Re: [GIT PULL] gcc-plugin updates for v4.19-rc1

[Kees Cook]

On Wed, Aug 15, 2018 at 9:41 AM, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Mon, Aug 13, 2018 at 2:43 PM Kees Cook <keescook@xxxxxxxxxxxx> wrote:
>>
>> Please pull these gcc-plugin changes for v4.19-rc1.
>
> No.
>
> It adds yet another BUG_ON() without having been merged.
>
> I'm not pulling this. Dammit, have you learnt *nothing*?

I swear I'm doing my best. Are you speaking of
stackleak_check_alloca() or stackleak_erase()? These were both
discussed on the list, and we weren't able to come up with
alternatives: in both cases we're off the stack, and recovery is
seemingly impossible. What would you prefer in these cases? If I need
to take a hard line of "never BUG", how do I handle legitimate system
corruption? (i.e. I have interpreted this as different from narrowing
copy_*_user() usage: if we let execution continue, we'll just crash
somewhere else with likely less information on how to handle it.)

> I'm, disappointed in the whole feature, but I'm also tired of having
> to go and even look for these things.

I am trying to make these patches easier to review. I even made sure
to get Ingo's Ack and Alexander implemented additional features Ingo
suggested, before sending them your way, as Ingo has a very
conservative eye on.

> Then actually *finding* them makes me just pissed off.

I'm sorry we've disappointed you. I've been pushing back on patches
that use BUG (with, I think, good success), but there are cases where
our imagination fails us.

I'd really like to find a way for this plugin to be acceptable, given
the coverage is provides. Even if we solve stack initialization and
finish VLA removal, we still would benefit from something doing
post-syscall stack poisoning just to keep future cache attacks against
the stack minimized.

In the meantime, I will send the gcc-plugin cleanups separately...

-Kees

-- 
Kees Cook
Pixel Security