On Thu, 2024-08-01 at 12:25 +0200, Paolo Abeni wrote:

On 8/1/24 11:49, Tariq Toukan wrote:

quoted

diff --git a/drivers/net/bonding/bond_main.c

b/drivers/net/bonding/bond_main.c
index e6514ef7ad89..0f8d1b29dc7f 100644

--- a/drivers/net/bonding/bond_main.c
+++ b/drivers/net/bonding/bond_main.c

@@ -436,41 +436,34 @@ static int bond_ipsec_add_sa(struct

xfrm_state *xs,
        if (!bond_dev)
                return -EINVAL;
  
-       rcu_read_lock();
        bond = netdev_priv(bond_dev);
        slave = rcu_dereference(bond->curr_active_slave);
-       if (!slave) {
-               rcu_read_unlock();

I'm sorry, I probably was not clear with my question on the previous 
revision.

I asked if this code is under RTNL lock already, if so we could
replace 
rcu_dereference with rtnl_dereference() and drop the rcu lock.

You stated this block is not under the RTNL lock, so we still need
the 
rcu lock around rcu_dereference().

Same thing in bond_ipsec_del_sa().

Sure, I will add back. Thanks!

Please have a run with CONFIG_PROVE_RCU, it should splat on such 
dereference.

Thanks,

Paolo