From: Pavel Machek Subject: Re: ext4 encryption trap Date: Mon, 29 Aug 2016 12:49:06 +0200 Message-ID: <20160829104906.GB16212@amd> References: <20160829100816.GA14524@amd> <1846364.OxLArdGf3j@t1700bs> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: tytso@mit.edu, adilger.kernel@dilger.ca, linux-ext4@vger.kernel.org To: Bernd Schubert Return-path: Received: from atrey.karlin.mff.cuni.cz ([195.113.26.193]:59669 "EHLO atrey.karlin.mff.cuni.cz" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750828AbcH2KtJ (ORCPT ); Mon, 29 Aug 2016 06:49:09 -0400 Content-Disposition: inline In-Reply-To: <1846364.OxLArdGf3j@t1700bs> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Mon 2016-08-29 12:40:24, Bernd Schubert wrote: > On Monday, August 29, 2016 12:08:16 PM CEST Pavel Machek wrote: > > Hi! > > > > You encrypt a directory -- sounds easy, right? Support is in 4.4 > > kernel, my machines run newer kernels than that. Encrypting root would > > be hard, but encrypting parts of data partition should be easy. > > > > Ok, lets follow howto... Need to do tune2fs. Right. Aha, still does > > not work, looks like I'll need to reboot. > > > > Hmm. Will not boot. Grub no longer recognizes my /data partition, and > > that's where new kernels are. Old kernels are in /boot, but those are > > now useless. Lets copy new kernel on machine using USB stick. Does not > > boot. Fun. > > > > tune2fs on root filesystem is useless, as it is too old. New one > > is ... on the data partition. Right. Ok, lets bring newer version of > > tune2fs in. "encryption" feature can not be cleared. > > > > Argh! Come on, I did not even create single encrypted directory on the > > partition. I want the damn bit to go off, so I can go back to working > > configuration. "Old kernels can not read encrypted files" sounds ok, > > but "old kernels can not mount filesystem at all" is not acceptable > > here :-(. > > > > Is there way to go back? Restoring 400GB from backups would not be fun > > I have not tried it myself, but this should work? > > debugfs -w -R "feature -encrypt" /dev/device > > > (assuming the feature flag is called "encrypt") Yes, I figured out debugfs could be used to do this. (But thanks for the command line). If all tunefs did was to set the bit, this is safe. Is it? [I guess I can do fsck -fn, debugfs, fsck -fn; if it passes I should be safe, if it does not I can turn +encrypt back on, and would be no worse than I'm now. Hmm?] Thanks, Pavel -- (english) http://www.livejournal.com/~pavelmachek (cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html