[PATCH v4 nf-next 2/2] netfilter: nf_flow_table_core: teardown direct xmit when destination changed

[Eric Woudstra <ericwouds@gmail.com>]

In case of a bridge in the forward-fastpath or bridge-fastpath the fdb is
used to create the tuple. In case of roaming at layer 2 level, for example
802.11r, the destination device is changed in the fdb. The destination
device of a direct transmitting tuple is no longer valid and traffic is
send to the wrong destination. Also the hardware offloaded fastpath is not
valid anymore.

In case of roaming, a switchdev notification is send to delete the old fdb
entry. Upon receiving this notification, mark all direct transmitting flows
with the same ifindex, vid and hardware address as the fdb entry to be
teared down. The hardware offloaded fastpath is still in effect, so
minimize the delay of the work queue by setting the delay to zero.

Signed-off-by: Eric Woudstra <ericwouds@gmail.com>
---
 net/netfilter/nf_flow_table_core.c | 88 ++++++++++++++++++++++++++++++
 1 file changed, 88 insertions(+)

diff --git a/net/netfilter/nf_flow_table_core.c b/net/netfilter/nf_flow_table_core.c
index 992958db4a19..e906f93f4abb 100644
--- a/net/netfilter/nf_flow_table_core.c
+++ b/net/netfilter/nf_flow_table_core.c
@@ -13,6 +13,7 @@
 #include <net/netfilter/nf_conntrack_core.h>
 #include <net/netfilter/nf_conntrack_l4proto.h>
 #include <net/netfilter/nf_conntrack_tuple.h>
+#include <net/switchdev.h>
 
 static DEFINE_MUTEX(flowtable_lock);
 static LIST_HEAD(flowtables);
@@ -745,6 +746,86 @@ void nf_flow_table_cleanup(struct net_device *dev)
 }
 EXPORT_SYMBOL_GPL(nf_flow_table_cleanup);
 
+struct flow_cleanup_data {
+	int ifindex;
+	u16 vid;
+	char addr[ETH_ALEN];
+	bool found;
+};
+
+struct flow_switchdev_event_work {
+	struct work_struct work;
+	struct flow_cleanup_data cud;
+};
+
+static void nf_flow_table_do_cleanup_addr(struct nf_flowtable *flow_table,
+					  struct flow_offload *flow, void *data)
+{
+	struct flow_cleanup_data *cud = data;
+
+	if ((flow->tuplehash[0].tuple.xmit_type == FLOW_OFFLOAD_XMIT_DIRECT &&
+	     flow->tuplehash[0].tuple.out.ifidx == cud->ifindex &&
+	     flow->tuplehash[0].tuple.out.bridge_vid == cud->vid &&
+	     ether_addr_equal(flow->tuplehash[0].tuple.out.h_dest, cud->addr)) ||
+	    (flow->tuplehash[1].tuple.xmit_type == FLOW_OFFLOAD_XMIT_DIRECT &&
+	     flow->tuplehash[1].tuple.out.ifidx == cud->ifindex &&
+	     flow->tuplehash[1].tuple.out.bridge_vid == cud->vid &&
+	     ether_addr_equal(flow->tuplehash[1].tuple.out.h_dest, cud->addr))) {
+		flow_offload_teardown(flow);
+		cud->found = true;
+	}
+}
+
+static void nf_flow_table_switchdev_event_work(struct work_struct *work)
+{
+	struct flow_switchdev_event_work *switchdev_work =
+		container_of(work, struct flow_switchdev_event_work, work);
+	struct nf_flowtable *flowtable;
+
+	mutex_lock(&flowtable_lock);
+
+	list_for_each_entry(flowtable, &flowtables, list) {
+		switchdev_work->cud.found = false;
+		nf_flow_table_iterate(flowtable, nf_flow_table_do_cleanup_addr,
+				      &switchdev_work->cud);
+		if (switchdev_work->cud.found)
+			mod_delayed_work(system_power_efficient_wq,
+					 &flowtable->gc_work, 0);
+	}
+
+	mutex_unlock(&flowtable_lock);
+
+	kfree(switchdev_work);
+}
+
+static int nf_flow_table_switchdev_event(struct notifier_block *unused,
+					 unsigned long event, void *ptr)
+{
+	struct flow_switchdev_event_work *switchdev_work;
+	struct switchdev_notifier_fdb_info *fdb_info;
+
+	if (event != SWITCHDEV_FDB_DEL_TO_DEVICE)
+		return NOTIFY_DONE;
+
+	switchdev_work = kzalloc(sizeof(*switchdev_work), GFP_ATOMIC);
+	if (WARN_ON(!switchdev_work))
+		return NOTIFY_BAD;
+
+	INIT_WORK(&switchdev_work->work, nf_flow_table_switchdev_event_work);
+	fdb_info = ptr;
+	switchdev_work->cud.ifindex = fdb_info->info.dev->ifindex;
+	switchdev_work->cud.vid = fdb_info->vid;
+	ether_addr_copy(switchdev_work->cud.addr, fdb_info->addr);
+
+	queue_work(system_long_wq, &switchdev_work->work);
+
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block nf_flow_table_switchdev_nb __read_mostly = {
+	.notifier_call = nf_flow_table_switchdev_event,
+};
+
 void nf_flow_table_free(struct nf_flowtable *flow_table)
 {
 	mutex_lock(&flowtable_lock);
@@ -818,6 +899,10 @@ static int __init nf_flow_table_module_init(void)
 	if (ret)
 		goto out_offload;
 
+	ret = register_switchdev_notifier(&nf_flow_table_switchdev_nb);
+	if (ret < 0)
+		goto out_sw_noti;
+
 	ret = nf_flow_register_bpf();
 	if (ret)
 		goto out_bpf;
@@ -825,6 +910,8 @@ static int __init nf_flow_table_module_init(void)
 	return 0;
 
 out_bpf:
+	unregister_switchdev_notifier(&nf_flow_table_switchdev_nb);
+out_sw_noti:
 	nf_flow_table_offload_exit();
 out_offload:
 	unregister_pernet_subsys(&nf_flow_table_net_ops);
@@ -833,6 +920,7 @@ static int __init nf_flow_table_module_init(void)
 
 static void __exit nf_flow_table_module_exit(void)
 {
+	unregister_switchdev_notifier(&nf_flow_table_switchdev_nb);
 	nf_flow_table_offload_exit();
 	unregister_pernet_subsys(&nf_flow_table_net_ops);
 }
-- 
2.50.0