Re: Support status of OpenBSD frontend drivers

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

On Thu, Mar 24, 2022 at 09:10:57PM -0400, Demi Marie Obenour wrote:
> On 3/24/22 18:21, Marek Marczykowski-Górecki wrote:
> > On Thu, Mar 24, 2022 at 11:49:14AM -0400, Demi Marie Obenour wrote:
> >> On 3/24/22 10:11, Roger Pau Monné wrote:
> >>> On Thu, Mar 24, 2022 at 09:56:29AM -0400, Demi Marie Obenour wrote:
> >>>> As per private discussion with Theo de Raadt, OpenBSD does not consider
> >>>> bugs in its xnf(4) that allow a backend to cause mischief to be security
> >>>> issues.  I believe the same applies to its xbf(4).  Should the support
> >>>> document be updated?
> >>>
> >>> I think that's already reflected in the support document:
> >>>
> >>> 'Status, OpenBSD: Supported, Security support external'
> >>>
> >>> Since the security support is external it's my understanding OpenBSD
> >>> security team gets to decide what's a security issue and what is not.
> >>>
> >>> That however creates differences in the level of support offered by
> >>> the different OSes, but I think that's unavoidable. It's also hard to
> >>> track the status here because those are external components in
> >>> separate code bases.
> >>>
> >>> Could be added as a mention together with the Windows note about
> >>> frontends trusting backends, but then I would fear this is likely to
> >>> get out of sync if OpenBSD ever changes their frontends to support
> >>> untrusted backends (even if not considered as a security issue).
> >>
> >> As a Qubes OS developer, I still think this is useful information and
> >> should be documented.  For instance, if I choose to add proper OpenBSD
> >> guest support to Qubes OS (as opposed to the current “you can run
> >> anything in an HVM” situation), I might decide to have OpenBSD
> >> guests use devices emulated by a Linux-based stubdomain, since the
> >> stubdomain’s netfront and blkfront drivers *are* security-supported
> >> against malicious backends.  I might also choose to have a warning in
> >> the GUI when switching the NetVM of an OpenBSD guest to something other
> >> than the empty string (meaning no network access) or the (normally
> >> fairly trusted) sys-firewall or sys-whonix qubes.
> > 
> > I'm with Roger on this - when security support is external, such
> > information in xen.git could easily become stale. If anything, there
> > could be a link to OpenBSD security status info, maintained by whoever
> > such support provides.
> 
> This ought to be on https://man.openbsd.org/xnf.4 and
> https://man.openbsd.org/xbf.4, but it is not.  Should I send a patch?

You should discuss with the OpenBSD people I think, I really have no
idea where those limitations should be listed. Introducing a man page
'Caveats' or 'Limitations' sections would seem suitable to me, but
it's ultimately up to them.

Thanks, Roger.