From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail.gmx.net (mailout-de.gmx.net [213.165.64.22]) by mail.saout.de (Postfix) with SMTP for ; Wed, 27 Oct 2010 17:14:33 +0200 (CEST) Message-ID: <4CC841D8.8000709@gmx.at> Date: Wed, 27 Oct 2010 17:14:32 +0200 From: Markus Krainz MIME-Version: 1.0 References: <20101027144102.GB10534@fancy-poultry.org> In-Reply-To: <20101027144102.GB10534@fancy-poultry.org> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Subject: Re: [dm-crypt] btrfs with raid5/6 List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: dm-crypt@saout.de > It's up to you. I'd prefer LUKS/dmcrypt. So do I. But I have a related concern: I currently do ext4 <-> dm-crypt <-> linux-software-raid <-> hdds However Linux Software Raid (mdadm) does not protect you from silent data loss, because it does not store checksums in it's metadata. So what mdadm does is, it compensates for failed devices, or blocks, if the device reports them as bad. But I can NOT correct data, that was e.g. corrupted by the SATA controller or backplane. See http://neil.brown.name/blog/20100211050355 "Smart or simple RAID recovery??" for more information. Btrfs (like ZFS) can be the solution here, as it stores CRC checksums with all data and metadata. RAID5 with protection from silent data loss is promised to be implemented in the future. A setup on top of dm-crypt would look as follows: btrfs <-> multiple dm-crypt partitions <-> multiple devices However in the past I experienced very disappointing results, if raid is running on top of dm-crypt partitions. If I remove hdds from such a setup the dm-crypt partition wont disappear. It won't propagate the error to the software raid on top of it (not familiar with the implementation details, maybe mdadm is to blame here). And finally it won't even let itself be deleted without restart (I think the last issue has been fixed in the meantime). I urge the dm-crypt developers to improve this situation with raid on top of dm-crypt, be it with mdadm or btrfs. Solutions for other long-time annoying issues like insufficient threading support, and the lack of discard support, would be appreciated too. :P I know the OSS pradigm is provide patches yourself. Would you be accepting patches? Sincerly, Markus