Re: [PATCH] generic/448: don't disable extent zeroing if extent_max_zeroout_kb isn't supported

[Eryu Guan <eguan@redhat.com>]

On Wed, Jul 19, 2017 at 10:35:01AM +0800, Xiao Yang wrote:
> On 2017/07/18 21:18, Eryu Guan wrote:
> > On Tue, Jul 18, 2017 at 04:27:50PM +0800, Xiao Yang wrote:
> > > On some old kernel(e.g. v3.5), this case fails because it can not create
> > > extent_max_zeroout_kb file, as below:
> > >    Silence is golden
> > >   +./tests/generic/448: line 54: /sys/fs/ext4/sda5/extent_max_zeroout_kb: No such file or directory
> > >    seek sanity check failed!
> > > 
> > > The extent_max_zeroout_kb file is introduced by:
> > >    67a5da564f97('ext4: make the zero-out chunk size tunable')
> > This is available since v3.7-rc1 kernel
> > $ git describe --contains 67a5da564f97
> > v3.7-rc1~91^2~63
> > 
> > But ext4 SEEK_DATA/SEEK_HOLE support was added in v3.8-rc1
> > $ git describe --contains c8c0df241cc2
> > v3.8-rc1~89^2~40
> > 
> > IMO, you shouldn't hit this error at all, because test won't pass
> > _require_seek_data_hole rule. (Unless you're testing some customized
> > kernel with seek_data/hole backported but not that ext4 sysfs entry.)
> Hi Eryu
> 
> Thanks for your explanation.
> 
> I test it in v3.5 kernel without seek_data/hole backported, please see the
> following message:
> [root@localhost xfstests]# uname -r
> 3.5.0
> [root@localhost xfstests]# src/seek_sanity_test -t
> /mnt/xfstests/test/testfile 2>&1
> File system magic#: 0xef53
> Allocation size: 4096
> File system supports the default behavior.

OK, I see why the test was *not* _notrun on 3.5 kernel now (without
seek_data/hole support added to ext4, but with SEEK_DATA/SEEK_HOLE flags
recognized by vfs, which was introduced by 982d81658 in v3.1-rc1),
because ext4 had the "default behavior" support, and that means "just
return i_size for SEEK_HOLE and will return the same offset for
SEEK_DATA".

> File system does not support unwritten extents.

This means the file system doesn't treat unwritten extents as a hole,
but data, which has nothing to do with SEEK_DATA/HOLE support status.

> 
> _require_seek_data_hole could not catch "File system does not support" if
> ext4 SEEK_DATA/SEEK_HOLE
> support was not added.
> 
> I think we should update  _require_seek_data_hole to catch both "Kernel does
> not support" and
> "File system does not support".   This case could be skipped when meetting
> either of them.

So this is wrong, IMO.

Thanks,
Eryu