[kernel-hardening] Re: [PATCH v7] printk: hash addresses printed with %p

["Tobin C. Harding" <me@tobin.cc>]

On Wed, Oct 25, 2017 at 06:00:21AM +0200, Jason A. Donenfeld wrote:
> On Wed, Oct 25, 2017 at 5:49 AM, Tobin C. Harding <me@tobin.cc> wrote:
> > static_branch_disable(&no_ptr_secret) : Doesn't sleep, just atomic read
> > and set and maybe a WARN_ONCE.
> 
> Are you sure about that? I just looked myself, and though there is a
> !HAVE_JUMP_LABEL ifdef that does what you described, there's also a
> HAVE_JUMP_LABEL that takes a mutex, which sleeps:
> 
> static_branch_disable
>   static_key_disable
>     cpus_read_lock
>       percpu_down_read
>         percpu_down_read_preempt_disable
>           might_sleep
> 
> > Now for the 'executes from process context' stuff.
> 
> Er, sorry, I meant to write non-process context in my original
> message, which is generally where you're worried about sleeping.
> 
> > If the callback mechanism is utilized (i.e print before randomness is
> > ready) then the call back will be executed the next time the randomness
> > pool gets added to
> 
> So it sounds to me like this might be called in non-process context.
> Disaster. I realize the static_key thing was my idea in the original
> email, so sorry for leading you astray. But moving to do this in
> early_initcall wound up fixing other issues too, so all and all a net
> good in going this direction.
> 
> Two options: you stick with static_branch, because it's cool and speed
> is fun, and work around all of the above with a call to queue_work so
> that static_branch_enable is called only from process context.

This definitely sounds more fun, the static_branch stuff is dead sexy.

> Or, you give up on static_key, because it's not actually super
> necessary, and instead just use an atomic, and reason that using `if
> (unlikely(!atomic_read(&whatever)))` is probably good enough. In this
> option, the code would be pretty much the same as v7, except you'd
> s/static_branch/atomic_t/, and change the helpers, etc. This is
> probably the more reasonable way.

How good is unlikely()?

It doesn't _feel_ right adding a check on every call to printk just to
check for a condition that was only true for the briefest time when the
kernel booted. But if unlikely() is good then I guess it doesn't hurt.

I'm leaning towards the option 1, but then all those text books I read
are telling me to implement the simplest solution first then if we need
to go faster implement the more complex solution.

This is a pretty airy fairy discussion now, but if you have an opinion
I'd love to hear it.

thanks,
Tobin.