From mboxrd@z Thu Jan 1 00:00:00 1970 From: "H. Peter Anvin" Subject: Re: e2fsck running extremely slowly Date: Sun, 21 Jul 2013 19:35:35 -0700 Message-ID: <51EC9A77.6090109@zytor.com> References: <51EC6480.6000200@zytor.com> <20130722012929.GA30902@thunk.org> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: linux-ext4@vger.kernel.org To: "Theodore Ts'o" Return-path: Received: from terminus.zytor.com ([198.137.202.10]:38873 "EHLO mail.zytor.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752743Ab3GVCf6 (ORCPT ); Sun, 21 Jul 2013 22:35:58 -0400 In-Reply-To: <20130722012929.GA30902@thunk.org> Sender: linux-ext4-owner@vger.kernel.org List-ID: On 07/21/2013 06:29 PM, Theodore Ts'o wrote: > On Sun, Jul 21, 2013 at 03:45:20PM -0700, H. Peter Anvin wrote: >> I have a large filesystem (14 TB) which suffered a RAID failure which >> seems to have corrupted some inodes. Unfortunately as a result there >> are now a number of inodes with "false extents" which result in a very >> large number of multiply claimed blocks. >> >> I have tried to run e2fsck on this filesystem, and it gets as far as >> phase 1D, at which point it starts running at a glacial pace. After 48 >> hours -- most of it sitting at 100% CPU executing no system calls at all >> -- it claims to have processed a single file out of almost 10000. > > What I usually do when I is to look at the inodes that are corrupted > in phases 1b, and examine them using debugfs. If they look insane, > nuke them using the debugfs clri command. > > Yes, this is horribly manual. The long term planned solution is that > the metadata checksum feature will allow us to determine the metadata > is corrupt, and then e2fsck will know which fs metadata it can trust, > and which it will have to discard. > Manual isn't really practical with almost 10,000 reported inodes... -hpa