From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Dr. Uwe Meyer-Gruhl" Subject: Problems with device-mapper on top of RAID-5 and RAID-6 Date: Fri, 02 Jun 2006 17:10:19 +0200 Message-ID: <448054DB.1020004@congenio.de> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-15; format=flowed Content-Transfer-Encoding: 7bit Return-path: Sender: linux-raid-owner@vger.kernel.org To: linux-raid@vger.kernel.org List-Id: linux-raid.ids Hi List, just to draw your attention to some discussion starting out here: http://thread.gmane.org/gmane.linux.kernel.device-mapper.dm-crypt/1576/focus=1576 and going on here: http://thread.gmane.org/gmane.linux.kernel.device-mapper.dm-crypt/1617/focus=1617 To recap, what has been found so far is: 1. There are problems with the combination of RAID and device mapper (e.g. for encrypted filesystems). The thread started off with this observation. 2. There are filesystem corruptions with heavy loads (i.e. copying big files or many files to the filesystem). The bug usually takes long to reproduce. 3. Problems occur with any filesystem type (ext3, reiser4 et. al.). 4. Problems occur with RAID-5 and RAID-6. Both are O.K. without dm-crypt. 5. Problems are unaffected by different ciphers under dm-crypt (at least AES, Serpent and Twofish expose the bug). dm-linear is reported to have failed, too. So, we suspect that neither dm-crypt nor ciphers are the culprit here, but rather the device mapper core functionality or RAID subsystem. 6. Bug seems to exist in at least kernel 2.6.13 to 2.6.16 (2.6.17 not yet tested, earlier versions may be affected). There have been discussions going on about his from earlier kernel versions, like here: http://lwn.net/Articles/150583/ Neil's suggestion indicates that there may be a race condition stacking md and dm over each other, but I have not yet tested that patch. I once had problems stacking cryptoloop over RAID-6, so it might really be a stacking problem. We don't know yet if LVM over RAID is affected as well. This bug is very critical and should be fixed as soon as possible, IMHO. Uwe