From mboxrd@z Thu Jan 1 00:00:00 1970 From: Dave Chinner Subject: Re: [Lsf-pc] [LSF/MM][ATTEND]filesystem -- reflink Date: Fri, 17 Jan 2014 11:07:38 +1100 Message-ID: <20140117000738.GC18112@dastard> References: <52D635C4.20201@oracle.com> <20140115090312.GA6732@quack.suse.cz> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: mingming cao , linux-fsdevel@vger.kernel.org, lsf-pc@lists.linux-foundation.org To: Jan Kara Return-path: Received: from ipmail07.adl2.internode.on.net ([150.101.137.131]:31537 "EHLO ipmail07.adl2.internode.on.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751379AbaAQAIW (ORCPT ); Thu, 16 Jan 2014 19:08:22 -0500 Content-Disposition: inline In-Reply-To: <20140115090312.GA6732@quack.suse.cz> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Wed, Jan 15, 2014 at 10:03:12AM +0100, Jan Kara wrote: > Hello, > > On Tue 14-01-14 23:16:20, mingming cao wrote: > > I'd like to attend Linux storage and filesystem summit. I am > > interested in discussion of general lockless direct io and more > > interested in discussion of reflink support for filesystem. Btrfs > > and OCFS2 has this support, My goal is explore what's best to way > > this support for ext4. If there is any crossover work between > > filesystem and vfs people, or filesystem and dm layer I am more than > > happy to discuss about it too. > There was a project for implementing COW for ext4 but it was a really > major surgery and in the end didn't get to an upstreamable state. Reflink > is somewhat simpler than general COW because it's only about fs data. In > particular implementing reflink with a file granularity (i.e., a type of > hardlink which is automatically converted to a copy when first opened for > writing) is relatively simple to do but I'm not sure how useful it is. > Doing reflink properly with block granularity is harder with block > refcounting etc. > > Do you have any particular usecase in mind? The use case I keep hearing for reflink in XFS is optimising distributed filesystem snapshot implementations, which tend to require file block level COW semantics on a per-file basis... Cheers, Dave. -- Dave Chinner david@fromorbit.com