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 <asm/uaccess.h>
  32 #include <linux/init.h>
  33 #include <linux/netfilter_ipv4.h>
  34 #include <linux/if_ether.h>
  35
  36 #include <net/sock.h>
  37 #include <net/snmp.h>
  38
  39 #include <net/ipv6.h>
  40 #include <net/protocol.h>
  41 #include <net/transp_v6.h>
  42 #include <net/ip6_fib.h>
  43 #include <net/ip6_route.h>
  44 #include <net/ndisc.h>
  45 #include <net/addrconf.h>
  46 #include <net/ip.h>
  47 #include <net/udp.h>
  48 #include <net/icmp.h>
  49 #include <net/ipip.h>
  50 #include <net/inet_ecn.h>
  51 #include <net/xfrm.h>
  52 #include <net/dsfield.h>
  53 #include <net/net_namespace.h>
  54 #include <net/netns/generic.h>
  55
  56 /*
  57    This version of net/ipv6/sit.c is cloned of net/ipv4/ip_gre.c
  58
  59    For comments look at net/ipv4/ip_gre.c --ANK
  60  */
  61
  62 #define HASH_SIZE  16
  63 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
  64
  65 static int ipip6_fb_tunnel_init(struct net_device *dev);
  66 static int ipip6_tunnel_init(struct net_device *dev);
  67 static void ipip6_tunnel_setup(struct net_device *dev);
  68
  69 static int sit_net_id;
  70 struct sit_net {
  71         struct ip_tunnel *tunnels_r_l[HASH_SIZE];
  72         struct ip_tunnel *tunnels_r[HASH_SIZE];
  73         struct ip_tunnel *tunnels_l[HASH_SIZE];
  74         struct ip_tunnel *tunnels_wc[1];
  75         struct ip_tunnel **tunnels[4];
  76
  77         struct net_device *fb_tunnel_dev;
  78 };
  79
  80 static DEFINE_RWLOCK(ipip6_lock);
  81
  82 static struct ip_tunnel * ipip6_tunnel_lookup(struct net *net,
  83                 __be32 remote, __be32 local)
  84 {
  85         unsigned h0 = HASH(remote);
  86         unsigned h1 = HASH(local);
  87         struct ip_tunnel *t;
  88         struct sit_net *sitn = net_generic(net, sit_net_id);
  89
  90         for (t = sitn->tunnels_r_l[h0^h1]; t; t = t->next) {
  91                 if (local == t->parms.iph.saddr &&
  92                     remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
  93                         return t;
  94         }
  95         for (t = sitn->tunnels_r[h0]; t; t = t->next) {
  96                 if (remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
  97                         return t;
  98         }
  99         for (t = sitn->tunnels_l[h1]; t; t = t->next) {
 100                 if (local == t->parms.iph.saddr && (t->dev->flags&IFF_UP))
 101                         return t;
 102         }
 103         if ((t = sitn->tunnels_wc[0]) != NULL && (t->dev->flags&IFF_UP))
 104                 return t;
 105         return NULL;
 106 }
 107
 108 static struct ip_tunnel **__ipip6_bucket(struct sit_net *sitn,
 109                 struct ip_tunnel_parm *parms)
 110 {
 111         __be32 remote = parms->iph.daddr;
 112         __be32 local = parms->iph.saddr;
 113         unsigned h = 0;
 114         int prio = 0;
 115
 116         if (remote) {
 117                 prio |= 2;
 118                 h ^= HASH(remote);
 119         }
 120         if (local) {
 121                 prio |= 1;
 122                 h ^= HASH(local);
 123         }
 124         return &sitn->tunnels[prio][h];
 125 }
 126
 127 static inline struct ip_tunnel **ipip6_bucket(struct sit_net *sitn,
 128                 struct ip_tunnel *t)
 129 {
 130         return __ipip6_bucket(sitn, &t->parms);
 131 }
 132
 133 static void ipip6_tunnel_unlink(struct sit_net *sitn, struct ip_tunnel *t)
 134 {
 135         struct ip_tunnel **tp;
 136
 137         for (tp = ipip6_bucket(sitn, t); *tp; tp = &(*tp)->next) {
 138                 if (t == *tp) {
 139                         write_lock_bh(&ipip6_lock);
 140                         *tp = t->next;
 141                         write_unlock_bh(&ipip6_lock);
 142                         break;
 143                 }
 144         }
 145 }
 146
 147 static void ipip6_tunnel_link(struct sit_net *sitn, struct ip_tunnel *t)
 148 {
 149         struct ip_tunnel **tp = ipip6_bucket(sitn, t);
 150
 151         t->next = *tp;
 152         write_lock_bh(&ipip6_lock);
 153         *tp = t;
 154         write_unlock_bh(&ipip6_lock);
 155 }
 156
 157 static struct ip_tunnel * ipip6_tunnel_locate(struct net *net,
 158                 struct ip_tunnel_parm *parms, int create)
 159 {
 160         __be32 remote = parms->iph.daddr;
 161         __be32 local = parms->iph.saddr;
 162         struct ip_tunnel *t, **tp, *nt;
 163         struct net_device *dev;
 164         char name[IFNAMSIZ];
 165         struct sit_net *sitn = net_generic(net, sit_net_id);
 166
 167         for (tp = __ipip6_bucket(sitn, parms); (t = *tp) != NULL; tp = &t->next) {
 168                 if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr)
 169                         return t;
 170         }
 171         if (!create)
 172                 goto failed;
 173
 174         if (parms->name[0])
 175                 strlcpy(name, parms->name, IFNAMSIZ);
 176         else
 177                 sprintf(name, "sit%%d");
 178
 179         dev = alloc_netdev(sizeof(*t), name, ipip6_tunnel_setup);
 180         if (dev == NULL)
 181                 return NULL;
 182
 183         dev_net_set(dev, net);
 184
 185         if (strchr(name, '%')) {
 186                 if (dev_alloc_name(dev, name) < 0)
 187                         goto failed_free;
 188         }
 189
 190         nt = netdev_priv(dev);
 191         dev->init = ipip6_tunnel_init;
 192         nt->parms = *parms;
 193
 194         if (parms->i_flags & SIT_ISATAP)
 195                 dev->priv_flags |= IFF_ISATAP;
 196
 197         if (register_netdevice(dev) < 0)
 198                 goto failed_free;
 199
 200         dev_hold(dev);
 201
 202         ipip6_tunnel_link(sitn, nt);
 203         return nt;
 204
 205 failed_free:
 206         free_netdev(dev);
 207 failed:
 208         return NULL;
 209 }
 210
 211 static struct ip_tunnel_prl_entry *
 212 __ipip6_tunnel_locate_prl(struct ip_tunnel *t, __be32 addr)
 213 {
 214         struct ip_tunnel_prl_entry *p = (struct ip_tunnel_prl_entry *)NULL;
 215
 216         for (p = t->prl; p; p = p->next)
 217                 if (p->addr == addr)
 218                         break;
 219         return p;
 220
 221 }
 222
 223 static int ipip6_tunnel_get_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a)
 224 {
 225         struct ip_tunnel_prl *kp;
 226         struct ip_tunnel_prl_entry *prl;
 227         unsigned int cmax, c = 0, ca, len;
 228         int ret = 0;
 229
 230         cmax = a->datalen / sizeof(*a);
 231         if (cmax > 1 && a->addr != htonl(INADDR_ANY))
 232                 cmax = 1;
 233
 234         /* For simple GET or for root users,
 235          * we try harder to allocate.
 236          */
 237         kp = (cmax <= 1 || capable(CAP_NET_ADMIN)) ?
 238                 kcalloc(cmax, sizeof(*kp), GFP_KERNEL) :
 239                 NULL;
 240
 241         read_lock(&ipip6_lock);
 242
 243         ca = t->prl_count < cmax ? t->prl_count : cmax;
 244
 245         if (!kp) {
 246                 /* We don't try hard to allocate much memory for
 247                  * non-root users.
 248                  * For root users, retry allocating enough memory for
 249                  * the answer.
 250                  */
 251                 kp = kcalloc(ca, sizeof(*kp), GFP_ATOMIC);
 252                 if (!kp) {
 253                         ret = -ENOMEM;
 254                         goto out;
 255                 }
 256         }
 257
 258         c = 0;
 259         for (prl = t->prl; prl; prl = prl->next) {
 260                 if (c > cmax)
 261                         break;
 262                 if (a->addr != htonl(INADDR_ANY) && prl->addr != a->addr)
 263                         continue;
 264                 kp[c].addr = prl->addr;
 265                 kp[c].flags = prl->flags;
 266                 c++;
 267                 if (a->addr != htonl(INADDR_ANY))
 268                         break;
 269         }
 270 out:
 271         read_unlock(&ipip6_lock);
 272
 273         len = sizeof(*kp) * c;
 274         ret = len ? copy_to_user(a->data, kp, len) : 0;
 275
 276         kfree(kp);
 277         if (ret)
 278                 return -EFAULT;
 279
 280         a->datalen = len;
 281         return 0;
 282 }
 283
 284 static int
 285 ipip6_tunnel_add_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a, int chg)
 286 {
 287         struct ip_tunnel_prl_entry *p;
 288         int err = 0;
 289
 290         if (a->addr == htonl(INADDR_ANY))
 291                 return -EINVAL;
 292
 293         write_lock(&ipip6_lock);
 294
 295         for (p = t->prl; p; p = p->next) {
 296                 if (p->addr == a->addr) {
 297                         if (chg)
 298                                 goto update;
 299                         err = -EEXIST;
 300                         goto out;
 301                 }
 302         }
 303
 304         if (chg) {
 305                 err = -ENXIO;
 306                 goto out;
 307         }
 308
 309         p = kzalloc(sizeof(struct ip_tunnel_prl_entry), GFP_KERNEL);
 310         if (!p) {
 311                 err = -ENOBUFS;
 312                 goto out;
 313         }
 314
 315         p->next = t->prl;
 316         t->prl = p;
 317         t->prl_count++;
 318 update:
 319         p->addr = a->addr;
 320         p->flags = a->flags;
 321 out:
 322         write_unlock(&ipip6_lock);
 323         return err;
 324 }
 325
 326 static int
 327 ipip6_tunnel_del_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a)
 328 {
 329         struct ip_tunnel_prl_entry *x, **p;
 330         int err = 0;
 331
 332         write_lock(&ipip6_lock);
 333
 334         if (a && a->addr != htonl(INADDR_ANY)) {
 335                 for (p = &t->prl; *p; p = &(*p)->next) {
 336                         if ((*p)->addr == a->addr) {
 337                                 x = *p;
 338                                 *p = x->next;
 339                                 kfree(x);
 340                                 t->prl_count--;
 341                                 goto out;
 342                         }
 343                 }
 344                 err = -ENXIO;
 345         } else {
 346                 while (t->prl) {
 347                         x = t->prl;
 348                         t->prl = t->prl->next;
 349                         kfree(x);
 350                         t->prl_count--;
 351                 }
 352         }
 353 out:
 354         write_unlock(&ipip6_lock);
 355         return 0;
 356 }
 357
 358 static int
 359 isatap_chksrc(struct sk_buff *skb, struct iphdr *iph, struct ip_tunnel *t)
 360 {
 361         struct ip_tunnel_prl_entry *p;
 362         int ok = 1;
 363
 364         read_lock(&ipip6_lock);
 365         p = __ipip6_tunnel_locate_prl(t, iph->saddr);
 366         if (p) {
 367                 if (p->flags & PRL_DEFAULT)
 368                         skb->ndisc_nodetype = NDISC_NODETYPE_DEFAULT;
 369                 else
 370                         skb->ndisc_nodetype = NDISC_NODETYPE_NODEFAULT;
 371         } else {
 372                 struct in6_addr *addr6 = &ipv6_hdr(skb)->saddr;
 373                 if (ipv6_addr_is_isatap(addr6) &&
 374                     (addr6->s6_addr32[3] == iph->saddr) &&
 375                     ipv6_chk_prefix(addr6, t->dev))
 376                         skb->ndisc_nodetype = NDISC_NODETYPE_HOST;
 377                 else
 378                         ok = 0;
 379         }
 380         read_unlock(&ipip6_lock);
 381         return ok;
 382 }
 383
 384 static void ipip6_tunnel_uninit(struct net_device *dev)
 385 {
 386         struct net *net = dev_net(dev);
 387         struct sit_net *sitn = net_generic(net, sit_net_id);
 388
 389         if (dev == sitn->fb_tunnel_dev) {
 390                 write_lock_bh(&ipip6_lock);
 391                 sitn->tunnels_wc[0] = NULL;
 392                 write_unlock_bh(&ipip6_lock);
 393                 dev_put(dev);
 394         } else {
 395                 ipip6_tunnel_unlink(sitn, netdev_priv(dev));
 396                 ipip6_tunnel_del_prl(netdev_priv(dev), NULL);
 397                 dev_put(dev);
 398         }
 399 }
 400
 401
 402 static int ipip6_err(struct sk_buff *skb, u32 info)
 403 {
 404
 405 /* All the routers (except for Linux) return only
 406    8 bytes of packet payload. It means, that precise relaying of
 407    ICMP in the real Internet is absolutely infeasible.
 408  */
 409         struct iphdr *iph = (struct iphdr*)skb->data;
 410         const int type = icmp_hdr(skb)->type;
 411         const int code = icmp_hdr(skb)->code;
 412         struct ip_tunnel *t;
 413         int err;
 414
 415         switch (type) {
 416         default:
 417         case ICMP_PARAMETERPROB:
 418                 return 0;
 419
 420         case ICMP_DEST_UNREACH:
 421                 switch (code) {
 422                 case ICMP_SR_FAILED:
 423                 case ICMP_PORT_UNREACH:
 424                         /* Impossible event. */
 425                         return 0;
 426                 case ICMP_FRAG_NEEDED:
 427                         /* Soft state for pmtu is maintained by IP core. */
 428                         return 0;
 429                 default:
 430                         /* All others are translated to HOST_UNREACH.
 431                            rfc2003 contains "deep thoughts" about NET_UNREACH,
 432                            I believe they are just ether pollution. --ANK
 433                          */
 434                         break;
 435                 }
 436                 break;
 437         case ICMP_TIME_EXCEEDED:
 438                 if (code != ICMP_EXC_TTL)
 439                         return 0;
 440                 break;
 441         }
 442
 443         err = -ENOENT;
 444
 445         read_lock(&ipip6_lock);
 446         t = ipip6_tunnel_lookup(dev_net(skb->dev), iph->daddr, iph->saddr);
 447         if (t == NULL || t->parms.iph.daddr == 0)
 448                 goto out;
 449
 450         err = 0;
 451         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
 452                 goto out;
 453
 454         if (jiffies - t->err_time < IPTUNNEL_ERR_TIMEO)
 455                 t->err_count++;
 456         else
 457                 t->err_count = 1;
 458         t->err_time = jiffies;
 459 out:
 460         read_unlock(&ipip6_lock);
 461         return err;
 462 }
 463
 464 static inline void ipip6_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb)
 465 {
 466         if (INET_ECN_is_ce(iph->tos))
 467                 IP6_ECN_set_ce(ipv6_hdr(skb));
 468 }
 469
 470 static int ipip6_rcv(struct sk_buff *skb)
 471 {
 472         struct iphdr *iph;
 473         struct ip_tunnel *tunnel;
 474
 475         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 476                 goto out;
 477
 478         iph = ip_hdr(skb);
 479
 480         read_lock(&ipip6_lock);
 481         if ((tunnel = ipip6_tunnel_lookup(dev_net(skb->dev),
 482                                         iph->saddr, iph->daddr)) != NULL) {
 483                 secpath_reset(skb);
 484                 skb->mac_header = skb->network_header;
 485                 skb_reset_network_header(skb);
 486                 IPCB(skb)->flags = 0;
 487                 skb->protocol = htons(ETH_P_IPV6);
 488                 skb->pkt_type = PACKET_HOST;
 489
 490                 if ((tunnel->dev->priv_flags & IFF_ISATAP) &&
 491                     !isatap_chksrc(skb, iph, tunnel)) {
 492                         tunnel->dev->stats.rx_errors++;
 493                         read_unlock(&ipip6_lock);
 494                         kfree_skb(skb);
 495                         return 0;
 496                 }
 497                 tunnel->dev->stats.rx_packets++;
 498                 tunnel->dev->stats.rx_bytes += skb->len;
 499                 skb->dev = tunnel->dev;
 500                 dst_release(skb->dst);
 501                 skb->dst = NULL;
 502                 nf_reset(skb);
 503                 ipip6_ecn_decapsulate(iph, skb);
 504                 netif_rx(skb);
 505                 read_unlock(&ipip6_lock);
 506                 return 0;
 507         }
 508
 509         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 510         read_unlock(&ipip6_lock);
 511 out:
 512         kfree_skb(skb);
 513         return 0;
 514 }
 515
 516 /* Returns the embedded IPv4 address if the IPv6 address
 517    comes from 6to4 (RFC 3056) addr space */
 518
 519 static inline __be32 try_6to4(struct in6_addr *v6dst)
 520 {
 521         __be32 dst = 0;
 522
 523         if (v6dst->s6_addr16[0] == htons(0x2002)) {
 524                 /* 6to4 v6 addr has 16 bits prefix, 32 v4addr, 16 SLA, ... */
 525                 memcpy(&dst, &v6dst->s6_addr16[1], 4);
 526         }
 527         return dst;
 528 }
 529
 530 /*
 531  *      This function assumes it is being called from dev_queue_xmit()
 532  *      and that skb is filled properly by that function.
 533  */
 534
 535 static int ipip6_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
 536 {
 537         struct ip_tunnel *tunnel = netdev_priv(dev);
 538         struct net_device_stats *stats = &tunnel->dev->stats;
 539         struct iphdr  *tiph = &tunnel->parms.iph;
 540         struct ipv6hdr *iph6 = ipv6_hdr(skb);
 541         u8     tos = tunnel->parms.iph.tos;
 542         struct rtable *rt;                      /* Route to the other host */
 543         struct net_device *tdev;                        /* Device to other host */
 544         struct iphdr  *iph;                     /* Our new IP header */
 545         unsigned int max_headroom;              /* The extra header space needed */
 546         __be32 dst = tiph->daddr;
 547         int    mtu;
 548         struct in6_addr *addr6;
 549         int addr_type;
 550
 551         if (tunnel->recursion++) {
 552                 stats->collisions++;
 553                 goto tx_error;
 554         }
 555
 556         if (skb->protocol != htons(ETH_P_IPV6))
 557                 goto tx_error;
 558
 559         /* ISATAP (RFC4214) - must come before 6to4 */
 560         if (dev->priv_flags & IFF_ISATAP) {
 561                 struct neighbour *neigh = NULL;
 562
 563                 if (skb->dst)
 564                         neigh = skb->dst->neighbour;
 565
 566                 if (neigh == NULL) {
 567                         if (net_ratelimit())
 568                                 printk(KERN_DEBUG "sit: nexthop == NULL\n");
 569                         goto tx_error;
 570                 }
 571
 572                 addr6 = (struct in6_addr*)&neigh->primary_key;
 573                 addr_type = ipv6_addr_type(addr6);
 574
 575                 if ((addr_type & IPV6_ADDR_UNICAST) &&
 576                      ipv6_addr_is_isatap(addr6))
 577                         dst = addr6->s6_addr32[3];
 578                 else
 579                         goto tx_error;
 580         }
 581
 582         if (!dst)
 583                 dst = try_6to4(&iph6->daddr);
 584
 585         if (!dst) {
 586                 struct neighbour *neigh = NULL;
 587
 588                 if (skb->dst)
 589                         neigh = skb->dst->neighbour;
 590
 591                 if (neigh == NULL) {
 592                         if (net_ratelimit())
 593                                 printk(KERN_DEBUG "sit: nexthop == NULL\n");
 594                         goto tx_error;
 595                 }
 596
 597                 addr6 = (struct in6_addr*)&neigh->primary_key;
 598                 addr_type = ipv6_addr_type(addr6);
 599
 600                 if (addr_type == IPV6_ADDR_ANY) {
 601                         addr6 = &ipv6_hdr(skb)->daddr;
 602                         addr_type = ipv6_addr_type(addr6);
 603                 }
 604
 605                 if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
 606                         goto tx_error_icmp;
 607
 608                 dst = addr6->s6_addr32[3];
 609         }
 610
 611         {
 612                 struct flowi fl = { .nl_u = { .ip4_u =
 613                                               { .daddr = dst,
 614                                                 .saddr = tiph->saddr,
 615                                                 .tos = RT_TOS(tos) } },
 616                                     .oif = tunnel->parms.link,
 617                                     .proto = IPPROTO_IPV6 };
 618                 if (ip_route_output_key(dev_net(dev), &rt, &fl)) {
 619                         stats->tx_carrier_errors++;
 620                         goto tx_error_icmp;
 621                 }
 622         }
 623         if (rt->rt_type != RTN_UNICAST) {
 624                 ip_rt_put(rt);
 625                 stats->tx_carrier_errors++;
 626                 goto tx_error_icmp;
 627         }
 628         tdev = rt->u.dst.dev;
 629
 630         if (tdev == dev) {
 631                 ip_rt_put(rt);
 632                 stats->collisions++;
 633                 goto tx_error;
 634         }
 635
 636         if (tiph->frag_off)
 637                 mtu = dst_mtu(&rt->u.dst) - sizeof(struct iphdr);
 638         else
 639                 mtu = skb->dst ? dst_mtu(skb->dst) : dev->mtu;
 640
 641         if (mtu < 68) {
 642                 stats->collisions++;
 643                 ip_rt_put(rt);
 644                 goto tx_error;
 645         }
 646         if (mtu < IPV6_MIN_MTU)
 647                 mtu = IPV6_MIN_MTU;
 648         if (tunnel->parms.iph.daddr && skb->dst)
 649                 skb->dst->ops->update_pmtu(skb->dst, mtu);
 650
 651         if (skb->len > mtu) {
 652                 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, dev);
 653                 ip_rt_put(rt);
 654                 goto tx_error;
 655         }
 656
 657         if (tunnel->err_count > 0) {
 658                 if (jiffies - tunnel->err_time < IPTUNNEL_ERR_TIMEO) {
 659                         tunnel->err_count--;
 660                         dst_link_failure(skb);
 661                 } else
 662                         tunnel->err_count = 0;
 663         }
 664
 665         /*
 666          * Okay, now see if we can stuff it in the buffer as-is.
 667          */
 668         max_headroom = LL_RESERVED_SPACE(tdev)+sizeof(struct iphdr);
 669
 670         if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
 671             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
 672                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
 673                 if (!new_skb) {
 674                         ip_rt_put(rt);
 675                         stats->tx_dropped++;
 676                         dev_kfree_skb(skb);
 677                         tunnel->recursion--;
 678                         return 0;
 679                 }
 680                 if (skb->sk)
 681                         skb_set_owner_w(new_skb, skb->sk);
 682                 dev_kfree_skb(skb);
 683                 skb = new_skb;
 684                 iph6 = ipv6_hdr(skb);
 685         }
 686
 687         skb->transport_header = skb->network_header;
 688         skb_push(skb, sizeof(struct iphdr));
 689         skb_reset_network_header(skb);
 690         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 691         IPCB(skb)->flags = 0;
 692         dst_release(skb->dst);
 693         skb->dst = &rt->u.dst;
 694
 695         /*
 696          *      Push down and install the IPIP header.
 697          */
 698
 699         iph                     =       ip_hdr(skb);
 700         iph->version            =       4;
 701         iph->ihl                =       sizeof(struct iphdr)>>2;
 702         if (mtu > IPV6_MIN_MTU)
 703                 iph->frag_off   =       htons(IP_DF);
 704         else
 705                 iph->frag_off   =       0;
 706
 707         iph->protocol           =       IPPROTO_IPV6;
 708         iph->tos                =       INET_ECN_encapsulate(tos, ipv6_get_dsfield(iph6));
 709         iph->daddr              =       rt->rt_dst;
 710         iph->saddr              =       rt->rt_src;
 711
 712         if ((iph->ttl = tiph->ttl) == 0)
 713                 iph->ttl        =       iph6->hop_limit;
 714
 715         nf_reset(skb);
 716
 717         IPTUNNEL_XMIT();
 718         tunnel->recursion--;
 719         return 0;
 720
 721 tx_error_icmp:
 722         dst_link_failure(skb);
 723 tx_error:
 724         stats->tx_errors++;
 725         dev_kfree_skb(skb);
 726         tunnel->recursion--;
 727         return 0;
 728 }
 729
 730 static void ipip6_tunnel_bind_dev(struct net_device *dev)
 731 {
 732         struct net_device *tdev = NULL;
 733         struct ip_tunnel *tunnel;
 734         struct iphdr *iph;
 735
 736         tunnel = netdev_priv(dev);
 737         iph = &tunnel->parms.iph;
 738
 739         if (iph->daddr) {
 740                 struct flowi fl = { .nl_u = { .ip4_u =
 741                                               { .daddr = iph->daddr,
 742                                                 .saddr = iph->saddr,
 743                                                 .tos = RT_TOS(iph->tos) } },
 744                                     .oif = tunnel->parms.link,
 745                                     .proto = IPPROTO_IPV6 };
 746                 struct rtable *rt;
 747                 if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
 748                         tdev = rt->u.dst.dev;
 749                         ip_rt_put(rt);
 750                 }
 751                 dev->flags |= IFF_POINTOPOINT;
 752         }
 753
 754         if (!tdev && tunnel->parms.link)
 755                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
 756
 757         if (tdev) {
 758                 dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
 759                 dev->mtu = tdev->mtu - sizeof(struct iphdr);
 760                 if (dev->mtu < IPV6_MIN_MTU)
 761                         dev->mtu = IPV6_MIN_MTU;
 762         }
 763         dev->iflink = tunnel->parms.link;
 764 }
 765
 766 static int
 767 ipip6_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
 768 {
 769         int err = 0;
 770         struct ip_tunnel_parm p;
 771         struct ip_tunnel_prl prl;
 772         struct ip_tunnel *t;
 773         struct net *net = dev_net(dev);
 774         struct sit_net *sitn = net_generic(net, sit_net_id);
 775
 776         switch (cmd) {
 777         case SIOCGETTUNNEL:
 778                 t = NULL;
 779                 if (dev == sitn->fb_tunnel_dev) {
 780                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
 781                                 err = -EFAULT;
 782                                 break;
 783                         }
 784                         t = ipip6_tunnel_locate(net, &p, 0);
 785                 }
 786                 if (t == NULL)
 787                         t = netdev_priv(dev);
 788                 memcpy(&p, &t->parms, sizeof(p));
 789                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
 790                         err = -EFAULT;
 791                 break;
 792
 793         case SIOCADDTUNNEL:
 794         case SIOCCHGTUNNEL:
 795                 err = -EPERM;
 796                 if (!capable(CAP_NET_ADMIN))
 797                         goto done;
 798
 799                 err = -EFAULT;
 800                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
 801                         goto done;
 802
 803                 err = -EINVAL;
 804                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPV6 ||
 805                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
 806                         goto done;
 807                 if (p.iph.ttl)
 808                         p.iph.frag_off |= htons(IP_DF);
 809
 810                 t = ipip6_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
 811
 812                 if (dev != sitn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
 813                         if (t != NULL) {
 814                                 if (t->dev != dev) {
 815                                         err = -EEXIST;
 816                                         break;
 817                                 }
 818                         } else {
 819                                 if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
 820                                     (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
 821                                         err = -EINVAL;
 822                                         break;
 823                                 }
 824                                 t = netdev_priv(dev);
 825                                 ipip6_tunnel_unlink(sitn, t);
 826                                 t->parms.iph.saddr = p.iph.saddr;
 827                                 t->parms.iph.daddr = p.iph.daddr;
 828                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
 829                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
 830                                 ipip6_tunnel_link(sitn, t);
 831                                 netdev_state_change(dev);
 832                         }
 833                 }
 834
 835                 if (t) {
 836                         err = 0;
 837                         if (cmd == SIOCCHGTUNNEL) {
 838                                 t->parms.iph.ttl = p.iph.ttl;
 839                                 t->parms.iph.tos = p.iph.tos;
 840                                 if (t->parms.link != p.link) {
 841                                         t->parms.link = p.link;
 842                                         ipip6_tunnel_bind_dev(dev);
 843                                         netdev_state_change(dev);
 844                                 }
 845                         }
 846                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
 847                                 err = -EFAULT;
 848                 } else
 849                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
 850                 break;
 851
 852         case SIOCDELTUNNEL:
 853                 err = -EPERM;
 854                 if (!capable(CAP_NET_ADMIN))
 855                         goto done;
 856
 857                 if (dev == sitn->fb_tunnel_dev) {
 858                         err = -EFAULT;
 859                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
 860                                 goto done;
 861                         err = -ENOENT;
 862                         if ((t = ipip6_tunnel_locate(net, &p, 0)) == NULL)
 863                                 goto done;
 864                         err = -EPERM;
 865                         if (t == netdev_priv(sitn->fb_tunnel_dev))
 866                                 goto done;
 867                         dev = t->dev;
 868                 }
 869                 unregister_netdevice(dev);
 870                 err = 0;
 871                 break;
 872
 873         case SIOCGETPRL:
 874         case SIOCADDPRL:
 875         case SIOCDELPRL:
 876         case SIOCCHGPRL:
 877                 err = -EPERM;
 878                 if (cmd != SIOCGETPRL && !capable(CAP_NET_ADMIN))
 879                         goto done;
 880                 err = -EINVAL;
 881                 if (dev == sitn->fb_tunnel_dev)
 882                         goto done;
 883                 err = -EFAULT;
 884                 if (copy_from_user(&prl, ifr->ifr_ifru.ifru_data, sizeof(prl)))
 885                         goto done;
 886                 err = -ENOENT;
 887                 if (!(t = netdev_priv(dev)))
 888                         goto done;
 889
 890                 switch (cmd) {
 891                 case SIOCGETPRL:
 892                         err = ipip6_tunnel_get_prl(t, &prl);
 893                         if (!err && copy_to_user(ifr->ifr_ifru.ifru_data,
 894                                                  &prl, sizeof(prl)))
 895                                 err = -EFAULT;
 896                         break;
 897                 case SIOCDELPRL:
 898                         err = ipip6_tunnel_del_prl(t, &prl);
 899                         break;
 900                 case SIOCADDPRL:
 901                 case SIOCCHGPRL:
 902                         err = ipip6_tunnel_add_prl(t, &prl, cmd == SIOCCHGPRL);
 903                         break;
 904                 }
 905                 if (cmd != SIOCGETPRL)
 906                         netdev_state_change(dev);
 907                 break;
 908
 909         default:
 910                 err = -EINVAL;
 911         }
 912
 913 done:
 914         return err;
 915 }
 916
 917 static int ipip6_tunnel_change_mtu(struct net_device *dev, int new_mtu)
 918 {
 919         if (new_mtu < IPV6_MIN_MTU || new_mtu > 0xFFF8 - sizeof(struct iphdr))
 920                 return -EINVAL;
 921         dev->mtu = new_mtu;
 922         return 0;
 923 }
 924
 925 static void ipip6_tunnel_setup(struct net_device *dev)
 926 {
 927         dev->uninit             = ipip6_tunnel_uninit;
 928         dev->destructor         = free_netdev;
 929         dev->hard_start_xmit    = ipip6_tunnel_xmit;
 930         dev->do_ioctl           = ipip6_tunnel_ioctl;
 931         dev->change_mtu         = ipip6_tunnel_change_mtu;
 932
 933         dev->type               = ARPHRD_SIT;
 934         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr);
 935         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr);
 936         dev->flags              = IFF_NOARP;
 937         dev->iflink             = 0;
 938         dev->addr_len           = 4;
 939         dev->features           |= NETIF_F_NETNS_LOCAL;
 940 }
 941
 942 static int ipip6_tunnel_init(struct net_device *dev)
 943 {
 944         struct ip_tunnel *tunnel;
 945
 946         tunnel = netdev_priv(dev);
 947
 948         tunnel->dev = dev;
 949         strcpy(tunnel->parms.name, dev->name);
 950
 951         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
 952         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
 953
 954         ipip6_tunnel_bind_dev(dev);
 955
 956         return 0;
 957 }
 958
 959 static int ipip6_fb_tunnel_init(struct net_device *dev)
 960 {
 961         struct ip_tunnel *tunnel = netdev_priv(dev);
 962         struct iphdr *iph = &tunnel->parms.iph;
 963         struct net *net = dev_net(dev);
 964         struct sit_net *sitn = net_generic(net, sit_net_id);
 965
 966         tunnel->dev = dev;
 967         strcpy(tunnel->parms.name, dev->name);
 968
 969         iph->version            = 4;
 970         iph->protocol           = IPPROTO_IPV6;
 971         iph->ihl                = 5;
 972         iph->ttl                = 64;
 973
 974         dev_hold(dev);
 975         sitn->tunnels_wc[0]     = tunnel;
 976         return 0;
 977 }
 978
 979 static struct xfrm_tunnel sit_handler = {
 980         .handler        =       ipip6_rcv,
 981         .err_handler    =       ipip6_err,
 982         .priority       =       1,
 983 };
 984
 985 static void sit_destroy_tunnels(struct sit_net *sitn)
 986 {
 987         int prio;
 988
 989         for (prio = 1; prio < 4; prio++) {
 990                 int h;
 991                 for (h = 0; h < HASH_SIZE; h++) {
 992                         struct ip_tunnel *t;
 993                         while ((t = sitn->tunnels[prio][h]) != NULL)
 994                                 unregister_netdevice(t->dev);
 995                 }
 996         }
 997 }
 998
 999 static int sit_init_net(struct net *net)
1000 {
1001         int err;
1002         struct sit_net *sitn;
1003
1004         err = -ENOMEM;
1005         sitn = kzalloc(sizeof(struct sit_net), GFP_KERNEL);
1006         if (sitn == NULL)
1007                 goto err_alloc;
1008
1009         err = net_assign_generic(net, sit_net_id, sitn);
1010         if (err < 0)
1011                 goto err_assign;
1012
1013         sitn->tunnels[0] = sitn->tunnels_wc;
1014         sitn->tunnels[1] = sitn->tunnels_l;
1015         sitn->tunnels[2] = sitn->tunnels_r;
1016         sitn->tunnels[3] = sitn->tunnels_r_l;
1017
1018         sitn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "sit0",
1019                                            ipip6_tunnel_setup);
1020         if (!sitn->fb_tunnel_dev) {
1021                 err = -ENOMEM;
1022                 goto err_alloc_dev;
1023         }
1024
1025         sitn->fb_tunnel_dev->init = ipip6_fb_tunnel_init;
1026         dev_net_set(sitn->fb_tunnel_dev, net);
1027
1028         if ((err = register_netdev(sitn->fb_tunnel_dev)))
1029                 goto err_reg_dev;
1030
1031         return 0;
1032
1033 err_reg_dev:
1034         free_netdev(sitn->fb_tunnel_dev);
1035 err_alloc_dev:
1036         /* nothing */
1037 err_assign:
1038         kfree(sitn);
1039 err_alloc:
1040         return err;
1041 }
1042
1043 static void sit_exit_net(struct net *net)
1044 {
1045         struct sit_net *sitn;
1046
1047         sitn = net_generic(net, sit_net_id);
1048         rtnl_lock();
1049         sit_destroy_tunnels(sitn);
1050         unregister_netdevice(sitn->fb_tunnel_dev);
1051         rtnl_unlock();
1052         kfree(sitn);
1053 }
1054
1055 static struct pernet_operations sit_net_ops = {
1056         .init = sit_init_net,
1057         .exit = sit_exit_net,
1058 };
1059
1060 static void __exit sit_cleanup(void)
1061 {
1062         xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
1063
1064         unregister_pernet_gen_device(sit_net_id, &sit_net_ops);
1065 }
1066
1067 static int __init sit_init(void)
1068 {
1069         int err;
1070
1071         printk(KERN_INFO "IPv6 over IPv4 tunneling driver\n");
1072
1073         if (xfrm4_tunnel_register(&sit_handler, AF_INET6) < 0) {
1074                 printk(KERN_INFO "sit init: Can't add protocol\n");
1075                 return -EAGAIN;
1076         }
1077
1078         err = register_pernet_gen_device(&sit_net_id, &sit_net_ops);
1079         if (err < 0)
1080                 xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
1081
1082         return err;
1083 }
1084
1085 module_init(sit_init);
1086 module_exit(sit_cleanup);
1087 MODULE_LICENSE("GPL");
1088 MODULE_ALIAS("sit0");