thp: fix MADV_DONTNEED vs. numa balancing race

In case prot_numa, we are under down_read(mmap_sem).  It's critical to
not clear pmd intermittently to avoid race with MADV_DONTNEED which is
also under down_read(mmap_sem):

	CPU0:				CPU1:
				change_huge_pmd(prot_numa=1)
				 pmdp_huge_get_and_clear_notify()
madvise_dontneed()
 zap_pmd_range()
  pmd_trans_huge(*pmd) == 0 (without ptl)
  // skip the pmd
				 set_pmd_at();
				 // pmd is re-established

The race makes MADV_DONTNEED miss the huge pmd and don't clear it
which may break userspace.

Found by code analysis, never saw triggered.

Link: http://lkml.kernel.org/r/20170302151034.27829-3-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hillf Danton <hillf.zj@alibaba-inc.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index a513861a903796d1639a49cff8c16f36bacf7fb9..26769465af63315766c5fe677c01d52fd48cd912 100644 (file)
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1746,7 +1746,39 @@ int change_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
        if (prot_numa && pmd_protnone(*pmd))
                goto unlock;
 
-       entry = pmdp_huge_get_and_clear_notify(mm, addr, pmd);
+       /*
+        * In case prot_numa, we are under down_read(mmap_sem). It's critical
+        * to not clear pmd intermittently to avoid race with MADV_DONTNEED
+        * which is also under down_read(mmap_sem):
+        *
+        *      CPU0:                           CPU1:
+        *                              change_huge_pmd(prot_numa=1)
+        *                               pmdp_huge_get_and_clear_notify()
+        * madvise_dontneed()
+        *  zap_pmd_range()
+        *   pmd_trans_huge(*pmd) == 0 (without ptl)
+        *   // skip the pmd
+        *                               set_pmd_at();
+        *                               // pmd is re-established
+        *
+        * The race makes MADV_DONTNEED miss the huge pmd and don't clear it
+        * which may break userspace.
+        *
+        * pmdp_invalidate() is required to make sure we don't miss
+        * dirty/young flags set by hardware.
+        */
+       entry = *pmd;
+       pmdp_invalidate(vma, addr, pmd);
+
+       /*
+        * Recover dirty/young flags.  It relies on pmdp_invalidate to not
+        * corrupt them.
+        */
+       if (pmd_dirty(*pmd))
+               entry = pmd_mkdirty(entry);
+       if (pmd_young(*pmd))
+               entry = pmd_mkyoung(entry);
+
        entry = pmd_modify(entry, newprot);
        if (preserve_write)
                entry = pmd_mk_savedwrite(entry);