net/ipv4/ipip.c

   1 /*
   2  *      Linux NET3:     IP/IP protocol decoder.
   3  *
   4  *      Authors:
   5  *              Sam Lantinga (slouken@cs.ucdavis.edu)  02/01/95
   6  *
   7  *      Fixes:
   8  *              Alan Cox        :       Merged and made usable non modular (its so tiny its silly as
   9  *                                      a module taking up 2 pages).
  10  *              Alan Cox        :       Fixed bug with 1.3.18 and IPIP not working (now needs to set skb->h.iph)
  11  *                                      to keep ip_forward happy.
  12  *              Alan Cox        :       More fixes for 1.3.21, and firewall fix. Maybe this will work soon 8).
  13  *              Kai Schulte     :       Fixed #defines for IP_FIREWALL->FIREWALL
  14  *              David Woodhouse :       Perform some basic ICMP handling.
  15  *                                      IPIP Routing without decapsulation.
  16  *              Carlos Picoto   :       GRE over IP support
  17  *              Alexey Kuznetsov:       Reworked. Really, now it is truncated version of ipv4/ip_gre.c.
  18  *                                      I do not want to merge them together.
  19  *
  20  *      This program is free software; you can redistribute it and/or
  21  *      modify it under the terms of the GNU General Public License
  22  *      as published by the Free Software Foundation; either version
  23  *      2 of the License, or (at your option) any later version.
  24  *
  25  */
  26
  27 /* tunnel.c: an IP tunnel driver
  28
  29         The purpose of this driver is to provide an IP tunnel through
  30         which you can tunnel network traffic transparently across subnets.
  31
  32         This was written by looking at Nick Holloway's dummy driver
  33         Thanks for the great code!
  34
  35                 -Sam Lantinga   (slouken@cs.ucdavis.edu)  02/01/95
  36
  37         Minor tweaks:
  38                 Cleaned up the code a little and added some pre-1.3.0 tweaks.
  39                 dev->hard_header/hard_header_len changed to use no headers.
  40                 Comments/bracketing tweaked.
  41                 Made the tunnels use dev->name not tunnel: when error reporting.
  42                 Added tx_dropped stat
  43
  44                 -Alan Cox       (alan@lxorguk.ukuu.org.uk) 21 March 95
  45
  46         Reworked:
  47                 Changed to tunnel to destination gateway in addition to the
  48                         tunnel's pointopoint address
  49                 Almost completely rewritten
  50                 Note:  There is currently no firewall or ICMP handling done.
  51
  52                 -Sam Lantinga   (slouken@cs.ucdavis.edu) 02/13/96
  53
  54 */
  55
  56 /* Things I wish I had known when writing the tunnel driver:
  57
  58         When the tunnel_xmit() function is called, the skb contains the
  59         packet to be sent (plus a great deal of extra info), and dev
  60         contains the tunnel device that _we_ are.
  61
  62         When we are passed a packet, we are expected to fill in the
  63         source address with our source IP address.
  64
  65         What is the proper way to allocate, copy and free a buffer?
  66         After you allocate it, it is a "0 length" chunk of memory
  67         starting at zero.  If you want to add headers to the buffer
  68         later, you'll have to call "skb_reserve(skb, amount)" with
  69         the amount of memory you want reserved.  Then, you call
  70         "skb_put(skb, amount)" with the amount of space you want in
  71         the buffer.  skb_put() returns a pointer to the top (#0) of
  72         that buffer.  skb->len is set to the amount of space you have
  73         "allocated" with skb_put().  You can then write up to skb->len
  74         bytes to that buffer.  If you need more, you can call skb_put()
  75         again with the additional amount of space you need.  You can
  76         find out how much more space you can allocate by calling
  77         "skb_tailroom(skb)".
  78         Now, to add header space, call "skb_push(skb, header_len)".
  79         This creates space at the beginning of the buffer and returns
  80         a pointer to this new space.  If later you need to strip a
  81         header from a buffer, call "skb_pull(skb, header_len)".
  82         skb_headroom() will return how much space is left at the top
  83         of the buffer (before the main data).  Remember, this headroom
  84         space must be reserved before the skb_put() function is called.
  85         */
  86
  87 /*
  88    This version of net/ipv4/ipip.c is cloned of net/ipv4/ip_gre.c
  89
  90    For comments look at net/ipv4/ip_gre.c --ANK
  91  */
  92
  93
  94 #include <linux/capability.h>
  95 #include <linux/module.h>
  96 #include <linux/types.h>
  97 #include <linux/kernel.h>
  98 #include <linux/slab.h>
  99 #include <asm/uaccess.h>
 100 #include <linux/skbuff.h>
 101 #include <linux/netdevice.h>
 102 #include <linux/in.h>
 103 #include <linux/tcp.h>
 104 #include <linux/udp.h>
 105 #include <linux/if_arp.h>
 106 #include <linux/mroute.h>
 107 #include <linux/init.h>
 108 #include <linux/netfilter_ipv4.h>
 109 #include <linux/if_ether.h>
 110
 111 #include <net/sock.h>
 112 #include <net/ip.h>
 113 #include <net/icmp.h>
 114 #include <net/ipip.h>
 115 #include <net/inet_ecn.h>
 116 #include <net/xfrm.h>
 117 #include <net/net_namespace.h>
 118 #include <net/netns/generic.h>
 119
 120 #define HASH_SIZE  16
 121 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
 122
 123 static int ipip_net_id __read_mostly;
 124 struct ipip_net {
 125         struct ip_tunnel __rcu *tunnels_r_l[HASH_SIZE];
 126         struct ip_tunnel __rcu *tunnels_r[HASH_SIZE];
 127         struct ip_tunnel __rcu *tunnels_l[HASH_SIZE];
 128         struct ip_tunnel __rcu *tunnels_wc[1];
 129         struct ip_tunnel __rcu **tunnels[4];
 130
 131         struct net_device *fb_tunnel_dev;
 132 };
 133
 134 static void ipip_tunnel_init(struct net_device *dev);
 135 static void ipip_tunnel_setup(struct net_device *dev);
 136
 137 /*
 138  * Locking : hash tables are protected by RCU and RTNL
 139  */
 140
 141 #define for_each_ip_tunnel_rcu(start) \
 142         for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
 143
 144 static struct ip_tunnel * ipip_tunnel_lookup(struct net *net,
 145                 __be32 remote, __be32 local)
 146 {
 147         unsigned int h0 = HASH(remote);
 148         unsigned int h1 = HASH(local);
 149         struct ip_tunnel *t;
 150         struct ipip_net *ipn = net_generic(net, ipip_net_id);
 151
 152         for_each_ip_tunnel_rcu(ipn->tunnels_r_l[h0 ^ h1])
 153                 if (local == t->parms.iph.saddr &&
 154                     remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
 155                         return t;
 156
 157         for_each_ip_tunnel_rcu(ipn->tunnels_r[h0])
 158                 if (remote == t->parms.iph.daddr && (t->dev->flags&IFF_UP))
 159                         return t;
 160
 161         for_each_ip_tunnel_rcu(ipn->tunnels_l[h1])
 162                 if (local == t->parms.iph.saddr && (t->dev->flags&IFF_UP))
 163                         return t;
 164
 165         t = rcu_dereference(ipn->tunnels_wc[0]);
 166         if (t && (t->dev->flags&IFF_UP))
 167                 return t;
 168         return NULL;
 169 }
 170
 171 static struct ip_tunnel __rcu **__ipip_bucket(struct ipip_net *ipn,
 172                 struct ip_tunnel_parm *parms)
 173 {
 174         __be32 remote = parms->iph.daddr;
 175         __be32 local = parms->iph.saddr;
 176         unsigned int h = 0;
 177         int prio = 0;
 178
 179         if (remote) {
 180                 prio |= 2;
 181                 h ^= HASH(remote);
 182         }
 183         if (local) {
 184                 prio |= 1;
 185                 h ^= HASH(local);
 186         }
 187         return &ipn->tunnels[prio][h];
 188 }
 189
 190 static inline struct ip_tunnel __rcu **ipip_bucket(struct ipip_net *ipn,
 191                 struct ip_tunnel *t)
 192 {
 193         return __ipip_bucket(ipn, &t->parms);
 194 }
 195
 196 static void ipip_tunnel_unlink(struct ipip_net *ipn, struct ip_tunnel *t)
 197 {
 198         struct ip_tunnel __rcu **tp;
 199         struct ip_tunnel *iter;
 200
 201         for (tp = ipip_bucket(ipn, t);
 202              (iter = rtnl_dereference(*tp)) != NULL;
 203              tp = &iter->next) {
 204                 if (t == iter) {
 205                         rcu_assign_pointer(*tp, t->next);
 206                         break;
 207                 }
 208         }
 209 }
 210
 211 static void ipip_tunnel_link(struct ipip_net *ipn, struct ip_tunnel *t)
 212 {
 213         struct ip_tunnel __rcu **tp = ipip_bucket(ipn, t);
 214
 215         rcu_assign_pointer(t->next, rtnl_dereference(*tp));
 216         rcu_assign_pointer(*tp, t);
 217 }
 218
 219 static struct ip_tunnel * ipip_tunnel_locate(struct net *net,
 220                 struct ip_tunnel_parm *parms, int create)
 221 {
 222         __be32 remote = parms->iph.daddr;
 223         __be32 local = parms->iph.saddr;
 224         struct ip_tunnel *t, *nt;
 225         struct ip_tunnel __rcu **tp;
 226         struct net_device *dev;
 227         char name[IFNAMSIZ];
 228         struct ipip_net *ipn = net_generic(net, ipip_net_id);
 229
 230         for (tp = __ipip_bucket(ipn, parms);
 231                  (t = rtnl_dereference(*tp)) != NULL;
 232                  tp = &t->next) {
 233                 if (local == t->parms.iph.saddr && remote == t->parms.iph.daddr)
 234                         return t;
 235         }
 236         if (!create)
 237                 return NULL;
 238
 239         if (parms->name[0])
 240                 strlcpy(name, parms->name, IFNAMSIZ);
 241         else
 242                 sprintf(name, "tunl%%d");
 243
 244         dev = alloc_netdev(sizeof(*t), name, ipip_tunnel_setup);
 245         if (dev == NULL)
 246                 return NULL;
 247
 248         dev_net_set(dev, net);
 249
 250         if (strchr(name, '%')) {
 251                 if (dev_alloc_name(dev, name) < 0)
 252                         goto failed_free;
 253         }
 254
 255         nt = netdev_priv(dev);
 256         nt->parms = *parms;
 257
 258         ipip_tunnel_init(dev);
 259
 260         if (register_netdevice(dev) < 0)
 261                 goto failed_free;
 262
 263         dev_hold(dev);
 264         ipip_tunnel_link(ipn, nt);
 265         return nt;
 266
 267 failed_free:
 268         free_netdev(dev);
 269         return NULL;
 270 }
 271
 272 /* called with RTNL */
 273 static void ipip_tunnel_uninit(struct net_device *dev)
 274 {
 275         struct net *net = dev_net(dev);
 276         struct ipip_net *ipn = net_generic(net, ipip_net_id);
 277
 278         if (dev == ipn->fb_tunnel_dev)
 279                 rcu_assign_pointer(ipn->tunnels_wc[0], NULL);
 280         else
 281                 ipip_tunnel_unlink(ipn, netdev_priv(dev));
 282         dev_put(dev);
 283 }
 284
 285 static int ipip_err(struct sk_buff *skb, u32 info)
 286 {
 287
 288 /* All the routers (except for Linux) return only
 289    8 bytes of packet payload. It means, that precise relaying of
 290    ICMP in the real Internet is absolutely infeasible.
 291  */
 292         struct iphdr *iph = (struct iphdr *)skb->data;
 293         const int type = icmp_hdr(skb)->type;
 294         const int code = icmp_hdr(skb)->code;
 295         struct ip_tunnel *t;
 296         int err;
 297
 298         switch (type) {
 299         default:
 300         case ICMP_PARAMETERPROB:
 301                 return 0;
 302
 303         case ICMP_DEST_UNREACH:
 304                 switch (code) {
 305                 case ICMP_SR_FAILED:
 306                 case ICMP_PORT_UNREACH:
 307                         /* Impossible event. */
 308                         return 0;
 309                 case ICMP_FRAG_NEEDED:
 310                         /* Soft state for pmtu is maintained by IP core. */
 311                         return 0;
 312                 default:
 313                         /* All others are translated to HOST_UNREACH.
 314                            rfc2003 contains "deep thoughts" about NET_UNREACH,
 315                            I believe they are just ether pollution. --ANK
 316                          */
 317                         break;
 318                 }
 319                 break;
 320         case ICMP_TIME_EXCEEDED:
 321                 if (code != ICMP_EXC_TTL)
 322                         return 0;
 323                 break;
 324         }
 325
 326         err = -ENOENT;
 327
 328         rcu_read_lock();
 329         t = ipip_tunnel_lookup(dev_net(skb->dev), iph->daddr, iph->saddr);
 330         if (t == NULL || t->parms.iph.daddr == 0)
 331                 goto out;
 332
 333         err = 0;
 334         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
 335                 goto out;
 336
 337         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
 338                 t->err_count++;
 339         else
 340                 t->err_count = 1;
 341         t->err_time = jiffies;
 342 out:
 343         rcu_read_unlock();
 344         return err;
 345 }
 346
 347 static inline void ipip_ecn_decapsulate(const struct iphdr *outer_iph,
 348                                         struct sk_buff *skb)
 349 {
 350         struct iphdr *inner_iph = ip_hdr(skb);
 351
 352         if (INET_ECN_is_ce(outer_iph->tos))
 353                 IP_ECN_set_ce(inner_iph);
 354 }
 355
 356 static int ipip_rcv(struct sk_buff *skb)
 357 {
 358         struct ip_tunnel *tunnel;
 359         const struct iphdr *iph = ip_hdr(skb);
 360
 361         rcu_read_lock();
 362         if ((tunnel = ipip_tunnel_lookup(dev_net(skb->dev),
 363                                         iph->saddr, iph->daddr)) != NULL) {
 364                 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
 365                         rcu_read_unlock();
 366                         kfree_skb(skb);
 367                         return 0;
 368                 }
 369
 370                 secpath_reset(skb);
 371
 372                 skb->mac_header = skb->network_header;
 373                 skb_reset_network_header(skb);
 374                 skb->protocol = htons(ETH_P_IP);
 375                 skb->pkt_type = PACKET_HOST;
 376
 377                 skb_tunnel_rx(skb, tunnel->dev);
 378
 379                 ipip_ecn_decapsulate(iph, skb);
 380
 381                 if (netif_rx(skb) == NET_RX_DROP)
 382                         tunnel->dev->stats.rx_dropped++;
 383
 384                 rcu_read_unlock();
 385                 return 0;
 386         }
 387         rcu_read_unlock();
 388
 389         return -1;
 390 }
 391
 392 /*
 393  *      This function assumes it is being called from dev_queue_xmit()
 394  *      and that skb is filled properly by that function.
 395  */
 396
 397 static netdev_tx_t ipip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
 398 {
 399         struct ip_tunnel *tunnel = netdev_priv(dev);
 400         struct net_device_stats *stats = &dev->stats;
 401         struct netdev_queue *txq = netdev_get_tx_queue(dev, 0);
 402         struct iphdr  *tiph = &tunnel->parms.iph;
 403         u8     tos = tunnel->parms.iph.tos;
 404         __be16 df = tiph->frag_off;
 405         struct rtable *rt;                      /* Route to the other host */
 406         struct net_device *tdev;                        /* Device to other host */
 407         struct iphdr  *old_iph = ip_hdr(skb);
 408         struct iphdr  *iph;                     /* Our new IP header */
 409         unsigned int max_headroom;              /* The extra header space needed */
 410         __be32 dst = tiph->daddr;
 411         int    mtu;
 412
 413         if (skb->protocol != htons(ETH_P_IP))
 414                 goto tx_error;
 415
 416         if (tos&1)
 417                 tos = old_iph->tos;
 418
 419         if (!dst) {
 420                 /* NBMA tunnel */
 421                 if ((rt = skb_rtable(skb)) == NULL) {
 422                         stats->tx_fifo_errors++;
 423                         goto tx_error;
 424                 }
 425                 if ((dst = rt->rt_gateway) == 0)
 426                         goto tx_error_icmp;
 427         }
 428
 429         {
 430                 struct flowi fl = { .oif = tunnel->parms.link,
 431                                     .nl_u = { .ip4_u =
 432                                               { .daddr = dst,
 433                                                 .saddr = tiph->saddr,
 434                                                 .tos = RT_TOS(tos) } },
 435                                     .proto = IPPROTO_IPIP };
 436                 if (ip_route_output_key(dev_net(dev), &rt, &fl)) {
 437                         stats->tx_carrier_errors++;
 438                         goto tx_error_icmp;
 439                 }
 440         }
 441         tdev = rt->dst.dev;
 442
 443         if (tdev == dev) {
 444                 ip_rt_put(rt);
 445                 stats->collisions++;
 446                 goto tx_error;
 447         }
 448
 449         df |= old_iph->frag_off & htons(IP_DF);
 450
 451         if (df) {
 452                 mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
 453
 454                 if (mtu < 68) {
 455                         stats->collisions++;
 456                         ip_rt_put(rt);
 457                         goto tx_error;
 458                 }
 459
 460                 if (skb_dst(skb))
 461                         skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
 462
 463                 if ((old_iph->frag_off & htons(IP_DF)) &&
 464                     mtu < ntohs(old_iph->tot_len)) {
 465                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
 466                                   htonl(mtu));
 467                         ip_rt_put(rt);
 468                         goto tx_error;
 469                 }
 470         }
 471
 472         if (tunnel->err_count > 0) {
 473                 if (time_before(jiffies,
 474                                 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
 475                         tunnel->err_count--;
 476                         dst_link_failure(skb);
 477                 } else
 478                         tunnel->err_count = 0;
 479         }
 480
 481         /*
 482          * Okay, now see if we can stuff it in the buffer as-is.
 483          */
 484         max_headroom = (LL_RESERVED_SPACE(tdev)+sizeof(struct iphdr));
 485
 486         if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
 487             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
 488                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
 489                 if (!new_skb) {
 490                         ip_rt_put(rt);
 491                         txq->tx_dropped++;
 492                         dev_kfree_skb(skb);
 493                         return NETDEV_TX_OK;
 494                 }
 495                 if (skb->sk)
 496                         skb_set_owner_w(new_skb, skb->sk);
 497                 dev_kfree_skb(skb);
 498                 skb = new_skb;
 499                 old_iph = ip_hdr(skb);
 500         }
 501
 502         skb->transport_header = skb->network_header;
 503         skb_push(skb, sizeof(struct iphdr));
 504         skb_reset_network_header(skb);
 505         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 506         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
 507                               IPSKB_REROUTED);
 508         skb_dst_drop(skb);
 509         skb_dst_set(skb, &rt->dst);
 510
 511         /*
 512          *      Push down and install the IPIP header.
 513          */
 514
 515         iph                     =       ip_hdr(skb);
 516         iph->version            =       4;
 517         iph->ihl                =       sizeof(struct iphdr)>>2;
 518         iph->frag_off           =       df;
 519         iph->protocol           =       IPPROTO_IPIP;
 520         iph->tos                =       INET_ECN_encapsulate(tos, old_iph->tos);
 521         iph->daddr              =       rt->rt_dst;
 522         iph->saddr              =       rt->rt_src;
 523
 524         if ((iph->ttl = tiph->ttl) == 0)
 525                 iph->ttl        =       old_iph->ttl;
 526
 527         nf_reset(skb);
 528
 529         IPTUNNEL_XMIT();
 530         return NETDEV_TX_OK;
 531
 532 tx_error_icmp:
 533         dst_link_failure(skb);
 534 tx_error:
 535         stats->tx_errors++;
 536         dev_kfree_skb(skb);
 537         return NETDEV_TX_OK;
 538 }
 539
 540 static void ipip_tunnel_bind_dev(struct net_device *dev)
 541 {
 542         struct net_device *tdev = NULL;
 543         struct ip_tunnel *tunnel;
 544         struct iphdr *iph;
 545
 546         tunnel = netdev_priv(dev);
 547         iph = &tunnel->parms.iph;
 548
 549         if (iph->daddr) {
 550                 struct flowi fl = { .oif = tunnel->parms.link,
 551                                     .nl_u = { .ip4_u =
 552                                               { .daddr = iph->daddr,
 553                                                 .saddr = iph->saddr,
 554                                                 .tos = RT_TOS(iph->tos) } },
 555                                     .proto = IPPROTO_IPIP };
 556                 struct rtable *rt;
 557                 if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
 558                         tdev = rt->dst.dev;
 559                         ip_rt_put(rt);
 560                 }
 561                 dev->flags |= IFF_POINTOPOINT;
 562         }
 563
 564         if (!tdev && tunnel->parms.link)
 565                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
 566
 567         if (tdev) {
 568                 dev->hard_header_len = tdev->hard_header_len + sizeof(struct iphdr);
 569                 dev->mtu = tdev->mtu - sizeof(struct iphdr);
 570         }
 571         dev->iflink = tunnel->parms.link;
 572 }
 573
 574 static int
 575 ipip_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
 576 {
 577         int err = 0;
 578         struct ip_tunnel_parm p;
 579         struct ip_tunnel *t;
 580         struct net *net = dev_net(dev);
 581         struct ipip_net *ipn = net_generic(net, ipip_net_id);
 582
 583         switch (cmd) {
 584         case SIOCGETTUNNEL:
 585                 t = NULL;
 586                 if (dev == ipn->fb_tunnel_dev) {
 587                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
 588                                 err = -EFAULT;
 589                                 break;
 590                         }
 591                         t = ipip_tunnel_locate(net, &p, 0);
 592                 }
 593                 if (t == NULL)
 594                         t = netdev_priv(dev);
 595                 memcpy(&p, &t->parms, sizeof(p));
 596                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
 597                         err = -EFAULT;
 598                 break;
 599
 600         case SIOCADDTUNNEL:
 601         case SIOCCHGTUNNEL:
 602                 err = -EPERM;
 603                 if (!capable(CAP_NET_ADMIN))
 604                         goto done;
 605
 606                 err = -EFAULT;
 607                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
 608                         goto done;
 609
 610                 err = -EINVAL;
 611                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_IPIP ||
 612                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)))
 613                         goto done;
 614                 if (p.iph.ttl)
 615                         p.iph.frag_off |= htons(IP_DF);
 616
 617                 t = ipip_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
 618
 619                 if (dev != ipn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
 620                         if (t != NULL) {
 621                                 if (t->dev != dev) {
 622                                         err = -EEXIST;
 623                                         break;
 624                                 }
 625                         } else {
 626                                 if (((dev->flags&IFF_POINTOPOINT) && !p.iph.daddr) ||
 627                                     (!(dev->flags&IFF_POINTOPOINT) && p.iph.daddr)) {
 628                                         err = -EINVAL;
 629                                         break;
 630                                 }
 631                                 t = netdev_priv(dev);
 632                                 ipip_tunnel_unlink(ipn, t);
 633                                 t->parms.iph.saddr = p.iph.saddr;
 634                                 t->parms.iph.daddr = p.iph.daddr;
 635                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
 636                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
 637                                 ipip_tunnel_link(ipn, t);
 638                                 netdev_state_change(dev);
 639                         }
 640                 }
 641
 642                 if (t) {
 643                         err = 0;
 644                         if (cmd == SIOCCHGTUNNEL) {
 645                                 t->parms.iph.ttl = p.iph.ttl;
 646                                 t->parms.iph.tos = p.iph.tos;
 647                                 t->parms.iph.frag_off = p.iph.frag_off;
 648                                 if (t->parms.link != p.link) {
 649                                         t->parms.link = p.link;
 650                                         ipip_tunnel_bind_dev(dev);
 651                                         netdev_state_change(dev);
 652                                 }
 653                         }
 654                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
 655                                 err = -EFAULT;
 656                 } else
 657                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
 658                 break;
 659
 660         case SIOCDELTUNNEL:
 661                 err = -EPERM;
 662                 if (!capable(CAP_NET_ADMIN))
 663                         goto done;
 664
 665                 if (dev == ipn->fb_tunnel_dev) {
 666                         err = -EFAULT;
 667                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
 668                                 goto done;
 669                         err = -ENOENT;
 670                         if ((t = ipip_tunnel_locate(net, &p, 0)) == NULL)
 671                                 goto done;
 672                         err = -EPERM;
 673                         if (t->dev == ipn->fb_tunnel_dev)
 674                                 goto done;
 675                         dev = t->dev;
 676                 }
 677                 unregister_netdevice(dev);
 678                 err = 0;
 679                 break;
 680
 681         default:
 682                 err = -EINVAL;
 683         }
 684
 685 done:
 686         return err;
 687 }
 688
 689 static int ipip_tunnel_change_mtu(struct net_device *dev, int new_mtu)
 690 {
 691         if (new_mtu < 68 || new_mtu > 0xFFF8 - sizeof(struct iphdr))
 692                 return -EINVAL;
 693         dev->mtu = new_mtu;
 694         return 0;
 695 }
 696
 697 static const struct net_device_ops ipip_netdev_ops = {
 698         .ndo_uninit     = ipip_tunnel_uninit,
 699         .ndo_start_xmit = ipip_tunnel_xmit,
 700         .ndo_do_ioctl   = ipip_tunnel_ioctl,
 701         .ndo_change_mtu = ipip_tunnel_change_mtu,
 702
 703 };
 704
 705 static void ipip_tunnel_setup(struct net_device *dev)
 706 {
 707         dev->netdev_ops         = &ipip_netdev_ops;
 708         dev->destructor         = free_netdev;
 709
 710         dev->type               = ARPHRD_TUNNEL;
 711         dev->hard_header_len    = LL_MAX_HEADER + sizeof(struct iphdr);
 712         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr);
 713         dev->flags              = IFF_NOARP;
 714         dev->iflink             = 0;
 715         dev->addr_len           = 4;
 716         dev->features           |= NETIF_F_NETNS_LOCAL;
 717         dev->priv_flags         &= ~IFF_XMIT_DST_RELEASE;
 718 }
 719
 720 static void ipip_tunnel_init(struct net_device *dev)
 721 {
 722         struct ip_tunnel *tunnel = netdev_priv(dev);
 723
 724         tunnel->dev = dev;
 725         strcpy(tunnel->parms.name, dev->name);
 726
 727         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
 728         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
 729
 730         ipip_tunnel_bind_dev(dev);
 731 }
 732
 733 static void __net_init ipip_fb_tunnel_init(struct net_device *dev)
 734 {
 735         struct ip_tunnel *tunnel = netdev_priv(dev);
 736         struct iphdr *iph = &tunnel->parms.iph;
 737         struct ipip_net *ipn = net_generic(dev_net(dev), ipip_net_id);
 738
 739         tunnel->dev = dev;
 740         strcpy(tunnel->parms.name, dev->name);
 741
 742         iph->version            = 4;
 743         iph->protocol           = IPPROTO_IPIP;
 744         iph->ihl                = 5;
 745
 746         dev_hold(dev);
 747         rcu_assign_pointer(ipn->tunnels_wc[0], tunnel);
 748 }
 749
 750 static struct xfrm_tunnel ipip_handler __read_mostly = {
 751         .handler        =       ipip_rcv,
 752         .err_handler    =       ipip_err,
 753         .priority       =       1,
 754 };
 755
 756 static const char banner[] __initconst =
 757         KERN_INFO "IPv4 over IPv4 tunneling driver\n";
 758
 759 static void ipip_destroy_tunnels(struct ipip_net *ipn, struct list_head *head)
 760 {
 761         int prio;
 762
 763         for (prio = 1; prio < 4; prio++) {
 764                 int h;
 765                 for (h = 0; h < HASH_SIZE; h++) {
 766                         struct ip_tunnel *t;
 767
 768                         t = rtnl_dereference(ipn->tunnels[prio][h]);
 769                         while (t != NULL) {
 770                                 unregister_netdevice_queue(t->dev, head);
 771                                 t = rtnl_dereference(t->next);
 772                         }
 773                 }
 774         }
 775 }
 776
 777 static int __net_init ipip_init_net(struct net *net)
 778 {
 779         struct ipip_net *ipn = net_generic(net, ipip_net_id);
 780         int err;
 781
 782         ipn->tunnels[0] = ipn->tunnels_wc;
 783         ipn->tunnels[1] = ipn->tunnels_l;
 784         ipn->tunnels[2] = ipn->tunnels_r;
 785         ipn->tunnels[3] = ipn->tunnels_r_l;
 786
 787         ipn->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel),
 788                                            "tunl0",
 789                                            ipip_tunnel_setup);
 790         if (!ipn->fb_tunnel_dev) {
 791                 err = -ENOMEM;
 792                 goto err_alloc_dev;
 793         }
 794         dev_net_set(ipn->fb_tunnel_dev, net);
 795
 796         ipip_fb_tunnel_init(ipn->fb_tunnel_dev);
 797
 798         if ((err = register_netdev(ipn->fb_tunnel_dev)))
 799                 goto err_reg_dev;
 800
 801         return 0;
 802
 803 err_reg_dev:
 804         free_netdev(ipn->fb_tunnel_dev);
 805 err_alloc_dev:
 806         /* nothing */
 807         return err;
 808 }
 809
 810 static void __net_exit ipip_exit_net(struct net *net)
 811 {
 812         struct ipip_net *ipn = net_generic(net, ipip_net_id);
 813         LIST_HEAD(list);
 814
 815         rtnl_lock();
 816         ipip_destroy_tunnels(ipn, &list);
 817         unregister_netdevice_queue(ipn->fb_tunnel_dev, &list);
 818         unregister_netdevice_many(&list);
 819         rtnl_unlock();
 820 }
 821
 822 static struct pernet_operations ipip_net_ops = {
 823         .init = ipip_init_net,
 824         .exit = ipip_exit_net,
 825         .id   = &ipip_net_id,
 826         .size = sizeof(struct ipip_net),
 827 };
 828
 829 static int __init ipip_init(void)
 830 {
 831         int err;
 832
 833         printk(banner);
 834
 835         err = register_pernet_device(&ipip_net_ops);
 836         if (err < 0)
 837                 return err;
 838         err = xfrm4_tunnel_register(&ipip_handler, AF_INET);
 839         if (err < 0) {
 840                 unregister_pernet_device(&ipip_net_ops);
 841                 printk(KERN_INFO "ipip init: can't register tunnel\n");
 842         }
 843         return err;
 844 }
 845
 846 static void __exit ipip_fini(void)
 847 {
 848         if (xfrm4_tunnel_deregister(&ipip_handler, AF_INET))
 849                 printk(KERN_INFO "ipip close: can't deregister tunnel\n");
 850
 851         unregister_pernet_device(&ipip_net_ops);
 852 }
 853
 854 module_init(ipip_init);
 855 module_exit(ipip_fini);
 856 MODULE_LICENSE("GPL");