[PATCH V2] mm/powerpc: Fix _PAGE_PTE breaking swapoff

Core kernel expect swp_entry_t to be consisting of
only swap type and swap offset. We should not leak pte bits to
swp_entry_t. This breaks swapoff which use the swap type and offset
to build a swp_entry_t and later compare that to the swp_entry_t
obtained from linux page table pte. Leaking pte bits to swp_entry_t
breaks that comparison and results in us looping in try_to_unuse.

The stack trace can be anywhere below try_to_unuse() in mm/swapfile.c,
since swapoff is circling around and around that function, reading from
each used swap block into a page, then trying to find where that page
belongs, looking at every non-file pte of every mm that ever swapped.

Reported-by: Hugh Dickins <hughd@google.com>
Suggested-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Aneesh Kumar K.V <redacted>
---
Changes from V1:
* improve change log and code comment

 arch/powerpc/include/asm/book3s/64/pgtable.h | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

diff --git a/arch/powerpc/include/asm/book3s/64/pgtable.h b/arch/powerpc/include/asm/book3s/64/pgtable.h
index 03c1a5a21c0c..cecb971674a8 100644
--- a/arch/powerpc/include/asm/book3s/64/pgtable.h
+++ b/arch/powerpc/include/asm/book3s/64/pgtable.h

@@ -158,9 +158,14 @@ static inline void pgd_set(pgd_t *pgdp, unsigned long val)
 #define __swp_entry(type, offset)	((swp_entry_t) { \
 					((type) << _PAGE_BIT_SWAP_TYPE) \
 					| ((offset) << PTE_RPN_SHIFT) })
-
-#define __pte_to_swp_entry(pte)		((swp_entry_t) { pte_val((pte)) })
-#define __swp_entry_to_pte(x)		__pte((x).val)
+/*
+ * swp_entry_t should be independent of pte bits. We build a swp_entry_t from
+ * swap type and offset we get from swap and convert that to pte to
+ * find a matching pte in linux page table.
+ * Clear bits not found in swap entries here
+ */
+#define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val((pte)) & ~_PAGE_PTE })
+#define __swp_entry_to_pte(x)	__pte((x).val | _PAGE_PTE)
 
 #ifdef CONFIG_HAVE_ARCH_SOFT_DIRTY
 #define _PAGE_SWP_SOFT_DIRTY   (1UL << (SWP_TYPE_BITS + _PAGE_BIT_SWAP_TYPE))

-- 
2.5.0

On Mon, 11 Jan 2016, Aneesh Kumar K.V wrote:

Core kernel expect swp_entry_t to be consisting of
only swap type and swap offset. We should not leak pte bits to
swp_entry_t. This breaks swapoff which use the swap type and offset
to build a swp_entry_t and later compare that to the swp_entry_t
obtained from linux page table pte. Leaking pte bits to swp_entry_t
breaks that comparison and results in us looping in try_to_unuse.

The stack trace can be anywhere below try_to_unuse() in mm/swapfile.c,
since swapoff is circling around and around that function, reading from
each used swap block into a page, then trying to find where that page
belongs, looking at every non-file pte of every mm that ever swapped.

Reported-by: Hugh Dickins <hughd@google.com>
Suggested-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Aneesh Kumar K.V <redacted>

I think we've seen enough of my name above, but if it helps further
Acked-by: Hugh Dickins <hughd@google.com>

Though I don't find the code comment below on swp_entry_t enlightening -
your commit description above is much more helpful.  If I were writing it,
I might say... hmm, it's too hard: given all the convolutions, I gave up.

On Mon, 2016-01-11 at 21:19 +0530, Aneesh Kumar K.V wrote:

Core kernel expect swp_entry_t to be consisting of
only swap type and swap offset. We should not leak pte bits to
swp_entry_t. This breaks swapoff which use the swap type and offset
to build a swp_entry_t and later compare that to the swp_entry_t
obtained from linux page table pte. Leaking pte bits to swp_entry_t
breaks that comparison and results in us looping in try_to_unuse.

The stack trace can be anywhere below try_to_unuse() in mm/swapfile.c,
since swapoff is circling around and around that function, reading from
each used swap block into a page, then trying to find where that page
belongs, looking at every non-file pte of every mm that ever swapped.

Reported-by: Hugh Dickins <hughd@google.com>
Suggested-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Aneesh Kumar K.V <redacted>

Thanks. I slightly edited the wording in the change log and added:

Fixes: 6a119eae942c ("powerpc/mm: Add a _PAGE_PTE bit")

cheers

On Mon, 2016-11-01 at 15:49:34 UTC, "Aneesh Kumar K.V" wrote:

Core kernel expect swp_entry_t to be consisting of
only swap type and swap offset. We should not leak pte bits to
swp_entry_t. This breaks swapoff which use the swap type and offset
to build a swp_entry_t and later compare that to the swp_entry_t
obtained from linux page table pte. Leaking pte bits to swp_entry_t
breaks that comparison and results in us looping in try_to_unuse.

The stack trace can be anywhere below try_to_unuse() in mm/swapfile.c,
since swapoff is circling around and around that function, reading from
each used swap block into a page, then trying to find where that page
belongs, looking at every non-file pte of every mm that ever swapped.

Reported-by: Hugh Dickins <hughd@google.com>
Suggested-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Aneesh Kumar K.V <redacted>
Acked-by: Hugh Dickins <hughd@google.com>

Applied to powerpc next, thanks.

https://git.kernel.org/powerpc/c/44734f23de2465c3c0d39e4a16

cheers