Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Message-ID: <1552607740.2571.15.camel@HansenPartnership.com>
Subject: Re: [PATCH 4/4] ubifs: Implement new mount option,
 fscrypt_key_required
From:   James Bottomley <James.Bottomley@HansenPartnership.com>
To:     Theodore Ts'o <tytso@mit.edu>
Cc:     Richard Weinberger <richard@nod.at>, linux-mtd@lists.infradead.org,
        linux-fscrypt@vger.kernel.org, jaegeuk@kernel.org,
        linux-unionfs@vger.kernel.org, miklos@szeredi.hu,
        amir73il@gmail.com, linux-fsdevel@vger.kernel.org,
        linux-kernel@vger.kernel.org, paullawrence@google.com
Date:   Thu, 14 Mar 2019 16:55:40 -0700
In-Reply-To: <20190314234230.GG6482@mit.edu>
References: <CAJfpegvX4ANfso-Jn1zaB+m4Q_0eK9i-+MCJ+sTH5QzD3PZFMQ@mail.gmail.com>
         <20190314171559.27584-1-richard@nod.at>
         <20190314171559.27584-5-richard@nod.at>
         <1552605311.2571.6.camel@HansenPartnership.com>
         <20190314234230.GG6482@mit.edu>
Content-Type: text/plain; charset="UTF-8"
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, 2019-03-14 at 19:42 -0400, Theodore Ts'o wrote:
> On Thu, Mar 14, 2019 at 04:15:11PM -0700, James Bottomley wrote:
> > On Thu, 2019-03-14 at 18:15 +0100, Richard Weinberger wrote:
> > > Usually fscrypt allows limited access to encrypted files even
> > > if no key is available.
> > > Encrypted filenames are shown and based on this names users
> > > can unlink and move files.
> > 
> > Shouldn't they be able to read/write and create as well (all with
> > the ciphertext name and contents, of course) ... otherwise how does
> > backup of encrypted files by admin without the key ever work?
> 
> That's not currently supported.  Michael Halcrow and I worked out
> some designs on how to do this, and I even had some prototype
> patches, but it's really, really, messy, and requires a lot of
> complex machinations in userspace on the save *and* restore, since
> there's a lot of crypto metadata that has to be saved, and you have
> to handle backing up the directory per-file keys, and restoring them
> when you recreate the directory on a restore, etc.
> 
> The simpler approach would have allowed backup of encrypted files
> without the key, but would require the user's key to do the restore,
> and if we ever actually tried to get this feature supported, that's
> the approach I'd suggest.

Well, as I said above encrypted file backup by an admin who doesn't
have the key was what I was thinking of.

> The fundamental reason why we never went did anything with this was
> that the original use case of fscrypt was for Chrome OS, where the
> original design premise was that you don't need to do backups, since
> everything is in the cloud.  A video from 2010:
> 
> 	https://www.youtube.com/watch?v=lm-Vnx58UYo
> 
> And with Android, backups happen automatically while you have the
> encryption key.

Heh, colour me paranoid, but if I backup my sensitive data to any
medium (including the cloud) I *want* the backup to be encrypted so
that if someone is careless with my backup data at least they don't get
the contents.

> There was talk about using fscrypt for Ubuntu desktops as an ecryptfs
> replacement, and in that case, we would have use case that would have
> required backups of a shared desktop where you don't have all of the
> encryption keys for all of the users.  Maybe someday as a Google
> Summer of Code project or maybe if some potential corporate user of
> fscrypt would be willing to fund the necessary engineering work?
> 
> But again, I'll repeat this because it's so important: In the vast
> majority of cases, including single-user laptops, desktops, etc., the
> real right answer is dm-crypt and *not* fscrypt.  Especially since
> time-sharing systems are *so* 1980's.  :-)
> 
> So I don't use fscrypt on my upstream development laptop (except for
> testing purposes) because it's simply not the right tool for the job.

I was thinking of the container use case again.  It's not really backup
per se but it is serialization: if you can't do a backup ... i.e. save
and restore an encrypted tar image, you can't use the filesystem for
container image protection.  But it also strikes me that inability to
do backup and restore without the key really restricts the use cases
for the entire filesystem.

James