[PATCH v3 13/29] riscv mmu: write protect and shadow stack

[Deepak Gupta]

On Thu, May 23, 2024 at 04:59:30PM +0200, Alexandre Ghiti wrote:
>Hi Deepak,
>
>On Mon, May 13, 2024 at 7:32 PM Deepak Gupta <debug at rivosinc.com> wrote:
>>
>> On Sun, May 12, 2024 at 06:31:24PM +0200, Alexandre Ghiti wrote:
>> >On 04/04/2024 01:35, Deepak Gupta wrote:
>> >>`fork` implements copy on write (COW) by making pages readonly in child
>> >>and parent both.
>> >>
>> >>ptep_set_wrprotect and pte_wrprotect clears _PAGE_WRITE in PTE.
>> >>Assumption is that page is readable and on fault copy on write happens.
>> >>
>> >>To implement COW on such pages,
>> >
>> >
>> >I guess you mean "shadow stack pages" here.
>>
>> Yes I meant shadow stack pages. Will fix the message.
>>
>> >
>> >
>> >>  clearing up W bit makes them XWR = 000.
>> >>This will result in wrong PTE setting which says no perms but V=1 and PFN
>> >>field pointing to final page. Instead desired behavior is to turn it into
>> >>a readable page, take an access (load/store) fault on sspush/sspop
>> >>(shadow stack) and then perform COW on such pages.
>> >>This way regular reads
>> >>would still be allowed and not lead to COW maintaining current behavior
>> >>of COW on non-shadow stack but writeable memory.
>> >>
>> >>On the other hand it doesn't interfere with existing COW for read-write
>> >>memory. Assumption is always that _PAGE_READ must have been set and thus
>> >>setting _PAGE_READ is harmless.
>> >>
>> >>Signed-off-by: Deepak Gupta <debug at rivosinc.com>
>> >>---
>> >>  arch/riscv/include/asm/pgtable.h | 12 ++++++++++--
>> >>  1 file changed, 10 insertions(+), 2 deletions(-)
>> >>
>> >>diff --git a/arch/riscv/include/asm/pgtable.h b/arch/riscv/include/asm/pgtable.h
>> >>index 9b837239d3e8..7a1c2a98d272 100644
>> >>--- a/arch/riscv/include/asm/pgtable.h
>> >>+++ b/arch/riscv/include/asm/pgtable.h
>> >>@@ -398,7 +398,7 @@ static inline int pte_special(pte_t pte)
>> >>  static inline pte_t pte_wrprotect(pte_t pte)
>> >>  {
>> >>-     return __pte(pte_val(pte) & ~(_PAGE_WRITE));
>> >>+     return __pte((pte_val(pte) & ~(_PAGE_WRITE)) | (_PAGE_READ));
>> >>  }
>> >>  /* static inline pte_t pte_mkread(pte_t pte) */
>> >>@@ -581,7 +581,15 @@ static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
>> >>  static inline void ptep_set_wrprotect(struct mm_struct *mm,
>> >>                                    unsigned long address, pte_t *ptep)
>> >>  {
>> >>-     atomic_long_and(~(unsigned long)_PAGE_WRITE, (atomic_long_t *)ptep);
>> >>+     volatile pte_t read_pte = *ptep;
>
>Sorry I missed this ^. You need to use ptep_get() to get the value of
>a pte. 

Noted. will fix it.

>And why do you need the volatile here?

I don't remember the reason. It's probably not needed here.
But I am sure I was debugging something and trying everything.
And this probably slipped sanitization before sending patches.

Will fix it.
>
>> >>+     /*
>> >>+      * ptep_set_wrprotect can be called for shadow stack ranges too.
>> >>+      * shadow stack memory is XWR = 010 and thus clearing _PAGE_WRITE will lead to
>> >>+      * encoding 000b which is wrong encoding with V = 1. This should lead to page fault
>> >>+      * but we dont want this wrong configuration to be set in page tables.
>> >>+      */
>> >>+     atomic_long_set((atomic_long_t *)ptep,
>> >>+                     ((pte_val(read_pte) & ~(unsigned long)_PAGE_WRITE) | _PAGE_READ));
>> >>  }
>> >>  #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
>> >
>> >
>> >Doesn't making the shadow stack page readable allow "normal" loads to
>> >access the page? If it does, isn't that an issue (security-wise)?
>>
>> When shadow stack permissions are there (i.e. R=0, W=1, X=0), then also shadow stack is
>> readable through "normal" loads. So nothing changes when it converts into a readonly page
>> from page permissions perspective.
>>
>> Security-wise it's not a concern because from threat modeling perspective, if attacker had
>> read-write primitives (via some bug in program) available to read and write address space
>> of process/task; then they would have availiblity of return addresses on normal stack. It's
>> the write primitive that is concerning and to be protected against. And that's why shadow stack
>> is not writeable using "normal" stores.
>>
>> >
>
>Thanks for the explanation!
>
>With the use of ptep_get(), you can add:
>
>Reviewed-by: Alexandre Ghiti <alexghiti at rivosinc.com>
>
>Thanks,
>
>Alex