[PATCH security-next v3 00/29] LSM: Explict LSM ordering

[Kees Cook]

On Sat, Sep 29, 2018 at 3:48 AM, Tetsuo Handa
<penguin-kernel at i-love.sakura.ne.jp> wrote:
> On 2018/09/29 5:01, Kees Cook wrote:
>> On Fri, Sep 28, 2018 at 8:55 AM, Casey Schaufler <casey at schaufler-ca.com> wrote:
>>> On 9/24/2018 5:18 PM, Kees Cook wrote:
>>>> v3:
>>>> - add CONFIG_LSM_ENABLE and refactor resulting logic
>>>
>>> Kees, you can add my
>>>
>>>         Reviewed-by:Casey Schaufler <casey at schaufler-ca.com>
>>>
>>> for this entire patch set. Thank you for taking this on, it's
>>> a significant and important chunk of the LSM infrastructure
>>> update.
>>
>> Thanks!
>>
>> John, you'd looked at this a bit too -- do the results line up with
>> your expectations?
>>
>> Any thoughts from SELinux, TOMOYO, or IMA folks?
>
> I'm OK with this approach. Thank you.

Thanks for looking it over!

> Just wondering what is "__lsm_name_##lsm" for...
>
> +#define DEFINE_LSM(lsm)                                                        \
> +       static const char __lsm_name_##lsm[] __initconst                \
> +               __aligned(1) = #lsm;                                    \
> +       static struct lsm_info __lsm_##lsm                              \
> +               __used __section(.lsm_info.init)                        \
> +               __aligned(sizeof(unsigned long))                        \
> +               = {                                                     \
> +                       .name = __lsm_name_##lsm,                       \
> +
> +#define END_LSM          }

I wasn't super happy with the END_LSM thing, but I wanted to be able
to declare the name as __initconst, otherwise it needlessly stays in
memory after init. That said, it's not a huge deal, and maybe
readability trumps a tiny meory savings?

> We could do something like below so that funny END_LSM is not required?
> I felt } like a typo error at the first glance. What we need is to
> gather into one section with appropriate alignment, isn't it?
>
> #define LSM_INFO                                                        \
>         static struct lsm_info __lsm_                                   \
>                 __used __section(.lsm_info.init)                        \
>                 __aligned(sizeof(unsigned long))                        \
>
> LSM_INFO = {
>         .name = "tomoyo",
>         .flags = LSM_FLAG_LEGACY_MAJOR | LSM_FLAG_EXCLUSIVE,
>         .init = tomoyo_init,
> };

I thought the structure instances would need a unique name, but it
seems the section naming removes that requirement. This seems only to
be needed if we had multiple LSMs defined in the same source file.
Though I wonder if this would be a problem for LTO in the future?

I'm happy to do whatever.

-Kees

-- 
Kees Cook
Pixel Security