macvlan: Avoid unnecessary multicast cloning

Currently we always queue a multicast packet for further processing,
even if none of the macvlan devices are subscribed to the address.

This patch optimises this by adding a global multicast filter for
a macvlan_port.

Note that this patch doesn't handle the broadcast addresses of the
individual macvlan devices correctly, if they are not all identical
to vlan->lowerdev.  However, this is already broken because there
is no mechanism in place to update the individual multicast filters
when you change the broadcast address.

If someone cares enough they should fix this by collecting all
broadcast addresses for a macvlan as we do for multicast and unicast.

Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/drivers/net/macvlan.c b/drivers/net/macvlan.c
index a71fa592b1fb47f5d05049ddebc2874203022619..0c65bd914aedaf6685925960801f9de3fed1a5cd 100644 (file)
--- a/drivers/net/macvlan.c
+++ b/drivers/net/macvlan.c
@@ -49,6 +49,7 @@ struct macvlan_port {
        bool                    passthru;
        int                     count;
        struct hlist_head       vlan_source_hash[MACVLAN_HASH_SIZE];
+       DECLARE_BITMAP(mc_filter, MACVLAN_MC_FILTER_SZ);
 };
 
 struct macvlan_source_entry {
@@ -419,6 +420,8 @@ static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
 
        port = macvlan_port_get_rcu(skb->dev);
        if (is_multicast_ether_addr(eth->h_dest)) {
+               unsigned int hash;
+
                skb = ip_check_defrag(dev_net(skb->dev), skb, IP_DEFRAG_MACVLAN);
                if (!skb)
                        return RX_HANDLER_CONSUMED;
@@ -436,7 +439,9 @@ static rx_handler_result_t macvlan_handle_frame(struct sk_buff **pskb)
                        goto out;
                }
 
-               macvlan_broadcast_enqueue(port, src, skb);
+               hash = mc_hash(NULL, eth->h_dest);
+               if (test_bit(hash, port->mc_filter))
+                       macvlan_broadcast_enqueue(port, src, skb);
 
                return RX_HANDLER_PASS;
        }
@@ -722,12 +727,12 @@ static void macvlan_change_rx_flags(struct net_device *dev, int change)
        }
 }
 
-static void macvlan_set_mac_lists(struct net_device *dev)
+static void macvlan_compute_filter(unsigned long *mc_filter,
+                                  struct net_device *dev,
+                                  struct macvlan_dev *vlan)
 {
-       struct macvlan_dev *vlan = netdev_priv(dev);
-
        if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
-               bitmap_fill(vlan->mc_filter, MACVLAN_MC_FILTER_SZ);
+               bitmap_fill(mc_filter, MACVLAN_MC_FILTER_SZ);
        } else {
                struct netdev_hw_addr *ha;
                DECLARE_BITMAP(filter, MACVLAN_MC_FILTER_SZ);
@@ -739,10 +744,33 @@ static void macvlan_set_mac_lists(struct net_device *dev)
 
                __set_bit(mc_hash(vlan, dev->broadcast), filter);
 
-               bitmap_copy(vlan->mc_filter, filter, MACVLAN_MC_FILTER_SZ);
+               bitmap_copy(mc_filter, filter, MACVLAN_MC_FILTER_SZ);
        }
+}
+
+static void macvlan_set_mac_lists(struct net_device *dev)
+{
+       struct macvlan_dev *vlan = netdev_priv(dev);
+
+       macvlan_compute_filter(vlan->mc_filter, dev, vlan);
+
        dev_uc_sync(vlan->lowerdev, dev);
        dev_mc_sync(vlan->lowerdev, dev);
+
+       /* This is slightly inaccurate as we're including the subscription
+        * list of vlan->lowerdev too.
+        *
+        * Bug alert: This only works if everyone has the same broadcast
+        * address as lowerdev.  As soon as someone changes theirs this
+        * will break.
+        *
+        * However, this is already broken as when you change your broadcast
+        * address we don't get called.
+        *
+        * The solution is to maintain a list of broadcast addresses like
+        * we do for uc/mc, if you care.
+        */
+       macvlan_compute_filter(vlan->port->mc_filter, vlan->lowerdev, NULL);
 }
 
 static int macvlan_change_mtu(struct net_device *dev, int new_mtu)