geneve: Support for PMTU discovery on directly bridged links
authorStefano Brivio <sbrivio@redhat.com>
Tue, 4 Aug 2020 05:53:45 +0000 (07:53 +0200)
committerDavid S. Miller <davem@davemloft.net>
Tue, 4 Aug 2020 20:01:45 +0000 (13:01 -0700)
If the interface is a bridge or Open vSwitch port, and we can't
forward a packet because it exceeds the local PMTU estimate,
trigger an ICMP or ICMPv6 reply to the sender, using the same
interface to forward it back.

If metadata collection is enabled, set destination and source
addresses for the flow as if we were receiving the packet, so that
Open vSwitch can match the ICMP error against the existing
association.

v2: Use netif_is_any_bridge_port() (David Ahern)

Signed-off-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/geneve.c

index de86b6d..c71f994 100644 (file)
@@ -893,8 +893,31 @@ static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev,
        if (IS_ERR(rt))
                return PTR_ERR(rt);
 
-       skb_tunnel_check_pmtu(skb, &rt->dst,
-                             GENEVE_IPV4_HLEN + info->options_len, false);
+       err = skb_tunnel_check_pmtu(skb, &rt->dst,
+                                   GENEVE_IPV4_HLEN + info->options_len,
+                                   netif_is_any_bridge_port(dev));
+       if (err < 0) {
+               dst_release(&rt->dst);
+               return err;
+       } else if (err) {
+               struct ip_tunnel_info *info;
+
+               info = skb_tunnel_info(skb);
+               if (info) {
+                       info->key.u.ipv4.dst = fl4.saddr;
+                       info->key.u.ipv4.src = fl4.daddr;
+               }
+
+               if (!pskb_may_pull(skb, ETH_HLEN)) {
+                       dst_release(&rt->dst);
+                       return -EINVAL;
+               }
+
+               skb->protocol = eth_type_trans(skb, geneve->dev);
+               netif_rx(skb);
+               dst_release(&rt->dst);
+               return -EMSGSIZE;
+       }
 
        sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
        if (geneve->cfg.collect_md) {
@@ -955,8 +978,30 @@ static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev,
        if (IS_ERR(dst))
                return PTR_ERR(dst);
 
-       skb_tunnel_check_pmtu(skb, dst, GENEVE_IPV6_HLEN + info->options_len,
-                             false);
+       err = skb_tunnel_check_pmtu(skb, dst,
+                                   GENEVE_IPV6_HLEN + info->options_len,
+                                   netif_is_any_bridge_port(dev));
+       if (err < 0) {
+               dst_release(dst);
+               return err;
+       } else if (err) {
+               struct ip_tunnel_info *info = skb_tunnel_info(skb);
+
+               if (info) {
+                       info->key.u.ipv6.dst = fl6.saddr;
+                       info->key.u.ipv6.src = fl6.daddr;
+               }
+
+               if (!pskb_may_pull(skb, ETH_HLEN)) {
+                       dst_release(dst);
+                       return -EINVAL;
+               }
+
+               skb->protocol = eth_type_trans(skb, geneve->dev);
+               netif_rx(skb);
+               dst_release(dst);
+               return -EMSGSIZE;
+       }
 
        sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true);
        if (geneve->cfg.collect_md) {
@@ -1013,7 +1058,8 @@ static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev)
        if (likely(!err))
                return NETDEV_TX_OK;
 
-       dev_kfree_skb(skb);
+       if (err != -EMSGSIZE)
+               dev_kfree_skb(skb);
 
        if (err == -ELOOP)
                dev->stats.collisions++;