Re: [PATCH] capabilities: introduce per-process capability bounding set (v10)

[serge@hallyn.com]

Quoting KaiGai Kohei (kaigai@ak.jp.nec.com):
> Andrew Morgan wrote:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>> KaiGai Kohei wrote:
>>> Serge,
>>>
>>> Please tell me the meanings of the following condition.
>>>
>>>> diff --git a/security/commoncap.c b/security/commoncap.c
>>>> index 3a95990..cb71bb0 100644
>>>> --- a/security/commoncap.c
>>>> +++ b/security/commoncap.c
>>>> @@ -133,6 +119,12 @@ int cap_capset_check (struct task_struct *target,
>>>> kernel_cap_t *effective,
>>>>          /* incapable of using this inheritable set */
>>>>          return -EPERM;
>>>>      }
>>>> +    if (!!cap_issubset(*inheritable,
>>>> +               cap_combine(target->cap_inheritable,
>>>> +                       current->cap_bset))) {
>>>> +        /* no new pI capabilities outside bounding set */
>>>> +        return -EPERM;
>>>> +    }
>>>>       /* verify restrictions on target's new Permitted set */
>>>>      if (!cap_issubset (*permitted,
>>> It seems to me this condition requires the new inheritable capability
>>> set must have a capability more than bounding set, at least.
>>> What is the purpose of this checking?
>> Yes, the !! was a bug. The correct check is a single !.
>
> I was in trouble with getting -EPERM at pam_cap.so :-)
>
>> (Thus, the correct check says no 'new' pI bits can be outside cap_bset.)
>
> If this condition intends to dominate 'new' pI bits by 'old' pI bits masked
> with bounding set, we should not apply cap_combine() here.
> I think applying cap_intersect() is correct for the purpose.

That would have been my first inclination, but Andrew actually
wanted to be able to keep a pI with bits not in the capability
bounding set.  And it's really not a big problem, since

	1. you can never grow cap_bset
	2. the capbound.c program just makes sure to call capset
	   to take the bit being removed from cap_bset out of
	   pI'
	3. It could be advantageous for some daemon to keep a bit
	   in pI which can never be gained through fP but can be
	   gained by a child through (fI&pI).

Does that seem reasonable to you?

(On the one hand man 7 capabilities shows cap_bset affecting
fP and not pP', on the other hand we're definately getting into
the territory where we'll have to rewrite the manpages anyway)

thanks,
-serge
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/