rmap: fix theoretical race between do_wp_page and shrink_active_list
authorVladimir Davydov <vdavydov@parallels.com>
Wed, 24 Jun 2015 23:56:56 +0000 (16:56 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 25 Jun 2015 00:49:42 +0000 (17:49 -0700)
As noted by Paul the compiler is free to store a temporary result in a
variable on stack, heap or global unless it is explicitly marked as
volatile, see:

  http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2015/n4455.html#sample-optimizations

This can result in a race between do_wp_page() and shrink_active_list()
as follows.

In do_wp_page() we can call page_move_anon_rmap(), which sets
page->mapping as follows:

  anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
  page->mapping = (struct address_space *) anon_vma;

The page in question may be on an LRU list, because nowhere in
do_wp_page() we remove it from the list, neither do we take any LRU
related locks.  Although the page is locked, shrink_active_list() can
still call page_referenced() on it concurrently, because the latter does
not require an anonymous page to be locked:

  CPU0                          CPU1
  ----                          ----
  do_wp_page                    shrink_active_list
   lock_page                     page_referenced
                                  PageAnon->yes, so skip trylock_page
   page_move_anon_rmap
    page->mapping = anon_vma
                                  rmap_walk
                                   PageAnon->no
                                   rmap_walk_file
                                    BUG
    page->mapping += PAGE_MAPPING_ANON

This patch fixes this race by explicitly forbidding the compiler to split
page->mapping store in page_move_anon_rmap() with the aid of WRITE_ONCE.

[akpm@linux-foundation.org: tweak comment, per Minchan]
Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Cc: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Rik van Riel <riel@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Acked-by: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/rmap.c

index 24dd3f9fee27dfe577c7f605bd0c260f0bc143c7..9f47f152b01e1ee5f155657837ae13a0e2b9f9f6 100644 (file)
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -950,7 +950,12 @@ void page_move_anon_rmap(struct page *page,
        VM_BUG_ON_PAGE(page->index != linear_page_index(vma, address), page);
 
        anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
-       page->mapping = (struct address_space *) anon_vma;
+       /*
+        * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written
+        * simultaneously, so a concurrent reader (eg page_referenced()'s
+        * PageAnon()) will not see one without the other.
+        */
+       WRITE_ONCE(page->mapping, (struct address_space *) anon_vma);
 }
 
 /**