ipv6: sit: add GSO/TSO support
authorEric Dumazet <edumazet@google.com>
Mon, 21 Oct 2013 03:47:30 +0000 (20:47 -0700)
committerDavid S. Miller <davem@davemloft.net>
Mon, 21 Oct 2013 22:49:39 +0000 (18:49 -0400)
Now ipv6_gso_segment() is stackable, its relatively easy to
implement GSO/TSO support for SIT tunnels

Performance results, when segmentation is done after tunnel
device (as no NIC is yet enabled for TSO SIT support) :

Before patch :

lpq84:~# ./netperf -H 2002:af6:1153:: -Cc
MIGRATED TCP STREAM TEST from ::0 (::) port 0 AF_INET6 to 2002:af6:1153:: () port 0 AF_INET6
Recv   Send    Send                          Utilization       Service Demand
Socket Socket  Message  Elapsed              Send     Recv     Send    Recv
Size   Size    Size     Time     Throughput  local    remote   local   remote
bytes  bytes   bytes    secs.    10^6bits/s  % S      % S      us/KB   us/KB

 87380  16384  16384    10.00      3168.31   4.81     4.64     2.988   2.877

After patch :

lpq84:~# ./netperf -H 2002:af6:1153:: -Cc
MIGRATED TCP STREAM TEST from ::0 (::) port 0 AF_INET6 to 2002:af6:1153:: () port 0 AF_INET6
Recv   Send    Send                          Utilization       Service Demand
Socket Socket  Message  Elapsed              Send     Recv     Send    Recv
Size   Size    Size     Time     Throughput  local    remote   local   remote
bytes  bytes   bytes    secs.    10^6bits/s  % S      % S      us/KB   us/KB

 87380  16384  16384    10.00      5525.00   7.76     5.17     2.763   1.840

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/netdev_features.h
include/linux/skbuff.h
net/core/ethtool.c
net/ipv4/af_inet.c
net/ipv4/tcp_offload.c
net/ipv6/ip6_offload.c
net/ipv6/sit.c
net/ipv6/udp_offload.c

index 8dad68c..b05a4b5 100644 (file)
@@ -43,6 +43,7 @@ enum {
        NETIF_F_FSO_BIT,                /* ... FCoE segmentation */
        NETIF_F_GSO_GRE_BIT,            /* ... GRE with TSO */
        NETIF_F_GSO_IPIP_BIT,           /* ... IPIP tunnel with TSO */
+       NETIF_F_GSO_SIT_BIT,            /* ... SIT tunnel with TSO */
        NETIF_F_GSO_UDP_TUNNEL_BIT,     /* ... UDP TUNNEL with TSO */
        NETIF_F_GSO_MPLS_BIT,           /* ... MPLS segmentation */
        /**/NETIF_F_GSO_LAST =          /* last bit, see GSO_MASK */
@@ -109,6 +110,7 @@ enum {
 #define NETIF_F_RXALL          __NETIF_F(RXALL)
 #define NETIF_F_GSO_GRE                __NETIF_F(GSO_GRE)
 #define NETIF_F_GSO_IPIP       __NETIF_F(GSO_IPIP)
+#define NETIF_F_GSO_SIT                __NETIF_F(GSO_SIT)
 #define NETIF_F_GSO_UDP_TUNNEL __NETIF_F(GSO_UDP_TUNNEL)
 #define NETIF_F_GSO_MPLS       __NETIF_F(GSO_MPLS)
 #define NETIF_F_HW_VLAN_STAG_FILTER __NETIF_F(HW_VLAN_STAG_FILTER)
index 6072913..2c15497 100644 (file)
@@ -320,9 +320,11 @@ enum {
 
        SKB_GSO_IPIP = 1 << 7,
 
-       SKB_GSO_UDP_TUNNEL = 1 << 8,
+       SKB_GSO_SIT = 1 << 8,
 
-       SKB_GSO_MPLS = 1 << 9,
+       SKB_GSO_UDP_TUNNEL = 1 << 9,
+
+       SKB_GSO_MPLS = 1 << 10,
 };
 
 #if BITS_PER_LONG > 32
index 8cab774..8629898 100644 (file)
@@ -82,6 +82,7 @@ static const char netdev_features_strings[NETDEV_FEATURE_COUNT][ETH_GSTRING_LEN]
        [NETIF_F_FSO_BIT] =              "tx-fcoe-segmentation",
        [NETIF_F_GSO_GRE_BIT] =          "tx-gre-segmentation",
        [NETIF_F_GSO_IPIP_BIT] =         "tx-ipip-segmentation",
+       [NETIF_F_GSO_SIT_BIT] =          "tx-sit-segmentation",
        [NETIF_F_GSO_UDP_TUNNEL_BIT] =   "tx-udp_tnl-segmentation",
        [NETIF_F_GSO_MPLS_BIT] =         "tx-mpls-segmentation",
 
index 24a53fc..f4a159e 100644 (file)
@@ -1265,6 +1265,7 @@ static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
                       SKB_GSO_TCP_ECN |
                       SKB_GSO_GRE |
                       SKB_GSO_IPIP |
+                      SKB_GSO_SIT |
                       SKB_GSO_TCPV6 |
                       SKB_GSO_UDP_TUNNEL |
                       SKB_GSO_MPLS |
index dfc96b0..a7a5583 100644 (file)
@@ -57,6 +57,7 @@ struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
                               SKB_GSO_TCPV6 |
                               SKB_GSO_GRE |
                               SKB_GSO_IPIP |
+                              SKB_GSO_SIT |
                               SKB_GSO_MPLS |
                               SKB_GSO_UDP_TUNNEL |
                               0) ||
index f9b33d8..4b85169 100644 (file)
@@ -98,6 +98,7 @@ static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
                       SKB_GSO_TCP_ECN |
                       SKB_GSO_GRE |
                       SKB_GSO_IPIP |
+                      SKB_GSO_SIT |
                       SKB_GSO_UDP_TUNNEL |
                       SKB_GSO_MPLS |
                       SKB_GSO_TCPV6 |
@@ -276,6 +277,13 @@ static struct packet_offload ipv6_packet_offload __read_mostly = {
        },
 };
 
+static const struct net_offload sit_offload = {
+       .callbacks = {
+               .gso_send_check = ipv6_gso_send_check,
+               .gso_segment    = ipv6_gso_segment,
+       },
+};
+
 static int __init ipv6_offload_init(void)
 {
 
@@ -287,6 +295,9 @@ static int __init ipv6_offload_init(void)
                pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
 
        dev_add_offload(&ipv6_packet_offload);
+
+       inet_add_offload(&sit_offload, IPPROTO_IPV6);
+
        return 0;
 }
 
index 1926945..3a9038d 100644 (file)
@@ -933,10 +933,9 @@ static netdev_tx_t ipip6_tunnel_xmit(struct sk_buff *skb,
                ttl = iph6->hop_limit;
        tos = INET_ECN_encapsulate(tos, ipv6_get_dsfield(iph6));
 
-       if (likely(!skb->encapsulation)) {
-               skb_reset_inner_headers(skb);
-               skb->encapsulation = 1;
-       }
+       skb = iptunnel_handle_offloads(skb, false, SKB_GSO_SIT);
+       if (IS_ERR(skb))
+               goto out;
 
        err = iptunnel_xmit(rt, skb, fl4.saddr, fl4.daddr, IPPROTO_IPV6, tos,
                            ttl, df, !net_eq(tunnel->net, dev_net(dev)));
@@ -946,8 +945,9 @@ static netdev_tx_t ipip6_tunnel_xmit(struct sk_buff *skb,
 tx_error_icmp:
        dst_link_failure(skb);
 tx_error:
-       dev->stats.tx_errors++;
        dev_kfree_skb(skb);
+out:
+       dev->stats.tx_errors++;
        return NETDEV_TX_OK;
 }
 
@@ -956,13 +956,15 @@ static netdev_tx_t ipip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
        struct ip_tunnel *tunnel = netdev_priv(dev);
        const struct iphdr  *tiph = &tunnel->parms.iph;
 
-       if (likely(!skb->encapsulation)) {
-               skb_reset_inner_headers(skb);
-               skb->encapsulation = 1;
-       }
+       skb = iptunnel_handle_offloads(skb, false, SKB_GSO_IPIP);
+       if (IS_ERR(skb))
+               goto out;
 
        ip_tunnel_xmit(skb, dev, tiph, IPPROTO_IPIP);
        return NETDEV_TX_OK;
+out:
+       dev->stats.tx_errors++;
+       return NETDEV_TX_OK;
 }
 
 static netdev_tx_t sit_tunnel_xmit(struct sk_buff *skb,
@@ -1292,6 +1294,12 @@ static void ipip6_dev_free(struct net_device *dev)
        free_netdev(dev);
 }
 
+#define SIT_FEATURES (NETIF_F_SG          | \
+                     NETIF_F_FRAGLIST     | \
+                     NETIF_F_HIGHDMA      | \
+                     NETIF_F_GSO_SOFTWARE | \
+                     NETIF_F_HW_CSUM)
+
 static void ipip6_tunnel_setup(struct net_device *dev)
 {
        dev->netdev_ops         = &ipip6_netdev_ops;
@@ -1305,6 +1313,8 @@ static void ipip6_tunnel_setup(struct net_device *dev)
        dev->iflink             = 0;
        dev->addr_len           = 4;
        dev->features           |= NETIF_F_LLTX;
+       dev->features           |= SIT_FEATURES;
+       dev->hw_features        |= SIT_FEATURES;
 }
 
 static int ipip6_tunnel_init(struct net_device *dev)
index f63780f..08e23b0 100644 (file)
@@ -65,6 +65,7 @@ static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb,
                                      SKB_GSO_UDP_TUNNEL |
                                      SKB_GSO_GRE |
                                      SKB_GSO_IPIP |
+                                     SKB_GSO_SIT |
                                      SKB_GSO_MPLS) ||
                             !(type & (SKB_GSO_UDP))))
                        goto out;