diff mbox series

[RFC,v9,16/27] mm: Update can_follow_write_pte() for Shadow Stack

Message ID 20200205181935.3712-17-yu-cheng.yu@intel.com (mailing list archive)
State New, archived
Headers show
Series Control-flow Enforcement: Shadow Stack | expand

Commit Message

Yu-cheng Yu Feb. 5, 2020, 6:19 p.m. UTC 
Can_follow_write_pte() verifies that a read-only page is the task's own
copy by ensuring the page has gone through faultin_page() and the PTE is
Dirty.

A Shadow Stack (SHSTK) PTE must be (read-only + _PAGE_DIRTY_HW).  When a
task does fork(), its SHSTK PTEs become (read-only + _PAGE_DIRTY_SW).  This
causes the next SHSTK access (i.e. CALL, RET, INCSSP) to trigger a fault;
the page is then copied, and (read-only + _PAGE_DIRTY_HW) is restored.

To update can_follow_write_pte() for SHSTK, introduce pte_exclusive().  It
verifies a data PTE is Dirty and a SHSTK PTE has _PAGE_DIRTY_HW.

Also rename can_follow_write_pte() to can_follow_write() to make its
meaning clear; i.e. "Can we write to the page?", not "Is the PTE writable?"

Also apply same changes to the huge memory case.

Signed-off-by: Yu-cheng Yu <yu-cheng.yu@intel.com>
---
 arch/x86/mm/pgtable.c         | 18 ++++++++++++++++++
 include/asm-generic/pgtable.h | 12 ++++++++++++
 mm/gup.c                      |  8 +++++---
 mm/huge_memory.c              |  8 +++++---
 4 files changed, 40 insertions(+), 6 deletions(-)

Comments

Dave Hansen Feb. 27, 2020, 12:34 a.m. UTC | #1 
> +inline bool pte_exclusive(pte_t pte, struct vm_area_struct *vma)
> +{
> +	if (vma->vm_flags & VM_SHSTK)
> +		return pte_dirty_hw(pte);
> +	else
> +		return pte_dirty(pte);
> +}

I'm not really getting the naming.  What is exclusive?

> diff --git a/mm/gup.c b/mm/gup.c
> index 7646bf993b25..d1dbfbde8443 100644
> --- a/mm/gup.c
> +++ b/mm/gup.c
> @@ -164,10 +164,12 @@ static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
>   * FOLL_FORCE can write to even unwritable pte's, but only
>   * after we've gone through a COW cycle and they are dirty.
>   */
> -static inline bool can_follow_write_pte(pte_t pte, unsigned int flags)
> +static inline bool can_follow_write(pte_t pte, unsigned int flags,
> +				    struct vm_area_struct *vma)

Having two identically named functions in two files in the same
subsystem seems like a recipe for confusion when I grep or cscope for
things.  It hardly seems worth the 4 characters of space savings IMNHO.

>  {
>  	return pte_write(pte) ||
> -		((flags & FOLL_FORCE) && (flags & FOLL_COW) && pte_dirty(pte));
> +		((flags & FOLL_FORCE) && (flags & FOLL_COW) &&
> +		 pte_exclusive(pte, vma));
>  }

FWIW, this is the hunk that fixed DirtyCOW.

The least this deserves is acknowledgement of that in the changelog and
a missive about how you're sure you didn't just introduce
ShadowDirtyCOW.  Don't bother.  I already registered the domain. ;)
diff mbox series

Patch

diff --git a/arch/x86/mm/pgtable.c b/arch/x86/mm/pgtable.c
index 3340b1d4e9da..fa8133f37918 100644
--- a/arch/x86/mm/pgtable.c
+++ b/arch/x86/mm/pgtable.c
@@ -887,6 +887,15 @@  inline pte_t pte_set_vma_features(pte_t pte, struct vm_area_struct *vma)
 		return pte;
 }
 
+inline bool pte_exclusive(pte_t pte, struct vm_area_struct *vma)
+{
+	if (vma->vm_flags & VM_SHSTK)
+		return pte_dirty_hw(pte);
+	else
+		return pte_dirty(pte);
+}
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 inline pmd_t pmd_set_vma_features(pmd_t pmd, struct vm_area_struct *vma)
 {
 	if (vma->vm_flags & VM_SHSTK)
@@ -894,4 +903,13 @@  inline pmd_t pmd_set_vma_features(pmd_t pmd, struct vm_area_struct *vma)
 	else
 		return pmd;
 }
+
+inline bool pmd_exclusive(pmd_t pmd, struct vm_area_struct *vma)
+{
+	if (vma->vm_flags & VM_SHSTK)
+		return pmd_dirty_hw(pmd);
+	else
+		return pmd_dirty(pmd);
+}
+#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 #endif /* CONFIG_X86_INTEL_SHADOW_STACK_USER */
diff --git a/include/asm-generic/pgtable.h b/include/asm-generic/pgtable.h
index a9df093fdf45..ae9a84fffc25 100644
--- a/include/asm-generic/pgtable.h
+++ b/include/asm-generic/pgtable.h
@@ -1202,18 +1202,30 @@  static inline pte_t pte_set_vma_features(pte_t pte, struct vm_area_struct *vma)
 	return pte;
 }
 
+static inline bool pte_exclusive(pte_t pte, struct vm_area_struct *vma)
+{
+	return pte_dirty(pte);
+}
+
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 static inline pmd_t pmd_set_vma_features(pmd_t pmd, struct vm_area_struct *vma)
 {
 	return pmd;
 }
+
+static inline bool pmd_exclusive(pmd_t pmd, struct vm_area_struct *vma)
+{
+	return pmd_dirty(pmd);
+}
 #endif
 #else
 bool arch_copy_pte_mapping(vm_flags_t vm_flags);
 pte_t pte_set_vma_features(pte_t pte, struct vm_area_struct *vma);
+bool pte_exclusive(pte_t pte, struct vm_area_struct *vma);
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 pmd_t pmd_set_vma_features(pmd_t pmd, struct vm_area_struct *vma);
+bool pmd_exclusive(pmd_t pmd, struct vm_area_struct *vma);
 #endif
 #endif
 #endif /* CONFIG_MMU */
diff --git a/mm/gup.c b/mm/gup.c
index 7646bf993b25..d1dbfbde8443 100644
--- a/mm/gup.c
+++ b/mm/gup.c
@@ -164,10 +164,12 @@  static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
  * FOLL_FORCE can write to even unwritable pte's, but only
  * after we've gone through a COW cycle and they are dirty.
  */
-static inline bool can_follow_write_pte(pte_t pte, unsigned int flags)
+static inline bool can_follow_write(pte_t pte, unsigned int flags,
+				    struct vm_area_struct *vma)
 {
 	return pte_write(pte) ||
-		((flags & FOLL_FORCE) && (flags & FOLL_COW) && pte_dirty(pte));
+		((flags & FOLL_FORCE) && (flags & FOLL_COW) &&
+		 pte_exclusive(pte, vma));
 }
 
 static struct page *follow_page_pte(struct vm_area_struct *vma,
@@ -205,7 +207,7 @@  static struct page *follow_page_pte(struct vm_area_struct *vma,
 	}
 	if ((flags & FOLL_NUMA) && pte_protnone(pte))
 		goto no_page;
-	if ((flags & FOLL_WRITE) && !can_follow_write_pte(pte, flags)) {
+	if ((flags & FOLL_WRITE) && !can_follow_write(pte, flags, vma)) {
 		pte_unmap_unlock(ptep, ptl);
 		return NULL;
 	}
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 93ef368df2dd..baad346e9f4a 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1469,10 +1469,12 @@  vm_fault_t do_huge_pmd_wp_page(struct vm_fault *vmf, pmd_t orig_pmd)
  * FOLL_FORCE can write to even unwritable pmd's, but only
  * after we've gone through a COW cycle and they are dirty.
  */
-static inline bool can_follow_write_pmd(pmd_t pmd, unsigned int flags)
+static inline bool can_follow_write(pmd_t pmd, unsigned int flags,
+				    struct vm_area_struct *vma)
 {
 	return pmd_write(pmd) ||
-	       ((flags & FOLL_FORCE) && (flags & FOLL_COW) && pmd_dirty(pmd));
+	       ((flags & FOLL_FORCE) && (flags & FOLL_COW) &&
+		pmd_exclusive(pmd, vma));
 }
 
 struct page *follow_trans_huge_pmd(struct vm_area_struct *vma,
@@ -1485,7 +1487,7 @@  struct page *follow_trans_huge_pmd(struct vm_area_struct *vma,
 
 	assert_spin_locked(pmd_lockptr(mm, pmd));
 
-	if (flags & FOLL_WRITE && !can_follow_write_pmd(*pmd, flags))
+	if (flags & FOLL_WRITE && !can_follow_write(*pmd, flags, vma))
 		goto out;
 
 	/* Avoid dumping huge zero page */