Re: [PATCH v3 05/19] x86/kasan/64: Teach KASAN about the cpu_entry_area

From: Andrey Ryabinin
Date: Fri Nov 24 2017 - 08:13:01 EST




On 11/24/2017 07:32 AM, Andy Lutomirski wrote:
> The cpu_entry_area will contain stacks. Make sure that KASAN has
> appropriate shadow mappings for them.
>
> Cc: Andrey Ryabinin <aryabinin@xxxxxxxxxxxxx>
> Cc: Alexander Potapenko <glider@xxxxxxxxxx>
> Cc: Dmitry Vyukov <dvyukov@xxxxxxxxxx>
> Cc: kasan-dev@xxxxxxxxxxxxxxxx
> Signed-off-by: Andy Lutomirski <luto@xxxxxxxxxx>
> ---
> arch/x86/mm/kasan_init_64.c | 13 ++++++++++++-
> 1 file changed, 12 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/mm/kasan_init_64.c b/arch/x86/mm/kasan_init_64.c
> index 99dfed6dfef8..54561dce742e 100644
> --- a/arch/x86/mm/kasan_init_64.c
> +++ b/arch/x86/mm/kasan_init_64.c
> @@ -277,6 +277,7 @@ void __init kasan_early_init(void)
> void __init kasan_init(void)
> {
> int i;
> + void *cpu_entry_area_begin, *cpu_entry_area_end;
>
> #ifdef CONFIG_KASAN_INLINE
> register_die_notifier(&kasan_die_notifier);
> @@ -329,8 +330,18 @@ void __init kasan_init(void)
> (unsigned long)kasan_mem_to_shadow(_end),
> early_pfn_to_nid(__pa(_stext)));
>
> + cpu_entry_area_begin = (void *)(__fix_to_virt(FIX_CPU_ENTRY_AREA_BOTTOM));
> + cpu_entry_area_end = (void *)(__fix_to_virt(FIX_CPU_ENTRY_AREA_TOP) + PAGE_SIZE);
> +
> kasan_populate_zero_shadow(kasan_mem_to_shadow((void *)MODULES_END),
> - (void *)KASAN_SHADOW_END);
> + kasan_mem_to_shadow(cpu_entry_area_begin));
> +
> + kasan_populate_shadow((unsigned long)kasan_mem_to_shadow(cpu_entry_area_begin),
> + (unsigned long)kasan_mem_to_shadow(cpu_entry_area_end),
> + 0);
> +
> + kasan_populate_zero_shadow(kasan_mem_to_shadow(cpu_entry_area_end),

Seems we need to round_up kasan_mem_to_shadow(cpu_entry_area_end) to the next page
(or alternatively - round_up(cpu_entry_area_end, KASAN_SHADOW_SCALE_SIZE*PAGE_SIZE)).
Otherwise, kasan_populate_zero_shadow() will overpopulate the last shadow page of cpu_entry area with kasan_zero_page.

We don't necessarily need to round_down(kasan_mem_to_shadow(cpu_entry_area_begin), PAGE_SIZE) because
kasan_populate_zero_shadow() will not populate the last 'incomplete' page and kasan_populate_shadow()
does round_down() internally, which is exactly what we want here. But it might be better to round_down()
explicitly anyway, to avoid relying on such subtle implementation details.

> + (void *)KASAN_SHADOW_END);
>
> load_cr3(init_top_pgt);
> __flush_tlb_all();
>