Re: [GIT PULL] SELinux patches for v5.8

From: Stephen Smalley
Date: Thu Jun 04 2020 - 11:33:30 EST


On Thu, Jun 4, 2020 at 11:28 AM Casey Schaufler <casey@xxxxxxxxxxxxxxxx> wrote:
>
> On 6/4/2020 5:45 AM, Stephen Smalley wrote:
> > On Wed, Jun 3, 2020 at 6:39 PM Casey Schaufler <casey@xxxxxxxxxxxxxxxx> wrote:
> >> On 6/3/2020 3:12 PM, James Morris wrote:
> >>> On Wed, 3 Jun 2020, Casey Schaufler wrote:
> >>>
> >>>> The use of security modules was expected to be rare.
> >>> This is not correct. Capabilities were ported to LSM and stacked from the
> >>> beginning, and several major distros worked on LSM so they could ship
> >>> their own security modules.
> >> Capabilities has always been a special case.
> >> Until Android adopted SELinux the actual use of LSMs was rare.
> > I don't think that is correct. Fedora/RHEL were enabling SELinux by
> > default since around 2004/2005 and for a while Fedora was tracking
> > SELinux status as part of their "smolt" hardware profiling project and
> > SELinux enablement was trending above 80% IIRC before they
> > de-commissioned smolt. SuSE/SLES and Ubuntu were enabling AppArmor by
> > default for quite some time too prior to SE Android.
>
> POSIX ACLs have been enabled just as long. Their use is still
> incredibly rare.
>
> > It is certainly
> > true that Android's adoption of SELinux massively increased the size
> > of the SELinux install base (and was the first to make SELinux usage
> > mandatory, not just default-enabled) but I don't think it is accurate
> > to say that LSM usage was rare prior to that.
>
> That will depend on whether you consider presence to be usage.
> That gets into the whole "transparent security" argument.

The distros were shipping policies for their respective LSMs that
confined some subset of the processes, and userspace was leveraging
those LSMs (both to get/set labels and to get policy decisions for
userspace enforcers) well before Android adopted SELinux. I think
that counts as usage. If by usage you mean end users were writing
their own policies, that certainly is a more specialized class of
users but that's even less so in Android, where end users aren't
allowed to modify the policy at all.

> Sorry I brought this up. I don't mean to disrespect the achievement
> of SELinux. My experience of the Orange Book and early Common
> Criteria era, including the Unix to Linux transition, seems to
> have differed somewhat from that others.