diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index f7b89743fab93e..165fa55cfb38a8 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/drivers/net/bonding/bond_main.c
@@ -401,6 +401,7 @@ static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 static int bond_ipsec_add_sa(struct xfrm_state *xs)
 {
 	struct net_device *bond_dev = xs->xso.dev;
+	struct bond_ipsec *ipsec;
 	struct bonding *bond;
 	struct slave *slave;
 	int err;
@@ -416,9 +417,6 @@ static int bond_ipsec_add_sa(struct xfrm_state *xs)
 		return -ENODEV;
 	}
 
-	xs->xso.real_dev = slave->dev;
-	bond->xs = xs;
-
 	if (!slave->dev->xfrmdev_ops ||
 	    !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
 	    netif_is_bond_master(slave->dev)) {
@@ -427,11 +425,63 @@ static int bond_ipsec_add_sa(struct xfrm_state *xs)
 		return -EINVAL;
 	}
 
+	ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC);
+	if (!ipsec) {
+		rcu_read_unlock();
+		return -ENOMEM;
+	}
+	xs->xso.real_dev = slave->dev;
+
 	err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs);
+	if (!err) {
+		ipsec->xs = xs;
+		INIT_LIST_HEAD(&ipsec->list);
+		spin_lock_bh(&bond->ipsec_lock);
+		list_add(&ipsec->list, &bond->ipsec_list);
+		spin_unlock_bh(&bond->ipsec_lock);
+	} else {
+		kfree(ipsec);
+	}
 	rcu_read_unlock();
 	return err;
 }
 
+static void bond_ipsec_add_sa_all(struct bonding *bond)
+{
+	struct net_device *bond_dev = bond->dev;
+	struct bond_ipsec *ipsec;
+	struct slave *slave;
+
+	rcu_read_lock();
+	slave = rcu_dereference(bond->curr_active_slave);
+	if (!slave)
+		goto out;
+
+	if (!slave->dev->xfrmdev_ops ||
+	    !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
+	    netif_is_bond_master(slave->dev)) {
+		spin_lock_bh(&bond->ipsec_lock);
+		if (!list_empty(&bond->ipsec_list))
+			slave_warn(bond_dev, slave->dev,
+				   "%s: no slave xdo_dev_state_add\n",
+				   __func__);
+		spin_unlock_bh(&bond->ipsec_lock);
+		goto out;
+	}
+
+	spin_lock_bh(&bond->ipsec_lock);
+	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
+		ipsec->xs->xso.real_dev = slave->dev;
+		if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs)) {
+			slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__);
+			ipsec->xs->xso.real_dev = NULL;
+		}
+	}
+	spin_unlock_bh(&bond->ipsec_lock);
+out:
+	rcu_read_unlock();
+}
+
 /**
  * bond_ipsec_del_sa - clear out this specific SA
  * @xs: pointer to transformer state struct
@@ -439,6 +489,7 @@ static int bond_ipsec_add_sa(struct xfrm_state *xs)
 static void bond_ipsec_del_sa(struct xfrm_state *xs)
 {
 	struct net_device *bond_dev = xs->xso.dev;
+	struct bond_ipsec *ipsec;
 	struct bonding *bond;
 	struct slave *slave;
 
@@ -452,7 +503,10 @@ static void bond_ipsec_del_sa(struct xfrm_state *xs)
 	if (!slave)
 		goto out;
 
-	xs->xso.real_dev = slave->dev;
+	if (!xs->xso.real_dev)
+		goto out;
+
+	WARN_ON(xs->xso.real_dev != slave->dev);
 
 	if (!slave->dev->xfrmdev_ops ||
 	    !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
@@ -463,6 +517,48 @@ static void bond_ipsec_del_sa(struct xfrm_state *xs)
 
 	slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
 out:
+	spin_lock_bh(&bond->ipsec_lock);
+	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
+		if (ipsec->xs == xs) {
+			list_del(&ipsec->list);
+			kfree(ipsec);
+			break;
+		}
+	}
+	spin_unlock_bh(&bond->ipsec_lock);
+	rcu_read_unlock();
+}
+
+static void bond_ipsec_del_sa_all(struct bonding *bond)
+{
+	struct net_device *bond_dev = bond->dev;
+	struct bond_ipsec *ipsec;
+	struct slave *slave;
+
+	rcu_read_lock();
+	slave = rcu_dereference(bond->curr_active_slave);
+	if (!slave) {
+		rcu_read_unlock();
+		return;
+	}
+
+	spin_lock_bh(&bond->ipsec_lock);
+	list_for_each_entry(ipsec, &bond->ipsec_list, list) {
+		if (!ipsec->xs->xso.real_dev)
+			continue;
+
+		if (!slave->dev->xfrmdev_ops ||
+		    !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
+		    netif_is_bond_master(slave->dev)) {
+			slave_warn(bond_dev, slave->dev,
+				   "%s: no slave xdo_dev_state_delete\n",
+				   __func__);
+		} else {
+			slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
+		}
+		ipsec->xs->xso.real_dev = NULL;
+	}
+	spin_unlock_bh(&bond->ipsec_lock);
 	rcu_read_unlock();
 }
 
@@ -474,22 +570,27 @@ static void bond_ipsec_del_sa(struct xfrm_state *xs)
 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
 {
 	struct net_device *bond_dev = xs->xso.dev;
-	struct bonding *bond = netdev_priv(bond_dev);
-	struct slave *curr_active = rcu_dereference(bond->curr_active_slave);
-	struct net_device *slave_dev = curr_active->dev;
+	struct net_device *real_dev;
+	struct slave *curr_active;
+	struct bonding *bond;
+
+	bond = netdev_priv(bond_dev);
+	curr_active = rcu_dereference(bond->curr_active_slave);
+	real_dev = curr_active->dev;
 
 	if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)
 		return true;
 
-	if (!slave_dev->xfrmdev_ops ||
-	    !slave_dev->xfrmdev_ops->xdo_dev_offload_ok ||
-	    netif_is_bond_master(slave_dev)) {
-		slave_warn(bond_dev, slave_dev, "%s: no slave xdo_dev_offload_ok\n", __func__);
+	if (!xs->xso.real_dev)
+		return false;
+
+	if (!real_dev->xfrmdev_ops ||
+	    !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
+	    netif_is_bond_master(real_dev)) {
 		return false;
 	}
 
-	xs->xso.real_dev = slave_dev;
-	return slave_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
+	return real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
 }
 
 static const struct xfrmdev_ops bond_xfrmdev_ops = {
@@ -1006,8 +1107,7 @@ void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 		return;
 
 #ifdef CONFIG_XFRM_OFFLOAD
-	if (old_active && bond->xs)
-		bond_ipsec_del_sa(bond->xs);
+	bond_ipsec_del_sa_all(bond);
 #endif /* CONFIG_XFRM_OFFLOAD */
 
 	if (new_active) {
@@ -1082,10 +1182,7 @@ void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 	}
 
 #ifdef CONFIG_XFRM_OFFLOAD
-	if (new_active && bond->xs) {
-		xfrm_dev_state_flush(dev_net(bond->dev), bond->dev, true);
-		bond_ipsec_add_sa(bond->xs);
-	}
+	bond_ipsec_add_sa_all(bond);
 #endif /* CONFIG_XFRM_OFFLOAD */
 
 	/* resend IGMP joins since active slave has changed or
@@ -3343,6 +3440,7 @@ static int bond_master_netdev_event(unsigned long event,
 		return bond_event_changename(event_bond);
 	case NETDEV_UNREGISTER:
 		bond_remove_proc_entry(event_bond);
+		xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
 		break;
 	case NETDEV_REGISTER:
 		bond_create_proc_entry(event_bond);
@@ -4910,7 +5008,8 @@ void bond_setup(struct net_device *bond_dev)
 #ifdef CONFIG_XFRM_OFFLOAD
 	/* set up xfrm device ops (only supported in active-backup right now) */
 	bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
-	bond->xs = NULL;
+	INIT_LIST_HEAD(&bond->ipsec_list);
+	spin_lock_init(&bond->ipsec_lock);
 #endif /* CONFIG_XFRM_OFFLOAD */
 
 	/* don't acquire bond device's netif_tx_lock when transmitting */
diff --git a/include/net/bonding.h b/include/net/bonding.h
index 15335732e16607..625d9c72dee37d 100644
--- a/include/net/bonding.h
+++ b/include/net/bonding.h
@@ -201,6 +201,11 @@ struct bond_up_slave {
  */
 #define BOND_LINK_NOCHANGE -1
 
+struct bond_ipsec {
+	struct list_head list;
+	struct xfrm_state *xs;
+};
+
 /*
  * Here are the locking policies for the two bonding locks:
  * Get rcu_read_lock when reading or RTNL when writing slave list.
@@ -249,7 +254,9 @@ struct bonding {
 #endif /* CONFIG_DEBUG_FS */
 	struct rtnl_link_stats64 bond_stats;
 #ifdef CONFIG_XFRM_OFFLOAD
-	struct xfrm_state *xs;
+	struct list_head ipsec_list;
+	/* protecting ipsec_list */
+	spinlock_t ipsec_lock;
 #endif /* CONFIG_XFRM_OFFLOAD */
 };