From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1763342AbXGaRRN (ORCPT ); Tue, 31 Jul 2007 13:17:13 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1756674AbXGaRQ6 (ORCPT ); Tue, 31 Jul 2007 13:16:58 -0400 Received: from filer.fsl.cs.sunysb.edu ([130.245.126.2]:46861 "EHLO filer.fsl.cs.sunysb.edu" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754210AbXGaRQ4 (ORCPT ); Tue, 31 Jul 2007 13:16:56 -0400 Date: Tue, 31 Jul 2007 13:16:44 -0400 From: Josef Sipek To: Mark Williamson Cc: Jan Blunck , linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org, Bharata B Rao Subject: Re: [RFC 12/26] ext2 white-out support Message-ID: <20070731171644.GB27234@filer.fsl.cs.sunysb.edu> References: <20070730161323.100048969@weierstrass.suse.de> <20070730161324.261652101@weierstrass.suse.de> <20070731163656.GC22350@filer.fsl.cs.sunysb.edu> <200707311803.08698.mark.williamson@cl.cam.ac.uk> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <200707311803.08698.mark.williamson@cl.cam.ac.uk> User-Agent: Mutt/1.4.1i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Tue, Jul 31, 2007 at 06:03:06PM +0100, Mark Williamson wrote: > > Really the only sane way of keeping track of whiteouts seems some external > > store. We did an experiment with Unionfs, and moving the whiteout handling > > to effectively a "library" that did all the dirty work cleaned up the code > > considerably [2,3]. > > What about keeping track of whiteouts in a special file (or files) in the top > level filesystem of the union? For instance, having a /.whiteouts file at > the root of the top FS in the stack, instead of storing union-specific data > in the flags / inode numbers of the lower levels. What is needed is a "filesystem" that has all the directory bits only. For ODF, we opted to "abuse" existing filesystems to see if it actually helped Unionfs, and I think it did help. Really, now what we (unionfs) need is a cleanup of the ODF code, with a bit better defined interface. ... > This might avoid requiring a store external to the stack of filesystems and I > believe it would solve the problem with shared branches and arbitrary > stacking that you described? We generally did a loopback mount on a file. Very similar to your idea. > I guess a rather similar effect could be had by somehow storing loopback > mountable ODF filesystems in the top layer of a union somewhere (e.g. with > the default path /.odf) and allowing the user to specify an alternate > location at mount time if necessary. So maybe these approaches are quite > similar after all... Very :) We forced the user to mount the fs in the odf loopback manually, but there's no reason why we couldn't do an in-kernel mount on unionfs mount time. Josef 'Jeff' Sipek. -- Once you have their hardware. Never give it back. (The First Rule of Hardware Acquisition)