Re: [PATCH 04/35] arm64/gcs: Document the ABI for Guarded Control Stacks

[Mike Rapoport]

On Wed, Jul 19, 2023 at 02:25:38PM +0100, Mark Brown wrote:
> On Wed, Jul 19, 2023 at 02:44:37PM +0300, Mike Rapoport wrote:
> > On Sun, Jul 16, 2023 at 10:51:00PM +0100, Mark Brown wrote:
> 
> > > +7.  ptrace extensions
> > > +---------------------
> > > +
> > > +* A new regset NT_ARM_GCS is defined for use with PTRACE_GETREGSET and
> > > +  PTRACE_SETREGSET.
> > > +
> > > +* Due to the complexity surrounding allocation and deallocation of stakcs and
> > > +  lack of practical application changes to the GCS configuration via ptrace
> > > +  are not supported.
> 
> > On x86 CRIU needed to be able to temporarily unlock shadow stack features
> > to recreate the shadow stack of the thread being restored. I presume CRIU
> > will need something like that on arm64 as well.
> 
> It would be good to understand why and what exactly is needed here.
> I'm guessing the main thing would be stores? 

Yes, CRIU has to modify the shadow stack.

CRIU uses sigreturn directly, so we had to update the shadow stack before
calling sigreturn.
On x86 ptrace(POKEDATA) and update of the shadow stack pointer were enough,
but it looks like ptrace(POKEDATA) won't work on arm64.

Another place that requires shadow stack modifications is the restore of
the shadow stack from the checkpoint. On x86 we had to enable WRSS and that
required to temporarily unlock the features.

> It's relatively easy to add features later, I think I'll just add support
> for everything except enable just now.

Fair enough :)

-- 
Sincerely yours,
Mike.