Re: [RFC PATCH v3 12/24] x86/mm: Modify ptep_set_wrprotect and pmdp_set_wrprotect for _PAGE_DIRTY_SW

On 08/30/2018 08:49 AM, Jann Horn wrote:

quoted

@@ -1203,7 +1203,28 @@ static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm,
 static inline void ptep_set_wrprotect(struct mm_struct *mm,
                                      unsigned long addr, pte_t *ptep)
 {
+       pte_t pte;
+
        clear_bit(_PAGE_BIT_RW, (unsigned long *)&ptep->pte);
+       pte = *ptep;
+
+       /*
+        * Some processors can start a write, but ending up seeing
+        * a read-only PTE by the time they get to the Dirty bit.
+        * In this case, they will set the Dirty bit, leaving a
+        * read-only, Dirty PTE which looks like a Shadow Stack PTE.
+        *
+        * However, this behavior has been improved and will not occur
+        * on processors supporting Shadow Stacks.  Without this
+        * guarantee, a transition to a non-present PTE and flush the
+        * TLB would be needed.
+        *
+        * When change a writable PTE to read-only and if the PTE has
+        * _PAGE_DIRTY_HW set, we move that bit to _PAGE_DIRTY_SW so
+        * that the PTE is not a valid Shadow Stack PTE.
+        */
+       pte = pte_move_flags(pte, _PAGE_DIRTY_HW, _PAGE_DIRTY_SW);
+       set_pte_at(mm, addr, ptep, pte);
 }

I don't understand why it's okay that you first atomically clear the
RW bit, then atomically switch from DIRTY_HW to DIRTY_SW. Doesn't that
mean that between the two atomic writes, another core can incorrectly
see a shadow stack?

Good point.

This could result in a spurious shadow-stack fault, or allow a
shadow-stack write to the page in the transient state.

But, the shadow-stack permissions are more restrictive than what could
be in the TLB at this point, so I don't think there's a real security
implication here.

The only trouble is handling the spurious shadow-stack fault.  The
alternative is to go !Present for a bit, which we would probably just
handle fine in the existing page fault code.