Pileus Git - ~andy/linux/blob - net/ipv4/ip_vti.c

   1 /*
   2  *      Linux NET3: IP/IP protocol decoder modified to support
   3  *                  virtual tunnel interface
   4  *
   5  *      Authors:
   6  *              Saurabh Mohan (saurabh.mohan@vyatta.com) 05/07/2012
   7  *
   8  *      This program is free software; you can redistribute it and/or
   9  *      modify it under the terms of the GNU General Public License
  10  *      as published by the Free Software Foundation; either version
  11  *      2 of the License, or (at your option) any later version.
  12  *
  13  */
  14
  15 /*
  16    This version of net/ipv4/ip_vti.c is cloned of net/ipv4/ipip.c
  17
  18    For comments look at net/ipv4/ip_gre.c --ANK
  19  */
  20
  21
  22 #include <linux/capability.h>
  23 #include <linux/module.h>
  24 #include <linux/types.h>
  25 #include <linux/kernel.h>
  26 #include <linux/uaccess.h>
  27 #include <linux/skbuff.h>
  28 #include <linux/netdevice.h>
  29 #include <linux/in.h>
  30 #include <linux/tcp.h>
  31 #include <linux/udp.h>
  32 #include <linux/if_arp.h>
  33 #include <linux/mroute.h>
  34 #include <linux/init.h>
  35 #include <linux/netfilter_ipv4.h>
  36 #include <linux/if_ether.h>
  37
  38 #include <net/sock.h>
  39 #include <net/ip.h>
  40 #include <net/icmp.h>
  41 #include <net/ipip.h>
  42 #include <net/inet_ecn.h>
  43 #include <net/xfrm.h>
  44 #include <net/net_namespace.h>
  45 #include <net/netns/generic.h>
  46
  47 #define HASH_SIZE  16
  48 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&(HASH_SIZE-1))
  49
  50 static struct rtnl_link_ops vti_link_ops __read_mostly;
  51
  52 static int vti_net_id __read_mostly;
  53 struct vti_net {
  54         struct ip_tunnel __rcu *tunnels_r_l[HASH_SIZE];
  55         struct ip_tunnel __rcu *tunnels_r[HASH_SIZE];
  56         struct ip_tunnel __rcu *tunnels_l[HASH_SIZE];
  57         struct ip_tunnel __rcu *tunnels_wc[1];
  58         struct ip_tunnel __rcu **tunnels[4];
  59
  60         struct net_device *fb_tunnel_dev;
  61 };
  62
  63 static int vti_fb_tunnel_init(struct net_device *dev);
  64 static int vti_tunnel_init(struct net_device *dev);
  65 static void vti_tunnel_setup(struct net_device *dev);
  66 static void vti_dev_free(struct net_device *dev);
  67 static int vti_tunnel_bind_dev(struct net_device *dev);
  68
  69 #define VTI_XMIT(stats1, stats2) do {                           \
  70         int err;                                                \
  71         int pkt_len = skb->len;                                 \
  72         err = dst_output(skb);                                  \
  73         if (net_xmit_eval(err) == 0) {                          \
  74                 u64_stats_update_begin(&(stats1)->syncp);       \
  75                 (stats1)->tx_bytes += pkt_len;                  \
  76                 (stats1)->tx_packets++;                         \
  77                 u64_stats_update_end(&(stats1)->syncp);         \
  78         } else {                                                \
  79                 (stats2)->tx_errors++;                          \
  80                 (stats2)->tx_aborted_errors++;                  \
  81         }                                                       \
  82 } while (0)
  83
  84
  85 static struct rtnl_link_stats64 *vti_get_stats64(struct net_device *dev,
  86                                                  struct rtnl_link_stats64 *tot)
  87 {
  88         int i;
  89
  90         for_each_possible_cpu(i) {
  91                 const struct pcpu_tstats *tstats = per_cpu_ptr(dev->tstats, i);
  92                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
  93                 unsigned int start;
  94
  95                 do {
  96                         start = u64_stats_fetch_begin_bh(&tstats->syncp);
  97                         rx_packets = tstats->rx_packets;
  98                         tx_packets = tstats->tx_packets;
  99                         rx_bytes = tstats->rx_bytes;
 100                         tx_bytes = tstats->tx_bytes;
 101                 } while (u64_stats_fetch_retry_bh(&tstats->syncp, start));
 102
 103                 tot->rx_packets += rx_packets;
 104                 tot->tx_packets += tx_packets;
 105                 tot->rx_bytes   += rx_bytes;
 106                 tot->tx_bytes   += tx_bytes;
 107         }
 108
 109         tot->multicast = dev->stats.multicast;
 110         tot->rx_crc_errors = dev->stats.rx_crc_errors;
 111         tot->rx_fifo_errors = dev->stats.rx_fifo_errors;
 112         tot->rx_length_errors = dev->stats.rx_length_errors;
 113         tot->rx_errors = dev->stats.rx_errors;
 114         tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
 115         tot->tx_carrier_errors = dev->stats.tx_carrier_errors;
 116         tot->tx_dropped = dev->stats.tx_dropped;
 117         tot->tx_aborted_errors = dev->stats.tx_aborted_errors;
 118         tot->tx_errors = dev->stats.tx_errors;
 119
 120         return tot;
 121 }
 122
 123 static struct ip_tunnel *vti_tunnel_lookup(struct net *net,
 124                                            __be32 remote, __be32 local)
 125 {
 126         unsigned h0 = HASH(remote);
 127         unsigned h1 = HASH(local);
 128         struct ip_tunnel *t;
 129         struct vti_net *ipn = net_generic(net, vti_net_id);
 130
 131         for_each_ip_tunnel_rcu(t, ipn->tunnels_r_l[h0 ^ h1])
 132                 if (local == t->parms.iph.saddr &&
 133                     remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
 134                         return t;
 135         for_each_ip_tunnel_rcu(t, ipn->tunnels_r[h0])
 136                 if (remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
 137                         return t;
 138
 139         for_each_ip_tunnel_rcu(t, ipn->tunnels_l[h1])
 140                 if (local == t->parms.iph.saddr && (t->dev->flags&IFF_UP))
 141                         return t;
 142
 143         for_each_ip_tunnel_rcu(t, ipn->tunnels_wc[0])
 144                 if (t && (t->dev->flags&IFF_UP))
 145                         return t;
 146         return NULL;
 147 }
 148
 149 static struct ip_tunnel __rcu **__vti_bucket(struct vti_net *ipn,
 150                                              struct ip_tunnel_parm *parms)
 151 {
 152         __be32 remote = parms->iph.daddr;
 153         __be32 local = parms->iph.saddr;
 154         unsigned h = 0;
 155         int prio = 0;
 156
 157         if (remote) {
 158                 prio |= 2;
 159                 h ^= HASH(remote);
 160         }
 161         if (local) {
 162                 prio |= 1;
 163                 h ^= HASH(local);
 164         }
 165         return &ipn->tunnels[prio][h];
 166 }
 167
 168 static inline struct ip_tunnel __rcu **vti_bucket(struct vti_net *ipn,
 169                                                   struct ip_tunnel *t)
 170 {
 171         return __vti_bucket(ipn, &t->parms);
 172 }
 173
 174 static void vti_tunnel_unlink(struct vti_net *ipn, struct ip_tunnel *t)
 175 {
 176         struct ip_tunnel __rcu **tp;
 177         struct ip_tunnel *iter;
 178
 179         for (tp = vti_bucket(ipn, t);
 180              (iter = rtnl_dereference(*tp)) != NULL;
 181              tp = &iter->next) {
 182                 if (t == iter) {
 183                         rcu_assign_pointer(*tp, t->next);
 184                         break;
 185                 }
 186         }
 187 }
 188
 189 static void vti_tunnel_link(struct vti_net *ipn, struct ip_tunnel *t)
 190 {
 191         struct ip_tunnel __rcu **tp = vti_bucket(ipn, t);
 192
 193         rcu_assign_pointer(t->next, rtnl_dereference(*tp));
 194         rcu_assign_pointer(*tp, t);
 195 }
 196
 197 static struct ip_tunnel *vti_tunnel_locate(struct net *net,
 198                                            struct ip_tunnel_parm *parms,
 199                                            int create)
 200 {
 201         __be32 remote = parms->iph.daddr;
 202         __be32 local = parms->iph.saddr;
 203         struct ip_tunnel *t, *nt;
 204         struct ip_tunnel __rcu **tp;
 205         struct net_device *dev;
 206         char name[IFNAMSIZ];
 207         struct vti_net *ipn = net_generic(net, vti_net_id);
 208
 209         for (tp = __vti_bucket(ipn, parms);
 210              (t = rtnl_dereference(*tp)) != NULL;
 211              tp = &t->next) {
 212                 if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr)
 213                         return t;
 214         }
 215         if (!create)
 216                 return NULL;
 217
 218         if (parms->name[0])
 219                 strlcpy(name, parms->name, IFNAMSIZ);
 220         else
 221                 strcpy(name, "vti%d");
 222
 223         dev = alloc_netdev(sizeof(*t), name, vti_tunnel_setup);
 224         if (dev == NULL)
 225                 return NULL;
 226
 227         dev_net_set(dev, net);
 228
 229         nt = netdev_priv(dev);
 230         nt->parms = *parms;
 231         dev->rtnl_link_ops = &vti_link_ops;
 232
 233         vti_tunnel_bind_dev(dev);
 234
 235         if (register_netdevice(dev) < 0)
 236                 goto failed_free;
 237
 238         dev_hold(dev);
 239         vti_tunnel_link(ipn, nt);
 240         return nt;
 241
 242 failed_free:
 243         free_netdev(dev);
 244         return NULL;
 245 }
 246
 247 static void vti_tunnel_uninit(struct net_device *dev)
 248 {
 249         struct net *net = dev_net(dev);
 250         struct vti_net *ipn = net_generic(net, vti_net_id);
 251
 252         vti_tunnel_unlink(ipn, netdev_priv(dev));
 253         dev_put(dev);
 254 }
 255
 256 static int vti_err(struct sk_buff *skb, u32 info)
 257 {
 258
 259         /* All the routers (except for Linux) return only
 260          * 8 bytes of packet payload. It means, that precise relaying of
 261          * ICMP in the real Internet is absolutely infeasible.
 262          */
 263         struct iphdr *iph = (struct iphdr *)skb->data;
 264         const int type = icmp_hdr(skb)->type;
 265         const int code = icmp_hdr(skb)->code;
 266         struct ip_tunnel *t;
 267         int err;
 268
 269         switch (type) {
 270         default:
 271         case ICMP_PARAMETERPROB:
 272                 return 0;
 273
 274         case ICMP_DEST_UNREACH:
 275                 switch (code) {
 276                 case ICMP_SR_FAILED:
 277                 case ICMP_PORT_UNREACH:
 278                         /* Impossible event. */
 279                         return 0;
 280                 default:
 281                         /* All others are translated to HOST_UNREACH. */
 282                         break;
 283                 }
 284                 break;
 285         case ICMP_TIME_EXCEEDED:
 286                 if (code != ICMP_EXC_TTL)
 287                         return 0;
 288                 break;
 289         }
 290
 291         err = -ENOENT;
 292
 293         t = vti_tunnel_lookup(dev_net(skb->dev), iph->daddr, iph->saddr);
 294         if (t == NULL)
 295                 goto out;
 296
 297         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
 298                 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
 299                                  t->parms.link, 0, IPPROTO_IPIP, 0);
 300                 err = 0;
 301                 goto out;
 302         }
 303
 304         err = 0;
 305         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
 306                 goto out;
 307
 308         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
 309                 t->err_count++;
 310         else
 311                 t->err_count = 1;
 312         t->err_time = jiffies;
 313 out:
 314         return err;
 315 }
 316
 317 /* We dont digest the packet therefore let the packet pass */
 318 static int vti_rcv(struct sk_buff *skb)
 319 {
 320         struct ip_tunnel *tunnel;
 321         const struct iphdr *iph = ip_hdr(skb);
 322
 323         tunnel = vti_tunnel_lookup(dev_net(skb->dev), iph->saddr, iph->daddr);
 324         if (tunnel != NULL) {
 325                 struct pcpu_tstats *tstats;
 326
 327                 tstats = this_cpu_ptr(tunnel->dev->tstats);
 328                 u64_stats_update_begin(&tstats->syncp);
 329                 tstats->rx_packets++;
 330                 tstats->rx_bytes += skb->len;
 331                 u64_stats_update_end(&tstats->syncp);
 332
 333                 skb->dev = tunnel->dev;
 334                 return 1;
 335         }
 336
 337         return -1;
 338 }
 339
 340 /* This function assumes it is being called from dev_queue_xmit()
 341  * and that skb is filled properly by that function.
 342  */
 343
 344 static netdev_tx_t vti_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
 345 {
 346         struct ip_tunnel *tunnel = netdev_priv(dev);
 347         struct pcpu_tstats *tstats;
 348         struct iphdr  *tiph = &tunnel->parms.iph;
 349         u8     tos;
 350         struct rtable *rt;              /* Route to the other host */
 351         struct net_device *tdev;        /* Device to other host */
 352         struct iphdr  *old_iph = ip_hdr(skb);
 353         __be32 dst = tiph->daddr;
 354         struct flowi4 fl4;
 355
 356         if (skb->protocol != htons(ETH_P_IP))
 357                 goto tx_error;
 358
 359         tos = old_iph->tos;
 360
 361         memset(&fl4, 0, sizeof(fl4));
 362         flowi4_init_output(&fl4, tunnel->parms.link,
 363                            be32_to_cpu(tunnel->parms.i_key), RT_TOS(tos),
 364                            RT_SCOPE_UNIVERSE,
 365                            IPPROTO_IPIP, 0,
 366                            dst, tiph->saddr, 0, 0);
 367         rt = ip_route_output_key(dev_net(dev), &fl4);
 368         if (IS_ERR(rt)) {
 369                 dev->stats.tx_carrier_errors++;
 370                 goto tx_error_icmp;
 371         }
 372         /* if there is no transform then this tunnel is not functional.
 373          * Or if the xfrm is not mode tunnel.
 374          */
 375         if (!rt->dst.xfrm ||
 376             rt->dst.xfrm->props.mode != XFRM_MODE_TUNNEL) {
 377                 dev->stats.tx_carrier_errors++;
 378                 goto tx_error_icmp;
 379         }
 380         tdev = rt->dst.dev;
 381
 382         if (tdev == dev) {
 383                 ip_rt_put(rt);
 384                 dev->stats.collisions++;
 385                 goto tx_error;
 386         }
 387
 388         if (tunnel->err_count > 0) {
 389                 if (time_before(jiffies,
 390                                 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
 391                         tunnel->err_count--;
 392                         dst_link_failure(skb);
 393                 } else
 394                         tunnel->err_count = 0;
 395         }
 396
 397         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
 398                               IPSKB_REROUTED);
 399         skb_dst_drop(skb);
 400         skb_dst_set(skb, &rt->dst);
 401         nf_reset(skb);
 402         skb->dev = skb_dst(skb)->dev;
 403
 404         tstats = this_cpu_ptr(dev->tstats);
 405         VTI_XMIT(tstats, &dev->stats);
 406         return NETDEV_TX_OK;
 407
 408 tx_error_icmp:
 409         dst_link_failure(skb);
 410 tx_error:
 411         dev->stats.tx_errors++;
 412         dev_kfree_skb(skb);
 413         return NETDEV_TX_OK;
 414 }
 415
 416 static int vti_tunnel_bind_dev(struct net_device *dev)
 417 {
 418         struct net_device *tdev = NULL;
 419         struct ip_tunnel *tunnel;
 420         struct iphdr *iph;
 421
 422         tunnel = netdev_priv(dev);
 423         iph = &tunnel->parms.iph;
 424
 425         if (iph->daddr) {
 426                 struct rtable *rt;
 427                 struct flowi4 fl4;
 428                 memset(&fl4, 0, sizeof(fl4));
 429                 flowi4_init_output(&fl4, tunnel->parms.link,
 430                                    be32_to_cpu(tunnel->parms.i_key),
 431                                    RT_TOS(iph->tos), RT_SCOPE_UNIVERSE,
 432                                    IPPROTO_IPIP, 0,
 433                                    iph->daddr, iph->saddr, 0, 0);
 434                 rt = ip_route_output_key(dev_net(dev), &fl4);
 435                 if (!IS_ERR(rt)) {
 436                         tdev = rt->dst.dev;
 437                         ip_rt_put(rt);
 438                 }
 439                 dev->flags |= IFF_POINTOPOINT;
 440         }
 441
 442         if (!tdev && tunnel->parms.link)
 443                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
 444
 445         if (tdev) {
 446                 dev->hard_header_len = tdev->hard_header_len +
 447                                        sizeof(struct iphdr);
 448                 dev->mtu = tdev->mtu;
 449         }
 450         dev->iflink = tunnel->parms.link;
 451         return dev->mtu;
 452 }
 453
 454 static int
 455 vti_tunnel_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 456 {
 457         int err = 0;
 458         struct ip_tunnel_parm p;
 459         struct ip_tunnel *t;
 460         struct net *net = dev_net(dev);
 461         struct vti_net *ipn = net_generic(net, vti_net_id);
 462
 463         switch (cmd) {
 464         case SIOCGETTUNNEL:
 465                 t = NULL;
 466                 if (dev == ipn->fb_tunnel_dev) {
 467                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data,
 468                                            sizeof(p))) {
 469                                 err = -EFAULT;
 470                                 break;
 471                         }
 472                         t = vti_tunnel_locate(net, &p, 0);
 473                 }
 474                 if (t == NULL)
 475                         t = netdev_priv(dev);
 476                 memcpy(&p, &t->parms, sizeof(p));
 477                 p.i_flags |= GRE_KEY | VTI_ISVTI;
 478                 p.o_flags |= GRE_KEY;
 479                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
 480                         err = -EFAULT;
 481                 break;
 482
 483         case SIOCADDTUNNEL:
 484         case SIOCCHGTUNNEL:
 485                 err = -EPERM;
 486                 if (!capable(CAP_NET_ADMIN))
 487                         goto done;
 488
 489                 err = -EFAULT;
 490                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
 491                         goto done;
 492
 493                 err = -EINVAL;
 494                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPIP ||
 495                     p.iph.ihl != 5)
 496                         goto done;
 497
 498                 t = vti_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
 499
 500                 if (dev != ipn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
 501                         if (t != NULL) {
 502                                 if (t->dev != dev) {
 503                                         err = -EEXIST;
 504                                         break;
 505                                 }
 506                         } else {
 507                                 if (((dev->flags&IFF_POINTOPOINT) &&
 508                                     !p.iph.daddr) ||
 509                                     (!(dev->flags&IFF_POINTOPOINT) &&
 510                                     p.iph.daddr)) {
 511                                         err = -EINVAL;
 512                                         break;
 513                                 }
 514                                 t = netdev_priv(dev);
 515                                 vti_tunnel_unlink(ipn, t);
 516                                 synchronize_net();
 517                                 t->parms.iph.saddr = p.iph.saddr;
 518                                 t->parms.iph.daddr = p.iph.daddr;
 519                                 t->parms.i_key = p.i_key;
 520                                 t->parms.o_key = p.o_key;
 521                                 t->parms.iph.protocol = IPPROTO_IPIP;
 522                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
 523                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
 524                                 vti_tunnel_link(ipn, t);
 525                                 netdev_state_change(dev);
 526                         }
 527                 }
 528
 529                 if (t) {
 530                         err = 0;
 531                         if (cmd == SIOCCHGTUNNEL) {
 532                                 t->parms.i_key = p.i_key;
 533                                 t->parms.o_key = p.o_key;
 534                                 if (t->parms.link != p.link) {
 535                                         t->parms.link = p.link;
 536                                         vti_tunnel_bind_dev(dev);
 537                                         netdev_state_change(dev);
 538                                 }
 539                         }
 540                         p.i_flags |= GRE_KEY | VTI_ISVTI;
 541                         p.o_flags |= GRE_KEY;
 542                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms,
 543                                          sizeof(p)))
 544                                 err = -EFAULT;
 545                 } else
 546                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
 547                 break;
 548
 549         case SIOCDELTUNNEL:
 550                 err = -EPERM;
 551                 if (!capable(CAP_NET_ADMIN))
 552                         goto done;
 553
 554                 if (dev == ipn->fb_tunnel_dev) {
 555                         err = -EFAULT;
 556                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data,
 557                                            sizeof(p)))
 558                                 goto done;
 559                         err = -ENOENT;
 560
 561                         t = vti_tunnel_locate(net, &p, 0);
 562                         if (t == NULL)
 563                                 goto done;
 564                         err = -EPERM;
 565                         if (t->dev == ipn->fb_tunnel_dev)
 566                                 goto done;
 567                         dev = t->dev;
 568                 }
 569                 unregister_netdevice(dev);
 570                 err = 0;
 571                 break;
 572
 573         default:
 574                 err = -EINVAL;
 575         }
 576
 577 done:
 578         return err;
 579 }
 580
 581 static int vti_tunnel_change_mtu(struct net_device *dev, int new_mtu)
 582 {
 583         if (new_mtu < 68 || new_mtu > 0xFFF8)
 584                 return -EINVAL;
 585         dev->mtu = new_mtu;
 586         return 0;
 587 }
 588
 589 static const struct net_device_ops vti_netdev_ops = {
 590         .ndo_init       = vti_tunnel_init,
 591         .ndo_uninit     = vti_tunnel_uninit,
 592         .ndo_start_xmit = vti_tunnel_xmit,
 593         .ndo_do_ioctl   = vti_tunnel_ioctl,
 594         .ndo_change_mtu = vti_tunnel_change_mtu,
 595         .ndo_get_stats64 = vti_get_stats64,
 596 };
 597
 598 static void vti_dev_free(struct net_device *dev)
 599 {
 600         free_percpu(dev->tstats);
 601         free_netdev(dev);
 602 }
 603
 604 static void vti_tunnel_setup(struct net_device *dev)
 605 {
 606         dev->netdev_ops         = &vti_netdev_ops;
 607         dev->destructor         = vti_dev_free;
 608
 609         dev->type               = ARPHRD_TUNNEL;
 610         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr);
 611         dev->mtu                = ETH_DATA_LEN;
 612         dev->flags              = IFF_NOARP;
 613         dev->iflink             = 0;
 614         dev->addr_len           = 4;
 615         dev->features           |= NETIF_F_NETNS_LOCAL;
 616         dev->features           |= NETIF_F_LLTX;
 617         dev->priv_flags         &= ~IFF_XMIT_DST_RELEASE;
 618 }
 619
 620 static int vti_tunnel_init(struct net_device *dev)
 621 {
 622         struct ip_tunnel *tunnel = netdev_priv(dev);
 623
 624         tunnel->dev = dev;
 625         strcpy(tunnel->parms.name, dev->name);
 626
 627         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
 628         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
 629
 630         dev->tstats = alloc_percpu(struct pcpu_tstats);
 631         if (!dev->tstats)
 632                 return -ENOMEM;
 633
 634         return 0;
 635 }
 636
 637 static int __net_init vti_fb_tunnel_init(struct net_device *dev)
 638 {
 639         struct ip_tunnel *tunnel = netdev_priv(dev);
 640         struct iphdr *iph = &tunnel->parms.iph;
 641         struct vti_net *ipn = net_generic(dev_net(dev), vti_net_id);
 642
 643         tunnel->dev = dev;
 644         strcpy(tunnel->parms.name, dev->name);
 645
 646         iph->version            = 4;
 647         iph->protocol           = IPPROTO_IPIP;
 648         iph->ihl                = 5;
 649
 650         dev->tstats = alloc_percpu(struct pcpu_tstats);
 651         if (!dev->tstats)
 652                 return -ENOMEM;
 653
 654         dev_hold(dev);
 655         rcu_assign_pointer(ipn->tunnels_wc[0], tunnel);
 656         return 0;
 657 }
 658
 659 static struct xfrm_tunnel vti_handler __read_mostly = {
 660         .handler        =       vti_rcv,
 661         .err_handler    =       vti_err,
 662         .priority       =       1,
 663 };
 664
 665 static void vti_destroy_tunnels(struct vti_net *ipn, struct list_head *head)
 666 {
 667         int prio;
 668
 669         for (prio = 1; prio < 4; prio++) {
 670                 int h;
 671                 for (h = 0; h < HASH_SIZE; h++) {
 672                         struct ip_tunnel *t;
 673
 674                         t = rtnl_dereference(ipn->tunnels[prio][h]);
 675                         while (t != NULL) {
 676                                 unregister_netdevice_queue(t->dev, head);
 677                                 t = rtnl_dereference(t->next);
 678                         }
 679                 }
 680         }
 681 }
 682
 683 static int __net_init vti_init_net(struct net *net)
 684 {
 685         int err;
 686         struct vti_net *ipn = net_generic(net, vti_net_id);
 687
 688         ipn->tunnels[0] = ipn->tunnels_wc;
 689         ipn->tunnels[1] = ipn->tunnels_l;
 690         ipn->tunnels[2] = ipn->tunnels_r;
 691         ipn->tunnels[3] = ipn->tunnels_r_l;
 692
 693         ipn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel),
 694                                           "ip_vti0",
 695                                           vti_tunnel_setup);
 696         if (!ipn->fb_tunnel_dev) {
 697                 err = -ENOMEM;
 698                 goto err_alloc_dev;
 699         }
 700         dev_net_set(ipn->fb_tunnel_dev, net);
 701
 702         err = vti_fb_tunnel_init(ipn->fb_tunnel_dev);
 703         if (err)
 704                 goto err_reg_dev;
 705         ipn->fb_tunnel_dev->rtnl_link_ops = &vti_link_ops;
 706
 707         err = register_netdev(ipn->fb_tunnel_dev);
 708         if (err)
 709                 goto err_reg_dev;
 710         return 0;
 711
 712 err_reg_dev:
 713         vti_dev_free(ipn->fb_tunnel_dev);
 714 err_alloc_dev:
 715         /* nothing */
 716         return err;
 717 }
 718
 719 static void __net_exit vti_exit_net(struct net *net)
 720 {
 721         struct vti_net *ipn = net_generic(net, vti_net_id);
 722         LIST_HEAD(list);
 723
 724         rtnl_lock();
 725         vti_destroy_tunnels(ipn, &list);
 726         unregister_netdevice_many(&list);
 727         rtnl_unlock();
 728 }
 729
 730 static struct pernet_operations vti_net_ops = {
 731         .init = vti_init_net,
 732         .exit = vti_exit_net,
 733         .id   = &vti_net_id,
 734         .size = sizeof(struct vti_net),
 735 };
 736
 737 static int vti_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
 738 {
 739         return 0;
 740 }
 741
 742 static void vti_netlink_parms(struct nlattr *data[],
 743                               struct ip_tunnel_parm *parms)
 744 {
 745         memset(parms, 0, sizeof(*parms));
 746
 747         parms->iph.protocol = IPPROTO_IPIP;
 748
 749         if (!data)
 750                 return;
 751
 752         if (data[IFLA_VTI_LINK])
 753                 parms->link = nla_get_u32(data[IFLA_VTI_LINK]);
 754
 755         if (data[IFLA_VTI_IKEY])
 756                 parms->i_key = nla_get_be32(data[IFLA_VTI_IKEY]);
 757
 758         if (data[IFLA_VTI_OKEY])
 759                 parms->o_key = nla_get_be32(data[IFLA_VTI_OKEY]);
 760
 761         if (data[IFLA_VTI_LOCAL])
 762                 parms->iph.saddr = nla_get_be32(data[IFLA_VTI_LOCAL]);
 763
 764         if (data[IFLA_VTI_REMOTE])
 765                 parms->iph.daddr = nla_get_be32(data[IFLA_VTI_REMOTE]);
 766
 767 }
 768
 769 static int vti_newlink(struct net *src_net, struct net_device *dev,
 770                        struct nlattr *tb[], struct nlattr *data[])
 771 {
 772         struct ip_tunnel *nt;
 773         struct net *net = dev_net(dev);
 774         struct vti_net *ipn = net_generic(net, vti_net_id);
 775         int mtu;
 776         int err;
 777
 778         nt = netdev_priv(dev);
 779         vti_netlink_parms(data, &nt->parms);
 780
 781         if (vti_tunnel_locate(net, &nt->parms, 0))
 782                 return -EEXIST;
 783
 784         mtu = vti_tunnel_bind_dev(dev);
 785         if (!tb[IFLA_MTU])
 786                 dev->mtu = mtu;
 787
 788         err = register_netdevice(dev);
 789         if (err)
 790                 goto out;
 791
 792         dev_hold(dev);
 793         vti_tunnel_link(ipn, nt);
 794
 795 out:
 796         return err;
 797 }
 798
 799 static int vti_changelink(struct net_device *dev, struct nlattr *tb[],
 800                           struct nlattr *data[])
 801 {
 802         struct ip_tunnel *t, *nt;
 803         struct net *net = dev_net(dev);
 804         struct vti_net *ipn = net_generic(net, vti_net_id);
 805         struct ip_tunnel_parm p;
 806         int mtu;
 807
 808         if (dev == ipn->fb_tunnel_dev)
 809                 return -EINVAL;
 810
 811         nt = netdev_priv(dev);
 812         vti_netlink_parms(data, &p);
 813
 814         t = vti_tunnel_locate(net, &p, 0);
 815
 816         if (t) {
 817                 if (t->dev != dev)
 818                         return -EEXIST;
 819         } else {
 820                 t = nt;
 821
 822                 vti_tunnel_unlink(ipn, t);
 823                 t->parms.iph.saddr = p.iph.saddr;
 824                 t->parms.iph.daddr = p.iph.daddr;
 825                 t->parms.i_key = p.i_key;
 826                 t->parms.o_key = p.o_key;
 827                 if (dev->type != ARPHRD_ETHER) {
 828                         memcpy(dev->dev_addr, &p.iph.saddr, 4);
 829                         memcpy(dev->broadcast, &p.iph.daddr, 4);
 830                 }
 831                 vti_tunnel_link(ipn, t);
 832                 netdev_state_change(dev);
 833         }
 834
 835         if (t->parms.link != p.link) {
 836                 t->parms.link = p.link;
 837                 mtu = vti_tunnel_bind_dev(dev);
 838                 if (!tb[IFLA_MTU])
 839                         dev->mtu = mtu;
 840                 netdev_state_change(dev);
 841         }
 842
 843         return 0;
 844 }
 845
 846 static size_t vti_get_size(const struct net_device *dev)
 847 {
 848         return
 849                 /* IFLA_VTI_LINK */
 850                 nla_total_size(4) +
 851                 /* IFLA_VTI_IKEY */
 852                 nla_total_size(4) +
 853                 /* IFLA_VTI_OKEY */
 854                 nla_total_size(4) +
 855                 /* IFLA_VTI_LOCAL */
 856                 nla_total_size(4) +
 857                 /* IFLA_VTI_REMOTE */
 858                 nla_total_size(4) +
 859                 0;
 860 }
 861
 862 static int vti_fill_info(struct sk_buff *skb, const struct net_device *dev)
 863 {
 864         struct ip_tunnel *t = netdev_priv(dev);
 865         struct ip_tunnel_parm *p = &t->parms;
 866
 867         nla_put_u32(skb, IFLA_VTI_LINK, p->link);
 868         nla_put_be32(skb, IFLA_VTI_IKEY, p->i_key);
 869         nla_put_be32(skb, IFLA_VTI_OKEY, p->o_key);
 870         nla_put_be32(skb, IFLA_VTI_LOCAL, p->iph.saddr);
 871         nla_put_be32(skb, IFLA_VTI_REMOTE, p->iph.daddr);
 872
 873         return 0;
 874 }
 875
 876 static const struct nla_policy vti_policy[IFLA_VTI_MAX + 1] = {
 877         [IFLA_VTI_LINK]         = { .type = NLA_U32 },
 878         [IFLA_VTI_IKEY]         = { .type = NLA_U32 },
 879         [IFLA_VTI_OKEY]         = { .type = NLA_U32 },
 880         [IFLA_VTI_LOCAL]        = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
 881         [IFLA_VTI_REMOTE]       = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
 882 };
 883
 884 static struct rtnl_link_ops vti_link_ops __read_mostly = {
 885         .kind           = "vti",
 886         .maxtype        = IFLA_VTI_MAX,
 887         .policy         = vti_policy,
 888         .priv_size      = sizeof(struct ip_tunnel),
 889         .setup          = vti_tunnel_setup,
 890         .validate       = vti_tunnel_validate,
 891         .newlink        = vti_newlink,
 892         .changelink     = vti_changelink,
 893         .get_size       = vti_get_size,
 894         .fill_info      = vti_fill_info,
 895 };
 896
 897 static int __init vti_init(void)
 898 {
 899         int err;
 900
 901         pr_info("IPv4 over IPSec tunneling driver\n");
 902
 903         err = register_pernet_device(&vti_net_ops);
 904         if (err < 0)
 905                 return err;
 906         err = xfrm4_mode_tunnel_input_register(&vti_handler);
 907         if (err < 0) {
 908                 unregister_pernet_device(&vti_net_ops);
 909                 pr_info(KERN_INFO "vti init: can't register tunnel\n");
 910         }
 911
 912         err = rtnl_link_register(&vti_link_ops);
 913         if (err < 0)
 914                 goto rtnl_link_failed;
 915
 916         return err;
 917
 918 rtnl_link_failed:
 919         xfrm4_mode_tunnel_input_deregister(&vti_handler);
 920         unregister_pernet_device(&vti_net_ops);
 921         return err;
 922 }
 923
 924 static void __exit vti_fini(void)
 925 {
 926         rtnl_link_unregister(&vti_link_ops);
 927         if (xfrm4_mode_tunnel_input_deregister(&vti_handler))
 928                 pr_info("vti close: can't deregister tunnel\n");
 929
 930         unregister_pernet_device(&vti_net_ops);
 931 }
 932
 933 module_init(vti_init);
 934 module_exit(vti_fini);
 935 MODULE_LICENSE("GPL");
 936 MODULE_ALIAS_RTNL_LINK("vti");
 937 MODULE_ALIAS_NETDEV("ip_vti0");