thp: fix corner case of munlock() of PTE-mapped THPs

The following program triggers BUG() in munlock_vma_pages_range():

	// autogenerated by syzkaller (http://github.com/google/syzkaller)
	#include <sys/mman.h>

	int main()
	{
	  mmap((void*)0x20105000ul, 0xc00000ul, 0x2ul, 0x2172ul, -1, 0);
	  mremap((void*)0x201fd000ul, 0x4000ul, 0xc00000ul, 0x3ul, 0x203f0000ul);
	  return 0;
	}

The test-case constructs the situation when munlock_vma_pages_range()
finds PTE-mapped THP-head in the middle of page table and, by mistake,
skips HPAGE_PMD_NR pages after that.

As result, on the next iteration it hits the middle of PMD-mapped THP
and gets upset seeing mlocked tail page.

The solution is only skip HPAGE_PMD_NR pages if the THP was mlocked
during munlock_vma_page().  It would guarantee that the page is
PMD-mapped as we never mlock PTE-mapeed THPs.

Fixes: e90309c ("thp: allow mlocked THP again")
Link: http://lkml.kernel.org/r/20161115132703.7s7rrgmwttegcdh4@black.fi.intel.com
Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Cc: Konstantin Khlebnikov <koct9i@gmail.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: syzkaller <syzkaller@googlegroups.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: <stable@vger.kernel.org>	[4.5+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mm/mlock.c

@@ -190,10 +190,13 @@ unsigned int munlock_vma_page(struct page *page)
 	 */
 	spin_lock_irq(zone_lru_lock(zone));
 
-	nr_pages = hpage_nr_pages(page);
-	if (!TestClearPageMlocked(page))
+	if (!TestClearPageMlocked(page)) {
+		/* Potentially, PTE-mapped THP: do not skip the rest PTEs */
+		nr_pages = 1;
 		goto unlock_out;
+	}
 
+	nr_pages = hpage_nr_pages(page);
 	__mod_zone_page_state(zone, NR_MLOCK, -nr_pages);
 
 	if (__munlock_isolate_lru_page(page, true)) {