ksm: fix potential missing rmap_item for stable_node
authorMiaohe Lin <linmiaohe@huawei.com>
Wed, 5 May 2021 01:37:45 +0000 (18:37 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Sat, 22 May 2021 08:40:31 +0000 (10:40 +0200)
[ Upstream commit c89a384e2551c692a9fe60d093fd7080f50afc51 ]

When removing rmap_item from stable tree, STABLE_FLAG of rmap_item is
cleared with head reserved.  So the following scenario might happen: For
ksm page with rmap_item1:

cmp_and_merge_page
  stable_node->head = &migrate_nodes;
  remove_rmap_item_from_tree, but head still equal to stable_node;
  try_to_merge_with_ksm_page failed;
  return;

For the same ksm page with rmap_item2, stable node migration succeed this
time.  The stable_node->head does not equal to migrate_nodes now.  For ksm
page with rmap_item1 again:

cmp_and_merge_page
 stable_node->head != &migrate_nodes && rmap_item->head == stable_node
 return;

We would miss the rmap_item for stable_node and might result in failed
rmap_walk_ksm().  Fix this by set rmap_item->head to NULL when rmap_item
is removed from stable tree.

Link: https://lkml.kernel.org/r/20210330140228.45635-5-linmiaohe@huawei.com
Fixes: 4146d2d673e8 ("ksm: make !merge_across_nodes migration safe")
Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
mm/ksm.c

index d6c81a5076a785f934ac6f4aec9899f5dae8c698..27ff68050d8567eb5677bbc91587cb758230c491 100644 (file)
--- a/mm/ksm.c
+++ b/mm/ksm.c
@@ -629,6 +629,7 @@ static void remove_rmap_item_from_tree(struct rmap_item *rmap_item)
                        ksm_pages_shared--;
 
                put_anon_vma(rmap_item->anon_vma);
+               rmap_item->head = NULL;
                rmap_item->address &= PAGE_MASK;
 
        } else if (rmap_item->address & UNSTABLE_FLAG) {