mm: rmap: call page_check_address() with sync enabled to avoid racy check

The previous patch addresses the race between split_huge_pmd_address()
and someone changing the pmd.  The fix is only for splitting of normal
thp (i.e.  pmd-mapped thp,) and for splitting of pte-mapped thp there
still is the similar race.

For splitting pte-mapped thp, the pte's conversion is done by
try_to_unmap_one(TTU_MIGRATION).  This function checks
page_check_address() to get the target pte, but it can return NULL under
some race, leading to VM_BUG_ON() in freeze_page().  Fortunately,
page_check_address() already has an argument to decide whether we do a
quick/racy check or not, so let's flip it when called from
freeze_page().

Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Naoya Horiguchi <[email protected]>
Cc: Kirill A. Shutemov <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

diff --git a/mm/rmap.c b/mm/rmap.c
index 0ea5d9071b32b967d012f36e600a2ee75acd8f3d..e4b713a6ed7ed791a3b8f5f30f1b464d85a3c9df 100644 (file)
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1427,7 +1427,8 @@ static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
                        goto out;
        }
 
-       pte = page_check_address(page, mm, address, &ptl, 0);
+       pte = page_check_address(page, mm, address, &ptl,
+                                PageTransCompound(page));
        if (!pte)
                goto out;