Re: modules: add ro_after_init support

[Kees Cook]

On Wed, Jun 29, 2016 at 9:56 PM, Rusty Russell <rusty@xxxxxxxxxxxxxxx> wrote:
> Jessica Yu <jeyu@xxxxxxxxxx> writes:
>> +++ Rusty Russell [29/06/16 10:38 +0930]:
>>>Jessica Yu <jeyu@xxxxxxxxxx> writes:
>>>> Add ro_after_init support for modules by adding a new page-aligned section
>>>> in the module layout (after rodata) for ro_after_init data and enabling RO
>>>> protection for that section after module init runs.
>>>>
>>>> Signed-off-by: Jessica Yu <jeyu@xxxxxxxxxx>
>>>
>>>I would prefer a "bool after_init" flag to module_enable_ro().  It's
>>>more explicit.
>>
>> Sure thing, I was just initially worried about the
>> module_{enable,disable}_ro() asymmetry. :)
>
> Yes, but I think compile-time-analyzable behaviour beats
> runtime-analyzable behaviour for clarity.
>
>>>Exposing the flags via uapi looks like a wart, but it's kind of a
>>>feature, since we don't *unset* it in any section; userspace may want to
>>>know about it.
>>
>> Hm, I'm still unsure about this. I'm starting to think it might be a
>> bit overkill to expose SHF_RO_AFTER_INIT through uapi (although that
>> is where all the other SHF_* flags are defined) SHF_RO_AFTER_INIT
>> would technically be used only internally in the kernel (i.e. module
>> loader), and it'd also be considered a non-standard flag, using a bit
>> from SHF_MASKOS (OS-specific range). What do you think?
>
> Some arch *could* use it by setting the flag in a section in their
> module I think; we don't stop them.  Since the other flags are there,
> I'd leave it.
>
> We don't expose the flags via sysfs, though, so that's the only
> exposure.

What's the state of this series? I'd love it if the functionality
could land for v4.8...

-Kees

-- 
Kees Cook
Brillo & Chrome OS Security