On 10/02/2018 12:17 PM, Kees Cook wrote:

On Tue, Oct 2, 2018 at 11:57 AM, John Johansen
[off-list ref] wrote:

quoted

Under the current scheme

lsm.enabled=selinux

could actually mean selinux,yama,loadpin,something_else are
enabled. If we extend this behavior to when full stacking lands

lsm.enabled=selinux,yama

might mean selinux,yama,apparmor,loadpin,something_else

and what that list is will vary from kernel to kernel, which I think
is harder for the user than the lsm.enabled list being what is
actually enabled at boot

Ah, I think I missed this in your earlier emails. What you don't like
here is that "lsm.enable=" is additive. You want it to be explicit.

Are you okay with lsm.order= having fallback?

yeah, if we are going to separate order, fallbacks are fine for
anything that isn't specified.

I am still not convinced that separating order from enablement is
right, but its generally something a user should care about so I can
live with it.

The situation we were trying to solve was with new LSMs getting
implicitly disabled if someone is booting with an explicit list. For
example:

lsm.enable=yama,apparmor

means when "landlock" gets added to the kernel, it will be implicitly disabled.

And here is the point of contention, I wouldn't call that implicitly
disabled. The user explicitly selected a set of LSMs to enable. Having
other LSMs enable when they aren't specified is confusing to a user,
as now they have to consider what is enabled by default in the
Kconfig.

I think requiring distros/builders to consider Kconfig options is
fine, but its a lot higher hurdle for regular users.

quoted

If we have to have multiple kernel parameter, I prefer a behvior where
if you hav conflicting kernel parameters specified

  apparmor=0 lsm.enabled=apparmor

that the conflict is logged and the lsm is left disabled, as I think
it is easier for users to understand than the overrides scheme of v3,
and sans logging of the conflict is effectively what we had in the
past

  apparmor=0 security=apparmor
or
  apparmor=1 security=selinux

would result in apparmor being disabed

Okay, so for this part you want per-LSM boot param to have priority
(which seems to match SELinux's concerns), possibly logging the

hrmmm I wouldn't call it priority :)

If you look at the above logic its a boolean AND operation. The LSM is
only enabled if $LSM=1 AND security=$LSM all other combinations result
in $LSM being disabled

conflict, but still accepting the apparmor= and selinux= state.

logging is nice for the user but certainly isn't required and is more
than we are doing today

security= would still driving initialization ordering (so I think the
behavior I have in the series would be correct).

quoted

That being said I get we have a mess currently, and there really
doesn't seem to be a good way to fix it. I think getting this right
for the user is important enough that I am willing to break current
apparmor userspace api. While apparmor=0 is documented we have also
documented security=X for years and apparmor=0 isn't used too often
so I think we can drop it to help clean this mess up abit.

I am not going to Nak, or block on v3 behavior if that is considered
the best path forward after this discussion/rant.

I could define CONFIG_LSM_ENABLE as being "additive" to
SECURITY_APPARMOR_BOOTPARAM_VALUE and
SECURITY_SELINUX_BOOTPARAM_VALUE?

Oh sure lets deal with my complaint about too many ways to configure
this beast by adding yet another config option :P

seriously though, please no. That just adds another layer of confusion
even if it is only being foisted on the distro/builder