[PATCH 2/6] bridge: Offload mrouter port forwarding to switchdev

[Joseph Huang]

Offload the mrouter port forwarding to switchdev also.

Currently multicast snooping fails to forward traffic in some cases
where there're multiple hardware-offloading bridges involved.

Consider the following scenario:

                 +--------------------+
                 |                    |
                 |      Snooping   +--|    +------------+
                 |      Bridge 1   |P1|----| Listener 1 |
                 |     (Querier)   +--|    +------------+
                 |                    |
                 +--------------------+
                           |
                           |
                 +--------------------+
                 |    | mrouter |     |
+-----------+    |    +---------+  +--|    +------------+
| MC Source |----|      Snooping   |P2|----| Listener 2 |
+-----------|    |      Bridge 2   +--|    +------------+
                 |    (Non-Querier)   |
                 +--------------------+

In this scenario, Listener 2 is able to receive multicast traffic
from MC Source while Listener 1 is not. The reason is that, on
Snooping Bridge 2, when the (soft) bridge attempts to forward
a packet to the mrouter port via br_multicast_flood, the effort
is blocked by nbp_switchdev_allowed_egress, since offload_fwd_mark
indicates that the packet should have been handled by the hardware
already. Listener 2 would receive the packets without any problem
since P2 is programmed into the hardware as a member of the group;
however the mrouter port would not since the mrouter port would
normally not be a member of any group, and thus will not be added
to the address database on the hardware switch chip.

This patch takes a simplistic approach: when an mrouter port is added/
deleted, it's added/deleted to all mdb groups; and similarly, when
an mdb group is added/deleted, all mrouter ports are added/deleted
to/from it.

Before this patch, switchdev programming matches exactly with mdb:
 +-----+
 | mdb |
 +-----+
    |
 +----------------------------------------------+
 |  |        +--------------------------------+ |
 |  |        | both in mdb and switchdev      | |
 |  |        | +------+   +------+   +------+ | |
 |  +--------|-| port |---| port |---| port | | |
 |           | +------+   +------+   +------+ | |
 | switchdev +--------------------------------+ |
 +----------------------------------------------+

After this patch, some entries will only exist in switchdev and not
in mdb:
 +-----+
 | mdb |
 +-----+
    |
 +---------------------------------------------------------------------+
 |  |        +--------------------------------++---------------------+ |
 |  |        |  both in mdb and switchdev     ||  only in switchdev  | |
 |  |        | +------+   +------+   +------+ || +------+   +------+ | |
 |  +--------|-| port |---| port |---| port | || |  mr  |---|  mr  | | |
 |           | +------+   +------+   +------+ || +------+   +------+ | |
 | switchdev +--------------------------------++---------------------+ |
 +---------------------------------------------------------------------+

Signed-off-by: Joseph Huang <Joseph.Huang@xxxxxxxxxx>
---
 net/bridge/br_multicast.c | 28 ++++++++++++++++++++++++++++
 1 file changed, 28 insertions(+)

diff --git a/net/bridge/br_multicast.c b/net/bridge/br_multicast.c
index 226bb05c3b42..5ed0d5efef09 100644
--- a/net/bridge/br_multicast.c
+++ b/net/bridge/br_multicast.c
@@ -522,10 +522,26 @@ static void br_multicast_destroy_mdb_entry(struct net_bridge_mcast_gc *gc)
 	kfree_rcu(mp, rcu);
 }
 
+/* Add/delete all mrouter ports to/from a group
+ * called while br->multicast_lock is held
+ */
+static void br_multicast_group_change(struct net_bridge_mdb_entry *mp,
+				      bool is_group_added)
+{
+	struct net_bridge_port *p;
+	struct hlist_node *n;
+
+	hlist_for_each_entry_safe(p, n, &mp->br->router_list, rlist)
+		br_mdb_switchdev_port(mp, p, is_group_added ?
+				      RTM_NEWMDB : RTM_DELMDB);
+}
+
 static void br_multicast_del_mdb_entry(struct net_bridge_mdb_entry *mp)
 {
 	struct net_bridge *br = mp->br;
 
+	br_multicast_group_change(mp, false);
+
 	rhashtable_remove_fast(&br->mdb_hash_tbl, &mp->rhnode,
 			       br_mdb_rht_params);
 	hlist_del_init_rcu(&mp->mdb_node);
@@ -1068,6 +1084,8 @@ struct net_bridge_mdb_entry *br_multicast_new_group(struct net_bridge *br,
 		hlist_add_head_rcu(&mp->mdb_node, &br->mdb_list);
 	}
 
+	br_multicast_group_change(mp, true);
+
 	return mp;
 }
 
@@ -2651,8 +2669,18 @@ static void br_port_mc_router_state_change(struct net_bridge_port *p,
 		.flags = SWITCHDEV_F_DEFER,
 		.u.mrouter = is_mc_router,
 	};
+	struct net_bridge_mdb_entry *mp;
+	struct hlist_node *n;
 
 	switchdev_port_attr_set(p->dev, &attr, NULL);
+
+	/* Add/delete the router port to/from all multicast group
+	 * called whle br->multicast_lock is held
+	 */
+	hlist_for_each_entry_safe(mp, n, &p->br->mdb_list, mdb_node) {
+		br_mdb_switchdev_port(mp, p, is_mc_router ?
+				      RTM_NEWMDB : RTM_DELMDB);
+	}
 }
 
 /*
-- 
2.17.1