net/ipv4/ip_gre.c

   1 /*
   2  *      Linux NET3:     GRE over IP protocol decoder.
   3  *
   4  *      Authors: Alexey Kuznetsov (kuznet@ms2.inr.ac.ru)
   5  *
   6  *      This program is free software; you can redistribute it and/or
   7  *      modify it under the terms of the GNU General Public License
   8  *      as published by the Free Software Foundation; either version
   9  *      2 of the License, or (at your option) any later version.
  10  *
  11  */
  12
  13 #include <linux/capability.h>
  14 #include <linux/module.h>
  15 #include <linux/types.h>
  16 #include <linux/kernel.h>
  17 #include <linux/slab.h>
  18 #include <asm/uaccess.h>
  19 #include <linux/skbuff.h>
  20 #include <linux/netdevice.h>
  21 #include <linux/in.h>
  22 #include <linux/tcp.h>
  23 #include <linux/udp.h>
  24 #include <linux/if_arp.h>
  25 #include <linux/mroute.h>
  26 #include <linux/init.h>
  27 #include <linux/in6.h>
  28 #include <linux/inetdevice.h>
  29 #include <linux/igmp.h>
  30 #include <linux/netfilter_ipv4.h>
  31 #include <linux/etherdevice.h>
  32 #include <linux/if_ether.h>
  33
  34 #include <net/sock.h>
  35 #include <net/ip.h>
  36 #include <net/icmp.h>
  37 #include <net/protocol.h>
  38 #include <net/ipip.h>
  39 #include <net/arp.h>
  40 #include <net/checksum.h>
  41 #include <net/dsfield.h>
  42 #include <net/inet_ecn.h>
  43 #include <net/xfrm.h>
  44 #include <net/net_namespace.h>
  45 #include <net/netns/generic.h>
  46 #include <net/rtnetlink.h>
  47 #include <net/gre.h>
  48
  49 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
  50 #include <net/ipv6.h>
  51 #include <net/ip6_fib.h>
  52 #include <net/ip6_route.h>
  53 #endif
  54
  55 /*
  56    Problems & solutions
  57    --------------------
  58
  59    1. The most important issue is detecting local dead loops.
  60    They would cause complete host lockup in transmit, which
  61    would be "resolved" by stack overflow or, if queueing is enabled,
  62    with infinite looping in net_bh.
  63
  64    We cannot track such dead loops during route installation,
  65    it is infeasible task. The most general solutions would be
  66    to keep skb->encapsulation counter (sort of local ttl),
  67    and silently drop packet when it expires. It is a good
  68    solution, but it supposes maintaing new variable in ALL
  69    skb, even if no tunneling is used.
  70
  71    Current solution: xmit_recursion breaks dead loops. This is a percpu
  72    counter, since when we enter the first ndo_xmit(), cpu migration is
  73    forbidden. We force an exit if this counter reaches RECURSION_LIMIT
  74
  75    2. Networking dead loops would not kill routers, but would really
  76    kill network. IP hop limit plays role of "t->recursion" in this case,
  77    if we copy it from packet being encapsulated to upper header.
  78    It is very good solution, but it introduces two problems:
  79
  80    - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
  81      do not work over tunnels.
  82    - traceroute does not work. I planned to relay ICMP from tunnel,
  83      so that this problem would be solved and traceroute output
  84      would even more informative. This idea appeared to be wrong:
  85      only Linux complies to rfc1812 now (yes, guys, Linux is the only
  86      true router now :-)), all routers (at least, in neighbourhood of mine)
  87      return only 8 bytes of payload. It is the end.
  88
  89    Hence, if we want that OSPF worked or traceroute said something reasonable,
  90    we should search for another solution.
  91
  92    One of them is to parse packet trying to detect inner encapsulation
  93    made by our node. It is difficult or even impossible, especially,
  94    taking into account fragmentation. TO be short, tt is not solution at all.
  95
  96    Current solution: The solution was UNEXPECTEDLY SIMPLE.
  97    We force DF flag on tunnels with preconfigured hop limit,
  98    that is ALL. :-) Well, it does not remove the problem completely,
  99    but exponential growth of network traffic is changed to linear
 100    (branches, that exceed pmtu are pruned) and tunnel mtu
 101    fastly degrades to value <68, where looping stops.
 102    Yes, it is not good if there exists a router in the loop,
 103    which does not force DF, even when encapsulating packets have DF set.
 104    But it is not our problem! Nobody could accuse us, we made
 105    all that we could make. Even if it is your gated who injected
 106    fatal route to network, even if it were you who configured
 107    fatal static route: you are innocent. :-)
 108
 109
 110
 111    3. Really, ipv4/ipip.c, ipv4/ip_gre.c and ipv6/sit.c contain
 112    practically identical code. It would be good to glue them
 113    together, but it is not very evident, how to make them modular.
 114    sit is integral part of IPv6, ipip and gre are naturally modular.
 115    We could extract common parts (hash table, ioctl etc)
 116    to a separate module (ip_tunnel.c).
 117
 118    Alexey Kuznetsov.
 119  */
 120
 121 static struct rtnl_link_ops ipgre_link_ops __read_mostly;
 122 static int ipgre_tunnel_init(struct net_device *dev);
 123 static void ipgre_tunnel_setup(struct net_device *dev);
 124 static int ipgre_tunnel_bind_dev(struct net_device *dev);
 125
 126 /* Fallback tunnel: no source, no destination, no key, no options */
 127
 128 #define HASH_SIZE  16
 129
 130 static int ipgre_net_id __read_mostly;
 131 struct ipgre_net {
 132         struct ip_tunnel __rcu *tunnels[4][HASH_SIZE];
 133
 134         struct net_device *fb_tunnel_dev;
 135 };
 136
 137 /* Tunnel hash table */
 138
 139 /*
 140    4 hash tables:
 141
 142    3: (remote,local)
 143    2: (remote,*)
 144    1: (*,local)
 145    0: (*,*)
 146
 147    We require exact key match i.e. if a key is present in packet
 148    it will match only tunnel with the same key; if it is not present,
 149    it will match only keyless tunnel.
 150
 151    All keysless packets, if not matched configured keyless tunnels
 152    will match fallback tunnel.
 153  */
 154
 155 #define HASH(addr) (((__force u32)addr^((__force u32)addr>>4))&0xF)
 156
 157 #define tunnels_r_l     tunnels[3]
 158 #define tunnels_r       tunnels[2]
 159 #define tunnels_l       tunnels[1]
 160 #define tunnels_wc      tunnels[0]
 161 /*
 162  * Locking : hash tables are protected by RCU and RTNL
 163  */
 164
 165 #define for_each_ip_tunnel_rcu(start) \
 166         for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
 167
 168 /* often modified stats are per cpu, other are shared (netdev->stats) */
 169 struct pcpu_tstats {
 170         unsigned long   rx_packets;
 171         unsigned long   rx_bytes;
 172         unsigned long   tx_packets;
 173         unsigned long   tx_bytes;
 174 };
 175
 176 static struct net_device_stats *ipgre_get_stats(struct net_device *dev)
 177 {
 178         struct pcpu_tstats sum = { 0 };
 179         int i;
 180
 181         for_each_possible_cpu(i) {
 182                 const struct pcpu_tstats *tstats = per_cpu_ptr(dev->tstats, i);
 183
 184                 sum.rx_packets += tstats->rx_packets;
 185                 sum.rx_bytes   += tstats->rx_bytes;
 186                 sum.tx_packets += tstats->tx_packets;
 187                 sum.tx_bytes   += tstats->tx_bytes;
 188         }
 189         dev->stats.rx_packets = sum.rx_packets;
 190         dev->stats.rx_bytes   = sum.rx_bytes;
 191         dev->stats.tx_packets = sum.tx_packets;
 192         dev->stats.tx_bytes   = sum.tx_bytes;
 193         return &dev->stats;
 194 }
 195
 196 /* Given src, dst and key, find appropriate for input tunnel. */
 197
 198 static struct ip_tunnel * ipgre_tunnel_lookup(struct net_device *dev,
 199                                               __be32 remote, __be32 local,
 200                                               __be32 key, __be16 gre_proto)
 201 {
 202         struct net *net = dev_net(dev);
 203         int link = dev->ifindex;
 204         unsigned int h0 = HASH(remote);
 205         unsigned int h1 = HASH(key);
 206         struct ip_tunnel *t, *cand = NULL;
 207         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
 208         int dev_type = (gre_proto == htons(ETH_P_TEB)) ?
 209                        ARPHRD_ETHER : ARPHRD_IPGRE;
 210         int score, cand_score = 4;
 211
 212         for_each_ip_tunnel_rcu(ign->tunnels_r_l[h0 ^ h1]) {
 213                 if (local != t->parms.iph.saddr ||
 214                     remote != t->parms.iph.daddr ||
 215                     key != t->parms.i_key ||
 216                     !(t->dev->flags & IFF_UP))
 217                         continue;
 218
 219                 if (t->dev->type != ARPHRD_IPGRE &&
 220                     t->dev->type != dev_type)
 221                         continue;
 222
 223                 score = 0;
 224                 if (t->parms.link != link)
 225                         score |= 1;
 226                 if (t->dev->type != dev_type)
 227                         score |= 2;
 228                 if (score == 0)
 229                         return t;
 230
 231                 if (score < cand_score) {
 232                         cand = t;
 233                         cand_score = score;
 234                 }
 235         }
 236
 237         for_each_ip_tunnel_rcu(ign->tunnels_r[h0 ^ h1]) {
 238                 if (remote != t->parms.iph.daddr ||
 239                     key != t->parms.i_key ||
 240                     !(t->dev->flags & IFF_UP))
 241                         continue;
 242
 243                 if (t->dev->type != ARPHRD_IPGRE &&
 244                     t->dev->type != dev_type)
 245                         continue;
 246
 247                 score = 0;
 248                 if (t->parms.link != link)
 249                         score |= 1;
 250                 if (t->dev->type != dev_type)
 251                         score |= 2;
 252                 if (score == 0)
 253                         return t;
 254
 255                 if (score < cand_score) {
 256                         cand = t;
 257                         cand_score = score;
 258                 }
 259         }
 260
 261         for_each_ip_tunnel_rcu(ign->tunnels_l[h1]) {
 262                 if ((local != t->parms.iph.saddr &&
 263                      (local != t->parms.iph.daddr ||
 264                       !ipv4_is_multicast(local))) ||
 265                     key != t->parms.i_key ||
 266                     !(t->dev->flags & IFF_UP))
 267                         continue;
 268
 269                 if (t->dev->type != ARPHRD_IPGRE &&
 270                     t->dev->type != dev_type)
 271                         continue;
 272
 273                 score = 0;
 274                 if (t->parms.link != link)
 275                         score |= 1;
 276                 if (t->dev->type != dev_type)
 277                         score |= 2;
 278                 if (score == 0)
 279                         return t;
 280
 281                 if (score < cand_score) {
 282                         cand = t;
 283                         cand_score = score;
 284                 }
 285         }
 286
 287         for_each_ip_tunnel_rcu(ign->tunnels_wc[h1]) {
 288                 if (t->parms.i_key != key ||
 289                     !(t->dev->flags & IFF_UP))
 290                         continue;
 291
 292                 if (t->dev->type != ARPHRD_IPGRE &&
 293                     t->dev->type != dev_type)
 294                         continue;
 295
 296                 score = 0;
 297                 if (t->parms.link != link)
 298                         score |= 1;
 299                 if (t->dev->type != dev_type)
 300                         score |= 2;
 301                 if (score == 0)
 302                         return t;
 303
 304                 if (score < cand_score) {
 305                         cand = t;
 306                         cand_score = score;
 307                 }
 308         }
 309
 310         if (cand != NULL)
 311                 return cand;
 312
 313         dev = ign->fb_tunnel_dev;
 314         if (dev->flags & IFF_UP)
 315                 return netdev_priv(dev);
 316
 317         return NULL;
 318 }
 319
 320 static struct ip_tunnel __rcu **__ipgre_bucket(struct ipgre_net *ign,
 321                 struct ip_tunnel_parm *parms)
 322 {
 323         __be32 remote = parms->iph.daddr;
 324         __be32 local = parms->iph.saddr;
 325         __be32 key = parms->i_key;
 326         unsigned int h = HASH(key);
 327         int prio = 0;
 328
 329         if (local)
 330                 prio |= 1;
 331         if (remote && !ipv4_is_multicast(remote)) {
 332                 prio |= 2;
 333                 h ^= HASH(remote);
 334         }
 335
 336         return &ign->tunnels[prio][h];
 337 }
 338
 339 static inline struct ip_tunnel __rcu **ipgre_bucket(struct ipgre_net *ign,
 340                 struct ip_tunnel *t)
 341 {
 342         return __ipgre_bucket(ign, &t->parms);
 343 }
 344
 345 static void ipgre_tunnel_link(struct ipgre_net *ign, struct ip_tunnel *t)
 346 {
 347         struct ip_tunnel __rcu **tp = ipgre_bucket(ign, t);
 348
 349         rcu_assign_pointer(t->next, rtnl_dereference(*tp));
 350         rcu_assign_pointer(*tp, t);
 351 }
 352
 353 static void ipgre_tunnel_unlink(struct ipgre_net *ign, struct ip_tunnel *t)
 354 {
 355         struct ip_tunnel __rcu **tp;
 356         struct ip_tunnel *iter;
 357
 358         for (tp = ipgre_bucket(ign, t);
 359              (iter = rtnl_dereference(*tp)) != NULL;
 360              tp = &iter->next) {
 361                 if (t == iter) {
 362                         rcu_assign_pointer(*tp, t->next);
 363                         break;
 364                 }
 365         }
 366 }
 367
 368 static struct ip_tunnel *ipgre_tunnel_find(struct net *net,
 369                                            struct ip_tunnel_parm *parms,
 370                                            int type)
 371 {
 372         __be32 remote = parms->iph.daddr;
 373         __be32 local = parms->iph.saddr;
 374         __be32 key = parms->i_key;
 375         int link = parms->link;
 376         struct ip_tunnel *t;
 377         struct ip_tunnel __rcu **tp;
 378         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
 379
 380         for (tp = __ipgre_bucket(ign, parms);
 381              (t = rtnl_dereference(*tp)) != NULL;
 382              tp = &t->next)
 383                 if (local == t->parms.iph.saddr &&
 384                     remote == t->parms.iph.daddr &&
 385                     key == t->parms.i_key &&
 386                     link == t->parms.link &&
 387                     type == t->dev->type)
 388                         break;
 389
 390         return t;
 391 }
 392
 393 static struct ip_tunnel *ipgre_tunnel_locate(struct net *net,
 394                 struct ip_tunnel_parm *parms, int create)
 395 {
 396         struct ip_tunnel *t, *nt;
 397         struct net_device *dev;
 398         char name[IFNAMSIZ];
 399         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
 400
 401         t = ipgre_tunnel_find(net, parms, ARPHRD_IPGRE);
 402         if (t || !create)
 403                 return t;
 404
 405         if (parms->name[0])
 406                 strlcpy(name, parms->name, IFNAMSIZ);
 407         else
 408                 sprintf(name, "gre%%d");
 409
 410         dev = alloc_netdev(sizeof(*t), name, ipgre_tunnel_setup);
 411         if (!dev)
 412           return NULL;
 413
 414         dev_net_set(dev, net);
 415
 416         if (strchr(name, '%')) {
 417                 if (dev_alloc_name(dev, name) < 0)
 418                         goto failed_free;
 419         }
 420
 421         nt = netdev_priv(dev);
 422         nt->parms = *parms;
 423         dev->rtnl_link_ops = &ipgre_link_ops;
 424
 425         dev->mtu = ipgre_tunnel_bind_dev(dev);
 426
 427         if (register_netdevice(dev) < 0)
 428                 goto failed_free;
 429
 430         dev_hold(dev);
 431         ipgre_tunnel_link(ign, nt);
 432         return nt;
 433
 434 failed_free:
 435         free_netdev(dev);
 436         return NULL;
 437 }
 438
 439 static void ipgre_tunnel_uninit(struct net_device *dev)
 440 {
 441         struct net *net = dev_net(dev);
 442         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
 443
 444         ipgre_tunnel_unlink(ign, netdev_priv(dev));
 445         dev_put(dev);
 446 }
 447
 448
 449 static void ipgre_err(struct sk_buff *skb, u32 info)
 450 {
 451
 452 /* All the routers (except for Linux) return only
 453    8 bytes of packet payload. It means, that precise relaying of
 454    ICMP in the real Internet is absolutely infeasible.
 455
 456    Moreover, Cisco "wise men" put GRE key to the third word
 457    in GRE header. It makes impossible maintaining even soft state for keyed
 458    GRE tunnels with enabled checksum. Tell them "thank you".
 459
 460    Well, I wonder, rfc1812 was written by Cisco employee,
 461    what the hell these idiots break standrads established
 462    by themself???
 463  */
 464
 465         struct iphdr *iph = (struct iphdr *)skb->data;
 466         __be16       *p = (__be16*)(skb->data+(iph->ihl<<2));
 467         int grehlen = (iph->ihl<<2) + 4;
 468         const int type = icmp_hdr(skb)->type;
 469         const int code = icmp_hdr(skb)->code;
 470         struct ip_tunnel *t;
 471         __be16 flags;
 472
 473         flags = p[0];
 474         if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
 475                 if (flags&(GRE_VERSION|GRE_ROUTING))
 476                         return;
 477                 if (flags&GRE_KEY) {
 478                         grehlen += 4;
 479                         if (flags&GRE_CSUM)
 480                                 grehlen += 4;
 481                 }
 482         }
 483
 484         /* If only 8 bytes returned, keyed message will be dropped here */
 485         if (skb_headlen(skb) < grehlen)
 486                 return;
 487
 488         switch (type) {
 489         default:
 490         case ICMP_PARAMETERPROB:
 491                 return;
 492
 493         case ICMP_DEST_UNREACH:
 494                 switch (code) {
 495                 case ICMP_SR_FAILED:
 496                 case ICMP_PORT_UNREACH:
 497                         /* Impossible event. */
 498                         return;
 499                 case ICMP_FRAG_NEEDED:
 500                         /* Soft state for pmtu is maintained by IP core. */
 501                         return;
 502                 default:
 503                         /* All others are translated to HOST_UNREACH.
 504                            rfc2003 contains "deep thoughts" about NET_UNREACH,
 505                            I believe they are just ether pollution. --ANK
 506                          */
 507                         break;
 508                 }
 509                 break;
 510         case ICMP_TIME_EXCEEDED:
 511                 if (code != ICMP_EXC_TTL)
 512                         return;
 513                 break;
 514         }
 515
 516         rcu_read_lock();
 517         t = ipgre_tunnel_lookup(skb->dev, iph->daddr, iph->saddr,
 518                                 flags & GRE_KEY ?
 519                                 *(((__be32 *)p) + (grehlen / 4) - 1) : 0,
 520                                 p[1]);
 521         if (t == NULL || t->parms.iph.daddr == 0 ||
 522             ipv4_is_multicast(t->parms.iph.daddr))
 523                 goto out;
 524
 525         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
 526                 goto out;
 527
 528         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
 529                 t->err_count++;
 530         else
 531                 t->err_count = 1;
 532         t->err_time = jiffies;
 533 out:
 534         rcu_read_unlock();
 535 }
 536
 537 static inline void ipgre_ecn_decapsulate(struct iphdr *iph, struct sk_buff *skb)
 538 {
 539         if (INET_ECN_is_ce(iph->tos)) {
 540                 if (skb->protocol == htons(ETH_P_IP)) {
 541                         IP_ECN_set_ce(ip_hdr(skb));
 542                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
 543                         IP6_ECN_set_ce(ipv6_hdr(skb));
 544                 }
 545         }
 546 }
 547
 548 static inline u8
 549 ipgre_ecn_encapsulate(u8 tos, struct iphdr *old_iph, struct sk_buff *skb)
 550 {
 551         u8 inner = 0;
 552         if (skb->protocol == htons(ETH_P_IP))
 553                 inner = old_iph->tos;
 554         else if (skb->protocol == htons(ETH_P_IPV6))
 555                 inner = ipv6_get_dsfield((struct ipv6hdr *)old_iph);
 556         return INET_ECN_encapsulate(tos, inner);
 557 }
 558
 559 static int ipgre_rcv(struct sk_buff *skb)
 560 {
 561         struct iphdr *iph;
 562         u8     *h;
 563         __be16    flags;
 564         __sum16   csum = 0;
 565         __be32 key = 0;
 566         u32    seqno = 0;
 567         struct ip_tunnel *tunnel;
 568         int    offset = 4;
 569         __be16 gre_proto;
 570
 571         if (!pskb_may_pull(skb, 16))
 572                 goto drop_nolock;
 573
 574         iph = ip_hdr(skb);
 575         h = skb->data;
 576         flags = *(__be16*)h;
 577
 578         if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
 579                 /* - Version must be 0.
 580                    - We do not support routing headers.
 581                  */
 582                 if (flags&(GRE_VERSION|GRE_ROUTING))
 583                         goto drop_nolock;
 584
 585                 if (flags&GRE_CSUM) {
 586                         switch (skb->ip_summed) {
 587                         case CHECKSUM_COMPLETE:
 588                                 csum = csum_fold(skb->csum);
 589                                 if (!csum)
 590                                         break;
 591                                 /* fall through */
 592                         case CHECKSUM_NONE:
 593                                 skb->csum = 0;
 594                                 csum = __skb_checksum_complete(skb);
 595                                 skb->ip_summed = CHECKSUM_COMPLETE;
 596                         }
 597                         offset += 4;
 598                 }
 599                 if (flags&GRE_KEY) {
 600                         key = *(__be32*)(h + offset);
 601                         offset += 4;
 602                 }
 603                 if (flags&GRE_SEQ) {
 604                         seqno = ntohl(*(__be32*)(h + offset));
 605                         offset += 4;
 606                 }
 607         }
 608
 609         gre_proto = *(__be16 *)(h + 2);
 610
 611         rcu_read_lock();
 612         if ((tunnel = ipgre_tunnel_lookup(skb->dev,
 613                                           iph->saddr, iph->daddr, key,
 614                                           gre_proto))) {
 615                 struct pcpu_tstats *tstats;
 616
 617                 secpath_reset(skb);
 618
 619                 skb->protocol = gre_proto;
 620                 /* WCCP version 1 and 2 protocol decoding.
 621                  * - Change protocol to IP
 622                  * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
 623                  */
 624                 if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) {
 625                         skb->protocol = htons(ETH_P_IP);
 626                         if ((*(h + offset) & 0xF0) != 0x40)
 627                                 offset += 4;
 628                 }
 629
 630                 skb->mac_header = skb->network_header;
 631                 __pskb_pull(skb, offset);
 632                 skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
 633                 skb->pkt_type = PACKET_HOST;
 634 #ifdef CONFIG_NET_IPGRE_BROADCAST
 635                 if (ipv4_is_multicast(iph->daddr)) {
 636                         /* Looped back packet, drop it! */
 637                         if (skb_rtable(skb)->fl.iif == 0)
 638                                 goto drop;
 639                         tunnel->dev->stats.multicast++;
 640                         skb->pkt_type = PACKET_BROADCAST;
 641                 }
 642 #endif
 643
 644                 if (((flags&GRE_CSUM) && csum) ||
 645                     (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
 646                         tunnel->dev->stats.rx_crc_errors++;
 647                         tunnel->dev->stats.rx_errors++;
 648                         goto drop;
 649                 }
 650                 if (tunnel->parms.i_flags&GRE_SEQ) {
 651                         if (!(flags&GRE_SEQ) ||
 652                             (tunnel->i_seqno && (s32)(seqno - tunnel->i_seqno) < 0)) {
 653                                 tunnel->dev->stats.rx_fifo_errors++;
 654                                 tunnel->dev->stats.rx_errors++;
 655                                 goto drop;
 656                         }
 657                         tunnel->i_seqno = seqno + 1;
 658                 }
 659
 660                 /* Warning: All skb pointers will be invalidated! */
 661                 if (tunnel->dev->type == ARPHRD_ETHER) {
 662                         if (!pskb_may_pull(skb, ETH_HLEN)) {
 663                                 tunnel->dev->stats.rx_length_errors++;
 664                                 tunnel->dev->stats.rx_errors++;
 665                                 goto drop;
 666                         }
 667
 668                         iph = ip_hdr(skb);
 669                         skb->protocol = eth_type_trans(skb, tunnel->dev);
 670                         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 671                 }
 672
 673                 tstats = this_cpu_ptr(tunnel->dev->tstats);
 674                 tstats->rx_packets++;
 675                 tstats->rx_bytes += skb->len;
 676
 677                 __skb_tunnel_rx(skb, tunnel->dev);
 678
 679                 skb_reset_network_header(skb);
 680                 ipgre_ecn_decapsulate(iph, skb);
 681
 682                 netif_rx(skb);
 683
 684                 rcu_read_unlock();
 685                 return 0;
 686         }
 687         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 688
 689 drop:
 690         rcu_read_unlock();
 691 drop_nolock:
 692         kfree_skb(skb);
 693         return 0;
 694 }
 695
 696 static netdev_tx_t ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev)
 697 {
 698         struct ip_tunnel *tunnel = netdev_priv(dev);
 699         struct pcpu_tstats *tstats;
 700         struct iphdr  *old_iph = ip_hdr(skb);
 701         struct iphdr  *tiph;
 702         u8     tos;
 703         __be16 df;
 704         struct rtable *rt;                      /* Route to the other host */
 705         struct net_device *tdev;                /* Device to other host */
 706         struct iphdr  *iph;                     /* Our new IP header */
 707         unsigned int max_headroom;              /* The extra header space needed */
 708         int    gre_hlen;
 709         __be32 dst;
 710         int    mtu;
 711
 712         if (dev->type == ARPHRD_ETHER)
 713                 IPCB(skb)->flags = 0;
 714
 715         if (dev->header_ops && dev->type == ARPHRD_IPGRE) {
 716                 gre_hlen = 0;
 717                 tiph = (struct iphdr *)skb->data;
 718         } else {
 719                 gre_hlen = tunnel->hlen;
 720                 tiph = &tunnel->parms.iph;
 721         }
 722
 723         if ((dst = tiph->daddr) == 0) {
 724                 /* NBMA tunnel */
 725
 726                 if (skb_dst(skb) == NULL) {
 727                         dev->stats.tx_fifo_errors++;
 728                         goto tx_error;
 729                 }
 730
 731                 if (skb->protocol == htons(ETH_P_IP)) {
 732                         rt = skb_rtable(skb);
 733                         if ((dst = rt->rt_gateway) == 0)
 734                                 goto tx_error_icmp;
 735                 }
 736 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 737                 else if (skb->protocol == htons(ETH_P_IPV6)) {
 738                         struct in6_addr *addr6;
 739                         int addr_type;
 740                         struct neighbour *neigh = skb_dst(skb)->neighbour;
 741
 742                         if (neigh == NULL)
 743                                 goto tx_error;
 744
 745                         addr6 = (struct in6_addr *)&neigh->primary_key;
 746                         addr_type = ipv6_addr_type(addr6);
 747
 748                         if (addr_type == IPV6_ADDR_ANY) {
 749                                 addr6 = &ipv6_hdr(skb)->daddr;
 750                                 addr_type = ipv6_addr_type(addr6);
 751                         }
 752
 753                         if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
 754                                 goto tx_error_icmp;
 755
 756                         dst = addr6->s6_addr32[3];
 757                 }
 758 #endif
 759                 else
 760                         goto tx_error;
 761         }
 762
 763         tos = tiph->tos;
 764         if (tos == 1) {
 765                 tos = 0;
 766                 if (skb->protocol == htons(ETH_P_IP))
 767                         tos = old_iph->tos;
 768                 else if (skb->protocol == htons(ETH_P_IPV6))
 769                         tos = ipv6_get_dsfield((struct ipv6hdr *)old_iph);
 770         }
 771
 772         {
 773                 struct flowi fl = {
 774                         .oif = tunnel->parms.link,
 775                         .nl_u = {
 776                                 .ip4_u = {
 777                                         .daddr = dst,
 778                                         .saddr = tiph->saddr,
 779                                         .tos = RT_TOS(tos)
 780                                 }
 781                         },
 782                         .proto = IPPROTO_GRE
 783                 }
 784 ;
 785                 if (ip_route_output_key(dev_net(dev), &rt, &fl)) {
 786                         dev->stats.tx_carrier_errors++;
 787                         goto tx_error;
 788                 }
 789         }
 790         tdev = rt->dst.dev;
 791
 792         if (tdev == dev) {
 793                 ip_rt_put(rt);
 794                 dev->stats.collisions++;
 795                 goto tx_error;
 796         }
 797
 798         df = tiph->frag_off;
 799         if (df)
 800                 mtu = dst_mtu(&rt->dst) - dev->hard_header_len - tunnel->hlen;
 801         else
 802                 mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
 803
 804         if (skb_dst(skb))
 805                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
 806
 807         if (skb->protocol == htons(ETH_P_IP)) {
 808                 df |= (old_iph->frag_off&htons(IP_DF));
 809
 810                 if ((old_iph->frag_off&htons(IP_DF)) &&
 811                     mtu < ntohs(old_iph->tot_len)) {
 812                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
 813                         ip_rt_put(rt);
 814                         goto tx_error;
 815                 }
 816         }
 817 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 818         else if (skb->protocol == htons(ETH_P_IPV6)) {
 819                 struct rt6_info *rt6 = (struct rt6_info *)skb_dst(skb);
 820
 821                 if (rt6 && mtu < dst_mtu(skb_dst(skb)) && mtu >= IPV6_MIN_MTU) {
 822                         if ((tunnel->parms.iph.daddr &&
 823                              !ipv4_is_multicast(tunnel->parms.iph.daddr)) ||
 824                             rt6->rt6i_dst.plen == 128) {
 825                                 rt6->rt6i_flags |= RTF_MODIFIED;
 826                                 skb_dst(skb)->metrics[RTAX_MTU-1] = mtu;
 827                         }
 828                 }
 829
 830                 if (mtu >= IPV6_MIN_MTU && mtu < skb->len - tunnel->hlen + gre_hlen) {
 831                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 832                         ip_rt_put(rt);
 833                         goto tx_error;
 834                 }
 835         }
 836 #endif
 837
 838         if (tunnel->err_count > 0) {
 839                 if (time_before(jiffies,
 840                                 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
 841                         tunnel->err_count--;
 842
 843                         dst_link_failure(skb);
 844                 } else
 845                         tunnel->err_count = 0;
 846         }
 847
 848         max_headroom = LL_RESERVED_SPACE(tdev) + gre_hlen + rt->dst.header_len;
 849
 850         if (skb_headroom(skb) < max_headroom || skb_shared(skb)||
 851             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
 852                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
 853                 if (max_headroom > dev->needed_headroom)
 854                         dev->needed_headroom = max_headroom;
 855                 if (!new_skb) {
 856                         ip_rt_put(rt);
 857                         dev->stats.tx_dropped++;
 858                         dev_kfree_skb(skb);
 859                         return NETDEV_TX_OK;
 860                 }
 861                 if (skb->sk)
 862                         skb_set_owner_w(new_skb, skb->sk);
 863                 dev_kfree_skb(skb);
 864                 skb = new_skb;
 865                 old_iph = ip_hdr(skb);
 866         }
 867
 868         skb_reset_transport_header(skb);
 869         skb_push(skb, gre_hlen);
 870         skb_reset_network_header(skb);
 871         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 872         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
 873                               IPSKB_REROUTED);
 874         skb_dst_drop(skb);
 875         skb_dst_set(skb, &rt->dst);
 876
 877         /*
 878          *      Push down and install the IPIP header.
 879          */
 880
 881         iph                     =       ip_hdr(skb);
 882         iph->version            =       4;
 883         iph->ihl                =       sizeof(struct iphdr) >> 2;
 884         iph->frag_off           =       df;
 885         iph->protocol           =       IPPROTO_GRE;
 886         iph->tos                =       ipgre_ecn_encapsulate(tos, old_iph, skb);
 887         iph->daddr              =       rt->rt_dst;
 888         iph->saddr              =       rt->rt_src;
 889
 890         if ((iph->ttl = tiph->ttl) == 0) {
 891                 if (skb->protocol == htons(ETH_P_IP))
 892                         iph->ttl = old_iph->ttl;
 893 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 894                 else if (skb->protocol == htons(ETH_P_IPV6))
 895                         iph->ttl = ((struct ipv6hdr *)old_iph)->hop_limit;
 896 #endif
 897                 else
 898                         iph->ttl = dst_metric(&rt->dst, RTAX_HOPLIMIT);
 899         }
 900
 901         ((__be16 *)(iph + 1))[0] = tunnel->parms.o_flags;
 902         ((__be16 *)(iph + 1))[1] = (dev->type == ARPHRD_ETHER) ?
 903                                    htons(ETH_P_TEB) : skb->protocol;
 904
 905         if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
 906                 __be32 *ptr = (__be32*)(((u8*)iph) + tunnel->hlen - 4);
 907
 908                 if (tunnel->parms.o_flags&GRE_SEQ) {
 909                         ++tunnel->o_seqno;
 910                         *ptr = htonl(tunnel->o_seqno);
 911                         ptr--;
 912                 }
 913                 if (tunnel->parms.o_flags&GRE_KEY) {
 914                         *ptr = tunnel->parms.o_key;
 915                         ptr--;
 916                 }
 917                 if (tunnel->parms.o_flags&GRE_CSUM) {
 918                         *ptr = 0;
 919                         *(__sum16*)ptr = ip_compute_csum((void*)(iph+1), skb->len - sizeof(struct iphdr));
 920                 }
 921         }
 922
 923         nf_reset(skb);
 924         tstats = this_cpu_ptr(dev->tstats);
 925         __IPTUNNEL_XMIT(tstats, &dev->stats);
 926         return NETDEV_TX_OK;
 927
 928 tx_error_icmp:
 929         dst_link_failure(skb);
 930
 931 tx_error:
 932         dev->stats.tx_errors++;
 933         dev_kfree_skb(skb);
 934         return NETDEV_TX_OK;
 935 }
 936
 937 static int ipgre_tunnel_bind_dev(struct net_device *dev)
 938 {
 939         struct net_device *tdev = NULL;
 940         struct ip_tunnel *tunnel;
 941         struct iphdr *iph;
 942         int hlen = LL_MAX_HEADER;
 943         int mtu = ETH_DATA_LEN;
 944         int addend = sizeof(struct iphdr) + 4;
 945
 946         tunnel = netdev_priv(dev);
 947         iph = &tunnel->parms.iph;
 948
 949         /* Guess output device to choose reasonable mtu and needed_headroom */
 950
 951         if (iph->daddr) {
 952                 struct flowi fl = {
 953                         .oif = tunnel->parms.link,
 954                         .nl_u = {
 955                                 .ip4_u = {
 956                                         .daddr = iph->daddr,
 957                                         .saddr = iph->saddr,
 958                                         .tos = RT_TOS(iph->tos)
 959                                 }
 960                         },
 961                         .proto = IPPROTO_GRE
 962                 };
 963                 struct rtable *rt;
 964
 965                 if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
 966                         tdev = rt->dst.dev;
 967                         ip_rt_put(rt);
 968                 }
 969
 970                 if (dev->type != ARPHRD_ETHER)
 971                         dev->flags |= IFF_POINTOPOINT;
 972         }
 973
 974         if (!tdev && tunnel->parms.link)
 975                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
 976
 977         if (tdev) {
 978                 hlen = tdev->hard_header_len + tdev->needed_headroom;
 979                 mtu = tdev->mtu;
 980         }
 981         dev->iflink = tunnel->parms.link;
 982
 983         /* Precalculate GRE options length */
 984         if (tunnel->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
 985                 if (tunnel->parms.o_flags&GRE_CSUM)
 986                         addend += 4;
 987                 if (tunnel->parms.o_flags&GRE_KEY)
 988                         addend += 4;
 989                 if (tunnel->parms.o_flags&GRE_SEQ)
 990                         addend += 4;
 991         }
 992         dev->needed_headroom = addend + hlen;
 993         mtu -= dev->hard_header_len + addend;
 994
 995         if (mtu < 68)
 996                 mtu = 68;
 997
 998         tunnel->hlen = addend;
 999
1000         return mtu;
1001 }
1002
1003 static int
1004 ipgre_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
1005 {
1006         int err = 0;
1007         struct ip_tunnel_parm p;
1008         struct ip_tunnel *t;
1009         struct net *net = dev_net(dev);
1010         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1011
1012         switch (cmd) {
1013         case SIOCGETTUNNEL:
1014                 t = NULL;
1015                 if (dev == ign->fb_tunnel_dev) {
1016                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1017                                 err = -EFAULT;
1018                                 break;
1019                         }
1020                         t = ipgre_tunnel_locate(net, &p, 0);
1021                 }
1022                 if (t == NULL)
1023                         t = netdev_priv(dev);
1024                 memcpy(&p, &t->parms, sizeof(p));
1025                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1026                         err = -EFAULT;
1027                 break;
1028
1029         case SIOCADDTUNNEL:
1030         case SIOCCHGTUNNEL:
1031                 err = -EPERM;
1032                 if (!capable(CAP_NET_ADMIN))
1033                         goto done;
1034
1035                 err = -EFAULT;
1036                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1037                         goto done;
1038
1039                 err = -EINVAL;
1040                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
1041                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) ||
1042                     ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING)))
1043                         goto done;
1044                 if (p.iph.ttl)
1045                         p.iph.frag_off |= htons(IP_DF);
1046
1047                 if (!(p.i_flags&GRE_KEY))
1048                         p.i_key = 0;
1049                 if (!(p.o_flags&GRE_KEY))
1050                         p.o_key = 0;
1051
1052                 t = ipgre_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
1053
1054                 if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1055                         if (t != NULL) {
1056                                 if (t->dev != dev) {
1057                                         err = -EEXIST;
1058                                         break;
1059                                 }
1060                         } else {
1061                                 unsigned int nflags = 0;
1062
1063                                 t = netdev_priv(dev);
1064
1065                                 if (ipv4_is_multicast(p.iph.daddr))
1066                                         nflags = IFF_BROADCAST;
1067                                 else if (p.iph.daddr)
1068                                         nflags = IFF_POINTOPOINT;
1069
1070                                 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) {
1071                                         err = -EINVAL;
1072                                         break;
1073                                 }
1074                                 ipgre_tunnel_unlink(ign, t);
1075                                 synchronize_net();
1076                                 t->parms.iph.saddr = p.iph.saddr;
1077                                 t->parms.iph.daddr = p.iph.daddr;
1078                                 t->parms.i_key = p.i_key;
1079                                 t->parms.o_key = p.o_key;
1080                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
1081                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
1082                                 ipgre_tunnel_link(ign, t);
1083                                 netdev_state_change(dev);
1084                         }
1085                 }
1086
1087                 if (t) {
1088                         err = 0;
1089                         if (cmd == SIOCCHGTUNNEL) {
1090                                 t->parms.iph.ttl = p.iph.ttl;
1091                                 t->parms.iph.tos = p.iph.tos;
1092                                 t->parms.iph.frag_off = p.iph.frag_off;
1093                                 if (t->parms.link != p.link) {
1094                                         t->parms.link = p.link;
1095                                         dev->mtu = ipgre_tunnel_bind_dev(dev);
1096                                         netdev_state_change(dev);
1097                                 }
1098                         }
1099                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
1100                                 err = -EFAULT;
1101                 } else
1102                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1103                 break;
1104
1105         case SIOCDELTUNNEL:
1106                 err = -EPERM;
1107                 if (!capable(CAP_NET_ADMIN))
1108                         goto done;
1109
1110                 if (dev == ign->fb_tunnel_dev) {
1111                         err = -EFAULT;
1112                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1113                                 goto done;
1114                         err = -ENOENT;
1115                         if ((t = ipgre_tunnel_locate(net, &p, 0)) == NULL)
1116                                 goto done;
1117                         err = -EPERM;
1118                         if (t == netdev_priv(ign->fb_tunnel_dev))
1119                                 goto done;
1120                         dev = t->dev;
1121                 }
1122                 unregister_netdevice(dev);
1123                 err = 0;
1124                 break;
1125
1126         default:
1127                 err = -EINVAL;
1128         }
1129
1130 done:
1131         return err;
1132 }
1133
1134 static int ipgre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1135 {
1136         struct ip_tunnel *tunnel = netdev_priv(dev);
1137         if (new_mtu < 68 ||
1138             new_mtu > 0xFFF8 - dev->hard_header_len - tunnel->hlen)
1139                 return -EINVAL;
1140         dev->mtu = new_mtu;
1141         return 0;
1142 }
1143
1144 /* Nice toy. Unfortunately, useless in real life :-)
1145    It allows to construct virtual multiprotocol broadcast "LAN"
1146    over the Internet, provided multicast routing is tuned.
1147
1148
1149    I have no idea was this bicycle invented before me,
1150    so that I had to set ARPHRD_IPGRE to a random value.
1151    I have an impression, that Cisco could make something similar,
1152    but this feature is apparently missing in IOS<=11.2(8).
1153
1154    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
1155    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
1156
1157    ping -t 255 224.66.66.66
1158
1159    If nobody answers, mbone does not work.
1160
1161    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
1162    ip addr add 10.66.66.<somewhat>/24 dev Universe
1163    ifconfig Universe up
1164    ifconfig Universe add fe80::<Your_real_addr>/10
1165    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
1166    ftp 10.66.66.66
1167    ...
1168    ftp fec0:6666:6666::193.233.7.65
1169    ...
1170
1171  */
1172
1173 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
1174                         unsigned short type,
1175                         const void *daddr, const void *saddr, unsigned int len)
1176 {
1177         struct ip_tunnel *t = netdev_priv(dev);
1178         struct iphdr *iph = (struct iphdr *)skb_push(skb, t->hlen);
1179         __be16 *p = (__be16*)(iph+1);
1180
1181         memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
1182         p[0]            = t->parms.o_flags;
1183         p[1]            = htons(type);
1184
1185         /*
1186          *      Set the source hardware address.
1187          */
1188
1189         if (saddr)
1190                 memcpy(&iph->saddr, saddr, 4);
1191         if (daddr)
1192                 memcpy(&iph->daddr, daddr, 4);
1193         if (iph->daddr)
1194                 return t->hlen;
1195
1196         return -t->hlen;
1197 }
1198
1199 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
1200 {
1201         struct iphdr *iph = (struct iphdr *) skb_mac_header(skb);
1202         memcpy(haddr, &iph->saddr, 4);
1203         return 4;
1204 }
1205
1206 static const struct header_ops ipgre_header_ops = {
1207         .create = ipgre_header,
1208         .parse  = ipgre_header_parse,
1209 };
1210
1211 #ifdef CONFIG_NET_IPGRE_BROADCAST
1212 static int ipgre_open(struct net_device *dev)
1213 {
1214         struct ip_tunnel *t = netdev_priv(dev);
1215
1216         if (ipv4_is_multicast(t->parms.iph.daddr)) {
1217                 struct flowi fl = {
1218                         .oif = t->parms.link,
1219                         .nl_u = {
1220                                 .ip4_u = {
1221                                         .daddr = t->parms.iph.daddr,
1222                                         .saddr = t->parms.iph.saddr,
1223                                         .tos = RT_TOS(t->parms.iph.tos)
1224                                 }
1225                         },
1226                         .proto = IPPROTO_GRE
1227                 };
1228                 struct rtable *rt;
1229
1230                 if (ip_route_output_key(dev_net(dev), &rt, &fl))
1231                         return -EADDRNOTAVAIL;
1232                 dev = rt->dst.dev;
1233                 ip_rt_put(rt);
1234                 if (__in_dev_get_rtnl(dev) == NULL)
1235                         return -EADDRNOTAVAIL;
1236                 t->mlink = dev->ifindex;
1237                 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
1238         }
1239         return 0;
1240 }
1241
1242 static int ipgre_close(struct net_device *dev)
1243 {
1244         struct ip_tunnel *t = netdev_priv(dev);
1245
1246         if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
1247                 struct in_device *in_dev;
1248                 in_dev = inetdev_by_index(dev_net(dev), t->mlink);
1249                 if (in_dev)
1250                         ip_mc_dec_group(in_dev, t->parms.iph.daddr);
1251         }
1252         return 0;
1253 }
1254
1255 #endif
1256
1257 static const struct net_device_ops ipgre_netdev_ops = {
1258         .ndo_init               = ipgre_tunnel_init,
1259         .ndo_uninit             = ipgre_tunnel_uninit,
1260 #ifdef CONFIG_NET_IPGRE_BROADCAST
1261         .ndo_open               = ipgre_open,
1262         .ndo_stop               = ipgre_close,
1263 #endif
1264         .ndo_start_xmit         = ipgre_tunnel_xmit,
1265         .ndo_do_ioctl           = ipgre_tunnel_ioctl,
1266         .ndo_change_mtu         = ipgre_tunnel_change_mtu,
1267         .ndo_get_stats          = ipgre_get_stats,
1268 };
1269
1270 static void ipgre_dev_free(struct net_device *dev)
1271 {
1272         free_percpu(dev->tstats);
1273         free_netdev(dev);
1274 }
1275
1276 static void ipgre_tunnel_setup(struct net_device *dev)
1277 {
1278         dev->netdev_ops         = &ipgre_netdev_ops;
1279         dev->destructor         = ipgre_dev_free;
1280
1281         dev->type               = ARPHRD_IPGRE;
1282         dev->needed_headroom    = LL_MAX_HEADER + sizeof(struct iphdr) + 4;
1283         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 4;
1284         dev->flags              = IFF_NOARP;
1285         dev->iflink             = 0;
1286         dev->addr_len           = 4;
1287         dev->features           |= NETIF_F_NETNS_LOCAL;
1288         dev->priv_flags         &= ~IFF_XMIT_DST_RELEASE;
1289 }
1290
1291 static int ipgre_tunnel_init(struct net_device *dev)
1292 {
1293         struct ip_tunnel *tunnel;
1294         struct iphdr *iph;
1295
1296         tunnel = netdev_priv(dev);
1297         iph = &tunnel->parms.iph;
1298
1299         tunnel->dev = dev;
1300         strcpy(tunnel->parms.name, dev->name);
1301
1302         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1303         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1304
1305         if (iph->daddr) {
1306 #ifdef CONFIG_NET_IPGRE_BROADCAST
1307                 if (ipv4_is_multicast(iph->daddr)) {
1308                         if (!iph->saddr)
1309                                 return -EINVAL;
1310                         dev->flags = IFF_BROADCAST;
1311                         dev->header_ops = &ipgre_header_ops;
1312                 }
1313 #endif
1314         } else
1315                 dev->header_ops = &ipgre_header_ops;
1316
1317         dev->tstats = alloc_percpu(struct pcpu_tstats);
1318         if (!dev->tstats)
1319                 return -ENOMEM;
1320
1321         return 0;
1322 }
1323
1324 static void ipgre_fb_tunnel_init(struct net_device *dev)
1325 {
1326         struct ip_tunnel *tunnel = netdev_priv(dev);
1327         struct iphdr *iph = &tunnel->parms.iph;
1328         struct ipgre_net *ign = net_generic(dev_net(dev), ipgre_net_id);
1329
1330         tunnel->dev = dev;
1331         strcpy(tunnel->parms.name, dev->name);
1332
1333         iph->version            = 4;
1334         iph->protocol           = IPPROTO_GRE;
1335         iph->ihl                = 5;
1336         tunnel->hlen            = sizeof(struct iphdr) + 4;
1337
1338         dev_hold(dev);
1339         rcu_assign_pointer(ign->tunnels_wc[0], tunnel);
1340 }
1341
1342
1343 static const struct gre_protocol ipgre_protocol = {
1344         .handler     = ipgre_rcv,
1345         .err_handler = ipgre_err,
1346 };
1347
1348 static void ipgre_destroy_tunnels(struct ipgre_net *ign, struct list_head *head)
1349 {
1350         int prio;
1351
1352         for (prio = 0; prio < 4; prio++) {
1353                 int h;
1354                 for (h = 0; h < HASH_SIZE; h++) {
1355                         struct ip_tunnel *t;
1356
1357                         t = rtnl_dereference(ign->tunnels[prio][h]);
1358
1359                         while (t != NULL) {
1360                                 unregister_netdevice_queue(t->dev, head);
1361                                 t = rtnl_dereference(t->next);
1362                         }
1363                 }
1364         }
1365 }
1366
1367 static int __net_init ipgre_init_net(struct net *net)
1368 {
1369         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1370         int err;
1371
1372         ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "gre0",
1373                                            ipgre_tunnel_setup);
1374         if (!ign->fb_tunnel_dev) {
1375                 err = -ENOMEM;
1376                 goto err_alloc_dev;
1377         }
1378         dev_net_set(ign->fb_tunnel_dev, net);
1379
1380         ipgre_fb_tunnel_init(ign->fb_tunnel_dev);
1381         ign->fb_tunnel_dev->rtnl_link_ops = &ipgre_link_ops;
1382
1383         if ((err = register_netdev(ign->fb_tunnel_dev)))
1384                 goto err_reg_dev;
1385
1386         return 0;
1387
1388 err_reg_dev:
1389         free_netdev(ign->fb_tunnel_dev);
1390 err_alloc_dev:
1391         return err;
1392 }
1393
1394 static void __net_exit ipgre_exit_net(struct net *net)
1395 {
1396         struct ipgre_net *ign;
1397         LIST_HEAD(list);
1398
1399         ign = net_generic(net, ipgre_net_id);
1400         rtnl_lock();
1401         ipgre_destroy_tunnels(ign, &list);
1402         unregister_netdevice_many(&list);
1403         rtnl_unlock();
1404 }
1405
1406 static struct pernet_operations ipgre_net_ops = {
1407         .init = ipgre_init_net,
1408         .exit = ipgre_exit_net,
1409         .id   = &ipgre_net_id,
1410         .size = sizeof(struct ipgre_net),
1411 };
1412
1413 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1414 {
1415         __be16 flags;
1416
1417         if (!data)
1418                 return 0;
1419
1420         flags = 0;
1421         if (data[IFLA_GRE_IFLAGS])
1422                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1423         if (data[IFLA_GRE_OFLAGS])
1424                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1425         if (flags & (GRE_VERSION|GRE_ROUTING))
1426                 return -EINVAL;
1427
1428         return 0;
1429 }
1430
1431 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1432 {
1433         __be32 daddr;
1434
1435         if (tb[IFLA_ADDRESS]) {
1436                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1437                         return -EINVAL;
1438                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1439                         return -EADDRNOTAVAIL;
1440         }
1441
1442         if (!data)
1443                 goto out;
1444
1445         if (data[IFLA_GRE_REMOTE]) {
1446                 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1447                 if (!daddr)
1448                         return -EINVAL;
1449         }
1450
1451 out:
1452         return ipgre_tunnel_validate(tb, data);
1453 }
1454
1455 static void ipgre_netlink_parms(struct nlattr *data[],
1456                                 struct ip_tunnel_parm *parms)
1457 {
1458         memset(parms, 0, sizeof(*parms));
1459
1460         parms->iph.protocol = IPPROTO_GRE;
1461
1462         if (!data)
1463                 return;
1464
1465         if (data[IFLA_GRE_LINK])
1466                 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1467
1468         if (data[IFLA_GRE_IFLAGS])
1469                 parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1470
1471         if (data[IFLA_GRE_OFLAGS])
1472                 parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1473
1474         if (data[IFLA_GRE_IKEY])
1475                 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1476
1477         if (data[IFLA_GRE_OKEY])
1478                 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1479
1480         if (data[IFLA_GRE_LOCAL])
1481                 parms->iph.saddr = nla_get_be32(data[IFLA_GRE_LOCAL]);
1482
1483         if (data[IFLA_GRE_REMOTE])
1484                 parms->iph.daddr = nla_get_be32(data[IFLA_GRE_REMOTE]);
1485
1486         if (data[IFLA_GRE_TTL])
1487                 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1488
1489         if (data[IFLA_GRE_TOS])
1490                 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1491
1492         if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC]))
1493                 parms->iph.frag_off = htons(IP_DF);
1494 }
1495
1496 static int ipgre_tap_init(struct net_device *dev)
1497 {
1498         struct ip_tunnel *tunnel;
1499
1500         tunnel = netdev_priv(dev);
1501
1502         tunnel->dev = dev;
1503         strcpy(tunnel->parms.name, dev->name);
1504
1505         ipgre_tunnel_bind_dev(dev);
1506
1507         dev->tstats = alloc_percpu(struct pcpu_tstats);
1508         if (!dev->tstats)
1509                 return -ENOMEM;
1510
1511         return 0;
1512 }
1513
1514 static const struct net_device_ops ipgre_tap_netdev_ops = {
1515         .ndo_init               = ipgre_tap_init,
1516         .ndo_uninit             = ipgre_tunnel_uninit,
1517         .ndo_start_xmit         = ipgre_tunnel_xmit,
1518         .ndo_set_mac_address    = eth_mac_addr,
1519         .ndo_validate_addr      = eth_validate_addr,
1520         .ndo_change_mtu         = ipgre_tunnel_change_mtu,
1521         .ndo_get_stats          = ipgre_get_stats,
1522 };
1523
1524 static void ipgre_tap_setup(struct net_device *dev)
1525 {
1526
1527         ether_setup(dev);
1528
1529         dev->netdev_ops         = &ipgre_tap_netdev_ops;
1530         dev->destructor         = ipgre_dev_free;
1531
1532         dev->iflink             = 0;
1533         dev->features           |= NETIF_F_NETNS_LOCAL;
1534 }
1535
1536 static int ipgre_newlink(struct net *src_net, struct net_device *dev, struct nlattr *tb[],
1537                          struct nlattr *data[])
1538 {
1539         struct ip_tunnel *nt;
1540         struct net *net = dev_net(dev);
1541         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1542         int mtu;
1543         int err;
1544
1545         nt = netdev_priv(dev);
1546         ipgre_netlink_parms(data, &nt->parms);
1547
1548         if (ipgre_tunnel_find(net, &nt->parms, dev->type))
1549                 return -EEXIST;
1550
1551         if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1552                 random_ether_addr(dev->dev_addr);
1553
1554         mtu = ipgre_tunnel_bind_dev(dev);
1555         if (!tb[IFLA_MTU])
1556                 dev->mtu = mtu;
1557
1558         /* Can use a lockless transmit, unless we generate output sequences */
1559         if (!(nt->parms.o_flags & GRE_SEQ))
1560                 dev->features |= NETIF_F_LLTX;
1561
1562         err = register_netdevice(dev);
1563         if (err)
1564                 goto out;
1565
1566         dev_hold(dev);
1567         ipgre_tunnel_link(ign, nt);
1568
1569 out:
1570         return err;
1571 }
1572
1573 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1574                             struct nlattr *data[])
1575 {
1576         struct ip_tunnel *t, *nt;
1577         struct net *net = dev_net(dev);
1578         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1579         struct ip_tunnel_parm p;
1580         int mtu;
1581
1582         if (dev == ign->fb_tunnel_dev)
1583                 return -EINVAL;
1584
1585         nt = netdev_priv(dev);
1586         ipgre_netlink_parms(data, &p);
1587
1588         t = ipgre_tunnel_locate(net, &p, 0);
1589
1590         if (t) {
1591                 if (t->dev != dev)
1592                         return -EEXIST;
1593         } else {
1594                 t = nt;
1595
1596                 if (dev->type != ARPHRD_ETHER) {
1597                         unsigned int nflags = 0;
1598
1599                         if (ipv4_is_multicast(p.iph.daddr))
1600                                 nflags = IFF_BROADCAST;
1601                         else if (p.iph.daddr)
1602                                 nflags = IFF_POINTOPOINT;
1603
1604                         if ((dev->flags ^ nflags) &
1605                             (IFF_POINTOPOINT | IFF_BROADCAST))
1606                                 return -EINVAL;
1607                 }
1608
1609                 ipgre_tunnel_unlink(ign, t);
1610                 t->parms.iph.saddr = p.iph.saddr;
1611                 t->parms.iph.daddr = p.iph.daddr;
1612                 t->parms.i_key = p.i_key;
1613                 if (dev->type != ARPHRD_ETHER) {
1614                         memcpy(dev->dev_addr, &p.iph.saddr, 4);
1615                         memcpy(dev->broadcast, &p.iph.daddr, 4);
1616                 }
1617                 ipgre_tunnel_link(ign, t);
1618                 netdev_state_change(dev);
1619         }
1620
1621         t->parms.o_key = p.o_key;
1622         t->parms.iph.ttl = p.iph.ttl;
1623         t->parms.iph.tos = p.iph.tos;
1624         t->parms.iph.frag_off = p.iph.frag_off;
1625
1626         if (t->parms.link != p.link) {
1627                 t->parms.link = p.link;
1628                 mtu = ipgre_tunnel_bind_dev(dev);
1629                 if (!tb[IFLA_MTU])
1630                         dev->mtu = mtu;
1631                 netdev_state_change(dev);
1632         }
1633
1634         return 0;
1635 }
1636
1637 static size_t ipgre_get_size(const struct net_device *dev)
1638 {
1639         return
1640                 /* IFLA_GRE_LINK */
1641                 nla_total_size(4) +
1642                 /* IFLA_GRE_IFLAGS */
1643                 nla_total_size(2) +
1644                 /* IFLA_GRE_OFLAGS */
1645                 nla_total_size(2) +
1646                 /* IFLA_GRE_IKEY */
1647                 nla_total_size(4) +
1648                 /* IFLA_GRE_OKEY */
1649                 nla_total_size(4) +
1650                 /* IFLA_GRE_LOCAL */
1651                 nla_total_size(4) +
1652                 /* IFLA_GRE_REMOTE */
1653                 nla_total_size(4) +
1654                 /* IFLA_GRE_TTL */
1655                 nla_total_size(1) +
1656                 /* IFLA_GRE_TOS */
1657                 nla_total_size(1) +
1658                 /* IFLA_GRE_PMTUDISC */
1659                 nla_total_size(1) +
1660                 0;
1661 }
1662
1663 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1664 {
1665         struct ip_tunnel *t = netdev_priv(dev);
1666         struct ip_tunnel_parm *p = &t->parms;
1667
1668         NLA_PUT_U32(skb, IFLA_GRE_LINK, p->link);
1669         NLA_PUT_BE16(skb, IFLA_GRE_IFLAGS, p->i_flags);
1670         NLA_PUT_BE16(skb, IFLA_GRE_OFLAGS, p->o_flags);
1671         NLA_PUT_BE32(skb, IFLA_GRE_IKEY, p->i_key);
1672         NLA_PUT_BE32(skb, IFLA_GRE_OKEY, p->o_key);
1673         NLA_PUT_BE32(skb, IFLA_GRE_LOCAL, p->iph.saddr);
1674         NLA_PUT_BE32(skb, IFLA_GRE_REMOTE, p->iph.daddr);
1675         NLA_PUT_U8(skb, IFLA_GRE_TTL, p->iph.ttl);
1676         NLA_PUT_U8(skb, IFLA_GRE_TOS, p->iph.tos);
1677         NLA_PUT_U8(skb, IFLA_GRE_PMTUDISC, !!(p->iph.frag_off & htons(IP_DF)));
1678
1679         return 0;
1680
1681 nla_put_failure:
1682         return -EMSGSIZE;
1683 }
1684
1685 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1686         [IFLA_GRE_LINK]         = { .type = NLA_U32 },
1687         [IFLA_GRE_IFLAGS]       = { .type = NLA_U16 },
1688         [IFLA_GRE_OFLAGS]       = { .type = NLA_U16 },
1689         [IFLA_GRE_IKEY]         = { .type = NLA_U32 },
1690         [IFLA_GRE_OKEY]         = { .type = NLA_U32 },
1691         [IFLA_GRE_LOCAL]        = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1692         [IFLA_GRE_REMOTE]       = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1693         [IFLA_GRE_TTL]          = { .type = NLA_U8 },
1694         [IFLA_GRE_TOS]          = { .type = NLA_U8 },
1695         [IFLA_GRE_PMTUDISC]     = { .type = NLA_U8 },
1696 };
1697
1698 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1699         .kind           = "gre",
1700         .maxtype        = IFLA_GRE_MAX,
1701         .policy         = ipgre_policy,
1702         .priv_size      = sizeof(struct ip_tunnel),
1703         .setup          = ipgre_tunnel_setup,
1704         .validate       = ipgre_tunnel_validate,
1705         .newlink        = ipgre_newlink,
1706         .changelink     = ipgre_changelink,
1707         .get_size       = ipgre_get_size,
1708         .fill_info      = ipgre_fill_info,
1709 };
1710
1711 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1712         .kind           = "gretap",
1713         .maxtype        = IFLA_GRE_MAX,
1714         .policy         = ipgre_policy,
1715         .priv_size      = sizeof(struct ip_tunnel),
1716         .setup          = ipgre_tap_setup,
1717         .validate       = ipgre_tap_validate,
1718         .newlink        = ipgre_newlink,
1719         .changelink     = ipgre_changelink,
1720         .get_size       = ipgre_get_size,
1721         .fill_info      = ipgre_fill_info,
1722 };
1723
1724 /*
1725  *      And now the modules code and kernel interface.
1726  */
1727
1728 static int __init ipgre_init(void)
1729 {
1730         int err;
1731
1732         printk(KERN_INFO "GRE over IPv4 tunneling driver\n");
1733
1734         err = register_pernet_device(&ipgre_net_ops);
1735         if (err < 0)
1736                 return err;
1737
1738         err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1739         if (err < 0) {
1740                 printk(KERN_INFO "ipgre init: can't add protocol\n");
1741                 goto add_proto_failed;
1742         }
1743
1744         err = rtnl_link_register(&ipgre_link_ops);
1745         if (err < 0)
1746                 goto rtnl_link_failed;
1747
1748         err = rtnl_link_register(&ipgre_tap_ops);
1749         if (err < 0)
1750                 goto tap_ops_failed;
1751
1752 out:
1753         return err;
1754
1755 tap_ops_failed:
1756         rtnl_link_unregister(&ipgre_link_ops);
1757 rtnl_link_failed:
1758         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1759 add_proto_failed:
1760         unregister_pernet_device(&ipgre_net_ops);
1761         goto out;
1762 }
1763
1764 static void __exit ipgre_fini(void)
1765 {
1766         rtnl_link_unregister(&ipgre_tap_ops);
1767         rtnl_link_unregister(&ipgre_link_ops);
1768         if (gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO) < 0)
1769                 printk(KERN_INFO "ipgre close: can't remove protocol\n");
1770         unregister_pernet_device(&ipgre_net_ops);
1771 }
1772
1773 module_init(ipgre_init);
1774 module_exit(ipgre_fini);
1775 MODULE_LICENSE("GPL");
1776 MODULE_ALIAS_RTNL_LINK("gre");
1777 MODULE_ALIAS_RTNL_LINK("gretap");