]> git.kernelconcepts.de Git - karo-tx-linux.git/commitdiff
vxlan: Add capability of Rx checksum offload for inner packet
authorJoseph Gasparakis <joseph.gasparakis@intel.com>
Fri, 7 Dec 2012 14:14:18 +0000 (14:14 +0000)
committerDavid S. Miller <davem@davemloft.net>
Sun, 9 Dec 2012 05:20:28 +0000 (00:20 -0500)
This patch adds capability in vxlan to identify received
checksummed inner packets and signal them to the upper layers of
the stack. The driver needs to set the skb->encapsulation bit
and also set the skb->ip_summed to CHECKSUM_UNNECESSARY.

Signed-off-by: Joseph Gasparakis <joseph.gasparakis@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/vxlan.c

index 88b31f2453dde72568242b8562cc81c2a80f3d11..3b3fdf648ea741267800a2b49eee87d835d2370a 100644 (file)
@@ -607,7 +607,17 @@ static int vxlan_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 
        __skb_tunnel_rx(skb, vxlan->dev);
        skb_reset_network_header(skb);
-       skb->ip_summed = CHECKSUM_NONE;
+
+       /* If the NIC driver gave us an encapsulated packet with
+        * CHECKSUM_UNNECESSARY and Rx checksum feature is enabled,
+        * leave the CHECKSUM_UNNECESSARY, the device checksummed it
+        * for us. Otherwise force the upper layers to verify it.
+        */
+       if (skb->ip_summed != CHECKSUM_UNNECESSARY || !skb->encapsulation ||
+           !(vxlan->dev->features & NETIF_F_RXCSUM))
+               skb->ip_summed = CHECKSUM_NONE;
+
+       skb->encapsulation = 0;
 
        err = IP_ECN_decapsulate(oip, skb);
        if (unlikely(err)) {
@@ -1175,7 +1185,9 @@ static void vxlan_setup(struct net_device *dev)
        dev->features   |= NETIF_F_LLTX;
        dev->features   |= NETIF_F_NETNS_LOCAL;
        dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
-       dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM;
+       dev->features   |= NETIF_F_RXCSUM;
+
+       dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
        dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
 
        spin_lock_init(&vxlan->hash_lock);