: softfail (google.com: domain of transitioning linux-kernel-owner@vger.kernel.org does not designate 23.128.96.19 as permitted sender) client-ip=23.128.96.19;
Date:   Tue, 19 Apr 2022 23:27:33 +0200
From:   Max Kellermann <mk@cm4all.com>
To:     David Howells <dhowells@redhat.com>
Cc:     Max Kellermann <mk@cm4all.com>, linux-cachefs@redhat.com,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: fscache corruption in Linux 5.17?
Message-ID: <Yl8pRVSW9y1o6MBV@rabbit.intern.cm-ag>
References: <Yl7d++G25sNXIR+p@rabbit.intern.cm-ag>
 <YlWWbpW5Foynjllo@rabbit.intern.cm-ag>
 <507518.1650383808@warthog.procyon.org.uk>
 <509961.1650386569@warthog.procyon.org.uk>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <509961.1650386569@warthog.procyon.org.uk>
Precedence: bulk

On 2022/04/19 18:42, David Howells <dhowells@redhat.com> wrote:
> Could the file have been modified by a third party?  If you're using NFS3
> there's a problem if two clients can modify a file at the same time.  The
> second write can mask the first write and the client has no way to detect it.
> The problem is inherent to the protocol design.  The NFS2 and NFS3 protocols
> don't support anything better than {ctime,mtime,filesize} - the change
> attribute only becomes available with NFS4.

I tried to write a script to stress-test writing and reading, but
found no clue so far.  I'll continue that tomorrow.

My latest theory is that this is a race condition; what if one process
writes to the file, which invalidates the cache; then, in the middle
of invalidating the local cache and sending the write to the NFS
server, another process (on the same server) reads the file; what
modification time and what data will it see?  What if the cache gets
filled with old data, while new data to-be-written is still in flight?

Max