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

[Kees Cook]

On Tue, Oct 2, 2018 at 11:33 AM, Stephen Smalley <sds at tycho.nsa.gov> wrote:
> On 10/02/2018 12:54 PM, Kees Cook wrote:
>>
>> On Tue, Oct 2, 2018 at 9:33 AM, Jordan Glover
>> <Golden_Miller83 at protonmail.ch> wrote:
>>>
>>> It's always documented as: "selinux=1 security=selinux" so security=
>>> should
>>> still do the job and selinux=1 become no-op, no?
>>
>>
>> The v3 patch set worked this way, yes. (The per-LSM enable defaults
>> were set by the LSM. Only in the case of "lsm.disable=selinux" would
>> the above stop working.)
>>
>> John did not like the separation of having two CONFIG and two
>> bootparams mixing the controls. The v3 resolution rules were:
>>
>> SECURITY_SELINUX_BOOTPARAM_VALUE overrides CONFIG_LSM_ENABLE.
>> SECURITY_APPARMOR_BOOTPARAM_VALUE overrides CONFIG_LSM_ENABLE.
>> selinux= overrides SECURITY_SELINUX_BOOTPARAM_VALUE.
>> apparmor.enabled= overrides SECURITY_APPARMOR_BOOTPARAM_VALUE.
>> apparmor= overrides apparmor.enabled=.
>> lsm.enable= overrides selinux=.
>> lsm.enable= overrides apparmor=.
>> lsm.disable= overrides lsm.enable=.
>> major LSM _omission_ from security= (if present) overrides lsm.enable.
>>
>> v4 removed the per-LSM boot params and CONFIGs at John's request, but
>> Paul and Stephen don't want this for SELinux.
>>
>> The pieces for reducing conflict with CONFIG_LSM_ENABLE and
>> lsm.{enable,disable}= were:
>>
>> 1- Remove SECURITY_APPARMOR_BOOTPARAM_VALUE.
>> 2- Remove apparmor= and apparmor.enabled=.
>> 3- Remove SECURITY_SELINUX_BOOTPARAM_VALUE.
>> 4- Remove selinux=.
>>
>> v4 used all of 1-4 above. SELinux says "4" cannot happen as it's too
>> commonly used. Would 3 be okay for SELinux?
>
>
> Let's say a user/packager/distro has been building kernels for the past 14
> years (*) with a config that has SECURITY_SELINUX_BOOTPARAM_VALUE=0, and now
> they build a new kernel that includes these patches using that same config.
> Won't SELinux be enabled by default because SECURITY_SELINUX_BOOTPARAM_VALUE
> is now ignored and LSM_ENABLE defaults to all?  Is it ok to require them to
> specify a new config option to preserve old behavior?

Yes, I think that's fine -- kernel CONFIGs change all the time. System
builders are used to examining these changes.

-Kees

-- 
Kees Cook
Pixel Security