Index | Thread | Search

From:
Crystal Kolipe <kolipe.c@exoticsilicon.com>
Subject:
Re: Replace Blowfish with AES in vnode disk driver
To:
Damien Miller <djm@mindrot.org>, Filip Cernoch <filipcernoch@posteo.net>, tech@openbsd.org
Date:
Wed, 17 Sep 2025 14:27:35 +0100

Download raw body.

Thread
On Wed, Sep 17, 2025 at 02:29:34PM +0200, Jeremie Courreges-Anglas wrote:
> On Wed, Sep 17, 2025 at 12:44:35PM +0100, Crystal Kolipe wrote:
> > On Wed, Sep 17, 2025 at 01:42:59AM +0200, Jeremie Courreges-Anglas wrote:
> > > On Wed, Sep 17, 2025 at 09:29:12AM +1000, Damien Miller wrote:
> > > > On Wed, 17 Sep 2025, Jeremie Courreges-Anglas wrote:
> > > > 
> > > > > Should vnconfig move from blowfish, it should probably be to a scheme
> > > > > actually designed for data storage like AES-XTS (like softraid CRYPTO)
> > > > > or similar.
> > > > > 
> > > > >   https://en.wikipedia.org/wiki/Disk_encryption_theory
> > > > > 
> > > > > I'm no crypto expert, but I doubt that moving from blowfish-CBC to
> > > > > AES-CBC would be a big win.
> > > > 
> > > > softraid already uses AES-XTS for encrypted volumes.
> > > 
> > > Yup.
> > > 
> > > > IMO vnconfig
> > > > crypto is just legacy and should be removed.
> > > 
> > > From looking at the code, vnconfig already says:
> > > 
> > >   WARNING: Consider using softraid crypto.
> > > 
> > > Maybe we should make it clear that we're going to remove this code,
> > > say, for 7.9?  If people actually wanted to keep using this, I guess
> > > someone would have stepped up by now.
> > 
> > Surely it's more likely that anyone who is already using it would just ignore
> > the warning and assume that it's aimed at new users?  Exactly because there is
> > no mention of the vnconfig encryption support being removed.
> > 
> > Why not start by making the related options undocumented in the manual?
> 
> You're saying people ignore the warning because it doesn't mention a
> pending removal, but instead of fixing that, you make it harder for
> users to get at the documentation needed to migrate their data.

Existing users of encrypted vnd will already know how to mount these volumes.

(Unless we considering the case where somebody finds such a file on an ancient
backup and has no idea what it is.)

We actively want to discourage the creation of new vnd encrypted volumes.

> It seems we all agree that these options should be removed, so ok for
> the diff below?

Are you sure that there is indeed sufficient agreement that encryption support
is going to be imminently removed from vnd?

I didn't think that decision had been met yet.

There is at least one use case where softraid crypto is NOT currently a
workable alternative.

If you have a read-only device with a vnd encrypted image on it, you can
happily mount and use it.  Currently, softraid crypto does not support
read-only devices.

I actually posted patches to -tech to add support for that, so if you want to
remove vnd encryption support and push people to softraid crypto, at least the
code has been written and tested.  But It's not in cvs, so such existing users
of vnd crypto would be left without a migration path if you push forward with
it's removal.