[RFC] Unicode/UTF-8 support for XFS

[Dave Chinner]

On Mon, Sep 15, 2014 at 09:16:24AM +0200, Olaf Weber wrote:
> On 12-09-14 22:55, Christoph Hellwig wrote:
> >On Fri, Sep 12, 2014 at 01:55:35PM +0200, Olaf Weber wrote:
> >>I looked up those discussions in the archives.  For example, here's
> >>Christoph about rejecting filenames if they're not well-formed unicode.
> >>    http://marc.info/?l=linux-fsdevel&m=120876935526856&w=2
> >>And Jamie Lokier making a similar point:
> >>    http://oss.sgi.com/archives/xfs/2008-04/msg01263.html
> >
> >And I might now disagree with my past self.  While non-ut8 characters
> >are perfectly valid unix filenames, and I think everyones life is easier
> >if we generally stay out of the utf8 business it seems that for this
> >particular use case (shared filesystem with Windows, right) just
> >accepting utf8 should be fine.  ZFS is doing, MacOS X apparently is,
> >and NFSv4 requires it, although as far as I know most implementations
> >ignore that requirement.
> >
> 
> One issue is working in environments that are not UTF-8 clean.  For
> example, unpacking a tarball with non-UTF-8 filenames in it. The
> names would have to be transcoded, which is only really possible if
> you know the original character set. And if the filesystem flat out
> rejects non-UTF-8 filenames, then you'd be unable to unpack the
> tarball at all.

So how do existing utf8/unicode enabled filesystems handle this? 

I think we should be consistent with ZFS, MacOS and others that
already deal with this problem if at all possible. However, this
really is a wider policy decision for the kernel/VFS as we want
consistent behaviour across all linux filesystems, hence this
patchset really needs to discussed at the lkml/-fsdevel level...

Cheers,

Dave.
-- 
Dave Chinner
david at fromorbit.com