From mboxrd@z Thu Jan  1 00:00:00 1970
Received: from mx2.suse.de ([195.135.220.15]:54671 "EHLO mx2.suse.de"
        rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
        id S933563AbeBVSRY (ORCPT <rfc822;fstests@vger.kernel.org>);
        Thu, 22 Feb 2018 13:17:24 -0500
Date: Thu, 22 Feb 2018 18:17:31 +0000
From: Luis Henriques <lhenriques@suse.com>
Subject: Re: [PATCH v2 6/8] fsstress: implement the clonerange/deduperange
 ioctls
Message-ID: <20180222181731.jlddocrjlaaauaf3@hermes.olymp>
References: <151314499003.18893.8687182548758898133.stgit@magnolia>
 <151314503583.18893.15475795025536691678.stgit@magnolia>
 <20171215020731.GM6896@magnolia>
 <20180222160614.t2afwtdsouxliqie@hermes.olymp>
 <20180222172741.GD9827@magnolia>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
In-Reply-To: <20180222172741.GD9827@magnolia>
Sender: fstests-owner@vger.kernel.org
Content-Transfer-Encoding: quoted-printable
To: "Darrick J. Wong" <darrick.wong@oracle.com>
Cc: eguan@redhat.com, linux-xfs@vger.kernel.org, fstests@vger.kernel.org
List-ID: <fstests@vger.kernel.org>

On Thu, Feb 22, 2018 at 09:27:41AM -0800, Darrick J. Wong wrote:
> On Thu, Feb 22, 2018 at 04:06:14PM +0000, Luis Henriques wrote:
> > On Thu, Dec 14, 2017 at 06:07:31PM -0800, Darrick J. Wong wrote:
> >=20
> > <snip>
> >=20
> > > +void
> > > +clonerange_f(
> > > +	int			opno,
> > > +	long			r)
> > > +{
> >=20
> > <snip>
> >=20
> > > +	/* Calculate offsets */
> > > +	len =3D (random() % FILELEN_MAX) + 1;
> > > +	len &=3D ~(stat1.st_blksize - 1);
> > > +	if (len =3D=3D 0)
> > > +		len =3D stat1.st_blksize;
> > > +	if (len > stat1.st_size)
> > > +		len =3D stat1.st_size;
> > > +
> > > +	lr =3D ((__int64_t)random() << 32) + random();
> > > +	if (stat1.st_size =3D=3D len)
> > > +		off1 =3D 0;
> > > +	else
> > > +		off1 =3D (off64_t)(lr % MIN(stat1.st_size - len, MAXFSIZE));
> > > +	off1 %=3D maxfsize;
> > > +	off1 &=3D ~(stat1.st_blksize - 1);
> > > +
> > > +	/*
> > > +	 * If srcfile =3D=3D destfile, randomly generate destination rang=
es
> > > +	 * until we find one that doesn't overlap the source range.
> > > +	 */
> > > +	do {
> > > +		lr =3D ((__int64_t)random() << 32) + random();
> > > +		off2 =3D (off64_t)(lr % MIN(stat2.st_size + (1024 * 1024), MAXFS=
IZE));
> > > +		off2 %=3D maxfsize;
> > > +		off2 &=3D ~(stat2.st_blksize - 1);
> > > +	} while (stat1.st_ino =3D=3D stat2.st_ino && llabs(off2 - off1) <=
 len);
> >=20
> > I started seeing hangs in generic/013 on cephfs.  After spending some
> > time looking, I found that this loops forever.  And the reason seems =
to
> > be that stat1.st_blksize is too big for this filesystem (4M) -- when
> > doing:
>=20
> "Too big for this filesystem"?
>=20
> Uh... maybe you'd better start by giving me more stat buffer info --
> what's st_size?
>=20
> > 	off1 &=3D ~(stat1.st_blksize - 1);
>=20
> These bits round the start offset down to block granularity, since clon=
e
> range implementations generally require that the ranges align to block
> boundaries.
>=20
> (Though AFAICT ceph doesn't support clone range anyway...)
>=20
> So reading between the lines, is the problem here that ceph advertises =
a
> blocksize of 4M and fsstress calls clonerange_f with files that are
> smaller than 4M in size, so the only possible offsets with a 4M
> blocksize are zero and that's why we end up looping forever?

Brilliantly described!  That is *exactly* what I'm seeing and failed to
describe.  I guess I could use FSSTRESS_AVOID to work around this issue,
but there are probably better options.

Cheers,
--
Lu=EDs

>=20
> --D
>=20
> >=20
> > off1 (and off2) will both end up with 0.  Does this make sense?  Woul=
d
> > something like:
> >=20
> > -	off1 &=3D ~(stat1.st_blksize - 1);
> > +	if (stat1.st_blksize <=3D stat1.st_size)
> > +		off1 &=3D ~(stat1.st_blksize - 1);
> >=20
> > be acceptable?  (and a similar change for off2, of course.)
>=20
> > Cheers,
> > --
> > Lu=EDs
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-xfs" =
in
> > the body of a message to majordomo@vger.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
>=20