Re: [PATCH v2] common/log: add -l su option in _xfs_log_config

From: "Darrick J. Wong" <darrick.wong@oracle.com>
To: Eryu Guan <guaneryu@gmail.com>
Cc: Yang Xu <xuyang2018.jy@cn.fujitsu.com>, fstests@vger.kernel.org
Subject: Re: [PATCH v2] common/log: add -l su option in _xfs_log_config
Date: Sun, 23 Feb 2020 08:37:10 -0800	[thread overview]
Message-ID: <20200223163710.GL9504@magnolia> (raw)
In-Reply-To: <20200223142208.GD3840@desktop>

On Sun, Feb 23, 2020 at 10:22:10PM +0800, Eryu Guan wrote:
> Hi Darrick,
> 
> On Thu, Feb 20, 2020 at 01:38:01PM +0800, Yang Xu wrote:
> > Currently, if we don't specify -l sunit or -l su option, mkfs.xfs
> > will get the stripe size from underlying device.
> > 
> > It works file on most situations. But on some machine, the size of
> > underlying device greater than logbsize of mount options, it will
> > report error like "logbuf size must be greater than or equal to log
> > stripe size". We can specify -l su=4096 to meet this requirement and
> > case can still run normally.
> > 
> > Also, from xfs manpage, version 2 also supports 16k log buf size for
> > mount option and case passed(only generic/054,055 used this api) on
> > my machine. So delete 32k and 64k with different sunit to be consistented

I don't understand why there's a need to be consistent, which means I
don't understand why we'd "delete 32k and 64k with different sunit".
ext4 tests should not be invoking _xfs_log_config()

> > with ext4 test num(10) and we can test all logbuf size.

Hm?  ext4/010 is an inode bitmap fuzz test.

> > 
> > Signed-off-by: Yang Xu <xuyang2018.jy@cn.fujitsu.com>
> 
> Would you like to review this v2 patch again? I feel more confident if
> xfs maintainer could ack it :)
> 
> Thanks,
> Eryu
> 
> > ---
> >  common/log | 12 ++++++------
> >  1 file changed, 6 insertions(+), 6 deletions(-)
> > 
> > diff --git a/common/log b/common/log
> > index c7921f50..9b5a2f6d 100644
> > --- a/common/log
> > +++ b/common/log
> > @@ -546,15 +546,15 @@ _xfs_log_config()
> >  {
> >      echo "# mkfs-opt             mount-opt"
> >      echo "# ------------------------------"
> > -    echo "  version=2            logbsize=32k"
> > +    echo "  version=2,su=4096    logbsize=16k"

The straight substitutions look fine to me, but then...

> > +    echo "  version=2,su=16k     logbsize=16k"

...I can't tell why we're adding this extra case here, or why this
has to be here and not in a separate patch justifying the addition?

> >      echo "  version=2,su=4096    logbsize=32k"
> > -    echo "  version=2,su=32768   logbsize=32k"
> > -    echo "  version=2,su=32768   logbsize=64k"

...I also don't think it's a good idea to reduce test coverage, and
definitely not buried in something that sounds like a fix patch.

--D

> > -    echo "  version=2            logbsize=64k"
> > +    echo "  version=2,su=32k     logbsize=32k"
> > +    echo "  version=2,su=4096    logbsize=64k"
> >      echo "  version=2,su=64k     logbsize=64k"
> > -    echo "  version=2            logbsize=128k"
> > +    echo "  version=2,su=4096    logbsize=128k"
> >      echo "  version=2,su=128k    logbsize=128k"
> > -    echo "  version=2            logbsize=256k"
> > +    echo "  version=2,su=4096    logbsize=256k"
> >      echo "  version=2,su=256k    logbsize=256k"
> >  }
> >  
> > -- 
> > 2.18.0
> > 
> > 
> > 