Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20200821015036.bn3yqiiuvunfxb42@xzhoux.usersys.redhat.com>
 <20200825212647.GB1955@fieldses.org> <20200825215357.GC1955@fieldses.org>
In-Reply-To: <20200825215357.GC1955@fieldses.org>
From:   Olga Kornievskaia <olga.kornievskaia@gmail.com>
Date:   Wed, 26 Aug 2020 11:59:20 -0400
Message-ID: <CAN-5tyH7SdzHVtS2zk5Md7ShmmTneWt8jFgjqNb0Bhdm1o140w@mail.gmail.com>
Subject: Re: 5.9 nfsd update breaks v4.2 copy_file_range
To:     "J. Bruce Fields" <bfields@fieldses.org>
Cc:     Murphy Zhou <jencce.kernel@gmail.com>,
        Chuck Lever <chuck.lever@oracle.com>,
        linux-nfs <linux-nfs@vger.kernel.org>,
        "J. Bruce Fields" <bfields@redhat.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-nfs-owner@vger.kernel.org
Precedence: bulk

On Tue, Aug 25, 2020 at 5:53 PM J. Bruce Fields <bfields@fieldses.org> wrote:
>
> On Tue, Aug 25, 2020 at 05:26:47PM -0400, J. Bruce Fields wrote:
> > On Fri, Aug 21, 2020 at 09:50:36AM +0800, Murphy Zhou wrote:
> > > It's easy to reproduce by running multiple xfstests testcases on localhost
> > > NFS shares. These testcases are:
> > >   generic/430 generic/431 generic/432 generic/433 generic/565
> > >
> > > This reproduces only on NFSv4.2.
> > >
> > > Error log diff sample:
> > >
> > > --- /dev/fd/63      2020-08-09 22:46:02.771745606 -0400
> > > +++ results/generic/431.out.bad     2020-08-09 22:46:02.546745248 -0400
> > > @@ -1,15 +1,22 @@
> > >  QA output created by 431
> > >  Create the original file and then copy
> > > +cmp: EOF on /mnt/testdir/test-431/copy which is empty
> > >  Original md5sums:
> > >  ab56b4d92b40713acc5af89985d4b786  TEST_DIR/test-431/file
> > > -ab56b4d92b40713acc5af89985d4b786  TEST_DIR/test-431/copy
> > > +d41d8cd98f00b204e9800998ecf8427e  TEST_DIR/test-431/copy
> >
> > When I check the files server-side after reproducing, the file "copy"
> > has the correct contents.  So I guess the problem is that the client
> > cache is out of date.  The difference with commit 94415b06e is that the
> > client holds read delegations on both source and destination, throughout
> > the COPY operation.
>
> Olga, do you know what the client's doing in this case?
>
> It seems to me that it should be invalidating its cache of the
> destination range after a COPY, regardless of whether it holds a
> delegation on the destination.  (Either that or updating the cache of
> the destination to hold the copied data, if it's confident its cache of
> the source range is up to date.)

It's on my todo list to reproduce this and see what's going on.

>
> --b.
>
> >
> > --b.
> >
> > >  Small copies from various points in the original file
> > > +cmp: EOF on /mnt/testdir/test-431/a which is empty
> > > +cmp: EOF on /mnt/testdir/test-431/b which is empty
> > > +cmp: EOF on /mnt/testdir/test-431/c which is empty
> > > +cmp: EOF on /mnt/testdir/test-431/d which is empty
> > > +cmp: EOF on /mnt/testdir/test-431/e which is empty
> > > +cmp: EOF on /mnt/testdir/test-431/f which is empty
> > >  md5sums after small copies
> > >  ab56b4d92b40713acc5af89985d4b786  TEST_DIR/test-431/file
> > > -0cc175b9c0f1b6a831c399e269772661  TEST_DIR/test-431/a
> > > -92eb5ffee6ae2fec3ad71c777531578f  TEST_DIR/test-431/b
> > > -4a8a08f09d37b73795649038408b5f33  TEST_DIR/test-431/c
> > > -8277e0910d750195b448797616e091ad  TEST_DIR/test-431/d
> > > -e1671797c52e15f763380b45e841ec32  TEST_DIR/test-431/e
> > > -2015eb238d706eceefc784742928054f  TEST_DIR/test-431/f
> > > +d41d8cd98f00b204e9800998ecf8427e  TEST_DIR/test-431/a
> > > +d41d8cd98f00b204e9800998ecf8427e  TEST_DIR/test-431/b
> > > +d41d8cd98f00b204e9800998ecf8427e  TEST_DIR/test-431/c
> > > +d41d8cd98f00b204e9800998ecf8427e  TEST_DIR/test-431/d
> > > +d41d8cd98f00b204e9800998ecf8427e  TEST_DIR/test-431/e
> > > +d41d8cd98f00b204e9800998ecf8427e  TEST_DIR/test-431/f
> > >  d41d8cd98f00b204e9800998ecf8427e  TEST_DIR/test-431/g
> > >
> > > Bisecting shows the first "bad" commit is:
> > >
> > > commit 94415b06eb8aed13481646026dc995f04a3a534a
> > > Author: J. Bruce Fields <bfields@redhat.com>
> > > Date:   Tue Jul 7 09:28:05 2020 -0400
> > >
> > >     nfsd4: a client's own opens needn't prevent delegations
> > >
> > > I'm wondering if you're already aware of it, this simple report is for
> > > your info.
> > >
> > > Thanks.
> > >
> > > --
> > > Murphy