mm: Fix NULL pointer dereference in madvise(MADV_WILLNEED) support
authorKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Fri, 20 Dec 2013 13:10:03 +0000 (15:10 +0200)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 20 Dec 2013 20:17:03 +0000 (12:17 -0800)
commitee53664bda169f519ce3c6a22d378f0b946c8178
treeacbf094289e6905013c66cb259e373c57af42cc8
parent4203d0eb3acc459d1e7737193b5684e71185dca7
mm: Fix NULL pointer dereference in madvise(MADV_WILLNEED) support

Sasha Levin found a NULL pointer dereference that is due to a missing
page table lock, which in turn is due to the pmd entry in question being
a transparent huge-table entry.

The code - introduced in commit 1998cc048901 ("mm: make
madvise(MADV_WILLNEED) support swap file prefetch") - correctly checks
for this situation using pmd_none_or_trans_huge_or_clear_bad(), but it
turns out that that function doesn't work correctly.

pmd_none_or_trans_huge_or_clear_bad() expected that pmd_bad() would
trigger if the transparent hugepage bit was set, but it doesn't do that
if pmd_numa() is also set. Note that the NUMA bit only gets set on real
NUMA machines, so people trying to reproduce this on most normal
development systems would never actually trigger this.

Fix it by removing the very subtle (and subtly incorrect) expectation,
and instead just checking pmd_trans_huge() explicitly.

Reported-by: Sasha Levin <sasha.levin@oracle.com>
Acked-by: Andrea Arcangeli <aarcange@redhat.com>
[ Additionally remove the now stale test for pmd_trans_huge() inside the
  pmd_bad() case - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/asm-generic/pgtable.h