vxlan: change vxlan to use UDP socket GRO
authorTom Herbert <tom@herbertland.com>
Tue, 5 Apr 2016 15:22:53 +0000 (08:22 -0700)
committerDavid S. Miller <davem@davemloft.net>
Thu, 7 Apr 2016 20:53:29 +0000 (16:53 -0400)
Adapt vxlan_gro_receive, vxlan_gro_complete to take a socket argument.
Set these functions in tunnel_config.  Don't set udp_offloads any more.

Signed-off-by: Tom Herbert <tom@herbertland.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/vxlan.c
include/net/vxlan.h

index 51cccdd..9f36340 100644 (file)
@@ -551,16 +551,15 @@ static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
        return vh;
 }
 
-static struct sk_buff **vxlan_gro_receive(struct sk_buff **head,
-                                         struct sk_buff *skb,
-                                         struct udp_offload *uoff)
+static struct sk_buff **vxlan_gro_receive(struct sock *sk,
+                                         struct sk_buff **head,
+                                         struct sk_buff *skb)
 {
        struct sk_buff *p, **pp = NULL;
        struct vxlanhdr *vh, *vh2;
        unsigned int hlen, off_vx;
        int flush = 1;
-       struct vxlan_sock *vs = container_of(uoff, struct vxlan_sock,
-                                            udp_offloads);
+       struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
        __be32 flags;
        struct gro_remcsum grc;
 
@@ -613,8 +612,7 @@ out:
        return pp;
 }
 
-static int vxlan_gro_complete(struct sk_buff *skb, int nhoff,
-                             struct udp_offload *uoff)
+static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 {
        udp_tunnel_gro_complete(skb, nhoff);
 
@@ -629,13 +627,6 @@ static void vxlan_notify_add_rx_port(struct vxlan_sock *vs)
        struct net *net = sock_net(sk);
        sa_family_t sa_family = vxlan_get_sk_family(vs);
        __be16 port = inet_sk(sk)->inet_sport;
-       int err;
-
-       if (sa_family == AF_INET) {
-               err = udp_add_offload(net, &vs->udp_offloads);
-               if (err)
-                       pr_warn("vxlan: udp_add_offload failed with status %d\n", err);
-       }
 
        rcu_read_lock();
        for_each_netdev_rcu(net, dev) {
@@ -662,9 +653,6 @@ static void vxlan_notify_del_rx_port(struct vxlan_sock *vs)
                                                            port);
        }
        rcu_read_unlock();
-
-       if (sa_family == AF_INET)
-               udp_del_offload(&vs->udp_offloads);
 }
 
 /* Add new entry to forwarding table -- assumes lock held */
@@ -2752,21 +2740,19 @@ static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
        atomic_set(&vs->refcnt, 1);
        vs->flags = (flags & VXLAN_F_RCV_FLAGS);
 
-       /* Initialize the vxlan udp offloads structure */
-       vs->udp_offloads.port = port;
-       vs->udp_offloads.callbacks.gro_receive  = vxlan_gro_receive;
-       vs->udp_offloads.callbacks.gro_complete = vxlan_gro_complete;
-
        spin_lock(&vn->sock_lock);
        hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
        vxlan_notify_add_rx_port(vs);
        spin_unlock(&vn->sock_lock);
 
        /* Mark socket as an encapsulation socket. */
+       memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
        tunnel_cfg.sk_user_data = vs;
        tunnel_cfg.encap_type = 1;
        tunnel_cfg.encap_rcv = vxlan_rcv;
        tunnel_cfg.encap_destroy = NULL;
+       tunnel_cfg.gro_receive = vxlan_gro_receive;
+       tunnel_cfg.gro_complete = vxlan_gro_complete;
 
        setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
 
index dcc6f40..2f168f0 100644 (file)
@@ -189,7 +189,6 @@ struct vxlan_sock {
        struct rcu_head   rcu;
        struct hlist_head vni_list[VNI_HASH_SIZE];
        atomic_t          refcnt;
-       struct udp_offload udp_offloads;
        u32               flags;
 };