From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from bombadil.infradead.org ([65.50.211.133]:43188 "EHLO bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751340AbdH3HvK (ORCPT ); Wed, 30 Aug 2017 03:51:10 -0400 Date: Wed, 30 Aug 2017 00:51:09 -0700 From: Christoph Hellwig Subject: Re: [PATCH] xfs: skip free eofblocks if inode is under written back Message-ID: <20170830075109.GA4711@infradead.org> References: <20170830042628.5592-1-eguan@redhat.com> <20170830072950.GJ10621@dastard> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20170830072950.GJ10621@dastard> Sender: linux-xfs-owner@vger.kernel.org List-ID: List-Id: xfs To: Dave Chinner Cc: Eryu Guan , linux-xfs@vger.kernel.org On Wed, Aug 30, 2017 at 05:29:50PM +1000, Dave Chinner wrote: > Quick patch below, doesn't work on reflink filesystems yet because > the way it handles cached COW mapping invalidation is unnecessarily > special and so doesn't work. Can you test it and see if it fixes the > problem? FYI, I thought about this problem a while ago in the context of rcu lookups of the extent tree (not, I'm not doing those quite yet, but I though of the possibilities once we've got a btree!) and then my answer was that we'd really need a seqlock around uses of the imaps. I didn't think far enough that we might need something like that even now..