Re: [PATCH security-next v2 26/26] LSM: Add all exclusive LSMs to ordered initialization

[Kees Cook]

On Thu, Sep 20, 2018 at 6:39 PM, John Johansen
<john.johansen@xxxxxxxxxxxxx> wrote:
> On 09/20/2018 06:10 PM, Casey Schaufler wrote:
>> On 9/20/2018 5:45 PM, Kees Cook wrote:
>>> On Thu, Sep 20, 2018 at 5:25 PM, Casey Schaufler <casey@xxxxxxxxxxxxxxxx> wrote:
>>>> On 9/20/2018 9:23 AM, Kees Cook wrote:
>>>>>  config LSM_ORDER
>>>>>       string "Default initialization order of builtin LSMs"
>>>>> -     default "yama,loadpin,integrity"
>>>>> +     default "yama,loadpin,integrity,selinux,smack,tomoyo,apparmor"
>>>> If I want to compile all the major modules into my kernel and use
>>>> AppArmor by default would I use
>>>>
>>>>         default "yama,loadpin,integrity,apparmor,selinux,smack,tomoyo"
>>>>
>>>> or
>>>>
>>>>         default "yama,loadpin,integrity,apparmor"
>>> I was expecting the former, but the latter will have the same result.
>
> t find having the two be equivalent violates expectations. At least
> when considering the end goal of full/extreme stacking, its trivially
> the same with current major lsms being exclusive

This mixes "enablement" with "ordering", though, and I think the past
threads have shown this to be largely problematic.

However, with CONFIG_LSM_ENABLED, we get the effect you're looking for, IIUC.

>>>> When we have "blob-sharing" how could I compile in tomoyo,
>>>> but exclude it without a boot line option?
>>> Ooh, yes, this series has no way to do that. Perhaps
>>> CONFIG_LSM_DISABLE in the same form as CONFIG_LSM_ORDER? I would
>>> totally remove LoadPin's CONFIG for this in favor it.
>>
>> I would generally prefer an optional CONFIG_LSM_ENABLE to
>> CONFIG_LSM_DISABLE, but I understand the logic behind your
>> approach. I would be looking for something like
>>
> +1 on the CONFIG_LSM_ENABLE ove DISABLE
>
>> CONFIG LSM_ENABLE
>>       string "Default set of enabled LSMs"
>>       default ""
>>
>> as opposed to
>>
>> CONFIG LSM_DISABLE
>>       string "Default set of disabled LSMs"
>>       default ""
>>
>> where an empty string is interpreted as "use 'em all"
>> in either case.

Yes, I like CONFIG_LSM_ENABLE if "empty" means "enable all". Should
CONFIG_LSM_ENABLE replace all the other CONFIG-based LSM
enabling/disabling?

>>>> When we have full stacking, how could I compile in selinux
>>>> but exclude it?
>>> Yup, same problem. Same suggested solution?
>>>
>>> Should lsm.enable/disable= also become a comma-separated list, or
>>> should I leave it as a multi-instance thing like I have it?
>>
>> I prefer the multi-instance
>>       lsm.disable=selinux lsm.disable=yama
>> to the list
>>       lsm.disable=selinux,yama
>>
>> but at this point I don't really care all that much.
>
> the comma separated list however is consistent with what is being
> done for default order

Yeah, and it would match the new CONFIG_LSM_ENABLE.

FWIW, it was tedious to type "lsm.enable" and "lsm.disable" over and
over when doing testing, so I almost prefer the comma separated list
at the end of the day. I'll ponder this a bit.

-Kees

-- 
Kees Cook
Pixel Security