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 /* Locking : hash tables are protected by RCU and RTNL */
  70
  71 #define for_each_ip_tunnel_rcu(start) \
  72         for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
  73
  74 /* often modified stats are per cpu, other are shared (netdev->stats) */
  75 struct pcpu_tstats {
  76         u64     rx_packets;
  77         u64     rx_bytes;
  78         u64     tx_packets;
  79         u64     tx_bytes;
  80         struct  u64_stats_sync  syncp;
  81 };
  82
  83 #define VTI_XMIT(stats1, stats2) do {                           \
  84         int err;                                                \
  85         int pkt_len = skb->len;                                 \
  86         err = dst_output(skb);                                  \
  87         if (net_xmit_eval(err) == 0) {                          \
  88                 u64_stats_update_begin(&(stats1)->syncp);       \
  89                 (stats1)->tx_bytes += pkt_len;                  \
  90                 (stats1)->tx_packets++;                         \
  91                 u64_stats_update_end(&(stats1)->syncp);         \
  92         } else {                                                \
  93                 (stats2)->tx_errors++;                          \
  94                 (stats2)->tx_aborted_errors++;                  \
  95         }                                                       \
  96 } while (0)
  97
  98
  99 static struct rtnl_link_stats64 *vti_get_stats64(struct net_device *dev,
 100                                                  struct rtnl_link_stats64 *tot)
 101 {
 102         int i;
 103
 104         for_each_possible_cpu(i) {
 105                 const struct pcpu_tstats *tstats = per_cpu_ptr(dev->tstats, i);
 106                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
 107                 unsigned int start;
 108
 109                 do {
 110                         start = u64_stats_fetch_begin_bh(&tstats->syncp);
 111                         rx_packets = tstats->rx_packets;
 112                         tx_packets = tstats->tx_packets;
 113                         rx_bytes = tstats->rx_bytes;
 114                         tx_bytes = tstats->tx_bytes;
 115                 } while (u64_stats_fetch_retry_bh(&tstats->syncp, start));
 116
 117                 tot->rx_packets += rx_packets;
 118                 tot->tx_packets += tx_packets;
 119                 tot->rx_bytes   += rx_bytes;
 120                 tot->tx_bytes   += tx_bytes;
 121         }
 122
 123         tot->multicast = dev->stats.multicast;
 124         tot->rx_crc_errors = dev->stats.rx_crc_errors;
 125         tot->rx_fifo_errors = dev->stats.rx_fifo_errors;
 126         tot->rx_length_errors = dev->stats.rx_length_errors;
 127         tot->rx_errors = dev->stats.rx_errors;
 128         tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
 129         tot->tx_carrier_errors = dev->stats.tx_carrier_errors;
 130         tot->tx_dropped = dev->stats.tx_dropped;
 131         tot->tx_aborted_errors = dev->stats.tx_aborted_errors;
 132         tot->tx_errors = dev->stats.tx_errors;
 133
 134         return tot;
 135 }
 136
 137 static struct ip_tunnel *vti_tunnel_lookup(struct net *net,
 138                                            __be32 remote, __be32 local)
 139 {
 140         unsigned h0 = HASH(remote);
 141         unsigned h1 = HASH(local);
 142         struct ip_tunnel *t;
 143         struct vti_net *ipn = net_generic(net, vti_net_id);
 144
 145         for_each_ip_tunnel_rcu(ipn->tunnels_r_l[h0 ^ h1])
 146                 if (local == t->parms.iph.saddr &&
 147                     remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
 148                         return t;
 149         for_each_ip_tunnel_rcu(ipn->tunnels_r[h0])
 150                 if (remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
 151                         return t;
 152
 153         for_each_ip_tunnel_rcu(ipn->tunnels_l[h1])
 154                 if (local == t->parms.iph.saddr && (t->dev->flags&IFF_UP))
 155                         return t;
 156
 157         for_each_ip_tunnel_rcu(ipn->tunnels_wc[0])
 158                 if (t && (t->dev->flags&IFF_UP))
 159                         return t;
 160         return NULL;
 161 }
 162
 163 static struct ip_tunnel __rcu **__vti_bucket(struct vti_net *ipn,
 164                                              struct ip_tunnel_parm *parms)
 165 {
 166         __be32 remote = parms->iph.daddr;
 167         __be32 local = parms->iph.saddr;
 168         unsigned h = 0;
 169         int prio = 0;
 170
 171         if (remote) {
 172                 prio |= 2;
 173                 h ^= HASH(remote);
 174         }
 175         if (local) {
 176                 prio |= 1;
 177                 h ^= HASH(local);
 178         }
 179         return &ipn->tunnels[prio][h];
 180 }
 181
 182 static inline struct ip_tunnel __rcu **vti_bucket(struct vti_net *ipn,
 183                                                   struct ip_tunnel *t)
 184 {
 185         return __vti_bucket(ipn, &t->parms);
 186 }
 187
 188 static void vti_tunnel_unlink(struct vti_net *ipn, struct ip_tunnel *t)
 189 {
 190         struct ip_tunnel __rcu **tp;
 191         struct ip_tunnel *iter;
 192
 193         for (tp = vti_bucket(ipn, t);
 194              (iter = rtnl_dereference(*tp)) != NULL;
 195              tp = &iter->next) {
 196                 if (t == iter) {
 197                         rcu_assign_pointer(*tp, t->next);
 198                         break;
 199                 }
 200         }
 201 }
 202
 203 static void vti_tunnel_link(struct vti_net *ipn, struct ip_tunnel *t)
 204 {
 205         struct ip_tunnel __rcu **tp = vti_bucket(ipn, t);
 206
 207         rcu_assign_pointer(t->next, rtnl_dereference(*tp));
 208         rcu_assign_pointer(*tp, t);
 209 }
 210
 211 static struct ip_tunnel *vti_tunnel_locate(struct net *net,
 212                                            struct ip_tunnel_parm *parms,
 213                                            int create)
 214 {
 215         __be32 remote = parms->iph.daddr;
 216         __be32 local = parms->iph.saddr;
 217         struct ip_tunnel *t, *nt;
 218         struct ip_tunnel __rcu **tp;
 219         struct net_device *dev;
 220         char name[IFNAMSIZ];
 221         struct vti_net *ipn = net_generic(net, vti_net_id);
 222
 223         for (tp = __vti_bucket(ipn, parms);
 224              (t = rtnl_dereference(*tp)) != NULL;
 225              tp = &t->next) {
 226                 if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr)
 227                         return t;
 228         }
 229         if (!create)
 230                 return NULL;
 231
 232         if (parms->name[0])
 233                 strlcpy(name, parms->name, IFNAMSIZ);
 234         else
 235                 strcpy(name, "vti%d");
 236
 237         dev = alloc_netdev(sizeof(*t), name, vti_tunnel_setup);
 238         if (dev == NULL)
 239                 return NULL;
 240
 241         dev_net_set(dev, net);
 242
 243         nt = netdev_priv(dev);
 244         nt->parms = *parms;
 245         dev->rtnl_link_ops = &vti_link_ops;
 246
 247         vti_tunnel_bind_dev(dev);
 248
 249         if (register_netdevice(dev) < 0)
 250                 goto failed_free;
 251
 252         dev_hold(dev);
 253         vti_tunnel_link(ipn, nt);
 254         return nt;
 255
 256 failed_free:
 257         free_netdev(dev);
 258         return NULL;
 259 }
 260
 261 static void vti_tunnel_uninit(struct net_device *dev)
 262 {
 263         struct net *net = dev_net(dev);
 264         struct vti_net *ipn = net_generic(net, vti_net_id);
 265
 266         vti_tunnel_unlink(ipn, netdev_priv(dev));
 267         dev_put(dev);
 268 }
 269
 270 static int vti_err(struct sk_buff *skb, u32 info)
 271 {
 272
 273         /* All the routers (except for Linux) return only
 274          * 8 bytes of packet payload. It means, that precise relaying of
 275          * ICMP in the real Internet is absolutely infeasible.
 276          */
 277         struct iphdr *iph = (struct iphdr *)skb->data;
 278         const int type = icmp_hdr(skb)->type;
 279         const int code = icmp_hdr(skb)->code;
 280         struct ip_tunnel *t;
 281         int err;
 282
 283         switch (type) {
 284         default:
 285         case ICMP_PARAMETERPROB:
 286                 return 0;
 287
 288         case ICMP_DEST_UNREACH:
 289                 switch (code) {
 290                 case ICMP_SR_FAILED:
 291                 case ICMP_PORT_UNREACH:
 292                         /* Impossible event. */
 293                         return 0;
 294                 default:
 295                         /* All others are translated to HOST_UNREACH. */
 296                         break;
 297                 }
 298                 break;
 299         case ICMP_TIME_EXCEEDED:
 300                 if (code != ICMP_EXC_TTL)
 301                         return 0;
 302                 break;
 303         }
 304
 305         err = -ENOENT;
 306
 307         t = vti_tunnel_lookup(dev_net(skb->dev), iph->daddr, iph->saddr);
 308         if (t == NULL)
 309                 goto out;
 310
 311         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
 312                 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
 313                                  t->parms.link, 0, IPPROTO_IPIP, 0);
 314                 err = 0;
 315                 goto out;
 316         }
 317
 318         err = 0;
 319         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
 320                 goto out;
 321
 322         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
 323                 t->err_count++;
 324         else
 325                 t->err_count = 1;
 326         t->err_time = jiffies;
 327 out:
 328         return err;
 329 }
 330
 331 /* We dont digest the packet therefore let the packet pass */
 332 static int vti_rcv(struct sk_buff *skb)
 333 {
 334         struct ip_tunnel *tunnel;
 335         const struct iphdr *iph = ip_hdr(skb);
 336
 337         tunnel = vti_tunnel_lookup(dev_net(skb->dev), iph->saddr, iph->daddr);
 338         if (tunnel != NULL) {
 339                 struct pcpu_tstats *tstats;
 340
 341                 tstats = this_cpu_ptr(tunnel->dev->tstats);
 342                 u64_stats_update_begin(&tstats->syncp);
 343                 tstats->rx_packets++;
 344                 tstats->rx_bytes += skb->len;
 345                 u64_stats_update_end(&tstats->syncp);
 346
 347                 skb->dev = tunnel->dev;
 348                 return 1;
 349         }
 350
 351         return -1;
 352 }
 353
 354 /* This function assumes it is being called from dev_queue_xmit()
 355  * and that skb is filled properly by that function.
 356  */
 357
 358 static netdev_tx_t vti_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
 359 {
 360         struct ip_tunnel *tunnel = netdev_priv(dev);
 361         struct pcpu_tstats *tstats;
 362         struct iphdr  *tiph = &tunnel->parms.iph;
 363         u8     tos;
 364         struct rtable *rt;              /* Route to the other host */
 365         struct net_device *tdev;        /* Device to other host */
 366         struct iphdr  *old_iph = ip_hdr(skb);
 367         __be32 dst = tiph->daddr;
 368         struct flowi4 fl4;
 369
 370         if (skb->protocol != htons(ETH_P_IP))
 371                 goto tx_error;
 372
 373         tos = old_iph->tos;
 374
 375         memset(&fl4, 0, sizeof(fl4));
 376         flowi4_init_output(&fl4, tunnel->parms.link,
 377                            be32_to_cpu(tunnel->parms.i_key), RT_TOS(tos),
 378                            RT_SCOPE_UNIVERSE,
 379                            IPPROTO_IPIP, 0,
 380                            dst, tiph->saddr, 0, 0);
 381         rt = ip_route_output_key(dev_net(dev), &fl4);
 382         if (IS_ERR(rt)) {
 383                 dev->stats.tx_carrier_errors++;
 384                 goto tx_error_icmp;
 385         }
 386         /* if there is no transform then this tunnel is not functional.
 387          * Or if the xfrm is not mode tunnel.
 388          */
 389         if (!rt->dst.xfrm ||
 390             rt->dst.xfrm->props.mode != XFRM_MODE_TUNNEL) {
 391                 dev->stats.tx_carrier_errors++;
 392                 goto tx_error_icmp;
 393         }
 394         tdev = rt->dst.dev;
 395
 396         if (tdev == dev) {
 397                 ip_rt_put(rt);
 398                 dev->stats.collisions++;
 399                 goto tx_error;
 400         }
 401
 402         if (tunnel->err_count > 0) {
 403                 if (time_before(jiffies,
 404                                 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
 405                         tunnel->err_count--;
 406                         dst_link_failure(skb);
 407                 } else
 408                         tunnel->err_count = 0;
 409         }
 410
 411         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
 412                               IPSKB_REROUTED);
 413         skb_dst_drop(skb);
 414         skb_dst_set(skb, &rt->dst);
 415         nf_reset(skb);
 416         skb->dev = skb_dst(skb)->dev;
 417
 418         tstats = this_cpu_ptr(dev->tstats);
 419         VTI_XMIT(tstats, &dev->stats);
 420         return NETDEV_TX_OK;
 421
 422 tx_error_icmp:
 423         dst_link_failure(skb);
 424 tx_error:
 425         dev->stats.tx_errors++;
 426         dev_kfree_skb(skb);
 427         return NETDEV_TX_OK;
 428 }
 429
 430 static int vti_tunnel_bind_dev(struct net_device *dev)
 431 {
 432         struct net_device *tdev = NULL;
 433         struct ip_tunnel *tunnel;
 434         struct iphdr *iph;
 435
 436         tunnel = netdev_priv(dev);
 437         iph = &tunnel->parms.iph;
 438
 439         if (iph->daddr) {
 440                 struct rtable *rt;
 441                 struct flowi4 fl4;
 442                 memset(&fl4, 0, sizeof(fl4));
 443                 flowi4_init_output(&fl4, tunnel->parms.link,
 444                                    be32_to_cpu(tunnel->parms.i_key),
 445                                    RT_TOS(iph->tos), RT_SCOPE_UNIVERSE,
 446                                    IPPROTO_IPIP, 0,
 447                                    iph->daddr, iph->saddr, 0, 0);
 448                 rt = ip_route_output_key(dev_net(dev), &fl4);
 449                 if (!IS_ERR(rt)) {
 450                         tdev = rt->dst.dev;
 451                         ip_rt_put(rt);
 452                 }
 453                 dev->flags |= IFF_POINTOPOINT;
 454         }
 455
 456         if (!tdev && tunnel->parms.link)
 457                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
 458
 459         if (tdev) {
 460                 dev->hard_header_len = tdev->hard_header_len +
 461                                        sizeof(struct iphdr);
 462                 dev->mtu = tdev->mtu;
 463         }
 464         dev->iflink = tunnel->parms.link;
 465         return dev->mtu;
 466 }
 467
 468 static int
 469 vti_tunnel_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 470 {
 471         int err = 0;
 472         struct ip_tunnel_parm p;
 473         struct ip_tunnel *t;
 474         struct net *net = dev_net(dev);
 475         struct vti_net *ipn = net_generic(net, vti_net_id);
 476
 477         switch (cmd) {
 478         case SIOCGETTUNNEL:
 479                 t = NULL;
 480                 if (dev == ipn->fb_tunnel_dev) {
 481                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data,
 482                                            sizeof(p))) {
 483                                 err = -EFAULT;
 484                                 break;
 485                         }
 486                         t = vti_tunnel_locate(net, &p, 0);
 487                 }
 488                 if (t == NULL)
 489                         t = netdev_priv(dev);
 490                 memcpy(&p, &t->parms, sizeof(p));
 491                 p.i_flags |= GRE_KEY | VTI_ISVTI;
 492                 p.o_flags |= GRE_KEY;
 493                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
 494                         err = -EFAULT;
 495                 break;
 496
 497         case SIOCADDTUNNEL:
 498         case SIOCCHGTUNNEL:
 499                 err = -EPERM;
 500                 if (!capable(CAP_NET_ADMIN))
 501                         goto done;
 502
 503                 err = -EFAULT;
 504                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
 505                         goto done;
 506
 507                 err = -EINVAL;
 508                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPIP ||
 509                     p.iph.ihl != 5)
 510                         goto done;
 511
 512                 t = vti_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
 513
 514                 if (dev != ipn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
 515                         if (t != NULL) {
 516                                 if (t->dev != dev) {
 517                                         err = -EEXIST;
 518                                         break;
 519                                 }
 520                         } else {
 521                                 if (((dev->flags&IFF_POINTOPOINT) &&
 522                                     !p.iph.daddr) ||
 523                                     (!(dev->flags&IFF_POINTOPOINT) &&
 524                                     p.iph.daddr)) {
 525                                         err = -EINVAL;
 526                                         break;
 527                                 }
 528                                 t = netdev_priv(dev);
 529                                 vti_tunnel_unlink(ipn, t);
 530                                 synchronize_net();
 531                                 t->parms.iph.saddr = p.iph.saddr;
 532                                 t->parms.iph.daddr = p.iph.daddr;
 533                                 t->parms.i_key = p.i_key;
 534                                 t->parms.o_key = p.o_key;
 535                                 t->parms.iph.protocol = IPPROTO_IPIP;
 536                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
 537                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
 538                                 vti_tunnel_link(ipn, t);
 539                                 netdev_state_change(dev);
 540                         }
 541                 }
 542
 543                 if (t) {
 544                         err = 0;
 545                         if (cmd == SIOCCHGTUNNEL) {
 546                                 t->parms.i_key = p.i_key;
 547                                 t->parms.o_key = p.o_key;
 548                                 if (t->parms.link != p.link) {
 549                                         t->parms.link = p.link;
 550                                         vti_tunnel_bind_dev(dev);
 551                                         netdev_state_change(dev);
 552                                 }
 553                         }
 554                         p.i_flags |= GRE_KEY | VTI_ISVTI;
 555                         p.o_flags |= GRE_KEY;
 556                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms,
 557                                          sizeof(p)))
 558                                 err = -EFAULT;
 559                 } else
 560                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
 561                 break;
 562
 563         case SIOCDELTUNNEL:
 564                 err = -EPERM;
 565                 if (!capable(CAP_NET_ADMIN))
 566                         goto done;
 567
 568                 if (dev == ipn->fb_tunnel_dev) {
 569                         err = -EFAULT;
 570                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data,
 571                                            sizeof(p)))
 572                                 goto done;
 573                         err = -ENOENT;
 574
 575                         t = vti_tunnel_locate(net, &p, 0);
 576                         if (t == NULL)
 577                                 goto done;
 578                         err = -EPERM;
 579                         if (t->dev == ipn->fb_tunnel_dev)
 580                                 goto done;
 581                         dev = t->dev;
 582                 }
 583                 unregister_netdevice(dev);
 584                 err = 0;
 585                 break;
 586
 587         default:
 588                 err = -EINVAL;
 589         }
 590
 591 done:
 592         return err;
 593 }
 594
 595 static int vti_tunnel_change_mtu(struct net_device *dev, int new_mtu)
 596 {
 597         if (new_mtu < 68 || new_mtu > 0xFFF8)
 598                 return -EINVAL;
 599         dev->mtu = new_mtu;
 600         return 0;
 601 }
 602
 603 static const struct net_device_ops vti_netdev_ops = {
 604         .ndo_init       = vti_tunnel_init,
 605         .ndo_uninit     = vti_tunnel_uninit,
 606         .ndo_start_xmit = vti_tunnel_xmit,
 607         .ndo_do_ioctl   = vti_tunnel_ioctl,
 608         .ndo_change_mtu = vti_tunnel_change_mtu,
 609         .ndo_get_stats64 = vti_get_stats64,
 610 };
 611
 612 static void vti_dev_free(struct net_device *dev)
 613 {
 614         free_percpu(dev->tstats);
 615         free_netdev(dev);
 616 }
 617
 618 static void vti_tunnel_setup(struct net_device *dev)
 619 {
 620         dev->netdev_ops         = &vti_netdev_ops;
 621         dev->destructor         = vti_dev_free;
 622
 623         dev->type               = ARPHRD_TUNNEL;
 624         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr);
 625         dev->mtu                = ETH_DATA_LEN;
 626         dev->flags              = IFF_NOARP;
 627         dev->iflink             = 0;
 628         dev->addr_len           = 4;
 629         dev->features           |= NETIF_F_NETNS_LOCAL;
 630         dev->features           |= NETIF_F_LLTX;
 631         dev->priv_flags         &= ~IFF_XMIT_DST_RELEASE;
 632 }
 633
 634 static int vti_tunnel_init(struct net_device *dev)
 635 {
 636         struct ip_tunnel *tunnel = netdev_priv(dev);
 637
 638         tunnel->dev = dev;
 639         strcpy(tunnel->parms.name, dev->name);
 640
 641         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
 642         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
 643
 644         dev->tstats = alloc_percpu(struct pcpu_tstats);
 645         if (!dev->tstats)
 646                 return -ENOMEM;
 647
 648         return 0;
 649 }
 650
 651 static int __net_init vti_fb_tunnel_init(struct net_device *dev)
 652 {
 653         struct ip_tunnel *tunnel = netdev_priv(dev);
 654         struct iphdr *iph = &tunnel->parms.iph;
 655         struct vti_net *ipn = net_generic(dev_net(dev), vti_net_id);
 656
 657         tunnel->dev = dev;
 658         strcpy(tunnel->parms.name, dev->name);
 659
 660         iph->version            = 4;
 661         iph->protocol           = IPPROTO_IPIP;
 662         iph->ihl                = 5;
 663
 664         dev->tstats = alloc_percpu(struct pcpu_tstats);
 665         if (!dev->tstats)
 666                 return -ENOMEM;
 667
 668         dev_hold(dev);
 669         rcu_assign_pointer(ipn->tunnels_wc[0], tunnel);
 670         return 0;
 671 }
 672
 673 static struct xfrm_tunnel vti_handler __read_mostly = {
 674         .handler        =       vti_rcv,
 675         .err_handler    =       vti_err,
 676         .priority       =       1,
 677 };
 678
 679 static void vti_destroy_tunnels(struct vti_net *ipn, struct list_head *head)
 680 {
 681         int prio;
 682
 683         for (prio = 1; prio < 4; prio++) {
 684                 int h;
 685                 for (h = 0; h < HASH_SIZE; h++) {
 686                         struct ip_tunnel *t;
 687
 688                         t = rtnl_dereference(ipn->tunnels[prio][h]);
 689                         while (t != NULL) {
 690                                 unregister_netdevice_queue(t->dev, head);
 691                                 t = rtnl_dereference(t->next);
 692                         }
 693                 }
 694         }
 695 }
 696
 697 static int __net_init vti_init_net(struct net *net)
 698 {
 699         int err;
 700         struct vti_net *ipn = net_generic(net, vti_net_id);
 701
 702         ipn->tunnels[0] = ipn->tunnels_wc;
 703         ipn->tunnels[1] = ipn->tunnels_l;
 704         ipn->tunnels[2] = ipn->tunnels_r;
 705         ipn->tunnels[3] = ipn->tunnels_r_l;
 706
 707         ipn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel),
 708                                           "ip_vti0",
 709                                           vti_tunnel_setup);
 710         if (!ipn->fb_tunnel_dev) {
 711                 err = -ENOMEM;
 712                 goto err_alloc_dev;
 713         }
 714         dev_net_set(ipn->fb_tunnel_dev, net);
 715
 716         err = vti_fb_tunnel_init(ipn->fb_tunnel_dev);
 717         if (err)
 718                 goto err_reg_dev;
 719         ipn->fb_tunnel_dev->rtnl_link_ops = &vti_link_ops;
 720
 721         err = register_netdev(ipn->fb_tunnel_dev);
 722         if (err)
 723                 goto err_reg_dev;
 724         return 0;
 725
 726 err_reg_dev:
 727         vti_dev_free(ipn->fb_tunnel_dev);
 728 err_alloc_dev:
 729         /* nothing */
 730         return err;
 731 }
 732
 733 static void __net_exit vti_exit_net(struct net *net)
 734 {
 735         struct vti_net *ipn = net_generic(net, vti_net_id);
 736         LIST_HEAD(list);
 737
 738         rtnl_lock();
 739         vti_destroy_tunnels(ipn, &list);
 740         unregister_netdevice_many(&list);
 741         rtnl_unlock();
 742 }
 743
 744 static struct pernet_operations vti_net_ops = {
 745         .init = vti_init_net,
 746         .exit = vti_exit_net,
 747         .id   = &vti_net_id,
 748         .size = sizeof(struct vti_net),
 749 };
 750
 751 static int vti_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
 752 {
 753         return 0;
 754 }
 755
 756 static void vti_netlink_parms(struct nlattr *data[],
 757                               struct ip_tunnel_parm *parms)
 758 {
 759         memset(parms, 0, sizeof(*parms));
 760
 761         parms->iph.protocol = IPPROTO_IPIP;
 762
 763         if (!data)
 764                 return;
 765
 766         if (data[IFLA_VTI_LINK])
 767                 parms->link = nla_get_u32(data[IFLA_VTI_LINK]);
 768
 769         if (data[IFLA_VTI_IKEY])
 770                 parms->i_key = nla_get_be32(data[IFLA_VTI_IKEY]);
 771
 772         if (data[IFLA_VTI_OKEY])
 773                 parms->o_key = nla_get_be32(data[IFLA_VTI_OKEY]);
 774
 775         if (data[IFLA_VTI_LOCAL])
 776                 parms->iph.saddr = nla_get_be32(data[IFLA_VTI_LOCAL]);
 777
 778         if (data[IFLA_VTI_REMOTE])
 779                 parms->iph.daddr = nla_get_be32(data[IFLA_VTI_REMOTE]);
 780
 781 }
 782
 783 static int vti_newlink(struct net *src_net, struct net_device *dev,
 784                        struct nlattr *tb[], struct nlattr *data[])
 785 {
 786         struct ip_tunnel *nt;
 787         struct net *net = dev_net(dev);
 788         struct vti_net *ipn = net_generic(net, vti_net_id);
 789         int mtu;
 790         int err;
 791
 792         nt = netdev_priv(dev);
 793         vti_netlink_parms(data, &nt->parms);
 794
 795         if (vti_tunnel_locate(net, &nt->parms, 0))
 796                 return -EEXIST;
 797
 798         mtu = vti_tunnel_bind_dev(dev);
 799         if (!tb[IFLA_MTU])
 800                 dev->mtu = mtu;
 801
 802         err = register_netdevice(dev);
 803         if (err)
 804                 goto out;
 805
 806         dev_hold(dev);
 807         vti_tunnel_link(ipn, nt);
 808
 809 out:
 810         return err;
 811 }
 812
 813 static int vti_changelink(struct net_device *dev, struct nlattr *tb[],
 814                           struct nlattr *data[])
 815 {
 816         struct ip_tunnel *t, *nt;
 817         struct net *net = dev_net(dev);
 818         struct vti_net *ipn = net_generic(net, vti_net_id);
 819         struct ip_tunnel_parm p;
 820         int mtu;
 821
 822         if (dev == ipn->fb_tunnel_dev)
 823                 return -EINVAL;
 824
 825         nt = netdev_priv(dev);
 826         vti_netlink_parms(data, &p);
 827
 828         t = vti_tunnel_locate(net, &p, 0);
 829
 830         if (t) {
 831                 if (t->dev != dev)
 832                         return -EEXIST;
 833         } else {
 834                 t = nt;
 835
 836                 vti_tunnel_unlink(ipn, t);
 837                 t->parms.iph.saddr = p.iph.saddr;
 838                 t->parms.iph.daddr = p.iph.daddr;
 839                 t->parms.i_key = p.i_key;
 840                 t->parms.o_key = p.o_key;
 841                 if (dev->type != ARPHRD_ETHER) {
 842                         memcpy(dev->dev_addr, &p.iph.saddr, 4);
 843                         memcpy(dev->broadcast, &p.iph.daddr, 4);
 844                 }
 845                 vti_tunnel_link(ipn, t);
 846                 netdev_state_change(dev);
 847         }
 848
 849         if (t->parms.link != p.link) {
 850                 t->parms.link = p.link;
 851                 mtu = vti_tunnel_bind_dev(dev);
 852                 if (!tb[IFLA_MTU])
 853                         dev->mtu = mtu;
 854                 netdev_state_change(dev);
 855         }
 856
 857         return 0;
 858 }
 859
 860 static size_t vti_get_size(const struct net_device *dev)
 861 {
 862         return
 863                 /* IFLA_VTI_LINK */
 864                 nla_total_size(4) +
 865                 /* IFLA_VTI_IKEY */
 866                 nla_total_size(4) +
 867                 /* IFLA_VTI_OKEY */
 868                 nla_total_size(4) +
 869                 /* IFLA_VTI_LOCAL */
 870                 nla_total_size(4) +
 871                 /* IFLA_VTI_REMOTE */
 872                 nla_total_size(4) +
 873                 0;
 874 }
 875
 876 static int vti_fill_info(struct sk_buff *skb, const struct net_device *dev)
 877 {
 878         struct ip_tunnel *t = netdev_priv(dev);
 879         struct ip_tunnel_parm *p = &t->parms;
 880
 881         nla_put_u32(skb, IFLA_VTI_LINK, p->link);
 882         nla_put_be32(skb, IFLA_VTI_IKEY, p->i_key);
 883         nla_put_be32(skb, IFLA_VTI_OKEY, p->o_key);
 884         nla_put_be32(skb, IFLA_VTI_LOCAL, p->iph.saddr);
 885         nla_put_be32(skb, IFLA_VTI_REMOTE, p->iph.daddr);
 886
 887         return 0;
 888 }
 889
 890 static const struct nla_policy vti_policy[IFLA_VTI_MAX + 1] = {
 891         [IFLA_VTI_LINK]         = { .type = NLA_U32 },
 892         [IFLA_VTI_IKEY]         = { .type = NLA_U32 },
 893         [IFLA_VTI_OKEY]         = { .type = NLA_U32 },
 894         [IFLA_VTI_LOCAL]        = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
 895         [IFLA_VTI_REMOTE]       = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
 896 };
 897
 898 static struct rtnl_link_ops vti_link_ops __read_mostly = {
 899         .kind           = "vti",
 900         .maxtype        = IFLA_VTI_MAX,
 901         .policy         = vti_policy,
 902         .priv_size      = sizeof(struct ip_tunnel),
 903         .setup          = vti_tunnel_setup,
 904         .validate       = vti_tunnel_validate,
 905         .newlink        = vti_newlink,
 906         .changelink     = vti_changelink,
 907         .get_size       = vti_get_size,
 908         .fill_info      = vti_fill_info,
 909 };
 910
 911 static int __init vti_init(void)
 912 {
 913         int err;
 914
 915         pr_info("IPv4 over IPSec tunneling driver\n");
 916
 917         err = register_pernet_device(&vti_net_ops);
 918         if (err < 0)
 919                 return err;
 920         err = xfrm4_mode_tunnel_input_register(&vti_handler);
 921         if (err < 0) {
 922                 unregister_pernet_device(&vti_net_ops);
 923                 pr_info(KERN_INFO "vti init: can't register tunnel\n");
 924         }
 925
 926         err = rtnl_link_register(&vti_link_ops);
 927         if (err < 0)
 928                 goto rtnl_link_failed;
 929
 930         return err;
 931
 932 rtnl_link_failed:
 933         xfrm4_mode_tunnel_input_deregister(&vti_handler);
 934         unregister_pernet_device(&vti_net_ops);
 935         return err;
 936 }
 937
 938 static void __exit vti_fini(void)
 939 {
 940         rtnl_link_unregister(&vti_link_ops);
 941         if (xfrm4_mode_tunnel_input_deregister(&vti_handler))
 942                 pr_info("vti close: can't deregister tunnel\n");
 943
 944         unregister_pernet_device(&vti_net_ops);
 945 }
 946
 947 module_init(vti_init);
 948 module_exit(vti_fini);
 949 MODULE_LICENSE("GPL");
 950 MODULE_ALIAS_RTNL_LINK("vti");
 951 MODULE_ALIAS_NETDEV("ip_vti0");