Re: [PATCH security-next v4 23/32] selinux: Remove boot parameter

[John Johansen]

On 10/03/2018 10:26 AM, Kees Cook wrote:
> On Wed, Oct 3, 2018 at 6:39 AM, Stephen Smalley <sds@xxxxxxxxxxxxx> wrote:
>> On 10/02/2018 07:54 PM, Kees Cook wrote:
>>>
>>> On Tue, Oct 2, 2018 at 4:46 PM, John Johansen
>>> <john.johansen@xxxxxxxxxxxxx> wrote:
>>>>
>>>> On 10/02/2018 04:06 PM, Kees Cook wrote:
>>>>>
>>>>> I think the current proposal (in the other thread) is likely the
>>>>> sanest approach:
>>>>>
>>>>> - Drop CONFIG_SECURITY_SELINUX_BOOTPARAM_VALUE
>>>>> - Drop CONFIG_SECURITY_APPARMOR_BOOTPARAM_VALUE
>>>>> - All enabled LSMs are listed at build-time in CONFIG_LSM_ENABLE
>>>>
>>>>
>>>> Hrrmmm isn't this a Kconfig selectable list, with each built-in LSM
>>>> available to be enabled by default at boot.
>>>
>>>
>>> That's not how I have it currently. It's a comma-separated a string,
>>> including the reserved name "all". The default would just be
>>> "CONFIG_LSM_ENABLE=all". Casey and I wanted this to have a way to
>>> capture new LSMs by default at build-time.
>>>
>>>>> - Boot time enabling for selinux= and apparmor= remain
>>>>> - lsm.enable= is explicit: overrides above and omissions are disabled
>>>>
>>>> wfm
>>>
>>>
>>> Okay, this is closer to v3 than v4. Paul or Stephen, how do you feel
>>> about losing the SELinux bootparam CONFIG? (i.e. CONFIG_LSM_ENABLE
>>> would be replacing its functionality.)
>>
>>
>> I'd like to know how distro kernel maintainers feel about it. They would
>> need to understand that if they were previously setting
>> CONFIG_SECURITY_SELINUX_BOOTPARAM_VALUE to 0 and want to preserve that
>> behavior, then they must set CONFIG_LSM_ENABLE explicitly to a list of
>> security modules (that does not include selinux, of course).  In practice,
> 
> That's not how it would be done. See below...
> 
>> this means that even the distros that choose to build all security modules
>> into their kernels must explicitly set CONFIG_LSM_ENABLE to a specific list
>> of security modules.  So no one would use "all" in practice.
> 
> This is why I had originally wanted to do CONFIG_LSM_DISABLE. Right
> now, distro kernel maintainers have two ways to trigger enablement:
> via the SELinux and AppArmor BOOTPARAM_VALUE _and_ DEFAULT_SECURITY
> (which is an implicit "enable" for Smack or TOMOYO). All the minors
> are on-if-built. So, really, the BOOTPARAM_VALUEs were only used for
> disabling. Distros would build what they wanted, then use
> DEFAULT_SECURITY for their desired major, and if their
> DEFAULT_SECURITY wasn't SELinux or AppArmor, they'd _also_ have to set
> those BOOTPARAM_VALUEs to 0.
> 
> The goal of the series is to split this more cleanly between "enable"
> and "order": the way to handle the LSMs is to enable _everything_ and
> then set the desired init order: the first exclusive "wins". So I *do*
> think the default would be CONFIG_LSM_ENALBE=all, since it's
> CONFIG_LSM_ORDER= that effectively replaces CONFIG_DEFAULT_SECURITY.
> 
but distinct of first exclusive (major) will likely be going away
once full lsm stacking land.

> Either a distro builds a very specific subset of LSMs, or they build
> in all LSMs (for the user to choose from). In both cases, they set an
> explicit order, which defines which exclusive LSM get selected.
> 
and when lsm stacking lands, that exlusive LSM goes away.

> AppArmor wants to drop BOOTPARAM_VALUE, which make sense, since it's
> even now redundant to CONFIG_DEFAULT_SECURITY. I think it makes sense
> to drop SELinux's BOOTPARAM_VALUE too. The current way to "enable" a
> major LSM is via CONFIG_DEFAULT_SECURITY. No sane distro kernel is
> going to set CONFIG_DEFAULT_SECURITY=selinux and
> CONFIG_SECURITY_SELINUX_BOOTPARAM_VALUE=0. If you wanted no major LSM
> (but still build them all in), you'd set CONFIG_DEFAULT_SECURITY="".
>