Re: [PATCH] arm64: Clear the stack

From: Mark Rutland
Date: Tue Jul 03 2018 - 08:14:54 EST


On Mon, Jul 02, 2018 at 11:48:05AM -0700, Laura Abbott wrote:
> On 07/02/2018 06:02 AM, Alexander Popov wrote:
> > On 29.06.2018 22:05, Laura Abbott wrote:
> > > Implementation of stackleak based heavily on the x86 version
> > >
> > > Signed-off-by: Laura Abbott <labbott@xxxxxxxxxx>
> > > ---
> > > Changes since last time:
> > > - Minor name change in entry.S
> > > - Converted to use the generic interfaces so there's minimal additions.
> > > - Added the fast syscall path.
> > > - Addition of on_thread_stack and current_top_of_stack
> > > - Disable stackleak on hyp per suggestion
> > > - Added a define for check_alloca. I'm still not sure about keeping it
> > > since the x86 version got reworked?
> > >
> > > I've mostly kept this as one patch with a minimal commit text. I can
> > > split it up and elaborate more before final merging.
> > > ---

[...]

> > > diff --git a/arch/arm64/kernel/entry.S b/arch/arm64/kernel/entry.S
> > > index ec2ee720e33e..31c9da7d401e 100644
> > > --- a/arch/arm64/kernel/entry.S
> > > +++ b/arch/arm64/kernel/entry.S
> > > @@ -401,6 +401,11 @@ tsk .req x28 // current thread_info
> > > .text
> > > + .macro stackleak_erase
> >
> > Could you rename the macro to STACKLEAK_ERASE for similarity with x86?
> >
>
> Mark Rutland had previously asked for this to be lowercase.
> I really don't care one way or the other so I'll defer to
> someone else to have the final word.

Will, Catalin, could you chime in either way?

I'd previously asked for lower-case for consistency with our other
assembly macros.

[...]

> > > diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c
> > > index f08a2ed9db0d..9f0f135f8b66 100644
> > > --- a/arch/arm64/kernel/process.c
> > > +++ b/arch/arm64/kernel/process.c
> > > @@ -493,3 +493,19 @@ void arch_setup_new_exec(void)
> > > {
> > > current->mm->context.flags = is_compat_task() ? MMCF_AARCH32 : 0;
> > > }
> > > +
> > > +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK
> > > +#define MIN_STACK_LEFT 256
> > > +
> > > +void __used stackleak_check_alloca(unsigned long size)
> > > +{
> > > + unsigned long sp, stack_left;
> > > +
> > > + sp = current_stack_pointer;
> > > +
> > > + stack_left = sp & (THREAD_SIZE - 1);
> > > + BUG_ON(stack_left < MIN_STACK_LEFT ||
> > > + size >= stack_left - MIN_STACK_LEFT);
> > > +}
> > > +EXPORT_SYMBOL(stackleak_check_alloca);
> > > +#endif
> >
> > This code should be updated.
> > You may remember the troubles I had with MIN_STACK_LEFT:
> > http://openwall.com/lists/kernel-hardening/2018/05/11/12
> > Please see that thread where Mark Rutland and I worked out the solution.
> >
>
> Ah yeah, I missed the details in that thread. Thanks for
> that pointer.
>
> > By the way, different stacks on x86_64 have different sizes. Is it false for arm64?
>
> Currently everything except the overflow stack looks to be
> the same size but there's also another stack I missed.

Assuming I've followed the code correctly, we currently have:

stack size alignment (minimum)
---------------------------------------------------
task THREAD_SIZE THREAD_ALIGN
irq THREAD_SIZE 16
overflow SZ_4K 16
sdei_normal THREAD_SIZE THREAD_ALIGN
sdei_critical THREAD_SIZE THREAD_ALIGN

... since IRQ_STACK_SIZE is defined as THREAD_SIZE, and SDEI_STACK_SIZE
is defined as IRQ_STACK_SIZE.

So we can't just mask the sp, unfortunately.

> It might be cleaner just to use on_accessible_stack and then another
> function to get the top of stack. This also might just be
> reimplementing what x86 already has? (Mark, Ard?)

It looks like we could build a get_stack_info() as they have.

We could probably clean up our stack traced atop of that, too.

Thanks,
Mark.