To avoid frequent TLB flushes, vmap_area structs are freed lazily, in
batches, once there are enough of them hanging around. As a result,
accounting them to a memcg pins the memcg itself and its kmem caches for
indefinitely long, which is not good. To avoid this, this patch makes
allocations of vmap_area's go through the root cgroup. Since they are
small and there cannot be many of them, not accounting them to memcg
won't open any security holes.

Signed-off-by: Vladimir Davydov <[email protected]>
---
 mm/vmalloc.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index 06a44461ccd5..ac32dca89d4f 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -357,7 +357,7 @@ static struct vmap_area *alloc_vmap_area(unsigned long size,
        BUG_ON(!is_power_of_2(align));
 
        va = kmalloc_node(sizeof(struct vmap_area),
-                       gfp_mask & GFP_RECLAIM_MASK, node);
+                       (gfp_mask & GFP_RECLAIM_MASK) | __GFP_NOACCOUNT, node);
        if (unlikely(!va))
                return ERR_PTR(-ENOMEM);
 
-- 
1.7.10.4

_______________________________________________
Devel mailing list
[email protected]
https://lists.openvz.org/mailman/listinfo/devel

Reply via email to