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 (rt_is_output_route(skb_rtable(skb)))
 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                         .fl4_dst = dst,
 776                         .fl4_src = tiph->saddr,
 777                         .fl4_tos = RT_TOS(tos),
 778                         .fl_gre_key = tunnel->parms.o_key
 779                 };
 780                 if (ip_route_output_key(dev_net(dev), &rt, &fl)) {
 781                         dev->stats.tx_carrier_errors++;
 782                         goto tx_error;
 783                 }
 784         }
 785         tdev = rt->dst.dev;
 786
 787         if (tdev == dev) {
 788                 ip_rt_put(rt);
 789                 dev->stats.collisions++;
 790                 goto tx_error;
 791         }
 792
 793         df = tiph->frag_off;
 794         if (df)
 795                 mtu = dst_mtu(&rt->dst) - dev->hard_header_len - tunnel->hlen;
 796         else
 797                 mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
 798
 799         if (skb_dst(skb))
 800                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), mtu);
 801
 802         if (skb->protocol == htons(ETH_P_IP)) {
 803                 df |= (old_iph->frag_off&htons(IP_DF));
 804
 805                 if ((old_iph->frag_off&htons(IP_DF)) &&
 806                     mtu < ntohs(old_iph->tot_len)) {
 807                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
 808                         ip_rt_put(rt);
 809                         goto tx_error;
 810                 }
 811         }
 812 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 813         else if (skb->protocol == htons(ETH_P_IPV6)) {
 814                 struct rt6_info *rt6 = (struct rt6_info *)skb_dst(skb);
 815
 816                 if (rt6 && mtu < dst_mtu(skb_dst(skb)) && mtu >= IPV6_MIN_MTU) {
 817                         if ((tunnel->parms.iph.daddr &&
 818                              !ipv4_is_multicast(tunnel->parms.iph.daddr)) ||
 819                             rt6->rt6i_dst.plen == 128) {
 820                                 rt6->rt6i_flags |= RTF_MODIFIED;
 821                                 skb_dst(skb)->metrics[RTAX_MTU-1] = mtu;
 822                         }
 823                 }
 824
 825                 if (mtu >= IPV6_MIN_MTU && mtu < skb->len - tunnel->hlen + gre_hlen) {
 826                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 827                         ip_rt_put(rt);
 828                         goto tx_error;
 829                 }
 830         }
 831 #endif
 832
 833         if (tunnel->err_count > 0) {
 834                 if (time_before(jiffies,
 835                                 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
 836                         tunnel->err_count--;
 837
 838                         dst_link_failure(skb);
 839                 } else
 840                         tunnel->err_count = 0;
 841         }
 842
 843         max_headroom = LL_RESERVED_SPACE(tdev) + gre_hlen + rt->dst.header_len;
 844
 845         if (skb_headroom(skb) < max_headroom || skb_shared(skb)||
 846             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
 847                 struct sk_buff *new_skb = skb_realloc_headroom(skb, max_headroom);
 848                 if (max_headroom > dev->needed_headroom)
 849                         dev->needed_headroom = max_headroom;
 850                 if (!new_skb) {
 851                         ip_rt_put(rt);
 852                         dev->stats.tx_dropped++;
 853                         dev_kfree_skb(skb);
 854                         return NETDEV_TX_OK;
 855                 }
 856                 if (skb->sk)
 857                         skb_set_owner_w(new_skb, skb->sk);
 858                 dev_kfree_skb(skb);
 859                 skb = new_skb;
 860                 old_iph = ip_hdr(skb);
 861         }
 862
 863         skb_reset_transport_header(skb);
 864         skb_push(skb, gre_hlen);
 865         skb_reset_network_header(skb);
 866         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 867         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
 868                               IPSKB_REROUTED);
 869         skb_dst_drop(skb);
 870         skb_dst_set(skb, &rt->dst);
 871
 872         /*
 873          *      Push down and install the IPIP header.
 874          */
 875
 876         iph                     =       ip_hdr(skb);
 877         iph->version            =       4;
 878         iph->ihl                =       sizeof(struct iphdr) >> 2;
 879         iph->frag_off           =       df;
 880         iph->protocol           =       IPPROTO_GRE;
 881         iph->tos                =       ipgre_ecn_encapsulate(tos, old_iph, skb);
 882         iph->daddr              =       rt->rt_dst;
 883         iph->saddr              =       rt->rt_src;
 884
 885         if ((iph->ttl = tiph->ttl) == 0) {
 886                 if (skb->protocol == htons(ETH_P_IP))
 887                         iph->ttl = old_iph->ttl;
 888 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 889                 else if (skb->protocol == htons(ETH_P_IPV6))
 890                         iph->ttl = ((struct ipv6hdr *)old_iph)->hop_limit;
 891 #endif
 892                 else
 893                         iph->ttl = dst_metric(&rt->dst, RTAX_HOPLIMIT);
 894         }
 895
 896         ((__be16 *)(iph + 1))[0] = tunnel->parms.o_flags;
 897         ((__be16 *)(iph + 1))[1] = (dev->type == ARPHRD_ETHER) ?
 898                                    htons(ETH_P_TEB) : skb->protocol;
 899
 900         if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
 901                 __be32 *ptr = (__be32*)(((u8*)iph) + tunnel->hlen - 4);
 902
 903                 if (tunnel->parms.o_flags&GRE_SEQ) {
 904                         ++tunnel->o_seqno;
 905                         *ptr = htonl(tunnel->o_seqno);
 906                         ptr--;
 907                 }
 908                 if (tunnel->parms.o_flags&GRE_KEY) {
 909                         *ptr = tunnel->parms.o_key;
 910                         ptr--;
 911                 }
 912                 if (tunnel->parms.o_flags&GRE_CSUM) {
 913                         *ptr = 0;
 914                         *(__sum16*)ptr = ip_compute_csum((void*)(iph+1), skb->len - sizeof(struct iphdr));
 915                 }
 916         }
 917
 918         nf_reset(skb);
 919         tstats = this_cpu_ptr(dev->tstats);
 920         __IPTUNNEL_XMIT(tstats, &dev->stats);
 921         return NETDEV_TX_OK;
 922
 923 tx_error_icmp:
 924         dst_link_failure(skb);
 925
 926 tx_error:
 927         dev->stats.tx_errors++;
 928         dev_kfree_skb(skb);
 929         return NETDEV_TX_OK;
 930 }
 931
 932 static int ipgre_tunnel_bind_dev(struct net_device *dev)
 933 {
 934         struct net_device *tdev = NULL;
 935         struct ip_tunnel *tunnel;
 936         struct iphdr *iph;
 937         int hlen = LL_MAX_HEADER;
 938         int mtu = ETH_DATA_LEN;
 939         int addend = sizeof(struct iphdr) + 4;
 940
 941         tunnel = netdev_priv(dev);
 942         iph = &tunnel->parms.iph;
 943
 944         /* Guess output device to choose reasonable mtu and needed_headroom */
 945
 946         if (iph->daddr) {
 947                 struct flowi fl = {
 948                         .oif = tunnel->parms.link,
 949                         .fl4_dst = iph->daddr,
 950                         .fl4_src = iph->saddr,
 951                         .fl4_tos = RT_TOS(iph->tos),
 952                         .proto = IPPROTO_GRE,
 953                         .fl_gre_key = tunnel->parms.o_key
 954                 };
 955                 struct rtable *rt;
 956
 957                 if (!ip_route_output_key(dev_net(dev), &rt, &fl)) {
 958                         tdev = rt->dst.dev;
 959                         ip_rt_put(rt);
 960                 }
 961
 962                 if (dev->type != ARPHRD_ETHER)
 963                         dev->flags |= IFF_POINTOPOINT;
 964         }
 965
 966         if (!tdev && tunnel->parms.link)
 967                 tdev = __dev_get_by_index(dev_net(dev), tunnel->parms.link);
 968
 969         if (tdev) {
 970                 hlen = tdev->hard_header_len + tdev->needed_headroom;
 971                 mtu = tdev->mtu;
 972         }
 973         dev->iflink = tunnel->parms.link;
 974
 975         /* Precalculate GRE options length */
 976         if (tunnel->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
 977                 if (tunnel->parms.o_flags&GRE_CSUM)
 978                         addend += 4;
 979                 if (tunnel->parms.o_flags&GRE_KEY)
 980                         addend += 4;
 981                 if (tunnel->parms.o_flags&GRE_SEQ)
 982                         addend += 4;
 983         }
 984         dev->needed_headroom = addend + hlen;
 985         mtu -= dev->hard_header_len + addend;
 986
 987         if (mtu < 68)
 988                 mtu = 68;
 989
 990         tunnel->hlen = addend;
 991
 992         return mtu;
 993 }
 994
 995 static int
 996 ipgre_tunnel_ioctl (struct net_device *dev, struct ifreq *ifr, int cmd)
 997 {
 998         int err = 0;
 999         struct ip_tunnel_parm p;
1000         struct ip_tunnel *t;
1001         struct net *net = dev_net(dev);
1002         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1003
1004         switch (cmd) {
1005         case SIOCGETTUNNEL:
1006                 t = NULL;
1007                 if (dev == ign->fb_tunnel_dev) {
1008                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1009                                 err = -EFAULT;
1010                                 break;
1011                         }
1012                         t = ipgre_tunnel_locate(net, &p, 0);
1013                 }
1014                 if (t == NULL)
1015                         t = netdev_priv(dev);
1016                 memcpy(&p, &t->parms, sizeof(p));
1017                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1018                         err = -EFAULT;
1019                 break;
1020
1021         case SIOCADDTUNNEL:
1022         case SIOCCHGTUNNEL:
1023                 err = -EPERM;
1024                 if (!capable(CAP_NET_ADMIN))
1025                         goto done;
1026
1027                 err = -EFAULT;
1028                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1029                         goto done;
1030
1031                 err = -EINVAL;
1032                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
1033                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) ||
1034                     ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING)))
1035                         goto done;
1036                 if (p.iph.ttl)
1037                         p.iph.frag_off |= htons(IP_DF);
1038
1039                 if (!(p.i_flags&GRE_KEY))
1040                         p.i_key = 0;
1041                 if (!(p.o_flags&GRE_KEY))
1042                         p.o_key = 0;
1043
1044                 t = ipgre_tunnel_locate(net, &p, cmd == SIOCADDTUNNEL);
1045
1046                 if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1047                         if (t != NULL) {
1048                                 if (t->dev != dev) {
1049                                         err = -EEXIST;
1050                                         break;
1051                                 }
1052                         } else {
1053                                 unsigned int nflags = 0;
1054
1055                                 t = netdev_priv(dev);
1056
1057                                 if (ipv4_is_multicast(p.iph.daddr))
1058                                         nflags = IFF_BROADCAST;
1059                                 else if (p.iph.daddr)
1060                                         nflags = IFF_POINTOPOINT;
1061
1062                                 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) {
1063                                         err = -EINVAL;
1064                                         break;
1065                                 }
1066                                 ipgre_tunnel_unlink(ign, t);
1067                                 synchronize_net();
1068                                 t->parms.iph.saddr = p.iph.saddr;
1069                                 t->parms.iph.daddr = p.iph.daddr;
1070                                 t->parms.i_key = p.i_key;
1071                                 t->parms.o_key = p.o_key;
1072                                 memcpy(dev->dev_addr, &p.iph.saddr, 4);
1073                                 memcpy(dev->broadcast, &p.iph.daddr, 4);
1074                                 ipgre_tunnel_link(ign, t);
1075                                 netdev_state_change(dev);
1076                         }
1077                 }
1078
1079                 if (t) {
1080                         err = 0;
1081                         if (cmd == SIOCCHGTUNNEL) {
1082                                 t->parms.iph.ttl = p.iph.ttl;
1083                                 t->parms.iph.tos = p.iph.tos;
1084                                 t->parms.iph.frag_off = p.iph.frag_off;
1085                                 if (t->parms.link != p.link) {
1086                                         t->parms.link = p.link;
1087                                         dev->mtu = ipgre_tunnel_bind_dev(dev);
1088                                         netdev_state_change(dev);
1089                                 }
1090                         }
1091                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &t->parms, sizeof(p)))
1092                                 err = -EFAULT;
1093                 } else
1094                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1095                 break;
1096
1097         case SIOCDELTUNNEL:
1098                 err = -EPERM;
1099                 if (!capable(CAP_NET_ADMIN))
1100                         goto done;
1101
1102                 if (dev == ign->fb_tunnel_dev) {
1103                         err = -EFAULT;
1104                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1105                                 goto done;
1106                         err = -ENOENT;
1107                         if ((t = ipgre_tunnel_locate(net, &p, 0)) == NULL)
1108                                 goto done;
1109                         err = -EPERM;
1110                         if (t == netdev_priv(ign->fb_tunnel_dev))
1111                                 goto done;
1112                         dev = t->dev;
1113                 }
1114                 unregister_netdevice(dev);
1115                 err = 0;
1116                 break;
1117
1118         default:
1119                 err = -EINVAL;
1120         }
1121
1122 done:
1123         return err;
1124 }
1125
1126 static int ipgre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1127 {
1128         struct ip_tunnel *tunnel = netdev_priv(dev);
1129         if (new_mtu < 68 ||
1130             new_mtu > 0xFFF8 - dev->hard_header_len - tunnel->hlen)
1131                 return -EINVAL;
1132         dev->mtu = new_mtu;
1133         return 0;
1134 }
1135
1136 /* Nice toy. Unfortunately, useless in real life :-)
1137    It allows to construct virtual multiprotocol broadcast "LAN"
1138    over the Internet, provided multicast routing is tuned.
1139
1140
1141    I have no idea was this bicycle invented before me,
1142    so that I had to set ARPHRD_IPGRE to a random value.
1143    I have an impression, that Cisco could make something similar,
1144    but this feature is apparently missing in IOS<=11.2(8).
1145
1146    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
1147    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
1148
1149    ping -t 255 224.66.66.66
1150
1151    If nobody answers, mbone does not work.
1152
1153    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
1154    ip addr add 10.66.66.<somewhat>/24 dev Universe
1155    ifconfig Universe up
1156    ifconfig Universe add fe80::<Your_real_addr>/10
1157    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
1158    ftp 10.66.66.66
1159    ...
1160    ftp fec0:6666:6666::193.233.7.65
1161    ...
1162
1163  */
1164
1165 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
1166                         unsigned short type,
1167                         const void *daddr, const void *saddr, unsigned int len)
1168 {
1169         struct ip_tunnel *t = netdev_priv(dev);
1170         struct iphdr *iph = (struct iphdr *)skb_push(skb, t->hlen);
1171         __be16 *p = (__be16*)(iph+1);
1172
1173         memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
1174         p[0]            = t->parms.o_flags;
1175         p[1]            = htons(type);
1176
1177         /*
1178          *      Set the source hardware address.
1179          */
1180
1181         if (saddr)
1182                 memcpy(&iph->saddr, saddr, 4);
1183         if (daddr)
1184                 memcpy(&iph->daddr, daddr, 4);
1185         if (iph->daddr)
1186                 return t->hlen;
1187
1188         return -t->hlen;
1189 }
1190
1191 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
1192 {
1193         struct iphdr *iph = (struct iphdr *) skb_mac_header(skb);
1194         memcpy(haddr, &iph->saddr, 4);
1195         return 4;
1196 }
1197
1198 static const struct header_ops ipgre_header_ops = {
1199         .create = ipgre_header,
1200         .parse  = ipgre_header_parse,
1201 };
1202
1203 #ifdef CONFIG_NET_IPGRE_BROADCAST
1204 static int ipgre_open(struct net_device *dev)
1205 {
1206         struct ip_tunnel *t = netdev_priv(dev);
1207
1208         if (ipv4_is_multicast(t->parms.iph.daddr)) {
1209                 struct flowi fl = {
1210                         .oif = t->parms.link,
1211                         .fl4_dst = t->parms.iph.daddr,
1212                         .fl4_src = t->parms.iph.saddr,
1213                         .fl4_tos = RT_TOS(t->parms.iph.tos),
1214                         .proto = IPPROTO_GRE,
1215                         .fl_gre_key = t->parms.o_key
1216                 };
1217                 struct rtable *rt;
1218
1219                 if (ip_route_output_key(dev_net(dev), &rt, &fl))
1220                         return -EADDRNOTAVAIL;
1221                 dev = rt->dst.dev;
1222                 ip_rt_put(rt);
1223                 if (__in_dev_get_rtnl(dev) == NULL)
1224                         return -EADDRNOTAVAIL;
1225                 t->mlink = dev->ifindex;
1226                 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
1227         }
1228         return 0;
1229 }
1230
1231 static int ipgre_close(struct net_device *dev)
1232 {
1233         struct ip_tunnel *t = netdev_priv(dev);
1234
1235         if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
1236                 struct in_device *in_dev;
1237                 in_dev = inetdev_by_index(dev_net(dev), t->mlink);
1238                 if (in_dev)
1239                         ip_mc_dec_group(in_dev, t->parms.iph.daddr);
1240         }
1241         return 0;
1242 }
1243
1244 #endif
1245
1246 static const struct net_device_ops ipgre_netdev_ops = {
1247         .ndo_init               = ipgre_tunnel_init,
1248         .ndo_uninit             = ipgre_tunnel_uninit,
1249 #ifdef CONFIG_NET_IPGRE_BROADCAST
1250         .ndo_open               = ipgre_open,
1251         .ndo_stop               = ipgre_close,
1252 #endif
1253         .ndo_start_xmit         = ipgre_tunnel_xmit,
1254         .ndo_do_ioctl           = ipgre_tunnel_ioctl,
1255         .ndo_change_mtu         = ipgre_tunnel_change_mtu,
1256         .ndo_get_stats          = ipgre_get_stats,
1257 };
1258
1259 static void ipgre_dev_free(struct net_device *dev)
1260 {
1261         free_percpu(dev->tstats);
1262         free_netdev(dev);
1263 }
1264
1265 static void ipgre_tunnel_setup(struct net_device *dev)
1266 {
1267         dev->netdev_ops         = &ipgre_netdev_ops;
1268         dev->destructor         = ipgre_dev_free;
1269
1270         dev->type               = ARPHRD_IPGRE;
1271         dev->needed_headroom    = LL_MAX_HEADER + sizeof(struct iphdr) + 4;
1272         dev->mtu                = ETH_DATA_LEN - sizeof(struct iphdr) - 4;
1273         dev->flags              = IFF_NOARP;
1274         dev->iflink             = 0;
1275         dev->addr_len           = 4;
1276         dev->features           |= NETIF_F_NETNS_LOCAL;
1277         dev->priv_flags         &= ~IFF_XMIT_DST_RELEASE;
1278 }
1279
1280 static int ipgre_tunnel_init(struct net_device *dev)
1281 {
1282         struct ip_tunnel *tunnel;
1283         struct iphdr *iph;
1284
1285         tunnel = netdev_priv(dev);
1286         iph = &tunnel->parms.iph;
1287
1288         tunnel->dev = dev;
1289         strcpy(tunnel->parms.name, dev->name);
1290
1291         memcpy(dev->dev_addr, &tunnel->parms.iph.saddr, 4);
1292         memcpy(dev->broadcast, &tunnel->parms.iph.daddr, 4);
1293
1294         if (iph->daddr) {
1295 #ifdef CONFIG_NET_IPGRE_BROADCAST
1296                 if (ipv4_is_multicast(iph->daddr)) {
1297                         if (!iph->saddr)
1298                                 return -EINVAL;
1299                         dev->flags = IFF_BROADCAST;
1300                         dev->header_ops = &ipgre_header_ops;
1301                 }
1302 #endif
1303         } else
1304                 dev->header_ops = &ipgre_header_ops;
1305
1306         dev->tstats = alloc_percpu(struct pcpu_tstats);
1307         if (!dev->tstats)
1308                 return -ENOMEM;
1309
1310         return 0;
1311 }
1312
1313 static void ipgre_fb_tunnel_init(struct net_device *dev)
1314 {
1315         struct ip_tunnel *tunnel = netdev_priv(dev);
1316         struct iphdr *iph = &tunnel->parms.iph;
1317
1318         tunnel->dev = dev;
1319         strcpy(tunnel->parms.name, dev->name);
1320
1321         iph->version            = 4;
1322         iph->protocol           = IPPROTO_GRE;
1323         iph->ihl                = 5;
1324         tunnel->hlen            = sizeof(struct iphdr) + 4;
1325
1326         dev_hold(dev);
1327 }
1328
1329
1330 static const struct gre_protocol ipgre_protocol = {
1331         .handler     = ipgre_rcv,
1332         .err_handler = ipgre_err,
1333 };
1334
1335 static void ipgre_destroy_tunnels(struct ipgre_net *ign, struct list_head *head)
1336 {
1337         int prio;
1338
1339         for (prio = 0; prio < 4; prio++) {
1340                 int h;
1341                 for (h = 0; h < HASH_SIZE; h++) {
1342                         struct ip_tunnel *t;
1343
1344                         t = rtnl_dereference(ign->tunnels[prio][h]);
1345
1346                         while (t != NULL) {
1347                                 unregister_netdevice_queue(t->dev, head);
1348                                 t = rtnl_dereference(t->next);
1349                         }
1350                 }
1351         }
1352 }
1353
1354 static int __net_init ipgre_init_net(struct net *net)
1355 {
1356         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1357         int err;
1358
1359         ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip_tunnel), "gre0",
1360                                            ipgre_tunnel_setup);
1361         if (!ign->fb_tunnel_dev) {
1362                 err = -ENOMEM;
1363                 goto err_alloc_dev;
1364         }
1365         dev_net_set(ign->fb_tunnel_dev, net);
1366
1367         ipgre_fb_tunnel_init(ign->fb_tunnel_dev);
1368         ign->fb_tunnel_dev->rtnl_link_ops = &ipgre_link_ops;
1369
1370         if ((err = register_netdev(ign->fb_tunnel_dev)))
1371                 goto err_reg_dev;
1372
1373         rcu_assign_pointer(ign->tunnels_wc[0],
1374                            netdev_priv(ign->fb_tunnel_dev));
1375         return 0;
1376
1377 err_reg_dev:
1378         ipgre_dev_free(ign->fb_tunnel_dev);
1379 err_alloc_dev:
1380         return err;
1381 }
1382
1383 static void __net_exit ipgre_exit_net(struct net *net)
1384 {
1385         struct ipgre_net *ign;
1386         LIST_HEAD(list);
1387
1388         ign = net_generic(net, ipgre_net_id);
1389         rtnl_lock();
1390         ipgre_destroy_tunnels(ign, &list);
1391         unregister_netdevice_many(&list);
1392         rtnl_unlock();
1393 }
1394
1395 static struct pernet_operations ipgre_net_ops = {
1396         .init = ipgre_init_net,
1397         .exit = ipgre_exit_net,
1398         .id   = &ipgre_net_id,
1399         .size = sizeof(struct ipgre_net),
1400 };
1401
1402 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1403 {
1404         __be16 flags;
1405
1406         if (!data)
1407                 return 0;
1408
1409         flags = 0;
1410         if (data[IFLA_GRE_IFLAGS])
1411                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1412         if (data[IFLA_GRE_OFLAGS])
1413                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1414         if (flags & (GRE_VERSION|GRE_ROUTING))
1415                 return -EINVAL;
1416
1417         return 0;
1418 }
1419
1420 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1421 {
1422         __be32 daddr;
1423
1424         if (tb[IFLA_ADDRESS]) {
1425                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1426                         return -EINVAL;
1427                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1428                         return -EADDRNOTAVAIL;
1429         }
1430
1431         if (!data)
1432                 goto out;
1433
1434         if (data[IFLA_GRE_REMOTE]) {
1435                 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1436                 if (!daddr)
1437                         return -EINVAL;
1438         }
1439
1440 out:
1441         return ipgre_tunnel_validate(tb, data);
1442 }
1443
1444 static void ipgre_netlink_parms(struct nlattr *data[],
1445                                 struct ip_tunnel_parm *parms)
1446 {
1447         memset(parms, 0, sizeof(*parms));
1448
1449         parms->iph.protocol = IPPROTO_GRE;
1450
1451         if (!data)
1452                 return;
1453
1454         if (data[IFLA_GRE_LINK])
1455                 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1456
1457         if (data[IFLA_GRE_IFLAGS])
1458                 parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1459
1460         if (data[IFLA_GRE_OFLAGS])
1461                 parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1462
1463         if (data[IFLA_GRE_IKEY])
1464                 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1465
1466         if (data[IFLA_GRE_OKEY])
1467                 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1468
1469         if (data[IFLA_GRE_LOCAL])
1470                 parms->iph.saddr = nla_get_be32(data[IFLA_GRE_LOCAL]);
1471
1472         if (data[IFLA_GRE_REMOTE])
1473                 parms->iph.daddr = nla_get_be32(data[IFLA_GRE_REMOTE]);
1474
1475         if (data[IFLA_GRE_TTL])
1476                 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1477
1478         if (data[IFLA_GRE_TOS])
1479                 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1480
1481         if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC]))
1482                 parms->iph.frag_off = htons(IP_DF);
1483 }
1484
1485 static int ipgre_tap_init(struct net_device *dev)
1486 {
1487         struct ip_tunnel *tunnel;
1488
1489         tunnel = netdev_priv(dev);
1490
1491         tunnel->dev = dev;
1492         strcpy(tunnel->parms.name, dev->name);
1493
1494         ipgre_tunnel_bind_dev(dev);
1495
1496         dev->tstats = alloc_percpu(struct pcpu_tstats);
1497         if (!dev->tstats)
1498                 return -ENOMEM;
1499
1500         return 0;
1501 }
1502
1503 static const struct net_device_ops ipgre_tap_netdev_ops = {
1504         .ndo_init               = ipgre_tap_init,
1505         .ndo_uninit             = ipgre_tunnel_uninit,
1506         .ndo_start_xmit         = ipgre_tunnel_xmit,
1507         .ndo_set_mac_address    = eth_mac_addr,
1508         .ndo_validate_addr      = eth_validate_addr,
1509         .ndo_change_mtu         = ipgre_tunnel_change_mtu,
1510         .ndo_get_stats          = ipgre_get_stats,
1511 };
1512
1513 static void ipgre_tap_setup(struct net_device *dev)
1514 {
1515
1516         ether_setup(dev);
1517
1518         dev->netdev_ops         = &ipgre_tap_netdev_ops;
1519         dev->destructor         = ipgre_dev_free;
1520
1521         dev->iflink             = 0;
1522         dev->features           |= NETIF_F_NETNS_LOCAL;
1523 }
1524
1525 static int ipgre_newlink(struct net *src_net, struct net_device *dev, struct nlattr *tb[],
1526                          struct nlattr *data[])
1527 {
1528         struct ip_tunnel *nt;
1529         struct net *net = dev_net(dev);
1530         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1531         int mtu;
1532         int err;
1533
1534         nt = netdev_priv(dev);
1535         ipgre_netlink_parms(data, &nt->parms);
1536
1537         if (ipgre_tunnel_find(net, &nt->parms, dev->type))
1538                 return -EEXIST;
1539
1540         if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1541                 random_ether_addr(dev->dev_addr);
1542
1543         mtu = ipgre_tunnel_bind_dev(dev);
1544         if (!tb[IFLA_MTU])
1545                 dev->mtu = mtu;
1546
1547         /* Can use a lockless transmit, unless we generate output sequences */
1548         if (!(nt->parms.o_flags & GRE_SEQ))
1549                 dev->features |= NETIF_F_LLTX;
1550
1551         err = register_netdevice(dev);
1552         if (err)
1553                 goto out;
1554
1555         dev_hold(dev);
1556         ipgre_tunnel_link(ign, nt);
1557
1558 out:
1559         return err;
1560 }
1561
1562 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1563                             struct nlattr *data[])
1564 {
1565         struct ip_tunnel *t, *nt;
1566         struct net *net = dev_net(dev);
1567         struct ipgre_net *ign = net_generic(net, ipgre_net_id);
1568         struct ip_tunnel_parm p;
1569         int mtu;
1570
1571         if (dev == ign->fb_tunnel_dev)
1572                 return -EINVAL;
1573
1574         nt = netdev_priv(dev);
1575         ipgre_netlink_parms(data, &p);
1576
1577         t = ipgre_tunnel_locate(net, &p, 0);
1578
1579         if (t) {
1580                 if (t->dev != dev)
1581                         return -EEXIST;
1582         } else {
1583                 t = nt;
1584
1585                 if (dev->type != ARPHRD_ETHER) {
1586                         unsigned int nflags = 0;
1587
1588                         if (ipv4_is_multicast(p.iph.daddr))
1589                                 nflags = IFF_BROADCAST;
1590                         else if (p.iph.daddr)
1591                                 nflags = IFF_POINTOPOINT;
1592
1593                         if ((dev->flags ^ nflags) &
1594                             (IFF_POINTOPOINT | IFF_BROADCAST))
1595                                 return -EINVAL;
1596                 }
1597
1598                 ipgre_tunnel_unlink(ign, t);
1599                 t->parms.iph.saddr = p.iph.saddr;
1600                 t->parms.iph.daddr = p.iph.daddr;
1601                 t->parms.i_key = p.i_key;
1602                 if (dev->type != ARPHRD_ETHER) {
1603                         memcpy(dev->dev_addr, &p.iph.saddr, 4);
1604                         memcpy(dev->broadcast, &p.iph.daddr, 4);
1605                 }
1606                 ipgre_tunnel_link(ign, t);
1607                 netdev_state_change(dev);
1608         }
1609
1610         t->parms.o_key = p.o_key;
1611         t->parms.iph.ttl = p.iph.ttl;
1612         t->parms.iph.tos = p.iph.tos;
1613         t->parms.iph.frag_off = p.iph.frag_off;
1614
1615         if (t->parms.link != p.link) {
1616                 t->parms.link = p.link;
1617                 mtu = ipgre_tunnel_bind_dev(dev);
1618                 if (!tb[IFLA_MTU])
1619                         dev->mtu = mtu;
1620                 netdev_state_change(dev);
1621         }
1622
1623         return 0;
1624 }
1625
1626 static size_t ipgre_get_size(const struct net_device *dev)
1627 {
1628         return
1629                 /* IFLA_GRE_LINK */
1630                 nla_total_size(4) +
1631                 /* IFLA_GRE_IFLAGS */
1632                 nla_total_size(2) +
1633                 /* IFLA_GRE_OFLAGS */
1634                 nla_total_size(2) +
1635                 /* IFLA_GRE_IKEY */
1636                 nla_total_size(4) +
1637                 /* IFLA_GRE_OKEY */
1638                 nla_total_size(4) +
1639                 /* IFLA_GRE_LOCAL */
1640                 nla_total_size(4) +
1641                 /* IFLA_GRE_REMOTE */
1642                 nla_total_size(4) +
1643                 /* IFLA_GRE_TTL */
1644                 nla_total_size(1) +
1645                 /* IFLA_GRE_TOS */
1646                 nla_total_size(1) +
1647                 /* IFLA_GRE_PMTUDISC */
1648                 nla_total_size(1) +
1649                 0;
1650 }
1651
1652 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1653 {
1654         struct ip_tunnel *t = netdev_priv(dev);
1655         struct ip_tunnel_parm *p = &t->parms;
1656
1657         NLA_PUT_U32(skb, IFLA_GRE_LINK, p->link);
1658         NLA_PUT_BE16(skb, IFLA_GRE_IFLAGS, p->i_flags);
1659         NLA_PUT_BE16(skb, IFLA_GRE_OFLAGS, p->o_flags);
1660         NLA_PUT_BE32(skb, IFLA_GRE_IKEY, p->i_key);
1661         NLA_PUT_BE32(skb, IFLA_GRE_OKEY, p->o_key);
1662         NLA_PUT_BE32(skb, IFLA_GRE_LOCAL, p->iph.saddr);
1663         NLA_PUT_BE32(skb, IFLA_GRE_REMOTE, p->iph.daddr);
1664         NLA_PUT_U8(skb, IFLA_GRE_TTL, p->iph.ttl);
1665         NLA_PUT_U8(skb, IFLA_GRE_TOS, p->iph.tos);
1666         NLA_PUT_U8(skb, IFLA_GRE_PMTUDISC, !!(p->iph.frag_off & htons(IP_DF)));
1667
1668         return 0;
1669
1670 nla_put_failure:
1671         return -EMSGSIZE;
1672 }
1673
1674 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1675         [IFLA_GRE_LINK]         = { .type = NLA_U32 },
1676         [IFLA_GRE_IFLAGS]       = { .type = NLA_U16 },
1677         [IFLA_GRE_OFLAGS]       = { .type = NLA_U16 },
1678         [IFLA_GRE_IKEY]         = { .type = NLA_U32 },
1679         [IFLA_GRE_OKEY]         = { .type = NLA_U32 },
1680         [IFLA_GRE_LOCAL]        = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1681         [IFLA_GRE_REMOTE]       = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1682         [IFLA_GRE_TTL]          = { .type = NLA_U8 },
1683         [IFLA_GRE_TOS]          = { .type = NLA_U8 },
1684         [IFLA_GRE_PMTUDISC]     = { .type = NLA_U8 },
1685 };
1686
1687 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1688         .kind           = "gre",
1689         .maxtype        = IFLA_GRE_MAX,
1690         .policy         = ipgre_policy,
1691         .priv_size      = sizeof(struct ip_tunnel),
1692         .setup          = ipgre_tunnel_setup,
1693         .validate       = ipgre_tunnel_validate,
1694         .newlink        = ipgre_newlink,
1695         .changelink     = ipgre_changelink,
1696         .get_size       = ipgre_get_size,
1697         .fill_info      = ipgre_fill_info,
1698 };
1699
1700 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1701         .kind           = "gretap",
1702         .maxtype        = IFLA_GRE_MAX,
1703         .policy         = ipgre_policy,
1704         .priv_size      = sizeof(struct ip_tunnel),
1705         .setup          = ipgre_tap_setup,
1706         .validate       = ipgre_tap_validate,
1707         .newlink        = ipgre_newlink,
1708         .changelink     = ipgre_changelink,
1709         .get_size       = ipgre_get_size,
1710         .fill_info      = ipgre_fill_info,
1711 };
1712
1713 /*
1714  *      And now the modules code and kernel interface.
1715  */
1716
1717 static int __init ipgre_init(void)
1718 {
1719         int err;
1720
1721         printk(KERN_INFO "GRE over IPv4 tunneling driver\n");
1722
1723         err = register_pernet_device(&ipgre_net_ops);
1724         if (err < 0)
1725                 return err;
1726
1727         err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1728         if (err < 0) {
1729                 printk(KERN_INFO "ipgre init: can't add protocol\n");
1730                 goto add_proto_failed;
1731         }
1732
1733         err = rtnl_link_register(&ipgre_link_ops);
1734         if (err < 0)
1735                 goto rtnl_link_failed;
1736
1737         err = rtnl_link_register(&ipgre_tap_ops);
1738         if (err < 0)
1739                 goto tap_ops_failed;
1740
1741 out:
1742         return err;
1743
1744 tap_ops_failed:
1745         rtnl_link_unregister(&ipgre_link_ops);
1746 rtnl_link_failed:
1747         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1748 add_proto_failed:
1749         unregister_pernet_device(&ipgre_net_ops);
1750         goto out;
1751 }
1752
1753 static void __exit ipgre_fini(void)
1754 {
1755         rtnl_link_unregister(&ipgre_tap_ops);
1756         rtnl_link_unregister(&ipgre_link_ops);
1757         if (gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO) < 0)
1758                 printk(KERN_INFO "ipgre close: can't remove protocol\n");
1759         unregister_pernet_device(&ipgre_net_ops);
1760 }
1761
1762 module_init(ipgre_init);
1763 module_exit(ipgre_fini);
1764 MODULE_LICENSE("GPL");
1765 MODULE_ALIAS_RTNL_LINK("gre");
1766 MODULE_ALIAS_RTNL_LINK("gretap");