]> git.karo-electronics.de Git - linux-beck.git/commitdiff
vti: switch to new ip tunnel code
authorAmerigo Wang <amwang@redhat.com>
Sun, 21 Jul 2013 02:46:25 +0000 (10:46 +0800)
committerDavid S. Miller <davem@davemloft.net>
Wed, 24 Jul 2013 00:27:32 +0000 (17:27 -0700)
GRE tunnel and IPIP tunnel already switched to the new
ip tunnel code, VTI tunnel can use it too.

Cc: Pravin B Shelar <pshelar@nicira.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Saurabh Mohan <saurabh.mohan@vyatta.com>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Cong Wang <amwang@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/ip_vti.c

index 17cc0ffa8c0d0ea2a3732b76b63be1f85c896778..79b263da416854b101043ad9a3047fa6d07e285e 100644 (file)
 #include <net/net_namespace.h>
 #include <net/netns/generic.h>
 
-#define HASH_SIZE  16
-#define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&(HASH_SIZE-1))
-
 static struct rtnl_link_ops vti_link_ops __read_mostly;
 
 static int vti_net_id __read_mostly;
-struct vti_net {
-       struct ip_tunnel __rcu *tunnels_r_l[HASH_SIZE];
-       struct ip_tunnel __rcu *tunnels_r[HASH_SIZE];
-       struct ip_tunnel __rcu *tunnels_l[HASH_SIZE];
-       struct ip_tunnel __rcu *tunnels_wc[1];
-       struct ip_tunnel __rcu **tunnels[4];
-
-       struct net_device *fb_tunnel_dev;
-};
-
-static int vti_fb_tunnel_init(struct net_device *dev);
 static int vti_tunnel_init(struct net_device *dev);
-static void vti_tunnel_setup(struct net_device *dev);
-static void vti_dev_free(struct net_device *dev);
-static int vti_tunnel_bind_dev(struct net_device *dev);
-
-#define VTI_XMIT(stats1, stats2) do {                          \
-       int err;                                                \
-       int pkt_len = skb->len;                                 \
-       err = dst_output(skb);                                  \
-       if (net_xmit_eval(err) == 0) {                          \
-               u64_stats_update_begin(&(stats1)->syncp);       \
-               (stats1)->tx_bytes += pkt_len;                  \
-               (stats1)->tx_packets++;                         \
-               u64_stats_update_end(&(stats1)->syncp);         \
-       } else {                                                \
-               (stats2)->tx_errors++;                          \
-               (stats2)->tx_aborted_errors++;                  \
-       }                                                       \
-} while (0)
-
-
-static struct ip_tunnel *vti_tunnel_lookup(struct net *net,
-                                          __be32 remote, __be32 local)
-{
-       unsigned h0 = HASH(remote);
-       unsigned h1 = HASH(local);
-       struct ip_tunnel *t;
-       struct vti_net *ipn = net_generic(net, vti_net_id);
-
-       for_each_ip_tunnel_rcu(t, ipn->tunnels_r_l[h0 ^ h1])
-               if (local == t->parms.iph.saddr &&
-                   remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
-                       return t;
-       for_each_ip_tunnel_rcu(t, ipn->tunnels_r[h0])
-               if (remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
-                       return t;
-
-       for_each_ip_tunnel_rcu(t, ipn->tunnels_l[h1])
-               if (local == t->parms.iph.saddr && (t->dev->flags&IFF_UP))
-                       return t;
-
-       for_each_ip_tunnel_rcu(t, ipn->tunnels_wc[0])
-               if (t && (t->dev->flags&IFF_UP))
-                       return t;
-       return NULL;
-}
-
-static struct ip_tunnel __rcu **__vti_bucket(struct vti_net *ipn,
-                                            struct ip_tunnel_parm *parms)
-{
-       __be32 remote = parms->iph.daddr;
-       __be32 local = parms->iph.saddr;
-       unsigned h = 0;
-       int prio = 0;
-
-       if (remote) {
-               prio |= 2;
-               h ^= HASH(remote);
-       }
-       if (local) {
-               prio |= 1;
-               h ^= HASH(local);
-       }
-       return &ipn->tunnels[prio][h];
-}
-
-static inline struct ip_tunnel __rcu **vti_bucket(struct vti_net *ipn,
-                                                 struct ip_tunnel *t)
-{
-       return __vti_bucket(ipn, &t->parms);
-}
-
-static void vti_tunnel_unlink(struct vti_net *ipn, struct ip_tunnel *t)
-{
-       struct ip_tunnel __rcu **tp;
-       struct ip_tunnel *iter;
-
-       for (tp = vti_bucket(ipn, t);
-            (iter = rtnl_dereference(*tp)) != NULL;
-            tp = &iter->next) {
-               if (t == iter) {
-                       rcu_assign_pointer(*tp, t->next);
-                       break;
-               }
-       }
-}
-
-static void vti_tunnel_link(struct vti_net *ipn, struct ip_tunnel *t)
-{
-       struct ip_tunnel __rcu **tp = vti_bucket(ipn, t);
-
-       rcu_assign_pointer(t->next, rtnl_dereference(*tp));
-       rcu_assign_pointer(*tp, t);
-}
-
-static struct ip_tunnel *vti_tunnel_locate(struct net *net,
-                                          struct ip_tunnel_parm *parms,
-                                          int create)
-{
-       __be32 remote = parms->iph.daddr;
-       __be32 local = parms->iph.saddr;
-       struct ip_tunnel *t, *nt;
-       struct ip_tunnel __rcu **tp;
-       struct net_device *dev;
-       char name[IFNAMSIZ];
-       struct vti_net *ipn = net_generic(net, vti_net_id);
-
-       for (tp = __vti_bucket(ipn, parms);
-            (t = rtnl_dereference(*tp)) != NULL;
-            tp = &t->next) {
-               if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr)
-                       return t;
-       }
-       if (!create)
-               return NULL;
-
-       if (parms->name[0])
-               strlcpy(name, parms->name, IFNAMSIZ);
-       else
-               strcpy(name, "vti%d");
-
-       dev = alloc_netdev(sizeof(*t), name, vti_tunnel_setup);
-       if (dev == NULL)
-               return NULL;
-
-       dev_net_set(dev, net);
-
-       nt = netdev_priv(dev);
-       nt->parms = *parms;
-       dev->rtnl_link_ops = &vti_link_ops;
-
-       vti_tunnel_bind_dev(dev);
-
-       if (register_netdevice(dev) < 0)
-               goto failed_free;
-
-       dev_hold(dev);
-       vti_tunnel_link(ipn, nt);
-       return nt;
-
-failed_free:
-       free_netdev(dev);
-       return NULL;
-}
-
-static void vti_tunnel_uninit(struct net_device *dev)
-{
-       struct net *net = dev_net(dev);
-       struct vti_net *ipn = net_generic(net, vti_net_id);
-
-       vti_tunnel_unlink(ipn, netdev_priv(dev));
-       dev_put(dev);
-}
 
 static int vti_err(struct sk_buff *skb, u32 info)
 {
@@ -222,6 +56,8 @@ static int vti_err(struct sk_buff *skb, u32 info)
         * 8 bytes of packet payload. It means, that precise relaying of
         * ICMP in the real Internet is absolutely infeasible.
         */
+       struct net *net = dev_net(skb->dev);
+       struct ip_tunnel_net *itn = net_generic(net, vti_net_id);
        struct iphdr *iph = (struct iphdr *)skb->data;
        const int type = icmp_hdr(skb)->type;
        const int code = icmp_hdr(skb)->code;
@@ -252,7 +88,8 @@ static int vti_err(struct sk_buff *skb, u32 info)
 
        err = -ENOENT;
 
-       t = vti_tunnel_lookup(dev_net(skb->dev), iph->daddr, iph->saddr);
+       t = ip_tunnel_lookup(itn, skb->dev->ifindex, TUNNEL_NO_KEY,
+                            iph->daddr, iph->saddr, 0);
        if (t == NULL)
                goto out;
 
@@ -281,8 +118,11 @@ static int vti_rcv(struct sk_buff *skb)
 {
        struct ip_tunnel *tunnel;
        const struct iphdr *iph = ip_hdr(skb);
+       struct net *net = dev_net(skb->dev);
+       struct ip_tunnel_net *itn = net_generic(net, vti_net_id);
 
-       tunnel = vti_tunnel_lookup(dev_net(skb->dev), iph->saddr, iph->daddr);
+       tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, TUNNEL_NO_KEY,
+                                 iph->saddr, iph->daddr, 0);
        if (tunnel != NULL) {
                struct pcpu_tstats *tstats;
 
@@ -311,7 +151,6 @@ static int vti_rcv(struct sk_buff *skb)
 static netdev_tx_t vti_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
 {
        struct ip_tunnel *tunnel = netdev_priv(dev);
-       struct pcpu_tstats *tstats;
        struct iphdr  *tiph = &tunnel->parms.iph;
        u8     tos;
        struct rtable *rt;              /* Route to the other host */
@@ -319,6 +158,7 @@ static netdev_tx_t vti_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
        struct iphdr  *old_iph = ip_hdr(skb);
        __be32 dst = tiph->daddr;
        struct flowi4 fl4;
+       int err;
 
        if (skb->protocol != htons(ETH_P_IP))
                goto tx_error;
@@ -367,8 +207,10 @@ static netdev_tx_t vti_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
        nf_reset(skb);
        skb->dev = skb_dst(skb)->dev;
 
-       tstats = this_cpu_ptr(dev->tstats);
-       VTI_XMIT(tstats, &dev->stats);
+       err = dst_output(skb);
+       if (net_xmit_eval(err) == 0)
+               err = skb->len;
+       iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
        return NETDEV_TX_OK;
 
 tx_error_icmp:
@@ -379,198 +221,57 @@ tx_error:
        return NETDEV_TX_OK;
 }
 
-static int vti_tunnel_bind_dev(struct net_device *dev)
-{
-       struct net_device *tdev = NULL;
-       struct ip_tunnel *tunnel;
-       struct iphdr *iph;
-
-       tunnel = netdev_priv(dev);
-       iph = &tunnel->parms.iph;
-
-       if (iph->daddr) {
-               struct rtable *rt;
-               struct flowi4 fl4;
-               memset(&fl4, 0, sizeof(fl4));
-               flowi4_init_output(&fl4, tunnel->parms.link,
-                                  be32_to_cpu(tunnel->parms.i_key),
-                                  RT_TOS(iph->tos), RT_SCOPE_UNIVERSE,
-                                  IPPROTO_IPIP, 0,
-                                  iph->daddr, iph->saddr, 0, 0);
-               rt = ip_route_output_key(dev_net(dev), &fl4);
-               if (!IS_ERR(rt)) {
-                       tdev = rt->dst.dev;
-                       ip_rt_put(rt);
-               }
-               dev->flags |= IFF_POINTOPOINT;
-       }
-
-       if (!tdev && tunnel->parms.link)
-               tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
-
-       if (tdev) {
-               dev->hard_header_len = tdev->hard_header_len +
-                                      sizeof(struct iphdr);
-               dev->mtu = tdev->mtu;
-       }
-       dev->iflink = tunnel->parms.link;
-       return dev->mtu;
-}
-
 static int
 vti_tunnel_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 {
        int err = 0;
        struct ip_tunnel_parm p;
-       struct ip_tunnel *t;
-       struct net *net = dev_net(dev);
-       struct vti_net *ipn = net_generic(net, vti_net_id);
-
-       switch (cmd) {
-       case SIOCGETTUNNEL:
-               t = NULL;
-               if (dev == ipn->fb_tunnel_dev) {
-                       if (copy_from_user(&p, ifr->ifr_ifru.ifru_data,
-                                          sizeof(p))) {
-                               err = -EFAULT;
-                               break;
-                       }
-                       t = vti_tunnel_locate(net, &p, 0);
-               }
-               if (t == NULL)
-                       t = netdev_priv(dev);
-               memcpy(&p, &t->parms, sizeof(p));
-               p.i_flags |= GRE_KEY | VTI_ISVTI;
-               p.o_flags |= GRE_KEY;
-               if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
-                       err = -EFAULT;
-               break;
-
-       case SIOCADDTUNNEL:
-       case SIOCCHGTUNNEL:
-               err = -EPERM;
-               if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
-                       goto done;
 
-               err = -EFAULT;
-               if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
-                       goto done;
+       if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
+               return -EFAULT;
 
-               err = -EINVAL;
+       if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
                if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPIP ||
                    p.iph.ihl != 5)
-                       goto done;
-
-               t = vti_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
-
-               if (dev != ipn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
-                       if (t != NULL) {
-                               if (t->dev != dev) {
-                                       err = -EEXIST;
-                                       break;
-                               }
-                       } else {
-                               if (((dev->flags&IFF_POINTOPOINT) &&
-                                   !p.iph.daddr) ||
-                                   (!(dev->flags&IFF_POINTOPOINT) &&
-                                   p.iph.daddr)) {
-                                       err = -EINVAL;
-                                       break;
-                               }
-                               t = netdev_priv(dev);
-                               vti_tunnel_unlink(ipn, t);
-                               synchronize_net();
-                               t->parms.iph.saddr = p.iph.saddr;
-                               t->parms.iph.daddr = p.iph.daddr;
-                               t->parms.i_key = p.i_key;
-                               t->parms.o_key = p.o_key;
-                               t->parms.iph.protocol = IPPROTO_IPIP;
-                               memcpy(dev->dev_addr, &p.iph.saddr, 4);
-                               memcpy(dev->broadcast, &p.iph.daddr, 4);
-                               vti_tunnel_link(ipn, t);
-                               netdev_state_change(dev);
-                       }
-               }
-
-               if (t) {
-                       err = 0;
-                       if (cmd == SIOCCHGTUNNEL) {
-                               t->parms.i_key = p.i_key;
-                               t->parms.o_key = p.o_key;
-                               if (t->parms.link != p.link) {
-                                       t->parms.link = p.link;
-                                       vti_tunnel_bind_dev(dev);
-                                       netdev_state_change(dev);
-                               }
-                       }
-                       p.i_flags |= GRE_KEY | VTI_ISVTI;
-                       p.o_flags |= GRE_KEY;
-                       if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms,
-                                        sizeof(p)))
-                               err = -EFAULT;
-               } else
-                       err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
-               break;
+                       return -EINVAL;
+       }
 
-       case SIOCDELTUNNEL:
-               err = -EPERM;
-               if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
-                       goto done;
-
-               if (dev == ipn->fb_tunnel_dev) {
-                       err = -EFAULT;
-                       if (copy_from_user(&p, ifr->ifr_ifru.ifru_data,
-                                          sizeof(p)))
-                               goto done;
-                       err = -ENOENT;
-
-                       t = vti_tunnel_locate(net, &p, 0);
-                       if (t == NULL)
-                               goto done;
-                       err = -EPERM;
-                       if (t->dev == ipn->fb_tunnel_dev)
-                               goto done;
-                       dev = t->dev;
-               }
-               unregister_netdevice(dev);
-               err = 0;
-               break;
+       err = ip_tunnel_ioctl(dev, &p, cmd);
+       if (err)
+               return err;
 
-       default:
-               err = -EINVAL;
+       if (cmd != SIOCDELTUNNEL) {
+               p.i_flags |= GRE_KEY | VTI_ISVTI;
+               p.o_flags |= GRE_KEY;
        }
 
-done:
-       return err;
-}
-
-static int vti_tunnel_change_mtu(struct net_device *dev, int new_mtu)
-{
-       if (new_mtu < 68 || new_mtu > 0xFFF8)
-               return -EINVAL;
-       dev->mtu = new_mtu;
+       if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
+               return -EFAULT;
        return 0;
 }
 
 static const struct net_device_ops vti_netdev_ops = {
        .ndo_init       = vti_tunnel_init,
-       .ndo_uninit     = vti_tunnel_uninit,
+       .ndo_uninit     = ip_tunnel_uninit,
        .ndo_start_xmit = vti_tunnel_xmit,
        .ndo_do_ioctl   = vti_tunnel_ioctl,
-       .ndo_change_mtu = vti_tunnel_change_mtu,
+       .ndo_change_mtu = ip_tunnel_change_mtu,
        .ndo_get_stats64 = ip_tunnel_get_stats64,
 };
 
-static void vti_dev_free(struct net_device *dev)
+static void vti_tunnel_setup(struct net_device *dev)
 {
-       free_percpu(dev->tstats);
-       free_netdev(dev);
+       dev->netdev_ops         = &vti_netdev_ops;
+       ip_tunnel_setup(dev, vti_net_id);
 }
 
-static void vti_tunnel_setup(struct net_device *dev)
+static int vti_tunnel_init(struct net_device *dev)
 {
-       dev->netdev_ops         = &vti_netdev_ops;
-       dev->destructor         = vti_dev_free;
+       struct ip_tunnel *tunnel = netdev_priv(dev);
+       struct iphdr *iph = &tunnel->parms.iph;
+
+       memcpy(dev->dev_addr, &iph->saddr, 4);
+       memcpy(dev->broadcast, &iph->daddr, 4);
 
        dev->type               = ARPHRD_TUNNEL;
        dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr);
@@ -581,38 +282,18 @@ static void vti_tunnel_setup(struct net_device *dev)
        dev->features           |= NETIF_F_NETNS_LOCAL;
        dev->features           |= NETIF_F_LLTX;
        dev->priv_flags         &= ~IFF_XMIT_DST_RELEASE;
-}
 
-static int vti_tunnel_init(struct net_device *dev)
-{
-       struct ip_tunnel *tunnel = netdev_priv(dev);
-
-       tunnel->dev = dev;
-       strcpy(tunnel->parms.name, dev->name);
-
-       memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
-       memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
-
-       dev->tstats = alloc_percpu(struct pcpu_tstats);
-       if (!dev->tstats)
-               return -ENOMEM;
-
-       return 0;
+       return ip_tunnel_init(dev);
 }
 
-static int __net_init vti_fb_tunnel_init(struct net_device *dev)
+static void __net_init vti_fb_tunnel_init(struct net_device *dev)
 {
        struct ip_tunnel *tunnel = netdev_priv(dev);
        struct iphdr *iph = &tunnel->parms.iph;
-       struct vti_net *ipn = net_generic(dev_net(dev), vti_net_id);
 
        iph->version            = 4;
        iph->protocol           = IPPROTO_IPIP;
        iph->ihl                = 5;
-
-       dev_hold(dev);
-       rcu_assign_pointer(ipn->tunnels_wc[0], tunnel);
-       return 0;
 }
 
 static struct xfrm_tunnel vti_handler __read_mostly = {
@@ -621,76 +302,30 @@ static struct xfrm_tunnel vti_handler __read_mostly = {
        .priority       =       1,
 };
 
-static void vti_destroy_tunnels(struct vti_net *ipn, struct list_head *head)
-{
-       int prio;
-
-       for (prio = 1; prio < 4; prio++) {
-               int h;
-               for (h = 0; h < HASH_SIZE; h++) {
-                       struct ip_tunnel *t;
-
-                       t = rtnl_dereference(ipn->tunnels[prio][h]);
-                       while (t != NULL) {
-                               unregister_netdevice_queue(t->dev, head);
-                               t = rtnl_dereference(t->next);
-                       }
-               }
-       }
-}
-
 static int __net_init vti_init_net(struct net *net)
 {
        int err;
-       struct vti_net *ipn = net_generic(net, vti_net_id);
-
-       ipn->tunnels[0] = ipn->tunnels_wc;
-       ipn->tunnels[1] = ipn->tunnels_l;
-       ipn->tunnels[2] = ipn->tunnels_r;
-       ipn->tunnels[3] = ipn->tunnels_r_l;
-
-       ipn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel),
-                                         "ip_vti0",
-                                         vti_tunnel_setup);
-       if (!ipn->fb_tunnel_dev) {
-               err = -ENOMEM;
-               goto err_alloc_dev;
-       }
-       dev_net_set(ipn->fb_tunnel_dev, net);
-
-       err = vti_fb_tunnel_init(ipn->fb_tunnel_dev);
-       if (err)
-               goto err_reg_dev;
-       ipn->fb_tunnel_dev->rtnl_link_ops = &vti_link_ops;
+       struct ip_tunnel_net *itn;
 
-       err = register_netdev(ipn->fb_tunnel_dev);
+       err = ip_tunnel_init_net(net, vti_net_id, &vti_link_ops, "ip_vti0");
        if (err)
-               goto err_reg_dev;
+               return err;
+       itn = net_generic(net, vti_net_id);
+       vti_fb_tunnel_init(itn->fb_tunnel_dev);
        return 0;
-
-err_reg_dev:
-       vti_dev_free(ipn->fb_tunnel_dev);
-err_alloc_dev:
-       /* nothing */
-       return err;
 }
 
 static void __net_exit vti_exit_net(struct net *net)
 {
-       struct vti_net *ipn = net_generic(net, vti_net_id);
-       LIST_HEAD(list);
-
-       rtnl_lock();
-       vti_destroy_tunnels(ipn, &list);
-       unregister_netdevice_many(&list);
-       rtnl_unlock();
+       struct ip_tunnel_net *itn = net_generic(net, vti_net_id);
+       ip_tunnel_delete_net(itn);
 }
 
 static struct pernet_operations vti_net_ops = {
        .init = vti_init_net,
        .exit = vti_exit_net,
        .id   = &vti_net_id,
-       .size = sizeof(struct vti_net),
+       .size = sizeof(struct ip_tunnel_net),
 };
 
 static int vti_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
@@ -728,78 +363,19 @@ static void vti_netlink_parms(struct nlattr *data[],
 static int vti_newlink(struct net *src_net, struct net_device *dev,
                       struct nlattr *tb[], struct nlattr *data[])
 {
-       struct ip_tunnel *nt;
-       struct net *net = dev_net(dev);
-       struct vti_net *ipn = net_generic(net, vti_net_id);
-       int mtu;
-       int err;
-
-       nt = netdev_priv(dev);
-       vti_netlink_parms(data, &nt->parms);
-
-       if (vti_tunnel_locate(net, &nt->parms, 0))
-               return -EEXIST;
+       struct ip_tunnel_parm parms;
 
-       mtu = vti_tunnel_bind_dev(dev);
-       if (!tb[IFLA_MTU])
-               dev->mtu = mtu;
-
-       err = register_netdevice(dev);
-       if (err)
-               goto out;
-
-       dev_hold(dev);
-       vti_tunnel_link(ipn, nt);
-
-out:
-       return err;
+       vti_netlink_parms(data, &parms);
+       return ip_tunnel_newlink(dev, tb, &parms);
 }
 
 static int vti_changelink(struct net_device *dev, struct nlattr *tb[],
                          struct nlattr *data[])
 {
-       struct ip_tunnel *t, *nt;
-       struct net *net = dev_net(dev);
-       struct vti_net *ipn = net_generic(net, vti_net_id);
        struct ip_tunnel_parm p;
-       int mtu;
-
-       if (dev == ipn->fb_tunnel_dev)
-               return -EINVAL;
 
-       nt = netdev_priv(dev);
        vti_netlink_parms(data, &p);
-
-       t = vti_tunnel_locate(net, &p, 0);
-
-       if (t) {
-               if (t->dev != dev)
-                       return -EEXIST;
-       } else {
-               t = nt;
-
-               vti_tunnel_unlink(ipn, t);
-               t->parms.iph.saddr = p.iph.saddr;
-               t->parms.iph.daddr = p.iph.daddr;
-               t->parms.i_key = p.i_key;
-               t->parms.o_key = p.o_key;
-               if (dev->type != ARPHRD_ETHER) {
-                       memcpy(dev->dev_addr, &p.iph.saddr, 4);
-                       memcpy(dev->broadcast, &p.iph.daddr, 4);
-               }
-               vti_tunnel_link(ipn, t);
-               netdev_state_change(dev);
-       }
-
-       if (t->parms.link != p.link) {
-               t->parms.link = p.link;
-               mtu = vti_tunnel_bind_dev(dev);
-               if (!tb[IFLA_MTU])
-                       dev->mtu = mtu;
-               netdev_state_change(dev);
-       }
-
-       return 0;
+       return ip_tunnel_changelink(dev, tb, &p);
 }
 
 static size_t vti_get_size(const struct net_device *dev)
@@ -865,7 +441,7 @@ static int __init vti_init(void)
        err = xfrm4_mode_tunnel_input_register(&vti_handler);
        if (err < 0) {
                unregister_pernet_device(&vti_net_ops);
-               pr_info(KERN_INFO "vti init: can't register tunnel\n");
+               pr_info("vti init: can't register tunnel\n");
        }
 
        err = rtnl_link_register(&vti_link_ops);