On Mon, Aug 07, 2023 at 11:00:16PM +0100, Mark Brown wrote:
> diff --git a/arch/arm64/mm/mmap.c b/arch/arm64/mm/mmap.c
> index 8f5b7ce857ed..8f40198cd44e 100644
> --- a/arch/arm64/mm/mmap.c
> +++ b/arch/arm64/mm/mmap.c
> @@ -79,8 +79,18 @@ arch_initcall(adjust_protection_map);
>
> pgprot_t vm_get_page_prot(unsigned long vm_flags)
> {
> - pteval_t prot = pgprot_val(protection_map[vm_flags &
> + pteval_t prot;
> +
> + /* If this is a GCS then only interpret VM_WRITE. */
> + if (system_supports_gcs() && (vm_flags & VM_SHADOW_STACK)) {
> + if (vm_flags & VM_WRITE)
> + prot = _PAGE_GCS;
> + else
> + prot = _PAGE_GCS_RO;
> + } else {
> + prot = pgprot_val(protection_map[vm_flags &
> (VM_READ|VM_WRITE|VM_EXEC|VM_SHARED)]);
> + }
>
> if (vm_flags & VM_ARM64_BTI)
> prot |= PTE_GP;
Some combinations here don't make sense like GCS + exec or BTI. I think
the code above (correctly) ignores exec but it still sets PTE_GP if BTI
(the architecture may allow this but you can't execute from the GCS page
anyway).
I haven't checked the x86 patches to see when VM_SHADOW_STACK is set but
if there's no additional check at a higher level, we should add
something to arch_validate_flags(), assuming it's called on those paths.
--
Catalin
