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             <[email protected]>
   7  *      Alexey Kuznetsov        <[email protected]>
   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 <[email protected]>:       6to4 support
  16  * Nate Thompson <[email protected]>:             6to4 support
  17  * Fred Templin <[email protected]>:    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 void ipip6_fb_tunnel_init(struct net_device *dev);
  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 *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 /*
  81  * Locking : hash tables are protected by RCU and a spinlock
  82  */
  83 static DEFINE_SPINLOCK(ipip6_lock);
  84
  85 #define for_each_ip_tunnel_rcu(start) \
  86         for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
  87
  88 /*
  89  * Must be invoked with rcu_read_lock
  90  */
  91 static struct ip_tunnel * ipip6_tunnel_lookup(struct net *net,
  92                 struct net_device *dev, __be32 remote, __be32 local)
  93 {
  94         unsigned h0 = HASH(remote);
  95         unsigned h1 = HASH(local);
  96         struct ip_tunnel *t;
  97         struct sit_net *sitn = net_generic(net, sit_net_id);
  98
  99         for_each_ip_tunnel_rcu(sitn->tunnels_r_l[h0 ^ h1]) {
 100                 if (local == t->parms.iph.saddr &&
 101                     remote == t->parms.iph.daddr &&
 102                     (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
 103                     (t->dev->flags & IFF_UP))
 104                         return t;
 105         }
 106         for_each_ip_tunnel_rcu(sitn->tunnels_r[h0]) {
 107                 if (remote == t->parms.iph.daddr &&
 108                     (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
 109                     (t->dev->flags & IFF_UP))
 110                         return t;
 111         }
 112         for_each_ip_tunnel_rcu(sitn->tunnels_l[h1]) {
 113                 if (local == t->parms.iph.saddr &&
 114                     (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
 115                     (t->dev->flags & IFF_UP))
 116                         return t;
 117         }
 118         t = rcu_dereference(sitn->tunnels_wc[0]);
 119         if ((t != NULL) && (t->dev->flags & IFF_UP))
 120                 return t;
 121         return NULL;
 122 }
 123
 124 static struct ip_tunnel **__ipip6_bucket(struct sit_net *sitn,
 125                 struct ip_tunnel_parm *parms)
 126 {
 127         __be32 remote = parms->iph.daddr;
 128         __be32 local = parms->iph.saddr;
 129         unsigned h = 0;
 130         int prio = 0;
 131
 132         if (remote) {
 133                 prio |= 2;
 134                 h ^= HASH(remote);
 135         }
 136         if (local) {
 137                 prio |= 1;
 138                 h ^= HASH(local);
 139         }
 140         return &sitn->tunnels[prio][h];
 141 }
 142
 143 static inline struct ip_tunnel **ipip6_bucket(struct sit_net *sitn,
 144                 struct ip_tunnel *t)
 145 {
 146         return __ipip6_bucket(sitn, &t->parms);
 147 }
 148
 149 static void ipip6_tunnel_unlink(struct sit_net *sitn, struct ip_tunnel *t)
 150 {
 151         struct ip_tunnel **tp;
 152
 153         for (tp = ipip6_bucket(sitn, t); *tp; tp = &(*tp)->next) {
 154                 if (t == *tp) {
 155                         spin_lock_bh(&ipip6_lock);
 156                         *tp = t->next;
 157                         spin_unlock_bh(&ipip6_lock);
 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 **tp = ipip6_bucket(sitn, t);
 166
 167         spin_lock_bh(&ipip6_lock);
 168         t->next = *tp;
 169         rcu_assign_pointer(*tp, t);
 170         spin_unlock_bh(&ipip6_lock);
 171 }
 172
 173 static void ipip6_tunnel_clone_6rd(struct net_device *dev, struct sit_net *sitn)
 174 {
 175 #ifdef CONFIG_IPV6_SIT_6RD
 176         struct ip_tunnel *t = netdev_priv(dev);
 177
 178         if (t->dev == sitn->fb_tunnel_dev) {
 179                 ipv6_addr_set(&t->ip6rd.prefix, htonl(0x20020000), 0, 0, 0);
 180                 t->ip6rd.relay_prefix = 0;
 181                 t->ip6rd.prefixlen = 16;
 182                 t->ip6rd.relay_prefixlen = 0;
 183         } else {
 184                 struct ip_tunnel *t0 = netdev_priv(sitn->fb_tunnel_dev);
 185                 memcpy(&t->ip6rd, &t0->ip6rd, sizeof(t->ip6rd));
 186         }
 187 #endif
 188 }
 189
 190 static struct ip_tunnel * ipip6_tunnel_locate(struct net *net,
 191                 struct ip_tunnel_parm *parms, int create)
 192 {
 193         __be32 remote = parms->iph.daddr;
 194         __be32 local = parms->iph.saddr;
 195         struct ip_tunnel *t, **tp, *nt;
 196         struct net_device *dev;
 197         char name[IFNAMSIZ];
 198         struct sit_net *sitn = net_generic(net, sit_net_id);
 199
 200         for (tp = __ipip6_bucket(sitn, parms); (t = *tp) != NULL; tp = &t->next) {
 201                 if (local == t->parms.iph.saddr &&
 202                     remote == t->parms.iph.daddr &&
 203                     parms->link == t->parms.link) {
 204                         if (create)
 205                                 return NULL;
 206                         else
 207                                 return t;
 208                 }
 209         }
 210         if (!create)
 211                 goto failed;
 212
 213         if (parms->name[0])
 214                 strlcpy(name, parms->name, IFNAMSIZ);
 215         else
 216                 sprintf(name, "sit%%d");
 217
 218         dev = alloc_netdev(sizeof(*t), name, ipip6_tunnel_setup);
 219         if (dev == NULL)
 220                 return NULL;
 221
 222         dev_net_set(dev, net);
 223
 224         if (strchr(name, '%')) {
 225                 if (dev_alloc_name(dev, name) < 0)
 226                         goto failed_free;
 227         }
 228
 229         nt = netdev_priv(dev);
 230
 231         nt->parms = *parms;
 232         ipip6_tunnel_init(dev);
 233         ipip6_tunnel_clone_6rd(dev, sitn);
 234
 235         if (parms->i_flags & SIT_ISATAP)
 236                 dev->priv_flags |= IFF_ISATAP;
 237
 238         if (register_netdevice(dev) < 0)
 239                 goto failed_free;
 240
 241         dev_hold(dev);
 242
 243         ipip6_tunnel_link(sitn, nt);
 244         return nt;
 245
 246 failed_free:
 247         free_netdev(dev);
 248 failed:
 249         return NULL;
 250 }
 251
 252 static DEFINE_SPINLOCK(ipip6_prl_lock);
 253
 254 #define for_each_prl_rcu(start)                 \
 255         for (prl = rcu_dereference(start);      \
 256              prl;                               \
 257              prl = rcu_dereference(prl->next))
 258
 259 static struct ip_tunnel_prl_entry *
 260 __ipip6_tunnel_locate_prl(struct ip_tunnel *t, __be32 addr)
 261 {
 262         struct ip_tunnel_prl_entry *prl;
 263
 264         for_each_prl_rcu(t->prl)
 265                 if (prl->addr == addr)
 266                         break;
 267         return prl;
 268
 269 }
 270
 271 static int ipip6_tunnel_get_prl(struct ip_tunnel *t,
 272                                 struct ip_tunnel_prl __user *a)
 273 {
 274         struct ip_tunnel_prl kprl, *kp;
 275         struct ip_tunnel_prl_entry *prl;
 276         unsigned int cmax, c = 0, ca, len;
 277         int ret = 0;
 278
 279         if (copy_from_user(&kprl, a, sizeof(kprl)))
 280                 return -EFAULT;
 281         cmax = kprl.datalen / sizeof(kprl);
 282         if (cmax > 1 && kprl.addr != htonl(INADDR_ANY))
 283                 cmax = 1;
 284
 285         /* For simple GET or for root users,
 286          * we try harder to allocate.
 287          */
 288         kp = (cmax <= 1 || capable(CAP_NET_ADMIN)) ?
 289                 kcalloc(cmax, sizeof(*kp), GFP_KERNEL) :
 290                 NULL;
 291
 292         rcu_read_lock();
 293
 294         ca = t->prl_count < cmax ? t->prl_count : cmax;
 295
 296         if (!kp) {
 297                 /* We don't try hard to allocate much memory for
 298                  * non-root users.
 299                  * For root users, retry allocating enough memory for
 300                  * the answer.
 301                  */
 302                 kp = kcalloc(ca, sizeof(*kp), GFP_ATOMIC);
 303                 if (!kp) {
 304                         ret = -ENOMEM;
 305                         goto out;
 306                 }
 307         }
 308
 309         c = 0;
 310         for_each_prl_rcu(t->prl) {
 311                 if (c >= cmax)
 312                         break;
 313                 if (kprl.addr != htonl(INADDR_ANY) && prl->addr != kprl.addr)
 314                         continue;
 315                 kp[c].addr = prl->addr;
 316                 kp[c].flags = prl->flags;
 317                 c++;
 318                 if (kprl.addr != htonl(INADDR_ANY))
 319                         break;
 320         }
 321 out:
 322         rcu_read_unlock();
 323
 324         len = sizeof(*kp) * c;
 325         ret = 0;
 326         if ((len && copy_to_user(a + 1, kp, len)) || put_user(len, &a->datalen))
 327                 ret = -EFAULT;
 328
 329         kfree(kp);
 330
 331         return ret;
 332 }
 333
 334 static int
 335 ipip6_tunnel_add_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a, int chg)
 336 {
 337         struct ip_tunnel_prl_entry *p;
 338         int err = 0;
 339
 340         if (a->addr == htonl(INADDR_ANY))
 341                 return -EINVAL;
 342
 343         spin_lock(&ipip6_prl_lock);
 344
 345         for (p = t->prl; p; p = p->next) {
 346                 if (p->addr == a->addr) {
 347                         if (chg) {
 348                                 p->flags = a->flags;
 349                                 goto out;
 350                         }
 351                         err = -EEXIST;
 352                         goto out;
 353                 }
 354         }
 355
 356         if (chg) {
 357                 err = -ENXIO;
 358                 goto out;
 359         }
 360
 361         p = kzalloc(sizeof(struct ip_tunnel_prl_entry), GFP_KERNEL);
 362         if (!p) {
 363                 err = -ENOBUFS;
 364                 goto out;
 365         }
 366
 367         INIT_RCU_HEAD(&p->rcu_head);
 368         p->next = t->prl;
 369         p->addr = a->addr;
 370         p->flags = a->flags;
 371         t->prl_count++;
 372         rcu_assign_pointer(t->prl, p);
 373 out:
 374         spin_unlock(&ipip6_prl_lock);
 375         return err;
 376 }
 377
 378 static void prl_entry_destroy_rcu(struct rcu_head *head)
 379 {
 380         kfree(container_of(head, struct ip_tunnel_prl_entry, rcu_head));
 381 }
 382
 383 static void prl_list_destroy_rcu(struct rcu_head *head)
 384 {
 385         struct ip_tunnel_prl_entry *p, *n;
 386
 387         p = container_of(head, struct ip_tunnel_prl_entry, rcu_head);
 388         do {
 389                 n = p->next;
 390                 kfree(p);
 391                 p = n;
 392         } while (p);
 393 }
 394
 395 static int
 396 ipip6_tunnel_del_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a)
 397 {
 398         struct ip_tunnel_prl_entry *x, **p;
 399         int err = 0;
 400
 401         spin_lock(&ipip6_prl_lock);
 402
 403         if (a && a->addr != htonl(INADDR_ANY)) {
 404                 for (p = &t->prl; *p; p = &(*p)->next) {
 405                         if ((*p)->addr == a->addr) {
 406                                 x = *p;
 407                                 *p = x->next;
 408                                 call_rcu(&x->rcu_head, prl_entry_destroy_rcu);
 409                                 t->prl_count--;
 410                                 goto out;
 411                         }
 412                 }
 413                 err = -ENXIO;
 414         } else {
 415                 if (t->prl) {
 416                         t->prl_count = 0;
 417                         x = t->prl;
 418                         call_rcu(&x->rcu_head, prl_list_destroy_rcu);
 419                         t->prl = NULL;
 420                 }
 421         }
 422 out:
 423         spin_unlock(&ipip6_prl_lock);
 424         return err;
 425 }
 426
 427 static int
 428 isatap_chksrc(struct sk_buff *skb, struct iphdr *iph, struct ip_tunnel *t)
 429 {
 430         struct ip_tunnel_prl_entry *p;
 431         int ok = 1;
 432
 433         rcu_read_lock();
 434         p = __ipip6_tunnel_locate_prl(t, iph->saddr);
 435         if (p) {
 436                 if (p->flags & PRL_DEFAULT)
 437                         skb->ndisc_nodetype = NDISC_NODETYPE_DEFAULT;
 438                 else
 439                         skb->ndisc_nodetype = NDISC_NODETYPE_NODEFAULT;
 440         } else {
 441                 struct in6_addr *addr6 = &ipv6_hdr(skb)->saddr;
 442                 if (ipv6_addr_is_isatap(addr6) &&
 443                     (addr6->s6_addr32[3] == iph->saddr) &&
 444                     ipv6_chk_prefix(addr6, t->dev))
 445                         skb->ndisc_nodetype = NDISC_NODETYPE_HOST;
 446                 else
 447                         ok = 0;
 448         }
 449         rcu_read_unlock();
 450         return ok;
 451 }
 452
 453 static void ipip6_tunnel_uninit(struct net_device *dev)
 454 {
 455         struct net *net = dev_net(dev);
 456         struct sit_net *sitn = net_generic(net, sit_net_id);
 457
 458         if (dev == sitn->fb_tunnel_dev) {
 459                 spin_lock_bh(&ipip6_lock);
 460                 sitn->tunnels_wc[0] = NULL;
 461                 spin_unlock_bh(&ipip6_lock);
 462                 dev_put(dev);
 463         } else {
 464                 ipip6_tunnel_unlink(sitn, netdev_priv(dev));
 465                 ipip6_tunnel_del_prl(netdev_priv(dev), NULL);
 466                 dev_put(dev);
 467         }
 468 }
 469
 470
 471 static int ipip6_err(struct sk_buff *skb, u32 info)
 472 {
 473
 474 /* All the routers (except for Linux) return only
 475    8 bytes of packet payload. It means, that precise relaying of
 476    ICMP in the real Internet is absolutely infeasible.
 477  */
 478         struct iphdr *iph = (struct iphdr*)skb->data;
 479         const int type = icmp_hdr(skb)->type;
 480         const int code = icmp_hdr(skb)->code;
 481         struct ip_tunnel *t;
 482         int err;
 483
 484         switch (type) {
 485         default:
 486         case ICMP_PARAMETERPROB:
 487                 return 0;
 488
 489         case ICMP_DEST_UNREACH:
 490                 switch (code) {
 491                 case ICMP_SR_FAILED:
 492                 case ICMP_PORT_UNREACH:
 493                         /* Impossible event. */
 494                         return 0;
 495                 case ICMP_FRAG_NEEDED:
 496                         /* Soft state for pmtu is maintained by IP core. */
 497                         return 0;
 498                 default:
 499                         /* All others are translated to HOST_UNREACH.
 500                            rfc2003 contains "deep thoughts" about NET_UNREACH,
 501                            I believe they are just ether pollution. --ANK
 502                          */
 503                         break;
 504                 }
 505                 break;
 506         case ICMP_TIME_EXCEEDED:
 507                 if (code != ICMP_EXC_TTL)
 508                         return 0;
 509                 break;
 510         }
 511
 512         err = -ENOENT;
 513
 514         rcu_read_lock();
 515         t = ipip6_tunnel_lookup(dev_net(skb->dev),
 516                                 skb->dev,
 517                                 iph->daddr,
 518                                 iph->saddr);
 519         if (t == NULL || t->parms.iph.daddr == 0)
 520                 goto out;
 521
 522         err = 0;
 523         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
 524                 goto out;
 525
 526         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
 527                 t->err_count++;
 528         else
 529                 t->err_count = 1;
 530         t->err_time = jiffies;
 531 out:
 532         rcu_read_unlock();
 533         return err;
 534 }
 535
 536 static inline void ipip6_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb)
 537 {
 538         if (INET_ECN_is_ce(iph->tos))
 539                 IP6_ECN_set_ce(ipv6_hdr(skb));
 540 }
 541
 542 static int ipip6_rcv(struct sk_buff *skb)
 543 {
 544         struct iphdr *iph;
 545         struct ip_tunnel *tunnel;
 546
 547         if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
 548                 goto out;
 549
 550         iph = ip_hdr(skb);
 551
 552         rcu_read_lock();
 553         tunnel = ipip6_tunnel_lookup(dev_net(skb->dev), skb->dev,
 554                                      iph->saddr, iph->daddr);
 555         if (tunnel != NULL) {
 556                 secpath_reset(skb);
 557                 skb->mac_header = skb->network_header;
 558                 skb_reset_network_header(skb);
 559                 IPCB(skb)->flags = 0;
 560                 skb->protocol = htons(ETH_P_IPV6);
 561                 skb->pkt_type = PACKET_HOST;
 562
 563                 if ((tunnel->dev->priv_flags & IFF_ISATAP) &&
 564                     !isatap_chksrc(skb, iph, tunnel)) {
 565                         tunnel->dev->stats.rx_errors++;
 566                         rcu_read_unlock();
 567                         kfree_skb(skb);
 568                         return 0;
 569                 }
 570                 tunnel->dev->stats.rx_packets++;
 571                 tunnel->dev->stats.rx_bytes += skb->len;
 572                 skb->dev = tunnel->dev;
 573                 skb_dst_drop(skb);
 574                 nf_reset(skb);
 575                 ipip6_ecn_decapsulate(iph, skb);
 576                 netif_rx(skb);
 577                 rcu_read_unlock();
 578                 return 0;
 579         }
 580
 581         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 582         rcu_read_unlock();
 583 out:
 584         kfree_skb(skb);
 585         return 0;
 586 }
 587
 588 /*
 589  * Returns the embedded IPv4 address if the IPv6 address
 590  * comes from 6rd / 6to4 (RFC 3056) addr space.
 591  */
 592 static inline
 593 __be32 try_6rd(struct in6_addr *v6dst, struct ip_tunnel *tunnel)
 594 {
 595         __be32 dst = 0;
 596
 597 #ifdef CONFIG_IPV6_SIT_6RD
 598         if (ipv6_prefix_equal(v6dst, &tunnel->ip6rd.prefix,
 599                               tunnel->ip6rd.prefixlen)) {
 600                 unsigned pbw0, pbi0;
 601                 int pbi1;
 602                 u32 d;
 603
 604                 pbw0 = tunnel->ip6rd.prefixlen >> 5;
 605                 pbi0 = tunnel->ip6rd.prefixlen & 0x1f;
 606
 607                 d = (ntohl(v6dst->s6_addr32[pbw0]) << pbi0) >>
 608                     tunnel->ip6rd.relay_prefixlen;
 609
 610                 pbi1 = pbi0 - tunnel->ip6rd.relay_prefixlen;
 611                 if (pbi1 > 0)
 612                         d |= ntohl(v6dst->s6_addr32[pbw0 + 1]) >>
 613                              (32 - pbi1);
 614
 615                 dst = tunnel->ip6rd.relay_prefix | htonl(d);
 616         }
 617 #else
 618         if (v6dst->s6_addr16[0] == htons(0x2002)) {
 619                 /* 6to4 v6 addr has 16 bits prefix, 32 v4addr, 16 SLA, ... */
 620                 memcpy(&dst, &v6dst->s6_addr16[1], 4);
 621         }
 622 #endif
 623         return dst;
 624 }
 625
 626 /*
 627  *      This function assumes it is being called from dev_queue_xmit()
 628  *      and that skb is filled properly by that function.
 629  */
 630
 631 static netdev_tx_t ipip6_tunnel_xmit(struct sk_buff *skb,
 632                                      struct net_device *dev)
 633 {
 634         struct ip_tunnel *tunnel = netdev_priv(dev);
 635         struct net_device_stats *stats = &dev->stats;
 636         struct netdev_queue *txq = netdev_get_tx_queue(dev, 0);
 637         struct iphdr  *tiph = &tunnel->parms.iph;
 638         struct ipv6hdr *iph6 = ipv6_hdr(skb);
 639         u8     tos = tunnel->parms.iph.tos;
 640         __be16 df = tiph->frag_off;
 641         struct rtable *rt;                      /* Route to the other host */
 642         struct net_device *tdev;                        /* Device to other host */
 643         struct iphdr  *iph;                     /* Our new IP header */
 644         unsigned int max_headroom;              /* The extra header space needed */
 645         __be32 dst = tiph->daddr;
 646         int    mtu;
 647         struct in6_addr *addr6;
 648         int addr_type;
 649
 650         if (skb->protocol != htons(ETH_P_IPV6))
 651                 goto tx_error;
 652
 653         /* ISATAP (RFC4214) - must come before 6to4 */
 654         if (dev->priv_flags & IFF_ISATAP) {
 655                 struct neighbour *neigh = NULL;
 656
 657                 if (skb_dst(skb))
 658                         neigh = skb_dst(skb)->neighbour;
 659
 660                 if (neigh == NULL) {
 661                         if (net_ratelimit())
 662                                 printk(KERN_DEBUG "sit: nexthop == NULL\n");
 663                         goto tx_error;
 664                 }
 665
 666                 addr6 = (struct in6_addr*)&neigh->primary_key;
 667                 addr_type = ipv6_addr_type(addr6);
 668
 669                 if ((addr_type & IPV6_ADDR_UNICAST) &&
 670                      ipv6_addr_is_isatap(addr6))
 671                         dst = addr6->s6_addr32[3];
 672                 else
 673                         goto tx_error;
 674         }
 675
 676         if (!dst)
 677                 dst = try_6rd(&iph6->daddr, tunnel);
 678
 679         if (!dst) {
 680                 struct neighbour *neigh = NULL;
 681
 682                 if (skb_dst(skb))
 683                         neigh = skb_dst(skb)->neighbour;
 684
 685                 if (neigh == NULL) {
 686                         if (net_ratelimit())
 687                                 printk(KERN_DEBUG "sit: nexthop == NULL\n");
 688                         goto tx_error;
 689                 }
 690
 691                 addr6 = (struct in6_addr*)&neigh->primary_key;
 692                 addr_type = ipv6_addr_type(addr6);
 693
 694                 if (addr_type == IPV6_ADDR_ANY) {
 695                         addr6 = &ipv6_hdr(skb)->daddr;
 696                         addr_type = ipv6_addr_type(addr6);
 697                 }
 698
 699                 if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
 700                         goto tx_error_icmp;
 701
 702                 dst = addr6->s6_addr32[3];
 703         }
 704
 705         {
 706                 struct flowi fl = { .nl_u = { .ip4_u =
 707                                               { .daddr = dst,
 708                                                 .saddr = tiph->saddr,
 709                                                 .tos = RT_TOS(tos) } },
 710                                     .oif = tunnel->parms.link,
 711                                     .proto = IPPROTO_IPV6 };
 712                 if (ip_route_output_key(dev_net(dev), &rt, &fl)) {
 713                         stats->tx_carrier_errors++;
 714                         goto tx_error_icmp;
 715                 }
 716         }
 717         if (rt->rt_type != RTN_UNICAST) {
 718                 ip_rt_put(rt);
 719                 stats->tx_carrier_errors++;
 720                 goto tx_error_icmp;
 721         }
 722         tdev = rt->u.dst.dev;
 723
 724         if (tdev == dev) {
 725                 ip_rt_put(rt);
 726                 stats->collisions++;
 727                 goto tx_error;
 728         }
 729
 730         if (df) {
 731                 mtu = dst_mtu(&rt->u.dst) - sizeof(struct iphdr);
 732
 733                 if (mtu < 68) {
 734                         stats->collisions++;
 735                         ip_rt_put(rt);
 736                         goto tx_error;
 737                 }
 738
 739                 if (mtu < IPV6_MIN_MTU) {
 740                         mtu = IPV6_MIN_MTU;
 741                         df = 0;
 742                 }
 743
 744                 if (tunnel->parms.iph.daddr && skb_dst(skb))
 745                         skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
 746
 747                 if (skb->len > mtu) {
 748                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu, dev);
 749                         ip_rt_put(rt);
 750                         goto tx_error;
 751                 }
 752         }
 753
 754         if (tunnel->err_count > 0) {
 755                 if (time_before(jiffies,
 756                                 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
 757                         tunnel->err_count--;
 758                         dst_link_failure(skb);
 759                 } else
 760                         tunnel->err_count = 0;
 761         }
 762
 763         /*
 764          * Okay, now see if we can stuff it in the buffer as-is.
 765          */
 766         max_headroom = LL_RESERVED_SPACE(tdev)+sizeof(struct iphdr);
 767
 768         if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
 769             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
 770                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
 771                 if (!new_skb) {
 772                         ip_rt_put(rt);
 773                         txq->tx_dropped++;
 774                         dev_kfree_skb(skb);
 775                         return NETDEV_TX_OK;
 776                 }
 777                 if (skb->sk)
 778                         skb_set_owner_w(new_skb, skb->sk);
 779                 dev_kfree_skb(skb);
 780                 skb = new_skb;
 781                 iph6 = ipv6_hdr(skb);
 782         }
 783
 784         skb->transport_header = skb->network_header;
 785         skb_push(skb, sizeof(struct iphdr));
 786         skb_reset_network_header(skb);
 787         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 788         IPCB(skb)->flags = 0;
 789         skb_dst_drop(skb);
 790         skb_dst_set(skb, &rt->u.dst);
 791
 792         /*
 793          *      Push down and install the IPIP header.
 794          */
 795
 796         iph                     =       ip_hdr(skb);
 797         iph->version            =       4;
 798         iph->ihl                =       sizeof(struct iphdr)>>2;
 799         iph->frag_off           =       df;
 800         iph->protocol           =       IPPROTO_IPV6;
 801         iph->tos                =       INET_ECN_encapsulate(tos, ipv6_get_dsfield(iph6));
 802         iph->daddr              =       rt->rt_dst;
 803         iph->saddr              =       rt->rt_src;
 804
 805         if ((iph->ttl = tiph->ttl) == 0)
 806                 iph->ttl        =       iph6->hop_limit;
 807
 808         nf_reset(skb);
 809
 810         IPTUNNEL_XMIT();
 811         return NETDEV_TX_OK;
 812
 813 tx_error_icmp:
 814         dst_link_failure(skb);
 815 tx_error:
 816         stats->tx_errors++;
 817         dev_kfree_skb(skb);
 818         return NETDEV_TX_OK;
 819 }
 820
 821 static void ipip6_tunnel_bind_dev(struct net_device *dev)
 822 {
 823         struct net_device *tdev = NULL;
 824         struct ip_tunnel *tunnel;
 825         struct iphdr *iph;
 826
 827         tunnel = netdev_priv(dev);
 828         iph = &tunnel->parms.iph;
 829
 830         if (iph->daddr) {
 831                 struct flowi fl = { .nl_u = { .ip4_u =
 832                                               { .daddr = iph->daddr,
 833                                                 .saddr = iph->saddr,
 834                                                 .tos = RT_TOS(iph->tos) } },
 835                                     .oif = tunnel->parms.link,
 836                                     .proto = IPPROTO_IPV6 };
 837                 struct rtable *rt;
 838                 if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
 839                         tdev = rt->u.dst.dev;
 840                         ip_rt_put(rt);
 841                 }
 842                 dev->flags |= IFF_POINTOPOINT;
 843         }
 844
 845         if (!tdev && tunnel->parms.link)
 846                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
 847
 848         if (tdev) {
 849                 dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
 850                 dev->mtu = tdev->mtu - sizeof(struct iphdr);
 851                 if (dev->mtu < IPV6_MIN_MTU)
 852                         dev->mtu = IPV6_MIN_MTU;
 853         }
 854         dev->iflink = tunnel->parms.link;
 855 }
 856
 857 static int
 858 ipip6_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
 859 {
 860         int err = 0;
 861         struct ip_tunnel_parm p;
 862         struct ip_tunnel_prl prl;
 863         struct ip_tunnel *t;
 864         struct net *net = dev_net(dev);
 865         struct sit_net *sitn = net_generic(net, sit_net_id);
 866 #ifdef CONFIG_IPV6_SIT_6RD
 867         struct ip_tunnel_6rd ip6rd;
 868 #endif
 869
 870         switch (cmd) {
 871         case SIOCGETTUNNEL:
 872 #ifdef CONFIG_IPV6_SIT_6RD
 873         case SIOCGET6RD:
 874 #endif
 875                 t = NULL;
 876                 if (dev == sitn->fb_tunnel_dev) {
 877                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
 878                                 err = -EFAULT;
 879                                 break;
 880                         }
 881                         t = ipip6_tunnel_locate(net, &p, 0);
 882                 }
 883                 if (t == NULL)
 884                         t = netdev_priv(dev);
 885
 886                 err = -EFAULT;
 887                 if (cmd == SIOCGETTUNNEL) {
 888                         memcpy(&p, &t->parms, sizeof(p));
 889                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &p,
 890                                          sizeof(p)))
 891                                 goto done;
 892 #ifdef CONFIG_IPV6_SIT_6RD
 893                 } else {
 894                         ipv6_addr_copy(&ip6rd.prefix, &t->ip6rd.prefix);
 895                         ip6rd.relay_prefix = t->ip6rd.relay_prefix;
 896                         ip6rd.prefixlen = t->ip6rd.prefixlen;
 897                         ip6rd.relay_prefixlen = t->ip6rd.relay_prefixlen;
 898                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &ip6rd,
 899                                          sizeof(ip6rd)))
 900                                 goto done;
 901 #endif
 902                 }
 903                 err = 0;
 904                 break;
 905
 906         case SIOCADDTUNNEL:
 907         case SIOCCHGTUNNEL:
 908                 err = -EPERM;
 909                 if (!capable(CAP_NET_ADMIN))
 910                         goto done;
 911
 912                 err = -EFAULT;
 913                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
 914                         goto done;
 915
 916                 err = -EINVAL;
 917                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPV6 ||
 918                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
 919                         goto done;
 920                 if (p.iph.ttl)
 921                         p.iph.frag_off |= htons(IP_DF);
 922
 923                 t = ipip6_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
 924
 925                 if (dev != sitn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
 926                         if (t != NULL) {
 927                                 if (t->dev != dev) {
 928                                         err = -EEXIST;
 929                                         break;
 930                                 }
 931                         } else {
 932                                 if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
 933                                     (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
 934                                         err = -EINVAL;
 935                                         break;
 936                                 }
 937                                 t = netdev_priv(dev);
 938                                 ipip6_tunnel_unlink(sitn, t);
 939                                 t->parms.iph.saddr = p.iph.saddr;
 940                                 t->parms.iph.daddr = p.iph.daddr;
 941                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
 942                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
 943                                 ipip6_tunnel_link(sitn, t);
 944                                 netdev_state_change(dev);
 945                         }
 946                 }
 947
 948                 if (t) {
 949                         err = 0;
 950                         if (cmd == SIOCCHGTUNNEL) {
 951                                 t->parms.iph.ttl = p.iph.ttl;
 952                                 t->parms.iph.tos = p.iph.tos;
 953                                 if (t->parms.link != p.link) {
 954                                         t->parms.link = p.link;
 955                                         ipip6_tunnel_bind_dev(dev);
 956                                         netdev_state_change(dev);
 957                                 }
 958                         }
 959                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
 960                                 err = -EFAULT;
 961                 } else
 962                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
 963                 break;
 964
 965         case SIOCDELTUNNEL:
 966                 err = -EPERM;
 967                 if (!capable(CAP_NET_ADMIN))
 968                         goto done;
 969
 970                 if (dev == sitn->fb_tunnel_dev) {
 971                         err = -EFAULT;
 972                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
 973                                 goto done;
 974                         err = -ENOENT;
 975                         if ((t = ipip6_tunnel_locate(net, &p, 0)) == NULL)
 976                                 goto done;
 977                         err = -EPERM;
 978                         if (t == netdev_priv(sitn->fb_tunnel_dev))
 979                                 goto done;
 980                         dev = t->dev;
 981                 }
 982                 unregister_netdevice(dev);
 983                 err = 0;
 984                 break;
 985
 986         case SIOCGETPRL:
 987                 err = -EINVAL;
 988                 if (dev == sitn->fb_tunnel_dev)
 989                         goto done;
 990                 err = -ENOENT;
 991                 if (!(t = netdev_priv(dev)))
 992                         goto done;
 993                 err = ipip6_tunnel_get_prl(t, ifr->ifr_ifru.ifru_data);
 994                 break;
 995
 996         case SIOCADDPRL:
 997         case SIOCDELPRL:
 998         case SIOCCHGPRL:
 999                 err = -EPERM;
1000                 if (!capable(CAP_NET_ADMIN))
1001                         goto done;
1002                 err = -EINVAL;
1003                 if (dev == sitn->fb_tunnel_dev)
1004                         goto done;
1005                 err = -EFAULT;
1006                 if (copy_from_user(&prl, ifr->ifr_ifru.ifru_data, sizeof(prl)))
1007                         goto done;
1008                 err = -ENOENT;
1009                 if (!(t = netdev_priv(dev)))
1010                         goto done;
1011
1012                 switch (cmd) {
1013                 case SIOCDELPRL:
1014                         err = ipip6_tunnel_del_prl(t, &prl);
1015                         break;
1016                 case SIOCADDPRL:
1017                 case SIOCCHGPRL:
1018                         err = ipip6_tunnel_add_prl(t, &prl, cmd == SIOCCHGPRL);
1019                         break;
1020                 }
1021                 netdev_state_change(dev);
1022                 break;
1023
1024 #ifdef CONFIG_IPV6_SIT_6RD
1025         case SIOCADD6RD:
1026         case SIOCCHG6RD:
1027         case SIOCDEL6RD:
1028                 err = -EPERM;
1029                 if (!capable(CAP_NET_ADMIN))
1030                         goto done;
1031
1032                 err = -EFAULT;
1033                 if (copy_from_user(&ip6rd, ifr->ifr_ifru.ifru_data,
1034                                    sizeof(ip6rd)))
1035                         goto done;
1036
1037                 t = netdev_priv(dev);
1038
1039                 if (cmd != SIOCDEL6RD) {
1040                         struct in6_addr prefix;
1041                         __be32 relay_prefix;
1042
1043                         err = -EINVAL;
1044                         if (ip6rd.relay_prefixlen > 32 ||
1045                             ip6rd.prefixlen + (32 - ip6rd.relay_prefixlen) > 64)
1046                                 goto done;
1047
1048                         ipv6_addr_prefix(&prefix, &ip6rd.prefix,
1049                                          ip6rd.prefixlen);
1050                         if (!ipv6_addr_equal(&prefix, &ip6rd.prefix))
1051                                 goto done;
1052                         if (ip6rd.relay_prefixlen)
1053                                 relay_prefix = ip6rd.relay_prefix &
1054                                                htonl(0xffffffffUL <<
1055                                                      (32 - ip6rd.relay_prefixlen));
1056                         else
1057                                 relay_prefix = 0;
1058                         if (relay_prefix != ip6rd.relay_prefix)
1059                                 goto done;
1060
1061                         ipv6_addr_copy(&t->ip6rd.prefix, &prefix);
1062                         t->ip6rd.relay_prefix = relay_prefix;
1063                         t->ip6rd.prefixlen = ip6rd.prefixlen;
1064                         t->ip6rd.relay_prefixlen = ip6rd.relay_prefixlen;
1065                 } else
1066                         ipip6_tunnel_clone_6rd(dev, sitn);
1067
1068                 err = 0;
1069                 break;
1070 #endif
1071
1072         default:
1073                 err = -EINVAL;
1074         }
1075
1076 done:
1077         return err;
1078 }
1079
1080 static int ipip6_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1081 {
1082         if (new_mtu < IPV6_MIN_MTU || new_mtu > 0xFFF8 - sizeof(struct iphdr))
1083                 return -EINVAL;
1084         dev->mtu = new_mtu;
1085         return 0;
1086 }
1087
1088 static const struct net_device_ops ipip6_netdev_ops = {
1089         .ndo_uninit     = ipip6_tunnel_uninit,
1090         .ndo_start_xmit = ipip6_tunnel_xmit,
1091         .ndo_do_ioctl   = ipip6_tunnel_ioctl,
1092         .ndo_change_mtu = ipip6_tunnel_change_mtu,
1093 };
1094
1095 static void ipip6_tunnel_setup(struct net_device *dev)
1096 {
1097         dev->netdev_ops         = &ipip6_netdev_ops;
1098         dev->destructor         = free_netdev;
1099
1100         dev->type               = ARPHRD_SIT;
1101         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr);
1102         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr);
1103         dev->flags              = IFF_NOARP;
1104         dev->priv_flags        &= ~IFF_XMIT_DST_RELEASE;
1105         dev->iflink             = 0;
1106         dev->addr_len           = 4;
1107         dev->features           |= NETIF_F_NETNS_LOCAL;
1108 }
1109
1110 static void ipip6_tunnel_init(struct net_device *dev)
1111 {
1112         struct ip_tunnel *tunnel = netdev_priv(dev);
1113
1114         tunnel->dev = dev;
1115         strcpy(tunnel->parms.name, dev->name);
1116
1117         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1118         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1119
1120         ipip6_tunnel_bind_dev(dev);
1121 }
1122
1123 static void ipip6_fb_tunnel_init(struct net_device *dev)
1124 {
1125         struct ip_tunnel *tunnel = netdev_priv(dev);
1126         struct iphdr *iph = &tunnel->parms.iph;
1127         struct net *net = dev_net(dev);
1128         struct sit_net *sitn = net_generic(net, sit_net_id);
1129
1130         tunnel->dev = dev;
1131         strcpy(tunnel->parms.name, dev->name);
1132
1133         iph->version            = 4;
1134         iph->protocol           = IPPROTO_IPV6;
1135         iph->ihl                = 5;
1136         iph->ttl                = 64;
1137
1138         dev_hold(dev);
1139         sitn->tunnels_wc[0]     = tunnel;
1140 }
1141
1142 static struct xfrm_tunnel sit_handler = {
1143         .handler        =       ipip6_rcv,
1144         .err_handler    =       ipip6_err,
1145         .priority       =       1,
1146 };
1147
1148 static void sit_destroy_tunnels(struct sit_net *sitn, struct list_head *head)
1149 {
1150         int prio;
1151
1152         for (prio = 1; prio < 4; prio++) {
1153                 int h;
1154                 for (h = 0; h < HASH_SIZE; h++) {
1155                         struct ip_tunnel *t = sitn->tunnels[prio][h];
1156
1157                         while (t != NULL) {
1158                                 unregister_netdevice_queue(t->dev, head);
1159                                 t = t->next;
1160                         }
1161                 }
1162         }
1163 }
1164
1165 static int sit_init_net(struct net *net)
1166 {
1167         struct sit_net *sitn = net_generic(net, sit_net_id);
1168         int err;
1169
1170         sitn->tunnels[0] = sitn->tunnels_wc;
1171         sitn->tunnels[1] = sitn->tunnels_l;
1172         sitn->tunnels[2] = sitn->tunnels_r;
1173         sitn->tunnels[3] = sitn->tunnels_r_l;
1174
1175         sitn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "sit0",
1176                                            ipip6_tunnel_setup);
1177         if (!sitn->fb_tunnel_dev) {
1178                 err = -ENOMEM;
1179                 goto err_alloc_dev;
1180         }
1181         dev_net_set(sitn->fb_tunnel_dev, net);
1182
1183         ipip6_fb_tunnel_init(sitn->fb_tunnel_dev);
1184         ipip6_tunnel_clone_6rd(sitn->fb_tunnel_dev, sitn);
1185
1186         if ((err = register_netdev(sitn->fb_tunnel_dev)))
1187                 goto err_reg_dev;
1188
1189         return 0;
1190
1191 err_reg_dev:
1192         dev_put(sitn->fb_tunnel_dev);
1193         free_netdev(sitn->fb_tunnel_dev);
1194 err_alloc_dev:
1195         return err;
1196 }
1197
1198 static void sit_exit_net(struct net *net)
1199 {
1200         struct sit_net *sitn = net_generic(net, sit_net_id);
1201         LIST_HEAD(list);
1202
1203         rtnl_lock();
1204         sit_destroy_tunnels(sitn, &list);
1205         unregister_netdevice_queue(sitn->fb_tunnel_dev, &list);
1206         unregister_netdevice_many(&list);
1207         rtnl_unlock();
1208 }
1209
1210 static struct pernet_operations sit_net_ops = {
1211         .init = sit_init_net,
1212         .exit = sit_exit_net,
1213         .id   = &sit_net_id,
1214         .size = sizeof(struct sit_net),
1215 };
1216
1217 static void __exit sit_cleanup(void)
1218 {
1219         xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
1220
1221         unregister_pernet_device(&sit_net_ops);
1222         rcu_barrier(); /* Wait for completion of call_rcu()'s */
1223 }
1224
1225 static int __init sit_init(void)
1226 {
1227         int err;
1228
1229         printk(KERN_INFO "IPv6 over IPv4 tunneling driver\n");
1230
1231         if (xfrm4_tunnel_register(&sit_handler, AF_INET6) < 0) {
1232                 printk(KERN_INFO "sit init: Can't add protocol\n");
1233                 return -EAGAIN;
1234         }
1235
1236         err = register_pernet_device(&sit_net_ops);
1237         if (err < 0)
1238                 xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
1239
1240         return err;
1241 }
1242
1243 module_init(sit_init);
1244 module_exit(sit_cleanup);
1245 MODULE_LICENSE("GPL");
1246 MODULE_ALIAS("sit0");