Re: [PATCH] lkdtm: Add CFI_BACKWARD to test ROP mitigations

[Dan Li]

Hi, Kees,
Thanks for the rewrite. I tested this patch, and it works fine for
me except for a few minor comments below :)

On 4/13/22 14:39, Kees Cook wrote:
> +/* The ultimate ROP gadget. */
> +static noinline __no_ret_protection
> +void set_return_addr_unchecked(unsigned long *expected, unsigned long *addr)
> +{
> +	/* Use of volatile is to make sure final write isn't seen as a dead store. */
> +	unsigned long * volatile *ret_addr = (unsigned long **)__builtin_frame_address(0) + 1;
> +
> +	/* Make sure we've found the right place on the stack before writing it. */
> +	if(*ret_addr == expected)
> +		*ret_addr = (addr);
> +	else
> +		/* Check architecture, stack layout, or compiler behavior... */
> +		pr_warn("Eek: return address mismatch! %px != %px\n",
> +			*ret_addr, addr);
> +}
> +
> +static noinline
> +void set_return_addr(unsigned long *expected, unsigned long *addr)
> +{
> +	/* Use of volatile is to make sure final write isn't seen as a dead store. */
> +	unsigned long * volatile *ret_addr = (unsigned long **)__builtin_frame_address(0) + 1;
> +
> +	/* Make sure we've found the right place on the stack before writing it. */
> +	if(*ret_addr == expected)
> +		*ret_addr = (addr);

When PAC is enabled, I get a mismatch as follows:

/kselftest_install/lkdtm # ./CFI_BACKWARD.sh
[  182.120133] lkdtm: Performing direct entry CFI_BACKWARD
[  182.120665] lkdtm: Attempting unchecked stack return address redirection ...
[  182.122543] lkdtm: ok: redirected stack return address.
[  182.123521] lkdtm: Attempting checked stack return address redirection ...
[  182.123964] lkdtm: Eek: return address mismatch! bfff800008fa8014 != ffff800008fa8030
[  182.124502] lkdtm: ok: control flow unchanged.
CFI_BACKWARD: saw 'call trace:|ok: control flow unchanged': ok

We may need to ignore the pac high bits of return address according
to TCR.T1SZ (or simply remove the high 16 bits before comparing).

> +	else
> +		/* Check architecture, stack layout, or compiler behavior... */
> +		pr_warn("Eek: return address mismatch! %px != %px\n",
> +			*ret_addr, addr);

According to the context, it might be "expected" here?

		pr_warn("Eek: return address mismatch! %px != %px\n",
			*ret_addr, expected);

I simply ignored the upper 16 bits, and tested it separately
in gcc/llvm 12 with SCS/PAC and all the four cases worked fine for me.

Thanks,
Dan.