Re: [PATCH v3 21/36] arm64/mm: Implement map_shadow_stack()

[Szabolcs Nagy]

The 08/07/2023 14:00, Mark Brown wrote:
> On Mon, Aug 07, 2023 at 11:20:58AM +0100, Szabolcs Nagy wrote:
> > The 07/31/2023 14:43, Mark Brown wrote:
> > > +SYSCALL_DEFINE3(map_shadow_stack, unsigned long, addr, unsigned long, size, unsigned int, flags)
> > > +{
> > > +	unsigned long alloc_size;
> > > +	unsigned long __user *cap_ptr;
> > > +	unsigned long cap_val;
> > > +	int ret;
> > > +
> > > +	if (!system_supports_gcs())
> > > +		return -EOPNOTSUPP;
> > > +
> > > +	if (flags)
> > > +		return -EINVAL;
> > > +
> > > +	if (addr % 16)
> > > +		return -EINVAL;
> 
> > mmap addr must be page aligned (and there is no align req on size).
> 
> > i'd expect similar api here.
> 
> That's not what the manual page or a quick check of the code suggest
> that mmap() does, they say that the kernel just takes it as a hint and

i should have said that i expect MAP_FIXED_NOREPLACE semantics
(since the x86 code seemed to use that) and then the mapped
address must match exactly thus page aligned.

> chooses a nearby page boundary, though I didn't test.  I'm not sure why
> I have that alignment check at all TBH, and to the extent it's needed I
> could just be 8 - this level of code doesn't really care.
> 
> > > +	if (size == 16 || size % 16)
> > > +		return -EINVAL;
> 
> > why %16 and not %8 ?
> 
> I don't think that's needed any more - there was some stuff in an
> earlier version of the code but no longer.

it's kind of important to know the exact logic so the cap token
location can be computed in userspace for arbitrary size.

(this is why i wanted to see the map_shadow_stack man page first
but i was told that comes separately on linux..)