Re: [PATCH] randomize_kstack: Remove non-functional per-arch entropy filtering

[Mark Rutland <mark.rutland@arm.com>]

On Thu, Jun 20, 2024 at 11:34:22AM -0700, Kees Cook wrote:
> On Thu, Jun 20, 2024 at 11:47:58AM +0800, liuyuntao (F) wrote:
> > 
> > 
> > On 2024/6/20 5:47, Kees Cook wrote:
> > > An unintended consequence of commit 9c573cd31343 ("randomize_kstack:
> > > Improve entropy diffusion") was that the per-architecture entropy size
> > > filtering reduced how many bits were being added to the mix, rather than
> > > how many bits were being used during the offsetting. All architectures
> > > fell back to the existing default of 0x3FF (10 bits), which will consume
> > > at most 1KiB of stack space. It seems that this is working just fine,
> > > so let's avoid the confusion and update everything to use the default.
> > > 
> > 
> > My original intent was indeed to do this, but I regret that not being more
> > explicit in the commit log..
> > 
> > Additionally, I've tested the stack entropy by applying the following patch,
> > the result was `Bits of stack entropy: 7` on arm64, too. It does not seem to
> > affect the entropy value, maybe removing it is OK, or there may be some
> > nuances of your intentions that I've overlooked.
> > 
> > --- a/include/linux/randomize_kstack.h
> > +++ b/include/linux/randomize_kstack.h
> > @@ -79,9 +79,7 @@ DECLARE_PER_CPU(u32, kstack_offset);
> >  #define choose_random_kstack_offset(rand) do {                         \
> >         if (static_branch_maybe(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT, \
> >                                 &randomize_kstack_offset)) {            \
> > -               u32 offset = raw_cpu_read(kstack_offset);               \
> > -               offset = ror32(offset, 5) ^ (rand);                     \
> > -               raw_cpu_write(kstack_offset, offset);                   \
> > +               raw_cpu_write(kstack_offset, rand);                     \
> >         }                                                               \
> >  } while (0)
> >  #else /* CONFIG_RANDOMIZE_KSTACK_OFFSET */
> 
> I blame the multiple applications of the word "entropy" in this feature. :)
> 
> So, there's both:
> 
> - "how many bits CAN be randomized?" (i.e. within what range can all
>   possible stack offsets be?)
> 
> and
> 
> - "is the randomization predictable?" (i.e. is the distribution of
>   selected positions with the above range evenly distributed?)
> 
> Commit 9c573cd31343 ("randomize_kstack: Improve entropy diffusion") was
> trying to improve the latter, but accidentally also grew the former.
> This patch is just trying to clean all this up now.
> 
> Thanks for testing! And I'm curious as to why arm64's stack offset
> entropy is 7 for you when we're expecting it to be 6. Anyway, that's not
> a problem I don't think. Just a greater offset range than expected.

Hmm....

I think this is due to the way the compiler aligns the stack in alloca(); it
rounds up the value of KSTACK_OFFSET_MAX(offset) and ends up spilling over an
additional bit (e.g. 0x3f1 to 0x3ff round up to 0x400).

Looking at v6.10-rc4 defconfig + CONFIG_RANDOMIZE_STACKOFFSET=y, the
disassembly for arm64's invoke_syscall() looks like:

	// offset = raw_cpu_read(kstack_offset)
	mov     x4, sp
	adrp    x0, kstack_offset
	mrs     x5, tpidr_el1
	add     x0, x0, #:lo12:kstack_offset
	ldr     w0, [x0, x5]

	// offset = KSTACK_OFFSET_MAX(offset)
	and     x0, x0, #0x3ff

	// alloca(offset)
	add     x0, x0, #0xf
	and     x0, x0, #0x7f0
	sub     sp, x4, x0

... which in C would be:

	offset = raw_cpu_read(kstack_offset)
	offset &= 0x3ff;			// [0x0, 0x3ff]
	offset += 0xf;				// [0xf, 0x40e]
	offset &= 0x7f0;			// [0x0,

... so when *all* bits [3:0] are 0, they'll have no impact, and when *any* of
bits [3:0] are 1 they'll trigger a carry into bit 4, which could ripple all the
way up and spill into bit 10.

I have no idea whether that's important. Kees, does that introduce a bias, and
if so do we need to care?

If I change the mask to discard the low bits:

	#define KSTACK_OFFSET_MAX(x)   ((x) & 0x3F0)

... then the assembly avoids the rounding:

	mov     x4, sp
	adrp    x0, 0 <kstack_offset>
	mrs     x5, tpidr_el1
	add     x0, x0, #:lo12:kstack_offset
	ldr     w0, [x0, x5]
	and     x0, x0, #0x3f0
	sub     sp, x4, x0

Mark.