From: Theodore Ts'o Subject: Re: ext4: journal has aborted Date: Sat, 5 Jul 2014 16:36:07 -0400 Message-ID: <20140705203607.GJ11103@thunk.org> References: <20140703134338.GE2374@thunk.org> <20140703161551.5fd13245@archvile> <87tx6yzdxz.fsf@openvz.org> <20140704114031.2915161a@archvile> <87r421zavi.fsf@openvz.org> <20140704132802.0d43b1fc@archvile> <20140704122022.GC10514@thunk.org> <20140704154559.026331ec@archvile> <20140704184539.GA11103@thunk.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Dmitry Monakhov , Matteo Croce , "Darrick J. Wong" , linux-ext4@vger.kernel.org, Pavel Machek To: David Jander Return-path: Received: from imap.thunk.org ([74.207.234.97]:46973 "EHLO imap.thunk.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932687AbaGEUgW (ORCPT ); Sat, 5 Jul 2014 16:36:22 -0400 Content-Disposition: inline In-Reply-To: <20140704184539.GA11103@thunk.org> Sender: linux-ext4-owner@vger.kernel.org List-ID: I've been auditing the ext4 code looking for a potential bug in our journalling code that might result in fs corruption after a crash/power failure. I did find one such instance: http://patchwork.ozlabs.org/patch/367346/ However, I'm fairly sure this is not the cause of most of the problems folks have been complaining about. For one thing, this bug has been around for a long, long time (since v3.5 or so). For another, hitting it is incredibly rare. It would only be noticeable if you crash immediately (within 5 seconds or so) after allocating the first inode in a block group where both the inode allocation bitmap and block allocation bitmap are uninitialized (i.e, no inodes or blocks have ever been allocated from that block group since the file system was first initialized). So for people who are indicating that they are seeing problems much more frequently in 3.15, but not in older kernels, it's almost certainly something else going on. But if you can reliably reproduce the problem, you might want to try applying the patch. It certainly won't hurt, and if it does help, please let me know. :-) Cheers, - Ted