From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jan Kara Subject: Re: REGRESSION: [PATCH 04/12] ext4: Disable merging of uninitialized extents Date: Thu, 14 Feb 2013 11:47:34 +0100 Message-ID: <20130214104734.GA605@quack.suse.cz> References: <1358510446-19174-1-git-send-email-jack@suse.cz> <1358510446-19174-5-git-send-email-jack@suse.cz> <20130209171015.GC8091@thunk.org> <20130212215832.GA25984@quack.suse.cz> <20130213045759.GE2614@thunk.org> <87ehgku11r.fsf@openvz.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Theodore Ts'o , Jan Kara , linux-ext4@vger.kernel.org To: Dmitry Monakhov Return-path: Received: from cantor2.suse.de ([195.135.220.15]:57535 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752757Ab3BNKrh (ORCPT ); Thu, 14 Feb 2013 05:47:37 -0500 Content-Disposition: inline In-Reply-To: <87ehgku11r.fsf@openvz.org> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Wed 13-02-13 11:26:40, Dmitry Monakhov wrote: > On Tue, 12 Feb 2013 23:57:59 -0500, Theodore Ts'o wrote: > > On Tue, Feb 12, 2013 at 10:58:32PM +0100, Jan Kara wrote: > > > The third patch is a fix which shouldn't cause any issues. So you can > > > take just that one and leave the other two aside until we are able to > > > resolve the issue. > > > > I thought the third patch depending on the first two? Certainly it > > doesn't apply cleanly without the first two patches... > My patch fix old issue, but i've prepared it on top of Jan's patches > only for simplicity. I'll send new version which not depend on his > patches today. > Over-all Jan's statment that split should not happen inside end_io > and it is clear sing of a bug is absolutely right decision. > This helps us to spot several hidden issues (number is still unknown) > so may be it is reasonable to split first patch in two parts: > 1) disable uninitialized extents merging itself. > 2) Print warning if split is required inside end_io(so only warning will > be printed, but w/o data corruption) > 3) Get rid of extent split machinery from end_io (because it is not > longer valid situation) > > (1) and (2) should be accepted ASAP and will help us to spot and fix > other hidden issues. And we fix all related issues it will be safe > to apply (3)'rd one. > I'll send patches soon. I agree with this plan, I just think the warnings are currently too easy to trigger (just random AIO DIO writing seems to trigger them with dioread_nolock) so as Ted said we'll probably have to hide them under EXT4_DEBUG or something like that. I'll try to debug the current problem while you prepare the patches :). Honza -- Jan Kara SUSE Labs, CR