[PATCH] mm/swap: Fix pte_same_as_swp() not removing uffd-wp bit when compare

[PATCH] mm/swap: Fix pte_same_as_swp() not removing uffd-wp bit when compare

[Peter Xu]

I found it by pure code review, that pte_same_as_swp() of unuse_vma() didn't
take uffd-wp bit into account when comparing ptes.  pte_same_as_swp() returning
false negative could cause failure to swapoff swap ptes that was wr-protected
by userfaultfd.

Cc: Hugh Dickins <hughd@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Peter Xu <peterx@redhat.com>
---
 include/linux/swapops.h | 15 +++++++++++----
 mm/swapfile.c           |  2 +-
 2 files changed, 12 insertions(+), 5 deletions(-)

diff --git a/include/linux/swapops.h b/include/linux/swapops.h
index af3d2661e41e..d356ab4047f7 100644
--- a/include/linux/swapops.h
+++ b/include/linux/swapops.h
@@ -23,6 +23,16 @@
 #define SWP_TYPE_SHIFT	(BITS_PER_XA_VALUE - MAX_SWAPFILES_SHIFT)
 #define SWP_OFFSET_MASK	((1UL << SWP_TYPE_SHIFT) - 1)
 
+/* Clear all flags but only keep swp_entry_t related information */
+static inline pte_t pte_swp_clear_flags(pte_t pte)
+{
+	if (pte_swp_soft_dirty(pte))
+		pte = pte_swp_clear_soft_dirty(pte);
+	if (pte_swp_uffd_wp(pte))
+		pte = pte_swp_clear_uffd_wp(pte);
+	return pte;
+}
+
 /*
  * Store a type+offset into a swp_entry_t in an arch-independent format
  */
@@ -66,10 +76,7 @@ static inline swp_entry_t pte_to_swp_entry(pte_t pte)
 {
 	swp_entry_t arch_entry;
 
-	if (pte_swp_soft_dirty(pte))
-		pte = pte_swp_clear_soft_dirty(pte);
-	if (pte_swp_uffd_wp(pte))
-		pte = pte_swp_clear_uffd_wp(pte);
+	pte = pte_swp_clear_flags(pte);
 	arch_entry = __pte_to_swp_entry(pte);
 	return swp_entry(__swp_type(arch_entry), __swp_offset(arch_entry));
 }
diff --git a/mm/swapfile.c b/mm/swapfile.c
index 6212bc033602..1e07d1c776f2 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -1887,7 +1887,7 @@ unsigned int count_swap_pages(int type, int free)
 
 static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
 {
-	return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
+	return pte_same(pte_swp_clear_flags(pte), swp_pte);
 }
 
 /*
-- 
2.31.1

Re: [PATCH] mm/swap: Fix pte_same_as_swp() not removing uffd-wp bit when compare

[Hugh Dickins]

On Thu, 3 Jun 2021, Peter Xu wrote:

> I found it by pure code review, that pte_same_as_swp() of unuse_vma() didn't

Yes, that is an odd corner, easily missed.

> take uffd-wp bit into account when comparing ptes.  pte_same_as_swp() returning
> false negative could cause failure to swapoff swap ptes that was wr-protected
> by userfaultfd.
> 
> Cc: Hugh Dickins <hughd@google.com>
> Cc: Andrea Arcangeli <aarcange@redhat.com>
> Cc: Andrew Morton <akpm@linux-foundation.org>
> Signed-off-by: Peter Xu <peterx@redhat.com>

I expect you're right: swapoff used to hang forever (but interruptibly)
when this went wrong on powerpc originally.  I don't know the uffd_wp
(nor the soft_dirty) end of it, but treating uffd_wp and soft_dirty
together looks a very good approach, so I'll venture an

Acked-by: Hugh Dickins <hughd@google.com>

But I think it should have a uffd_wp Fixes tag and be Cc stable.

> ---
>  include/linux/swapops.h | 15 +++++++++++----
>  mm/swapfile.c           |  2 +-
>  2 files changed, 12 insertions(+), 5 deletions(-)
> 
> diff --git a/include/linux/swapops.h b/include/linux/swapops.h
> index af3d2661e41e..d356ab4047f7 100644
> --- a/include/linux/swapops.h
> +++ b/include/linux/swapops.h
> @@ -23,6 +23,16 @@
>  #define SWP_TYPE_SHIFT	(BITS_PER_XA_VALUE - MAX_SWAPFILES_SHIFT)
>  #define SWP_OFFSET_MASK	((1UL << SWP_TYPE_SHIFT) - 1)
>  
> +/* Clear all flags but only keep swp_entry_t related information */
> +static inline pte_t pte_swp_clear_flags(pte_t pte)
> +{
> +	if (pte_swp_soft_dirty(pte))
> +		pte = pte_swp_clear_soft_dirty(pte);
> +	if (pte_swp_uffd_wp(pte))
> +		pte = pte_swp_clear_uffd_wp(pte);
> +	return pte;
> +}
> +
>  /*
>   * Store a type+offset into a swp_entry_t in an arch-independent format
>   */
> @@ -66,10 +76,7 @@ static inline swp_entry_t pte_to_swp_entry(pte_t pte)
>  {
>  	swp_entry_t arch_entry;
>  
> -	if (pte_swp_soft_dirty(pte))
> -		pte = pte_swp_clear_soft_dirty(pte);
> -	if (pte_swp_uffd_wp(pte))
> -		pte = pte_swp_clear_uffd_wp(pte);
> +	pte = pte_swp_clear_flags(pte);
>  	arch_entry = __pte_to_swp_entry(pte);
>  	return swp_entry(__swp_type(arch_entry), __swp_offset(arch_entry));
>  }
> diff --git a/mm/swapfile.c b/mm/swapfile.c
> index 6212bc033602..1e07d1c776f2 100644
> --- a/mm/swapfile.c
> +++ b/mm/swapfile.c
> @@ -1887,7 +1887,7 @@ unsigned int count_swap_pages(int type, int free)
>  
>  static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte)
>  {
> -	return pte_same(pte_swp_clear_soft_dirty(pte), swp_pte);
> +	return pte_same(pte_swp_clear_flags(pte), swp_pte);
>  }
>  
>  /*
> -- 
> 2.31.1

Re: [PATCH] mm/swap: Fix pte_same_as_swp() not removing uffd-wp bit when compare

[Peter Xu]

On Thu, Jun 03, 2021 at 08:26:02PM -0700, Hugh Dickins wrote:
> On Thu, 3 Jun 2021, Peter Xu wrote:
> 
> > I found it by pure code review, that pte_same_as_swp() of unuse_vma() didn't
> 
> Yes, that is an odd corner, easily missed.
> 
> > take uffd-wp bit into account when comparing ptes.  pte_same_as_swp() returning
> > false negative could cause failure to swapoff swap ptes that was wr-protected
> > by userfaultfd.
> > 
> > Cc: Hugh Dickins <hughd@google.com>
> > Cc: Andrea Arcangeli <aarcange@redhat.com>
> > Cc: Andrew Morton <akpm@linux-foundation.org>
> > Signed-off-by: Peter Xu <peterx@redhat.com>
> 
> I expect you're right: swapoff used to hang forever (but interruptibly)
> when this went wrong on powerpc originally.  I don't know the uffd_wp
> (nor the soft_dirty) end of it, but treating uffd_wp and soft_dirty
> together looks a very good approach, so I'll venture an
> 
> Acked-by: Hugh Dickins <hughd@google.com>

Thanks!

> 
> But I think it should have a uffd_wp Fixes tag and be Cc stable.

Yes, should be:

Cc: stable@vger.kernel.org # v5.7+
Fixes: f45ec5ff16a7 ("userfaultfd: wp: support swap and page migration")

-- 
Peter Xu