[PATCH] rmap: anon_vma_prepare() can leak anon_vma_chain

[PATCH] rmap: anon_vma_prepare() can leak anon_vma_chain

[Oleg Nesterov]

If find_mergeable_anon_vma() succeeds but another thread installs
->anon_vma before we take ptl, then allocated == NULL but avc should
be freed. Change the code to check avc != NULL to detect this case.

Also, a couple of whitespace changes to make the critical section more
visible.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
---

 mm/rmap.c |    9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

--- 34-rc1/mm/rmap.c~AVP_AVC_LEAK	2010-03-11 13:11:50.000000000 +0100
+++ 34-rc1/mm/rmap.c	2010-04-20 21:05:24.000000000 +0200
@@ -133,8 +133,8 @@ int anon_vma_prepare(struct vm_area_stru
 				goto out_enomem_free_avc;
 			allocated = anon_vma;
 		}
-		spin_lock(&anon_vma->lock);
 
+		spin_lock(&anon_vma->lock);
 		/* page_table_lock to protect against threads */
 		spin_lock(&mm->page_table_lock);
 		if (likely(!vma->anon_vma)) {
@@ -144,14 +144,15 @@ int anon_vma_prepare(struct vm_area_stru
 			list_add(&avc->same_vma, &vma->anon_vma_chain);
 			list_add(&avc->same_anon_vma, &anon_vma->head);
 			allocated = NULL;
+			avc = NULL;
 		}
 		spin_unlock(&mm->page_table_lock);
-
 		spin_unlock(&anon_vma->lock);
-		if (unlikely(allocated)) {
+
+		if (unlikely(allocated))
 			anon_vma_free(allocated);
+		if (unlikely(avc))
 			anon_vma_chain_free(avc);
-		}
 	}
 	return 0;
 

Re: [PATCH] rmap: anon_vma_prepare() can leak anon_vma_chain

[Rik van Riel]

On 04/20/2010 03:24 PM, Oleg Nesterov wrote:
> If find_mergeable_anon_vma() succeeds but another thread installs
> ->anon_vma before we take ptl, then allocated == NULL but avc should
> be freed. Change the code to check avc != NULL to detect this case.

Good catch!

> Signed-off-by: Oleg Nesterov<oleg@redhat.com>

Reviewed-by: Rik van Riel <riel@redhat.com>