batman-adv: Correct rcu refcounting for neigh_node
It might be possible that 2 threads access the same data in the same
rcu grace period. The first thread calls call_rcu() to decrement the
refcount and free the data while the second thread increases the
refcount to use the data. To avoid this race condition all refcount
operations have to be atomic.
Reported-by: Sven Eckelmann <sven@narfation.org>
Signed-off-by: Marek Lindner <lindner_marek@yahoo.de>
diff --git a/net/batman-adv/vis.c b/net/batman-adv/vis.c
index 8092ead..9832d8f 100644
--- a/net/batman-adv/vis.c
+++ b/net/batman-adv/vis.c
@@ -764,21 +764,35 @@
struct vis_info *info)
{
struct orig_node *orig_node;
+ struct neigh_node *neigh_node = NULL;
struct sk_buff *skb;
struct vis_packet *packet;
struct batman_if *batman_if;
uint8_t dstaddr[ETH_ALEN];
- spin_lock_bh(&bat_priv->orig_hash_lock);
packet = (struct vis_packet *)info->skb_packet->data;
+
+ spin_lock_bh(&bat_priv->orig_hash_lock);
rcu_read_lock();
orig_node = ((struct orig_node *)hash_find(bat_priv->orig_hash,
compare_orig, choose_orig,
packet->target_orig));
- rcu_read_unlock();
- if ((!orig_node) || (!orig_node->router))
- goto out;
+ if (!orig_node)
+ goto unlock;
+
+ kref_get(&orig_node->refcount);
+ neigh_node = orig_node->router;
+
+ if (!neigh_node)
+ goto unlock;
+
+ if (!atomic_inc_not_zero(&neigh_node->refcount)) {
+ neigh_node = NULL;
+ goto unlock;
+ }
+
+ rcu_read_unlock();
/* don't lock while sending the packets ... we therefore
* copy the required data before sending */
@@ -790,10 +804,17 @@
if (skb)
send_skb_packet(skb, batman_if, dstaddr);
- return;
+ goto out;
-out:
+unlock:
+ rcu_read_unlock();
spin_unlock_bh(&bat_priv->orig_hash_lock);
+out:
+ if (neigh_node)
+ neigh_node_free_ref(neigh_node);
+ if (orig_node)
+ kref_put(&orig_node->refcount, orig_node_free_ref);
+ return;
}
/* only send one vis packet. called from send_vis_packets() */