net/ipv6/sit.c

   1 /*
   2  *      IPv6 over IPv4 tunnel device - Simple Internet Transition (SIT)
   3  *      Linux INET6 implementation
   4  *
   5  *      Authors:
   6  *      Pedro Roque             <roque@di.fc.ul.pt>
   7  *      Alexey Kuznetsov        <kuznet@ms2.inr.ac.ru>
   8  *
   9  *      This program is free software; you can redistribute it and/or
  10  *      modify it under the terms of the GNU General Public License
  11  *      as published by the Free Software Foundation; either version
  12  *      2 of the License, or (at your option) any later version.
  13  *
  14  *      Changes:
  15  * Roger Venning <r.venning@telstra.com>:       6to4 support
  16  * Nate Thompson <nate@thebog.net>:             6to4 support
  17  * Fred Templin <fred.l.templin@boeing.com>:    isatap support
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/capability.h>
  22 #include <linux/errno.h>
  23 #include <linux/types.h>
  24 #include <linux/socket.h>
  25 #include <linux/sockios.h>
  26 #include <linux/net.h>
  27 #include <linux/in6.h>
  28 #include <linux/netdevice.h>
  29 #include <linux/if_arp.h>
  30 #include <linux/icmp.h>
  31 #include <linux/slab.h>
  32 #include <asm/uaccess.h>
  33 #include <linux/init.h>
  34 #include <linux/netfilter_ipv4.h>
  35 #include <linux/if_ether.h>
  36
  37 #include <net/sock.h>
  38 #include <net/snmp.h>
  39
  40 #include <net/ipv6.h>
  41 #include <net/protocol.h>
  42 #include <net/transp_v6.h>
  43 #include <net/ip6_fib.h>
  44 #include <net/ip6_route.h>
  45 #include <net/ndisc.h>
  46 #include <net/addrconf.h>
  47 #include <net/ip.h>
  48 #include <net/udp.h>
  49 #include <net/icmp.h>
  50 #include <net/ipip.h>
  51 #include <net/inet_ecn.h>
  52 #include <net/xfrm.h>
  53 #include <net/dsfield.h>
  54 #include <net/net_namespace.h>
  55 #include <net/netns/generic.h>
  56
  57 /*
  58    This version of net/ipv6/sit.c is cloned of net/ipv4/ip_gre.c
  59
  60    For comments look at net/ipv4/ip_gre.c --ANK
  61  */
  62
  63 #define HASH_SIZE  16
  64 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
  65
  66 static void ipip6_tunnel_init(struct net_device *dev);
  67 static void ipip6_tunnel_setup(struct net_device *dev);
  68
  69 static int sit_net_id __read_mostly;
  70 struct sit_net {
  71         struct ip_tunnel __rcu *tunnels_r_l[HASH_SIZE];
  72         struct ip_tunnel __rcu *tunnels_r[HASH_SIZE];
  73         struct ip_tunnel __rcu *tunnels_l[HASH_SIZE];
  74         struct ip_tunnel __rcu *tunnels_wc[1];
  75         struct ip_tunnel __rcu **tunnels[4];
  76
  77         struct net_device *fb_tunnel_dev;
  78 };
  79
  80 /*
  81  * Locking : hash tables are protected by RCU and RTNL
  82  */
  83
  84 #define for_each_ip_tunnel_rcu(start) \
  85         for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
  86
  87 /*
  88  * Must be invoked with rcu_read_lock
  89  */
  90 static struct ip_tunnel * ipip6_tunnel_lookup(struct net *net,
  91                 struct net_device *dev, __be32 remote, __be32 local)
  92 {
  93         unsigned int h0 = HASH(remote);
  94         unsigned int h1 = HASH(local);
  95         struct ip_tunnel *t;
  96         struct sit_net *sitn = net_generic(net, sit_net_id);
  97
  98         for_each_ip_tunnel_rcu(sitn->tunnels_r_l[h0 ^ h1]) {
  99                 if (local == t->parms.iph.saddr &&
 100                     remote == t->parms.iph.daddr &&
 101                     (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
 102                     (t->dev->flags & IFF_UP))
 103                         return t;
 104         }
 105         for_each_ip_tunnel_rcu(sitn->tunnels_r[h0]) {
 106                 if (remote == t->parms.iph.daddr &&
 107                     (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
 108                     (t->dev->flags & IFF_UP))
 109                         return t;
 110         }
 111         for_each_ip_tunnel_rcu(sitn->tunnels_l[h1]) {
 112                 if (local == t->parms.iph.saddr &&
 113                     (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
 114                     (t->dev->flags & IFF_UP))
 115                         return t;
 116         }
 117         t = rcu_dereference(sitn->tunnels_wc[0]);
 118         if ((t != NULL) && (t->dev->flags & IFF_UP))
 119                 return t;
 120         return NULL;
 121 }
 122
 123 static struct ip_tunnel __rcu **__ipip6_bucket(struct sit_net *sitn,
 124                 struct ip_tunnel_parm *parms)
 125 {
 126         __be32 remote = parms->iph.daddr;
 127         __be32 local = parms->iph.saddr;
 128         unsigned int h = 0;
 129         int prio = 0;
 130
 131         if (remote) {
 132                 prio |= 2;
 133                 h ^= HASH(remote);
 134         }
 135         if (local) {
 136                 prio |= 1;
 137                 h ^= HASH(local);
 138         }
 139         return &sitn->tunnels[prio][h];
 140 }
 141
 142 static inline struct ip_tunnel __rcu **ipip6_bucket(struct sit_net *sitn,
 143                 struct ip_tunnel *t)
 144 {
 145         return __ipip6_bucket(sitn, &t->parms);
 146 }
 147
 148 static void ipip6_tunnel_unlink(struct sit_net *sitn, struct ip_tunnel *t)
 149 {
 150         struct ip_tunnel __rcu **tp;
 151         struct ip_tunnel *iter;
 152
 153         for (tp = ipip6_bucket(sitn, t);
 154              (iter = rtnl_dereference(*tp)) != NULL;
 155              tp = &iter->next) {
 156                 if (t == iter) {
 157                         rcu_assign_pointer(*tp, t->next);
 158                         break;
 159                 }
 160         }
 161 }
 162
 163 static void ipip6_tunnel_link(struct sit_net *sitn, struct ip_tunnel *t)
 164 {
 165         struct ip_tunnel __rcu **tp = ipip6_bucket(sitn, t);
 166
 167         rcu_assign_pointer(t->next, rtnl_dereference(*tp));
 168         rcu_assign_pointer(*tp, t);
 169 }
 170
 171 static void ipip6_tunnel_clone_6rd(struct net_device *dev, struct sit_net *sitn)
 172 {
 173 #ifdef CONFIG_IPV6_SIT_6RD
 174         struct ip_tunnel *t = netdev_priv(dev);
 175
 176         if (t->dev == sitn->fb_tunnel_dev) {
 177                 ipv6_addr_set(&t->ip6rd.prefix, htonl(0x20020000), 0, 0, 0);
 178                 t->ip6rd.relay_prefix = 0;
 179                 t->ip6rd.prefixlen = 16;
 180                 t->ip6rd.relay_prefixlen = 0;
 181         } else {
 182                 struct ip_tunnel *t0 = netdev_priv(sitn->fb_tunnel_dev);
 183                 memcpy(&t->ip6rd, &t0->ip6rd, sizeof(t->ip6rd));
 184         }
 185 #endif
 186 }
 187
 188 static struct ip_tunnel *ipip6_tunnel_locate(struct net *net,
 189                 struct ip_tunnel_parm *parms, int create)
 190 {
 191         __be32 remote = parms->iph.daddr;
 192         __be32 local = parms->iph.saddr;
 193         struct ip_tunnel *t, *nt;
 194         struct ip_tunnel __rcu **tp;
 195         struct net_device *dev;
 196         char name[IFNAMSIZ];
 197         struct sit_net *sitn = net_generic(net, sit_net_id);
 198
 199         for (tp = __ipip6_bucket(sitn, parms);
 200             (t = rtnl_dereference(*tp)) != NULL;
 201              tp = &t->next) {
 202                 if (local == t->parms.iph.saddr &&
 203                     remote == t->parms.iph.daddr &&
 204                     parms->link == t->parms.link) {
 205                         if (create)
 206                                 return NULL;
 207                         else
 208                                 return t;
 209                 }
 210         }
 211         if (!create)
 212                 goto failed;
 213
 214         if (parms->name[0])
 215                 strlcpy(name, parms->name, IFNAMSIZ);
 216         else
 217                 sprintf(name, "sit%%d");
 218
 219         dev = alloc_netdev(sizeof(*t), name, ipip6_tunnel_setup);
 220         if (dev == NULL)
 221                 return NULL;
 222
 223         dev_net_set(dev, net);
 224
 225         if (strchr(name, '%')) {
 226                 if (dev_alloc_name(dev, name) < 0)
 227                         goto failed_free;
 228         }
 229
 230         nt = netdev_priv(dev);
 231
 232         nt->parms = *parms;
 233         ipip6_tunnel_init(dev);
 234         ipip6_tunnel_clone_6rd(dev, sitn);
 235
 236         if (parms->i_flags & SIT_ISATAP)
 237                 dev->priv_flags |= IFF_ISATAP;
 238
 239         if (register_netdevice(dev) < 0)
 240                 goto failed_free;
 241
 242         dev_hold(dev);
 243
 244         ipip6_tunnel_link(sitn, nt);
 245         return nt;
 246
 247 failed_free:
 248         free_netdev(dev);
 249 failed:
 250         return NULL;
 251 }
 252
 253 #define for_each_prl_rcu(start)                 \
 254         for (prl = rcu_dereference(start);      \
 255              prl;                               \
 256              prl = rcu_dereference(prl->next))
 257
 258 static struct ip_tunnel_prl_entry *
 259 __ipip6_tunnel_locate_prl(struct ip_tunnel *t, __be32 addr)
 260 {
 261         struct ip_tunnel_prl_entry *prl;
 262
 263         for_each_prl_rcu(t->prl)
 264                 if (prl->addr == addr)
 265                         break;
 266         return prl;
 267
 268 }
 269
 270 static int ipip6_tunnel_get_prl(struct ip_tunnel *t,
 271                                 struct ip_tunnel_prl __user *a)
 272 {
 273         struct ip_tunnel_prl kprl, *kp;
 274         struct ip_tunnel_prl_entry *prl;
 275         unsigned int cmax, c = 0, ca, len;
 276         int ret = 0;
 277
 278         if (copy_from_user(&kprl, a, sizeof(kprl)))
 279                 return -EFAULT;
 280         cmax = kprl.datalen / sizeof(kprl);
 281         if (cmax > 1 && kprl.addr != htonl(INADDR_ANY))
 282                 cmax = 1;
 283
 284         /* For simple GET or for root users,
 285          * we try harder to allocate.
 286          */
 287         kp = (cmax <= 1 || capable(CAP_NET_ADMIN)) ?
 288                 kcalloc(cmax, sizeof(*kp), GFP_KERNEL) :
 289                 NULL;
 290
 291         rcu_read_lock();
 292
 293         ca = t->prl_count < cmax ? t->prl_count : cmax;
 294
 295         if (!kp) {
 296                 /* We don't try hard to allocate much memory for
 297                  * non-root users.
 298                  * For root users, retry allocating enough memory for
 299                  * the answer.
 300                  */
 301                 kp = kcalloc(ca, sizeof(*kp), GFP_ATOMIC);
 302                 if (!kp) {
 303                         ret = -ENOMEM;
 304                         goto out;
 305                 }
 306         }
 307
 308         c = 0;
 309         for_each_prl_rcu(t->prl) {
 310                 if (c >= cmax)
 311                         break;
 312                 if (kprl.addr != htonl(INADDR_ANY) && prl->addr != kprl.addr)
 313                         continue;
 314                 kp[c].addr = prl->addr;
 315                 kp[c].flags = prl->flags;
 316                 c++;
 317                 if (kprl.addr != htonl(INADDR_ANY))
 318                         break;
 319         }
 320 out:
 321         rcu_read_unlock();
 322
 323         len = sizeof(*kp) * c;
 324         ret = 0;
 325         if ((len && copy_to_user(a + 1, kp, len)) || put_user(len, &a->datalen))
 326                 ret = -EFAULT;
 327
 328         kfree(kp);
 329
 330         return ret;
 331 }
 332
 333 static int
 334 ipip6_tunnel_add_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a, int chg)
 335 {
 336         struct ip_tunnel_prl_entry *p;
 337         int err = 0;
 338
 339         if (a->addr == htonl(INADDR_ANY))
 340                 return -EINVAL;
 341
 342         ASSERT_RTNL();
 343
 344         for (p = rtnl_dereference(t->prl); p; p = rtnl_dereference(p->next)) {
 345                 if (p->addr == a->addr) {
 346                         if (chg) {
 347                                 p->flags = a->flags;
 348                                 goto out;
 349                         }
 350                         err = -EEXIST;
 351                         goto out;
 352                 }
 353         }
 354
 355         if (chg) {
 356                 err = -ENXIO;
 357                 goto out;
 358         }
 359
 360         p = kzalloc(sizeof(struct ip_tunnel_prl_entry), GFP_KERNEL);
 361         if (!p) {
 362                 err = -ENOBUFS;
 363                 goto out;
 364         }
 365
 366         p->next = t->prl;
 367         p->addr = a->addr;
 368         p->flags = a->flags;
 369         t->prl_count++;
 370         rcu_assign_pointer(t->prl, p);
 371 out:
 372         return err;
 373 }
 374
 375 static void prl_entry_destroy_rcu(struct rcu_head *head)
 376 {
 377         kfree(container_of(head, struct ip_tunnel_prl_entry, rcu_head));
 378 }
 379
 380 static void prl_list_destroy_rcu(struct rcu_head *head)
 381 {
 382         struct ip_tunnel_prl_entry *p, *n;
 383
 384         p = container_of(head, struct ip_tunnel_prl_entry, rcu_head);
 385         do {
 386                 n = p->next;
 387                 kfree(p);
 388                 p = n;
 389         } while (p);
 390 }
 391
 392 static int
 393 ipip6_tunnel_del_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a)
 394 {
 395         struct ip_tunnel_prl_entry *x, **p;
 396         int err = 0;
 397
 398         ASSERT_RTNL();
 399
 400         if (a && a->addr != htonl(INADDR_ANY)) {
 401                 for (p = &t->prl; *p; p = &(*p)->next) {
 402                         if ((*p)->addr == a->addr) {
 403                                 x = *p;
 404                                 *p = x->next;
 405                                 call_rcu(&x->rcu_head, prl_entry_destroy_rcu);
 406                                 t->prl_count--;
 407                                 goto out;
 408                         }
 409                 }
 410                 err = -ENXIO;
 411         } else {
 412                 if (t->prl) {
 413                         t->prl_count = 0;
 414                         x = t->prl;
 415                         call_rcu(&x->rcu_head, prl_list_destroy_rcu);
 416                         t->prl = NULL;
 417                 }
 418         }
 419 out:
 420         return err;
 421 }
 422
 423 static int
 424 isatap_chksrc(struct sk_buff *skb, struct iphdr *iph, struct ip_tunnel *t)
 425 {
 426         struct ip_tunnel_prl_entry *p;
 427         int ok = 1;
 428
 429         rcu_read_lock();
 430         p = __ipip6_tunnel_locate_prl(t, iph->saddr);
 431         if (p) {
 432                 if (p->flags & PRL_DEFAULT)
 433                         skb->ndisc_nodetype = NDISC_NODETYPE_DEFAULT;
 434                 else
 435                         skb->ndisc_nodetype = NDISC_NODETYPE_NODEFAULT;
 436         } else {
 437                 struct in6_addr *addr6 = &ipv6_hdr(skb)->saddr;
 438                 if (ipv6_addr_is_isatap(addr6) &&
 439                     (addr6->s6_addr32[3] == iph->saddr) &&
 440                     ipv6_chk_prefix(addr6, t->dev))
 441                         skb->ndisc_nodetype = NDISC_NODETYPE_HOST;
 442                 else
 443                         ok = 0;
 444         }
 445         rcu_read_unlock();
 446         return ok;
 447 }
 448
 449 static void ipip6_tunnel_uninit(struct net_device *dev)
 450 {
 451         struct net *net = dev_net(dev);
 452         struct sit_net *sitn = net_generic(net, sit_net_id);
 453
 454         if (dev == sitn->fb_tunnel_dev) {
 455                 rcu_assign_pointer(sitn->tunnels_wc[0], NULL);
 456         } else {
 457                 ipip6_tunnel_unlink(sitn, netdev_priv(dev));
 458                 ipip6_tunnel_del_prl(netdev_priv(dev), NULL);
 459         }
 460         dev_put(dev);
 461 }
 462
 463
 464 static int ipip6_err(struct sk_buff *skb, u32 info)
 465 {
 466
 467 /* All the routers (except for Linux) return only
 468    8 bytes of packet payload. It means, that precise relaying of
 469    ICMP in the real Internet is absolutely infeasible.
 470  */
 471         struct iphdr *iph = (struct iphdr*)skb->data;
 472         const int type = icmp_hdr(skb)->type;
 473         const int code = icmp_hdr(skb)->code;
 474         struct ip_tunnel *t;
 475         int err;
 476
 477         switch (type) {
 478         default:
 479         case ICMP_PARAMETERPROB:
 480                 return 0;
 481
 482         case ICMP_DEST_UNREACH:
 483                 switch (code) {
 484                 case ICMP_SR_FAILED:
 485                 case ICMP_PORT_UNREACH:
 486                         /* Impossible event. */
 487                         return 0;
 488                 case ICMP_FRAG_NEEDED:
 489                         /* Soft state for pmtu is maintained by IP core. */
 490                         return 0;
 491                 default:
 492                         /* All others are translated to HOST_UNREACH.
 493                            rfc2003 contains "deep thoughts" about NET_UNREACH,
 494                            I believe they are just ether pollution. --ANK
 495                          */
 496                         break;
 497                 }
 498                 break;
 499         case ICMP_TIME_EXCEEDED:
 500                 if (code != ICMP_EXC_TTL)
 501                         return 0;
 502                 break;
 503         }
 504
 505         err = -ENOENT;
 506
 507         rcu_read_lock();
 508         t = ipip6_tunnel_lookup(dev_net(skb->dev),
 509                                 skb->dev,
 510                                 iph->daddr,
 511                                 iph->saddr);
 512         if (t == NULL || t->parms.iph.daddr == 0)
 513                 goto out;
 514
 515         err = 0;
 516         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
 517                 goto out;
 518
 519         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
 520                 t->err_count++;
 521         else
 522                 t->err_count = 1;
 523         t->err_time = jiffies;
 524 out:
 525         rcu_read_unlock();
 526         return err;
 527 }
 528
 529 static inline void ipip6_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb)
 530 {
 531         if (INET_ECN_is_ce(iph->tos))
 532                 IP6_ECN_set_ce(ipv6_hdr(skb));
 533 }
 534
 535 static int ipip6_rcv(struct sk_buff *skb)
 536 {
 537         struct iphdr *iph;
 538         struct ip_tunnel *tunnel;
 539
 540         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 541                 goto out;
 542
 543         iph = ip_hdr(skb);
 544
 545         rcu_read_lock();
 546         tunnel = ipip6_tunnel_lookup(dev_net(skb->dev), skb->dev,
 547                                      iph->saddr, iph->daddr);
 548         if (tunnel != NULL) {
 549                 secpath_reset(skb);
 550                 skb->mac_header = skb->network_header;
 551                 skb_reset_network_header(skb);
 552                 IPCB(skb)->flags = 0;
 553                 skb->protocol = htons(ETH_P_IPV6);
 554                 skb->pkt_type = PACKET_HOST;
 555
 556                 if ((tunnel->dev->priv_flags & IFF_ISATAP) &&
 557                     !isatap_chksrc(skb, iph, tunnel)) {
 558                         tunnel->dev->stats.rx_errors++;
 559                         rcu_read_unlock();
 560                         kfree_skb(skb);
 561                         return 0;
 562                 }
 563
 564                 skb_tunnel_rx(skb, tunnel->dev);
 565
 566                 ipip6_ecn_decapsulate(iph, skb);
 567
 568                 if (netif_rx(skb) == NET_RX_DROP)
 569                         tunnel->dev->stats.rx_dropped++;
 570
 571                 rcu_read_unlock();
 572                 return 0;
 573         }
 574
 575         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 576         rcu_read_unlock();
 577 out:
 578         kfree_skb(skb);
 579         return 0;
 580 }
 581
 582 /*
 583  * Returns the embedded IPv4 address if the IPv6 address
 584  * comes from 6rd / 6to4 (RFC 3056) addr space.
 585  */
 586 static inline
 587 __be32 try_6rd(struct in6_addr *v6dst, struct ip_tunnel *tunnel)
 588 {
 589         __be32 dst = 0;
 590
 591 #ifdef CONFIG_IPV6_SIT_6RD
 592         if (ipv6_prefix_equal(v6dst, &tunnel->ip6rd.prefix,
 593                               tunnel->ip6rd.prefixlen)) {
 594                 unsigned int pbw0, pbi0;
 595                 int pbi1;
 596                 u32 d;
 597
 598                 pbw0 = tunnel->ip6rd.prefixlen >> 5;
 599                 pbi0 = tunnel->ip6rd.prefixlen & 0x1f;
 600
 601                 d = (ntohl(v6dst->s6_addr32[pbw0]) << pbi0) >>
 602                     tunnel->ip6rd.relay_prefixlen;
 603
 604                 pbi1 = pbi0 - tunnel->ip6rd.relay_prefixlen;
 605                 if (pbi1 > 0)
 606                         d |= ntohl(v6dst->s6_addr32[pbw0 + 1]) >>
 607                              (32 - pbi1);
 608
 609                 dst = tunnel->ip6rd.relay_prefix | htonl(d);
 610         }
 611 #else
 612         if (v6dst->s6_addr16[0] == htons(0x2002)) {
 613                 /* 6to4 v6 addr has 16 bits prefix, 32 v4addr, 16 SLA, ... */
 614                 memcpy(&dst, &v6dst->s6_addr16[1], 4);
 615         }
 616 #endif
 617         return dst;
 618 }
 619
 620 /*
 621  *      This function assumes it is being called from dev_queue_xmit()
 622  *      and that skb is filled properly by that function.
 623  */
 624
 625 static netdev_tx_t ipip6_tunnel_xmit(struct sk_buff *skb,
 626                                      struct net_device *dev)
 627 {
 628         struct ip_tunnel *tunnel = netdev_priv(dev);
 629         struct net_device_stats *stats = &dev->stats;
 630         struct netdev_queue *txq = netdev_get_tx_queue(dev, 0);
 631         struct iphdr  *tiph = &tunnel->parms.iph;
 632         struct ipv6hdr *iph6 = ipv6_hdr(skb);
 633         u8     tos = tunnel->parms.iph.tos;
 634         __be16 df = tiph->frag_off;
 635         struct rtable *rt;                      /* Route to the other host */
 636         struct net_device *tdev;                        /* Device to other host */
 637         struct iphdr  *iph;                     /* Our new IP header */
 638         unsigned int max_headroom;              /* The extra header space needed */
 639         __be32 dst = tiph->daddr;
 640         int    mtu;
 641         struct in6_addr *addr6;
 642         int addr_type;
 643
 644         if (skb->protocol != htons(ETH_P_IPV6))
 645                 goto tx_error;
 646
 647         /* ISATAP (RFC4214) - must come before 6to4 */
 648         if (dev->priv_flags & IFF_ISATAP) {
 649                 struct neighbour *neigh = NULL;
 650
 651                 if (skb_dst(skb))
 652                         neigh = skb_dst(skb)->neighbour;
 653
 654                 if (neigh == NULL) {
 655                         if (net_ratelimit())
 656                                 printk(KERN_DEBUG "sit: nexthop == NULL\n");
 657                         goto tx_error;
 658                 }
 659
 660                 addr6 = (struct in6_addr*)&neigh->primary_key;
 661                 addr_type = ipv6_addr_type(addr6);
 662
 663                 if ((addr_type & IPV6_ADDR_UNICAST) &&
 664                      ipv6_addr_is_isatap(addr6))
 665                         dst = addr6->s6_addr32[3];
 666                 else
 667                         goto tx_error;
 668         }
 669
 670         if (!dst)
 671                 dst = try_6rd(&iph6->daddr, tunnel);
 672
 673         if (!dst) {
 674                 struct neighbour *neigh = NULL;
 675
 676                 if (skb_dst(skb))
 677                         neigh = skb_dst(skb)->neighbour;
 678
 679                 if (neigh == NULL) {
 680                         if (net_ratelimit())
 681                                 printk(KERN_DEBUG "sit: nexthop == NULL\n");
 682                         goto tx_error;
 683                 }
 684
 685                 addr6 = (struct in6_addr*)&neigh->primary_key;
 686                 addr_type = ipv6_addr_type(addr6);
 687
 688                 if (addr_type == IPV6_ADDR_ANY) {
 689                         addr6 = &ipv6_hdr(skb)->daddr;
 690                         addr_type = ipv6_addr_type(addr6);
 691                 }
 692
 693                 if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
 694                         goto tx_error_icmp;
 695
 696                 dst = addr6->s6_addr32[3];
 697         }
 698
 699         {
 700                 struct flowi fl = { .nl_u = { .ip4_u =
 701                                               { .daddr = dst,
 702                                                 .saddr = tiph->saddr,
 703                                                 .tos = RT_TOS(tos) } },
 704                                     .oif = tunnel->parms.link,
 705                                     .proto = IPPROTO_IPV6 };
 706                 if (ip_route_output_key(dev_net(dev), &rt, &fl)) {
 707                         stats->tx_carrier_errors++;
 708                         goto tx_error_icmp;
 709                 }
 710         }
 711         if (rt->rt_type != RTN_UNICAST) {
 712                 ip_rt_put(rt);
 713                 stats->tx_carrier_errors++;
 714                 goto tx_error_icmp;
 715         }
 716         tdev = rt->dst.dev;
 717
 718         if (tdev == dev) {
 719                 ip_rt_put(rt);
 720                 stats->collisions++;
 721                 goto tx_error;
 722         }
 723
 724         if (df) {
 725                 mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
 726
 727                 if (mtu < 68) {
 728                         stats->collisions++;
 729                         ip_rt_put(rt);
 730                         goto tx_error;
 731                 }
 732
 733                 if (mtu < IPV6_MIN_MTU) {
 734                         mtu = IPV6_MIN_MTU;
 735                         df = 0;
 736                 }
 737
 738                 if (tunnel->parms.iph.daddr && skb_dst(skb))
 739                         skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
 740
 741                 if (skb->len > mtu) {
 742                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 743                         ip_rt_put(rt);
 744                         goto tx_error;
 745                 }
 746         }
 747
 748         if (tunnel->err_count > 0) {
 749                 if (time_before(jiffies,
 750                                 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
 751                         tunnel->err_count--;
 752                         dst_link_failure(skb);
 753                 } else
 754                         tunnel->err_count = 0;
 755         }
 756
 757         /*
 758          * Okay, now see if we can stuff it in the buffer as-is.
 759          */
 760         max_headroom = LL_RESERVED_SPACE(tdev)+sizeof(struct iphdr);
 761
 762         if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
 763             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
 764                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
 765                 if (!new_skb) {
 766                         ip_rt_put(rt);
 767                         txq->tx_dropped++;
 768                         dev_kfree_skb(skb);
 769                         return NETDEV_TX_OK;
 770                 }
 771                 if (skb->sk)
 772                         skb_set_owner_w(new_skb, skb->sk);
 773                 dev_kfree_skb(skb);
 774                 skb = new_skb;
 775                 iph6 = ipv6_hdr(skb);
 776         }
 777
 778         skb->transport_header = skb->network_header;
 779         skb_push(skb, sizeof(struct iphdr));
 780         skb_reset_network_header(skb);
 781         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 782         IPCB(skb)->flags = 0;
 783         skb_dst_drop(skb);
 784         skb_dst_set(skb, &rt->dst);
 785
 786         /*
 787          *      Push down and install the IPIP header.
 788          */
 789
 790         iph                     =       ip_hdr(skb);
 791         iph->version            =       4;
 792         iph->ihl                =       sizeof(struct iphdr)>>2;
 793         iph->frag_off           =       df;
 794         iph->protocol           =       IPPROTO_IPV6;
 795         iph->tos                =       INET_ECN_encapsulate(tos, ipv6_get_dsfield(iph6));
 796         iph->daddr              =       rt->rt_dst;
 797         iph->saddr              =       rt->rt_src;
 798
 799         if ((iph->ttl = tiph->ttl) == 0)
 800                 iph->ttl        =       iph6->hop_limit;
 801
 802         nf_reset(skb);
 803
 804         IPTUNNEL_XMIT();
 805         return NETDEV_TX_OK;
 806
 807 tx_error_icmp:
 808         dst_link_failure(skb);
 809 tx_error:
 810         stats->tx_errors++;
 811         dev_kfree_skb(skb);
 812         return NETDEV_TX_OK;
 813 }
 814
 815 static void ipip6_tunnel_bind_dev(struct net_device *dev)
 816 {
 817         struct net_device *tdev = NULL;
 818         struct ip_tunnel *tunnel;
 819         struct iphdr *iph;
 820
 821         tunnel = netdev_priv(dev);
 822         iph = &tunnel->parms.iph;
 823
 824         if (iph->daddr) {
 825                 struct flowi fl = { .nl_u = { .ip4_u =
 826                                               { .daddr = iph->daddr,
 827                                                 .saddr = iph->saddr,
 828                                                 .tos = RT_TOS(iph->tos) } },
 829                                     .oif = tunnel->parms.link,
 830                                     .proto = IPPROTO_IPV6 };
 831                 struct rtable *rt;
 832                 if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
 833                         tdev = rt->dst.dev;
 834                         ip_rt_put(rt);
 835                 }
 836                 dev->flags |= IFF_POINTOPOINT;
 837         }
 838
 839         if (!tdev && tunnel->parms.link)
 840                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
 841
 842         if (tdev) {
 843                 dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
 844                 dev->mtu = tdev->mtu - sizeof(struct iphdr);
 845                 if (dev->mtu < IPV6_MIN_MTU)
 846                         dev->mtu = IPV6_MIN_MTU;
 847         }
 848         dev->iflink = tunnel->parms.link;
 849 }
 850
 851 static int
 852 ipip6_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
 853 {
 854         int err = 0;
 855         struct ip_tunnel_parm p;
 856         struct ip_tunnel_prl prl;
 857         struct ip_tunnel *t;
 858         struct net *net = dev_net(dev);
 859         struct sit_net *sitn = net_generic(net, sit_net_id);
 860 #ifdef CONFIG_IPV6_SIT_6RD
 861         struct ip_tunnel_6rd ip6rd;
 862 #endif
 863
 864         switch (cmd) {
 865         case SIOCGETTUNNEL:
 866 #ifdef CONFIG_IPV6_SIT_6RD
 867         case SIOCGET6RD:
 868 #endif
 869                 t = NULL;
 870                 if (dev == sitn->fb_tunnel_dev) {
 871                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
 872                                 err = -EFAULT;
 873                                 break;
 874                         }
 875                         t = ipip6_tunnel_locate(net, &p, 0);
 876                 }
 877                 if (t == NULL)
 878                         t = netdev_priv(dev);
 879
 880                 err = -EFAULT;
 881                 if (cmd == SIOCGETTUNNEL) {
 882                         memcpy(&p, &t->parms, sizeof(p));
 883                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &p,
 884                                          sizeof(p)))
 885                                 goto done;
 886 #ifdef CONFIG_IPV6_SIT_6RD
 887                 } else {
 888                         ipv6_addr_copy(&ip6rd.prefix, &t->ip6rd.prefix);
 889                         ip6rd.relay_prefix = t->ip6rd.relay_prefix;
 890                         ip6rd.prefixlen = t->ip6rd.prefixlen;
 891                         ip6rd.relay_prefixlen = t->ip6rd.relay_prefixlen;
 892                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &ip6rd,
 893                                          sizeof(ip6rd)))
 894                                 goto done;
 895 #endif
 896                 }
 897                 err = 0;
 898                 break;
 899
 900         case SIOCADDTUNNEL:
 901         case SIOCCHGTUNNEL:
 902                 err = -EPERM;
 903                 if (!capable(CAP_NET_ADMIN))
 904                         goto done;
 905
 906                 err = -EFAULT;
 907                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
 908                         goto done;
 909
 910                 err = -EINVAL;
 911                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPV6 ||
 912                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
 913                         goto done;
 914                 if (p.iph.ttl)
 915                         p.iph.frag_off |= htons(IP_DF);
 916
 917                 t = ipip6_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
 918
 919                 if (dev != sitn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
 920                         if (t != NULL) {
 921                                 if (t->dev != dev) {
 922                                         err = -EEXIST;
 923                                         break;
 924                                 }
 925                         } else {
 926                                 if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
 927                                     (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
 928                                         err = -EINVAL;
 929                                         break;
 930                                 }
 931                                 t = netdev_priv(dev);
 932                                 ipip6_tunnel_unlink(sitn, t);
 933                                 t->parms.iph.saddr = p.iph.saddr;
 934                                 t->parms.iph.daddr = p.iph.daddr;
 935                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
 936                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
 937                                 ipip6_tunnel_link(sitn, t);
 938                                 netdev_state_change(dev);
 939                         }
 940                 }
 941
 942                 if (t) {
 943                         err = 0;
 944                         if (cmd == SIOCCHGTUNNEL) {
 945                                 t->parms.iph.ttl = p.iph.ttl;
 946                                 t->parms.iph.tos = p.iph.tos;
 947                                 if (t->parms.link != p.link) {
 948                                         t->parms.link = p.link;
 949                                         ipip6_tunnel_bind_dev(dev);
 950                                         netdev_state_change(dev);
 951                                 }
 952                         }
 953                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
 954                                 err = -EFAULT;
 955                 } else
 956                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
 957                 break;
 958
 959         case SIOCDELTUNNEL:
 960                 err = -EPERM;
 961                 if (!capable(CAP_NET_ADMIN))
 962                         goto done;
 963
 964                 if (dev == sitn->fb_tunnel_dev) {
 965                         err = -EFAULT;
 966                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
 967                                 goto done;
 968                         err = -ENOENT;
 969                         if ((t = ipip6_tunnel_locate(net, &p, 0)) == NULL)
 970                                 goto done;
 971                         err = -EPERM;
 972                         if (t == netdev_priv(sitn->fb_tunnel_dev))
 973                                 goto done;
 974                         dev = t->dev;
 975                 }
 976                 unregister_netdevice(dev);
 977                 err = 0;
 978                 break;
 979
 980         case SIOCGETPRL:
 981                 err = -EINVAL;
 982                 if (dev == sitn->fb_tunnel_dev)
 983                         goto done;
 984                 err = -ENOENT;
 985                 if (!(t = netdev_priv(dev)))
 986                         goto done;
 987                 err = ipip6_tunnel_get_prl(t, ifr->ifr_ifru.ifru_data);
 988                 break;
 989
 990         case SIOCADDPRL:
 991         case SIOCDELPRL:
 992         case SIOCCHGPRL:
 993                 err = -EPERM;
 994                 if (!capable(CAP_NET_ADMIN))
 995                         goto done;
 996                 err = -EINVAL;
 997                 if (dev == sitn->fb_tunnel_dev)
 998                         goto done;
 999                 err = -EFAULT;
1000                 if (copy_from_user(&prl, ifr->ifr_ifru.ifru_data, sizeof(prl)))
1001                         goto done;
1002                 err = -ENOENT;
1003                 if (!(t = netdev_priv(dev)))
1004                         goto done;
1005
1006                 switch (cmd) {
1007                 case SIOCDELPRL:
1008                         err = ipip6_tunnel_del_prl(t, &prl);
1009                         break;
1010                 case SIOCADDPRL:
1011                 case SIOCCHGPRL:
1012                         err = ipip6_tunnel_add_prl(t, &prl, cmd == SIOCCHGPRL);
1013                         break;
1014                 }
1015                 netdev_state_change(dev);
1016                 break;
1017
1018 #ifdef CONFIG_IPV6_SIT_6RD
1019         case SIOCADD6RD:
1020         case SIOCCHG6RD:
1021         case SIOCDEL6RD:
1022                 err = -EPERM;
1023                 if (!capable(CAP_NET_ADMIN))
1024                         goto done;
1025
1026                 err = -EFAULT;
1027                 if (copy_from_user(&ip6rd, ifr->ifr_ifru.ifru_data,
1028                                    sizeof(ip6rd)))
1029                         goto done;
1030
1031                 t = netdev_priv(dev);
1032
1033                 if (cmd != SIOCDEL6RD) {
1034                         struct in6_addr prefix;
1035                         __be32 relay_prefix;
1036
1037                         err = -EINVAL;
1038                         if (ip6rd.relay_prefixlen > 32 ||
1039                             ip6rd.prefixlen + (32 - ip6rd.relay_prefixlen) > 64)
1040                                 goto done;
1041
1042                         ipv6_addr_prefix(&prefix, &ip6rd.prefix,
1043                                          ip6rd.prefixlen);
1044                         if (!ipv6_addr_equal(&prefix, &ip6rd.prefix))
1045                                 goto done;
1046                         if (ip6rd.relay_prefixlen)
1047                                 relay_prefix = ip6rd.relay_prefix &
1048                                                htonl(0xffffffffUL <<
1049                                                      (32 - ip6rd.relay_prefixlen));
1050                         else
1051                                 relay_prefix = 0;
1052                         if (relay_prefix != ip6rd.relay_prefix)
1053                                 goto done;
1054
1055                         ipv6_addr_copy(&t->ip6rd.prefix, &prefix);
1056                         t->ip6rd.relay_prefix = relay_prefix;
1057                         t->ip6rd.prefixlen = ip6rd.prefixlen;
1058                         t->ip6rd.relay_prefixlen = ip6rd.relay_prefixlen;
1059                 } else
1060                         ipip6_tunnel_clone_6rd(dev, sitn);
1061
1062                 err = 0;
1063                 break;
1064 #endif
1065
1066         default:
1067                 err = -EINVAL;
1068         }
1069
1070 done:
1071         return err;
1072 }
1073
1074 static int ipip6_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1075 {
1076         if (new_mtu < IPV6_MIN_MTU || new_mtu > 0xFFF8 - sizeof(struct iphdr))
1077                 return -EINVAL;
1078         dev->mtu = new_mtu;
1079         return 0;
1080 }
1081
1082 static const struct net_device_ops ipip6_netdev_ops = {
1083         .ndo_uninit     = ipip6_tunnel_uninit,
1084         .ndo_start_xmit = ipip6_tunnel_xmit,
1085         .ndo_do_ioctl   = ipip6_tunnel_ioctl,
1086         .ndo_change_mtu = ipip6_tunnel_change_mtu,
1087 };
1088
1089 static void ipip6_tunnel_setup(struct net_device *dev)
1090 {
1091         dev->netdev_ops         = &ipip6_netdev_ops;
1092         dev->destructor         = free_netdev;
1093
1094         dev->type               = ARPHRD_SIT;
1095         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr);
1096         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr);
1097         dev->flags              = IFF_NOARP;
1098         dev->priv_flags        &= ~IFF_XMIT_DST_RELEASE;
1099         dev->iflink             = 0;
1100         dev->addr_len           = 4;
1101         dev->features           |= NETIF_F_NETNS_LOCAL;
1102 }
1103
1104 static void ipip6_tunnel_init(struct net_device *dev)
1105 {
1106         struct ip_tunnel *tunnel = netdev_priv(dev);
1107
1108         tunnel->dev = dev;
1109         strcpy(tunnel->parms.name, dev->name);
1110
1111         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1112         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1113
1114         ipip6_tunnel_bind_dev(dev);
1115 }
1116
1117 static void __net_init ipip6_fb_tunnel_init(struct net_device *dev)
1118 {
1119         struct ip_tunnel *tunnel = netdev_priv(dev);
1120         struct iphdr *iph = &tunnel->parms.iph;
1121         struct net *net = dev_net(dev);
1122         struct sit_net *sitn = net_generic(net, sit_net_id);
1123
1124         tunnel->dev = dev;
1125         strcpy(tunnel->parms.name, dev->name);
1126
1127         iph->version            = 4;
1128         iph->protocol           = IPPROTO_IPV6;
1129         iph->ihl                = 5;
1130         iph->ttl                = 64;
1131
1132         dev_hold(dev);
1133         sitn->tunnels_wc[0]     = tunnel;
1134 }
1135
1136 static struct xfrm_tunnel sit_handler __read_mostly = {
1137         .handler        =       ipip6_rcv,
1138         .err_handler    =       ipip6_err,
1139         .priority       =       1,
1140 };
1141
1142 static void __net_exit sit_destroy_tunnels(struct sit_net *sitn, struct list_head *head)
1143 {
1144         int prio;
1145
1146         for (prio = 1; prio < 4; prio++) {
1147                 int h;
1148                 for (h = 0; h < HASH_SIZE; h++) {
1149                         struct ip_tunnel *t = sitn->tunnels[prio][h];
1150
1151                         while (t != NULL) {
1152                                 unregister_netdevice_queue(t->dev, head);
1153                                 t = t->next;
1154                         }
1155                 }
1156         }
1157 }
1158
1159 static int __net_init sit_init_net(struct net *net)
1160 {
1161         struct sit_net *sitn = net_generic(net, sit_net_id);
1162         int err;
1163
1164         sitn->tunnels[0] = sitn->tunnels_wc;
1165         sitn->tunnels[1] = sitn->tunnels_l;
1166         sitn->tunnels[2] = sitn->tunnels_r;
1167         sitn->tunnels[3] = sitn->tunnels_r_l;
1168
1169         sitn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "sit0",
1170                                            ipip6_tunnel_setup);
1171         if (!sitn->fb_tunnel_dev) {
1172                 err = -ENOMEM;
1173                 goto err_alloc_dev;
1174         }
1175         dev_net_set(sitn->fb_tunnel_dev, net);
1176
1177         ipip6_fb_tunnel_init(sitn->fb_tunnel_dev);
1178         ipip6_tunnel_clone_6rd(sitn->fb_tunnel_dev, sitn);
1179
1180         if ((err = register_netdev(sitn->fb_tunnel_dev)))
1181                 goto err_reg_dev;
1182
1183         return 0;
1184
1185 err_reg_dev:
1186         dev_put(sitn->fb_tunnel_dev);
1187         free_netdev(sitn->fb_tunnel_dev);
1188 err_alloc_dev:
1189         return err;
1190 }
1191
1192 static void __net_exit sit_exit_net(struct net *net)
1193 {
1194         struct sit_net *sitn = net_generic(net, sit_net_id);
1195         LIST_HEAD(list);
1196
1197         rtnl_lock();
1198         sit_destroy_tunnels(sitn, &list);
1199         unregister_netdevice_queue(sitn->fb_tunnel_dev, &list);
1200         unregister_netdevice_many(&list);
1201         rtnl_unlock();
1202 }
1203
1204 static struct pernet_operations sit_net_ops = {
1205         .init = sit_init_net,
1206         .exit = sit_exit_net,
1207         .id   = &sit_net_id,
1208         .size = sizeof(struct sit_net),
1209 };
1210
1211 static void __exit sit_cleanup(void)
1212 {
1213         xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
1214
1215         unregister_pernet_device(&sit_net_ops);
1216         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1217 }
1218
1219 static int __init sit_init(void)
1220 {
1221         int err;
1222
1223         printk(KERN_INFO "IPv6 over IPv4 tunneling driver\n");
1224
1225         err = register_pernet_device(&sit_net_ops);
1226         if (err < 0)
1227                 return err;
1228         err = xfrm4_tunnel_register(&sit_handler, AF_INET6);
1229         if (err < 0) {
1230                 unregister_pernet_device(&sit_net_ops);
1231                 printk(KERN_INFO "sit init: Can't add protocol\n");
1232         }
1233         return err;
1234 }
1235
1236 module_init(sit_init);
1237 module_exit(sit_cleanup);
1238 MODULE_LICENSE("GPL");
1239 MODULE_ALIAS("sit0");