kernel/net/ipv6/ip6_gre.c

   1 /*
   2  *      GRE over IPv6 protocol decoder.
   3  *
   4  *      Authors: Dmitry Kozlov (xeb@mail.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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  14
  15 #include <linux/capability.h>
  16 #include <linux/module.h>
  17 #include <linux/types.h>
  18 #include <linux/kernel.h>
  19 #include <linux/slab.h>
  20 #include <linux/uaccess.h>
  21 #include <linux/skbuff.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/in.h>
  24 #include <linux/tcp.h>
  25 #include <linux/udp.h>
  26 #include <linux/if_arp.h>
  27 #include <linux/mroute.h>
  28 #include <linux/init.h>
  29 #include <linux/in6.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/igmp.h>
  32 #include <linux/netfilter_ipv4.h>
  33 #include <linux/etherdevice.h>
  34 #include <linux/if_ether.h>
  35 #include <linux/hash.h>
  36 #include <linux/if_tunnel.h>
  37 #include <linux/ip6_tunnel.h>
  38
  39 #include <net/sock.h>
  40 #include <net/ip.h>
  41 #include <net/ip_tunnels.h>
  42 #include <net/icmp.h>
  43 #include <net/protocol.h>
  44 #include <net/addrconf.h>
  45 #include <net/arp.h>
  46 #include <net/checksum.h>
  47 #include <net/dsfield.h>
  48 #include <net/inet_ecn.h>
  49 #include <net/xfrm.h>
  50 #include <net/net_namespace.h>
  51 #include <net/netns/generic.h>
  52 #include <net/rtnetlink.h>
  53
  54 #include <net/ipv6.h>
  55 #include <net/ip6_fib.h>
  56 #include <net/ip6_route.h>
  57 #include <net/ip6_tunnel.h>
  58 #include <net/gre.h>
  59
  60
  61 static bool log_ecn_error = true;
  62 module_param(log_ecn_error, bool, 0644);
  63 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  64
  65 #define HASH_SIZE_SHIFT  5
  66 #define HASH_SIZE (1 << HASH_SIZE_SHIFT)
  67
  68 static int ip6gre_net_id __read_mostly;
  69 struct ip6gre_net {
  70         struct ip6_tnl __rcu *tunnels[4][HASH_SIZE];
  71
  72         struct net_device *fb_tunnel_dev;
  73 };
  74
  75 static struct rtnl_link_ops ip6gre_link_ops __read_mostly;
  76 static struct rtnl_link_ops ip6gre_tap_ops __read_mostly;
  77 static int ip6gre_tunnel_init(struct net_device *dev);
  78 static void ip6gre_tunnel_setup(struct net_device *dev);
  79 static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t);
  80 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu);
  81
  82 /* Tunnel hash table */
  83
  84 /*
  85    4 hash tables:
  86
  87    3: (remote,local)
  88    2: (remote,*)
  89    1: (*,local)
  90    0: (*,*)
  91
  92    We require exact key match i.e. if a key is present in packet
  93    it will match only tunnel with the same key; if it is not present,
  94    it will match only keyless tunnel.
  95
  96    All keysless packets, if not matched configured keyless tunnels
  97    will match fallback tunnel.
  98  */
  99
 100 #define HASH_KEY(key) (((__force u32)key^((__force u32)key>>4))&(HASH_SIZE - 1))
 101 static u32 HASH_ADDR(const struct in6_addr *addr)
 102 {
 103         u32 hash = ipv6_addr_hash(addr);
 104
 105         return hash_32(hash, HASH_SIZE_SHIFT);
 106 }
 107
 108 #define tunnels_r_l     tunnels[3]
 109 #define tunnels_r       tunnels[2]
 110 #define tunnels_l       tunnels[1]
 111 #define tunnels_wc      tunnels[0]
 112
 113 /* Given src, dst and key, find appropriate for input tunnel. */
 114
 115 static struct ip6_tnl *ip6gre_tunnel_lookup(struct net_device *dev,
 116                 const struct in6_addr *remote, const struct in6_addr *local,
 117                 __be32 key, __be16 gre_proto)
 118 {
 119         struct net *net = dev_net(dev);
 120         int link = dev->ifindex;
 121         unsigned int h0 = HASH_ADDR(remote);
 122         unsigned int h1 = HASH_KEY(key);
 123         struct ip6_tnl *t, *cand = NULL;
 124         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
 125         int dev_type = (gre_proto == htons(ETH_P_TEB)) ?
 126                        ARPHRD_ETHER : ARPHRD_IP6GRE;
 127         int score, cand_score = 4;
 128
 129         for_each_ip_tunnel_rcu(t, ign->tunnels_r_l[h0 ^ h1]) {
 130                 if (!ipv6_addr_equal(local, &t->parms.laddr) ||
 131                     !ipv6_addr_equal(remote, &t->parms.raddr) ||
 132                     key != t->parms.i_key ||
 133                     !(t->dev->flags & IFF_UP))
 134                         continue;
 135
 136                 if (t->dev->type != ARPHRD_IP6GRE &&
 137                     t->dev->type != dev_type)
 138                         continue;
 139
 140                 score = 0;
 141                 if (t->parms.link != link)
 142                         score |= 1;
 143                 if (t->dev->type != dev_type)
 144                         score |= 2;
 145                 if (score == 0)
 146                         return t;
 147
 148                 if (score < cand_score) {
 149                         cand = t;
 150                         cand_score = score;
 151                 }
 152         }
 153
 154         for_each_ip_tunnel_rcu(t, ign->tunnels_r[h0 ^ h1]) {
 155                 if (!ipv6_addr_equal(remote, &t->parms.raddr) ||
 156                     key != t->parms.i_key ||
 157                     !(t->dev->flags & IFF_UP))
 158                         continue;
 159
 160                 if (t->dev->type != ARPHRD_IP6GRE &&
 161                     t->dev->type != dev_type)
 162                         continue;
 163
 164                 score = 0;
 165                 if (t->parms.link != link)
 166                         score |= 1;
 167                 if (t->dev->type != dev_type)
 168                         score |= 2;
 169                 if (score == 0)
 170                         return t;
 171
 172                 if (score < cand_score) {
 173                         cand = t;
 174                         cand_score = score;
 175                 }
 176         }
 177
 178         for_each_ip_tunnel_rcu(t, ign->tunnels_l[h1]) {
 179                 if ((!ipv6_addr_equal(local, &t->parms.laddr) &&
 180                           (!ipv6_addr_equal(local, &t->parms.raddr) ||
 181                                  !ipv6_addr_is_multicast(local))) ||
 182                     key != t->parms.i_key ||
 183                     !(t->dev->flags & IFF_UP))
 184                         continue;
 185
 186                 if (t->dev->type != ARPHRD_IP6GRE &&
 187                     t->dev->type != dev_type)
 188                         continue;
 189
 190                 score = 0;
 191                 if (t->parms.link != link)
 192                         score |= 1;
 193                 if (t->dev->type != dev_type)
 194                         score |= 2;
 195                 if (score == 0)
 196                         return t;
 197
 198                 if (score < cand_score) {
 199                         cand = t;
 200                         cand_score = score;
 201                 }
 202         }
 203
 204         for_each_ip_tunnel_rcu(t, ign->tunnels_wc[h1]) {
 205                 if (t->parms.i_key != key ||
 206                     !(t->dev->flags & IFF_UP))
 207                         continue;
 208
 209                 if (t->dev->type != ARPHRD_IP6GRE &&
 210                     t->dev->type != dev_type)
 211                         continue;
 212
 213                 score = 0;
 214                 if (t->parms.link != link)
 215                         score |= 1;
 216                 if (t->dev->type != dev_type)
 217                         score |= 2;
 218                 if (score == 0)
 219                         return t;
 220
 221                 if (score < cand_score) {
 222                         cand = t;
 223                         cand_score = score;
 224                 }
 225         }
 226
 227         if (cand)
 228                 return cand;
 229
 230         dev = ign->fb_tunnel_dev;
 231         if (dev->flags & IFF_UP)
 232                 return netdev_priv(dev);
 233
 234         return NULL;
 235 }
 236
 237 static struct ip6_tnl __rcu **__ip6gre_bucket(struct ip6gre_net *ign,
 238                 const struct __ip6_tnl_parm *p)
 239 {
 240         const struct in6_addr *remote = &p->raddr;
 241         const struct in6_addr *local = &p->laddr;
 242         unsigned int h = HASH_KEY(p->i_key);
 243         int prio = 0;
 244
 245         if (!ipv6_addr_any(local))
 246                 prio |= 1;
 247         if (!ipv6_addr_any(remote) && !ipv6_addr_is_multicast(remote)) {
 248                 prio |= 2;
 249                 h ^= HASH_ADDR(remote);
 250         }
 251
 252         return &ign->tunnels[prio][h];
 253 }
 254
 255 static inline struct ip6_tnl __rcu **ip6gre_bucket(struct ip6gre_net *ign,
 256                 const struct ip6_tnl *t)
 257 {
 258         return __ip6gre_bucket(ign, &t->parms);
 259 }
 260
 261 static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t)
 262 {
 263         struct ip6_tnl __rcu **tp = ip6gre_bucket(ign, t);
 264
 265         rcu_assign_pointer(t->next, rtnl_dereference(*tp));
 266         rcu_assign_pointer(*tp, t);
 267 }
 268
 269 static void ip6gre_tunnel_unlink(struct ip6gre_net *ign, struct ip6_tnl *t)
 270 {
 271         struct ip6_tnl __rcu **tp;
 272         struct ip6_tnl *iter;
 273
 274         for (tp = ip6gre_bucket(ign, t);
 275              (iter = rtnl_dereference(*tp)) != NULL;
 276              tp = &iter->next) {
 277                 if (t == iter) {
 278                         rcu_assign_pointer(*tp, t->next);
 279                         break;
 280                 }
 281         }
 282 }
 283
 284 static struct ip6_tnl *ip6gre_tunnel_find(struct net *net,
 285                                            const struct __ip6_tnl_parm *parms,
 286                                            int type)
 287 {
 288         const struct in6_addr *remote = &parms->raddr;
 289         const struct in6_addr *local = &parms->laddr;
 290         __be32 key = parms->i_key;
 291         int link = parms->link;
 292         struct ip6_tnl *t;
 293         struct ip6_tnl __rcu **tp;
 294         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
 295
 296         for (tp = __ip6gre_bucket(ign, parms);
 297              (t = rtnl_dereference(*tp)) != NULL;
 298              tp = &t->next)
 299                 if (ipv6_addr_equal(local, &t->parms.laddr) &&
 300                     ipv6_addr_equal(remote, &t->parms.raddr) &&
 301                     key == t->parms.i_key &&
 302                     link == t->parms.link &&
 303                     type == t->dev->type)
 304                         break;
 305
 306         return t;
 307 }
 308
 309 static struct ip6_tnl *ip6gre_tunnel_locate(struct net *net,
 310                 const struct __ip6_tnl_parm *parms, int create)
 311 {
 312         struct ip6_tnl *t, *nt;
 313         struct net_device *dev;
 314         char name[IFNAMSIZ];
 315         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
 316
 317         t = ip6gre_tunnel_find(net, parms, ARPHRD_IP6GRE);
 318         if (t && create)
 319                 return NULL;
 320         if (t || !create)
 321                 return t;
 322
 323         if (parms->name[0])
 324                 strlcpy(name, parms->name, IFNAMSIZ);
 325         else
 326                 strcpy(name, "ip6gre%d");
 327
 328         dev = alloc_netdev(sizeof(*t), name, NET_NAME_UNKNOWN,
 329                            ip6gre_tunnel_setup);
 330         if (!dev)
 331                 return NULL;
 332
 333         dev_net_set(dev, net);
 334
 335         nt = netdev_priv(dev);
 336         nt->parms = *parms;
 337         dev->rtnl_link_ops = &ip6gre_link_ops;
 338
 339         nt->dev = dev;
 340         nt->net = dev_net(dev);
 341         ip6gre_tnl_link_config(nt, 1);
 342
 343         if (register_netdevice(dev) < 0)
 344                 goto failed_free;
 345
 346         /* Can use a lockless transmit, unless we generate output sequences */
 347         if (!(nt->parms.o_flags & GRE_SEQ))
 348                 dev->features |= NETIF_F_LLTX;
 349
 350         dev_hold(dev);
 351         ip6gre_tunnel_link(ign, nt);
 352         return nt;
 353
 354 failed_free:
 355         free_netdev(dev);
 356         return NULL;
 357 }
 358
 359 static void ip6gre_tunnel_uninit(struct net_device *dev)
 360 {
 361         struct ip6_tnl *t = netdev_priv(dev);
 362         struct ip6gre_net *ign = net_generic(t->net, ip6gre_net_id);
 363
 364         ip6gre_tunnel_unlink(ign, t);
 365         ip6_tnl_dst_reset(t);
 366         dev_put(dev);
 367 }
 368
 369
 370 static void ip6gre_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 371                        u8 type, u8 code, int offset, __be32 info)
 372 {
 373         const struct gre_base_hdr *greh;
 374         const struct ipv6hdr *ipv6h;
 375         int grehlen = sizeof(*greh);
 376         struct ip6_tnl *t;
 377         int key_off = 0;
 378         __be16 flags;
 379         __be32 key;
 380
 381         if (!pskb_may_pull(skb, offset + grehlen))
 382                 return;
 383         greh = (const struct gre_base_hdr *)(skb->data + offset);
 384         flags = greh->flags;
 385         if (flags & (GRE_VERSION | GRE_ROUTING))
 386                 return;
 387         if (flags & GRE_CSUM)
 388                 grehlen += 4;
 389         if (flags & GRE_KEY) {
 390                 key_off = grehlen + offset;
 391                 grehlen += 4;
 392         }
 393
 394         if (!pskb_may_pull(skb, offset + grehlen))
 395                 return;
 396         ipv6h = (const struct ipv6hdr *)skb->data;
 397         greh = (const struct gre_base_hdr *)(skb->data + offset);
 398         key = key_off ? *(__be32 *)(skb->data + key_off) : 0;
 399
 400         t = ip6gre_tunnel_lookup(skb->dev, &ipv6h->daddr, &ipv6h->saddr,
 401                                  key, greh->protocol);
 402         if (!t)
 403                 return;
 404
 405         switch (type) {
 406                 __u32 teli;
 407                 struct ipv6_tlv_tnl_enc_lim *tel;
 408                 __u32 mtu;
 409         case ICMPV6_DEST_UNREACH:
 410                 net_dbg_ratelimited("%s: Path to destination invalid or inactive!\n",
 411                                     t->parms.name);
 412                 break;
 413         case ICMPV6_TIME_EXCEED:
 414                 if (code == ICMPV6_EXC_HOPLIMIT) {
 415                         net_dbg_ratelimited("%s: Too small hop limit or routing loop in tunnel!\n",
 416                                             t->parms.name);
 417                 }
 418                 break;
 419         case ICMPV6_PARAMPROB:
 420                 teli = 0;
 421                 if (code == ICMPV6_HDR_FIELD)
 422                         teli = ip6_tnl_parse_tlv_enc_lim(skb, skb->data);
 423
 424                 if (teli && teli == be32_to_cpu(info) - 2) {
 425                         tel = (struct ipv6_tlv_tnl_enc_lim *) &skb->data[teli];
 426                         if (tel->encap_limit == 0) {
 427                                 net_dbg_ratelimited("%s: Too small encapsulation limit or routing loop in tunnel!\n",
 428                                                     t->parms.name);
 429                         }
 430                 } else {
 431                         net_dbg_ratelimited("%s: Recipient unable to parse tunneled packet!\n",
 432                                             t->parms.name);
 433                 }
 434                 break;
 435         case ICMPV6_PKT_TOOBIG:
 436                 mtu = be32_to_cpu(info) - offset;
 437                 if (mtu < IPV6_MIN_MTU)
 438                         mtu = IPV6_MIN_MTU;
 439                 t->dev->mtu = mtu;
 440                 break;
 441         }
 442
 443         if (time_before(jiffies, t->err_time + IP6TUNNEL_ERR_TIMEO))
 444                 t->err_count++;
 445         else
 446                 t->err_count = 1;
 447         t->err_time = jiffies;
 448 }
 449
 450 static int ip6gre_rcv(struct sk_buff *skb)
 451 {
 452         const struct ipv6hdr *ipv6h;
 453         u8     *h;
 454         __be16    flags;
 455         __sum16   csum = 0;
 456         __be32 key = 0;
 457         u32    seqno = 0;
 458         struct ip6_tnl *tunnel;
 459         int    offset = 4;
 460         __be16 gre_proto;
 461         int err;
 462
 463         if (!pskb_may_pull(skb, sizeof(struct in6_addr)))
 464                 goto drop;
 465
 466         ipv6h = ipv6_hdr(skb);
 467         h = skb->data;
 468         flags = *(__be16 *)h;
 469
 470         if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
 471                 /* - Version must be 0.
 472                    - We do not support routing headers.
 473                  */
 474                 if (flags&(GRE_VERSION|GRE_ROUTING))
 475                         goto drop;
 476
 477                 if (flags&GRE_CSUM) {
 478                         csum = skb_checksum_simple_validate(skb);
 479                         offset += 4;
 480                 }
 481                 if (flags&GRE_KEY) {
 482                         key = *(__be32 *)(h + offset);
 483                         offset += 4;
 484                 }
 485                 if (flags&GRE_SEQ) {
 486                         seqno = ntohl(*(__be32 *)(h + offset));
 487                         offset += 4;
 488                 }
 489         }
 490
 491         gre_proto = *(__be16 *)(h + 2);
 492
 493         tunnel = ip6gre_tunnel_lookup(skb->dev,
 494                                           &ipv6h->saddr, &ipv6h->daddr, key,
 495                                           gre_proto);
 496         if (tunnel) {
 497                 struct pcpu_sw_netstats *tstats;
 498
 499                 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
 500                         goto drop;
 501
 502                 if (!ip6_tnl_rcv_ctl(tunnel, &ipv6h->daddr, &ipv6h->saddr)) {
 503                         tunnel->dev->stats.rx_dropped++;
 504                         goto drop;
 505                 }
 506
 507                 skb->protocol = gre_proto;
 508                 /* WCCP version 1 and 2 protocol decoding.
 509                  * - Change protocol to IPv6
 510                  * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
 511                  */
 512                 if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) {
 513                         skb->protocol = htons(ETH_P_IPV6);
 514                         if ((*(h + offset) & 0xF0) != 0x40)
 515                                 offset += 4;
 516                 }
 517
 518                 skb->mac_header = skb->network_header;
 519                 __pskb_pull(skb, offset);
 520                 skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
 521
 522                 if (((flags&GRE_CSUM) && csum) ||
 523                     (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
 524                         tunnel->dev->stats.rx_crc_errors++;
 525                         tunnel->dev->stats.rx_errors++;
 526                         goto drop;
 527                 }
 528                 if (tunnel->parms.i_flags&GRE_SEQ) {
 529                         if (!(flags&GRE_SEQ) ||
 530                             (tunnel->i_seqno &&
 531                                         (s32)(seqno - tunnel->i_seqno) < 0)) {
 532                                 tunnel->dev->stats.rx_fifo_errors++;
 533                                 tunnel->dev->stats.rx_errors++;
 534                                 goto drop;
 535                         }
 536                         tunnel->i_seqno = seqno + 1;
 537                 }
 538
 539                 /* Warning: All skb pointers will be invalidated! */
 540                 if (tunnel->dev->type == ARPHRD_ETHER) {
 541                         if (!pskb_may_pull(skb, ETH_HLEN)) {
 542                                 tunnel->dev->stats.rx_length_errors++;
 543                                 tunnel->dev->stats.rx_errors++;
 544                                 goto drop;
 545                         }
 546
 547                         ipv6h = ipv6_hdr(skb);
 548                         skb->protocol = eth_type_trans(skb, tunnel->dev);
 549                         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 550                 }
 551
 552                 __skb_tunnel_rx(skb, tunnel->dev, tunnel->net);
 553
 554                 skb_reset_network_header(skb);
 555
 556                 err = IP6_ECN_decapsulate(ipv6h, skb);
 557                 if (unlikely(err)) {
 558                         if (log_ecn_error)
 559                                 net_info_ratelimited("non-ECT from %pI6 with dsfield=%#x\n",
 560                                                      &ipv6h->saddr,
 561                                                      ipv6_get_dsfield(ipv6h));
 562                         if (err > 1) {
 563                                 ++tunnel->dev->stats.rx_frame_errors;
 564                                 ++tunnel->dev->stats.rx_errors;
 565                                 goto drop;
 566                         }
 567                 }
 568
 569                 tstats = this_cpu_ptr(tunnel->dev->tstats);
 570                 u64_stats_update_begin(&tstats->syncp);
 571                 tstats->rx_packets++;
 572                 tstats->rx_bytes += skb->len;
 573                 u64_stats_update_end(&tstats->syncp);
 574
 575                 netif_rx(skb);
 576
 577                 return 0;
 578         }
 579         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 580
 581 drop:
 582         kfree_skb(skb);
 583         return 0;
 584 }
 585
 586 struct ipv6_tel_txoption {
 587         struct ipv6_txoptions ops;
 588         __u8 dst_opt[8];
 589 };
 590
 591 static void init_tel_txopt(struct ipv6_tel_txoption *opt, __u8 encap_limit)
 592 {
 593         memset(opt, 0, sizeof(struct ipv6_tel_txoption));
 594
 595         opt->dst_opt[2] = IPV6_TLV_TNL_ENCAP_LIMIT;
 596         opt->dst_opt[3] = 1;
 597         opt->dst_opt[4] = encap_limit;
 598         opt->dst_opt[5] = IPV6_TLV_PADN;
 599         opt->dst_opt[6] = 1;
 600
 601         opt->ops.dst0opt = (struct ipv6_opt_hdr *) opt->dst_opt;
 602         opt->ops.opt_nflen = 8;
 603 }
 604
 605 static netdev_tx_t ip6gre_xmit2(struct sk_buff *skb,
 606                          struct net_device *dev,
 607                          __u8 dsfield,
 608                          struct flowi6 *fl6,
 609                          int encap_limit,
 610                          __u32 *pmtu)
 611 {
 612         struct ip6_tnl *tunnel = netdev_priv(dev);
 613         struct net *net = tunnel->net;
 614         struct net_device *tdev;    /* Device to other host */
 615         struct ipv6hdr  *ipv6h;     /* Our new IP header */
 616         unsigned int max_headroom = 0; /* The extra header space needed */
 617         int    gre_hlen;
 618         struct ipv6_tel_txoption opt;
 619         int    mtu;
 620         struct dst_entry *dst = NULL, *ndst = NULL;
 621         struct net_device_stats *stats = &tunnel->dev->stats;
 622         int err = -1;
 623         u8 proto;
 624         struct sk_buff *new_skb;
 625         __be16 protocol;
 626
 627         if (dev->type == ARPHRD_ETHER)
 628                 IPCB(skb)->flags = 0;
 629
 630         if (dev->header_ops && dev->type == ARPHRD_IP6GRE) {
 631                 gre_hlen = 0;
 632                 ipv6h = (struct ipv6hdr *)skb->data;
 633                 fl6->daddr = ipv6h->daddr;
 634         } else {
 635                 gre_hlen = tunnel->hlen;
 636                 fl6->daddr = tunnel->parms.raddr;
 637         }
 638
 639         if (!fl6->flowi6_mark)
 640                 dst = ip6_tnl_dst_get(tunnel);
 641
 642         if (!dst) {
 643                 dst = ip6_route_output(net, NULL, fl6);
 644
 645                 if (dst->error)
 646                         goto tx_err_link_failure;
 647                 dst = xfrm_lookup(net, dst, flowi6_to_flowi(fl6), NULL, 0);
 648                 if (IS_ERR(dst)) {
 649                         err = PTR_ERR(dst);
 650                         dst = NULL;
 651                         goto tx_err_link_failure;
 652                 }
 653                 ndst = dst;
 654         }
 655
 656         tdev = dst->dev;
 657
 658         if (tdev == dev) {
 659                 stats->collisions++;
 660                 net_warn_ratelimited("%s: Local routing loop detected!\n",
 661                                      tunnel->parms.name);
 662                 goto tx_err_dst_release;
 663         }
 664
 665         mtu = dst_mtu(dst) - sizeof(*ipv6h);
 666         if (encap_limit >= 0) {
 667                 max_headroom += 8;
 668                 mtu -= 8;
 669         }
 670         if (mtu < IPV6_MIN_MTU)
 671                 mtu = IPV6_MIN_MTU;
 672         if (skb_dst(skb))
 673                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), NULL, skb, mtu);
 674         if (skb->len > mtu) {
 675                 *pmtu = mtu;
 676                 err = -EMSGSIZE;
 677                 goto tx_err_dst_release;
 678         }
 679
 680         if (tunnel->err_count > 0) {
 681                 if (time_before(jiffies,
 682                                 tunnel->err_time + IP6TUNNEL_ERR_TIMEO)) {
 683                         tunnel->err_count--;
 684
 685                         dst_link_failure(skb);
 686                 } else
 687                         tunnel->err_count = 0;
 688         }
 689
 690         skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(dev)));
 691
 692         max_headroom += LL_RESERVED_SPACE(tdev) + gre_hlen + dst->header_len;
 693
 694         if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
 695             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
 696                 new_skb = skb_realloc_headroom(skb, max_headroom);
 697                 if (max_headroom > dev->needed_headroom)
 698                         dev->needed_headroom = max_headroom;
 699                 if (!new_skb)
 700                         goto tx_err_dst_release;
 701
 702                 if (skb->sk)
 703                         skb_set_owner_w(new_skb, skb->sk);
 704                 consume_skb(skb);
 705                 skb = new_skb;
 706         }
 707
 708         if (!fl6->flowi6_mark && ndst)
 709                 ip6_tnl_dst_set(tunnel, ndst);
 710         skb_dst_set(skb, dst);
 711
 712         proto = NEXTHDR_GRE;
 713         if (encap_limit >= 0) {
 714                 init_tel_txopt(&opt, encap_limit);
 715                 ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
 716         }
 717
 718         if (likely(!skb->encapsulation)) {
 719                 skb_reset_inner_headers(skb);
 720                 skb->encapsulation = 1;
 721         }
 722
 723         skb_push(skb, gre_hlen);
 724         skb_reset_network_header(skb);
 725         skb_set_transport_header(skb, sizeof(*ipv6h));
 726
 727         /*
 728          *      Push down and install the IP header.
 729          */
 730         ipv6h = ipv6_hdr(skb);
 731         ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield),
 732                      ip6_make_flowlabel(net, skb, fl6->flowlabel, true, fl6));
 733         ipv6h->hop_limit = tunnel->parms.hop_limit;
 734         ipv6h->nexthdr = proto;
 735         ipv6h->saddr = fl6->saddr;
 736         ipv6h->daddr = fl6->daddr;
 737
 738         ((__be16 *)(ipv6h + 1))[0] = tunnel->parms.o_flags;
 739         protocol = (dev->type == ARPHRD_ETHER) ?
 740                     htons(ETH_P_TEB) : skb->protocol;
 741         ((__be16 *)(ipv6h + 1))[1] = protocol;
 742
 743         if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
 744                 __be32 *ptr = (__be32 *)(((u8 *)ipv6h) + tunnel->hlen - 4);
 745
 746                 if (tunnel->parms.o_flags&GRE_SEQ) {
 747                         ++tunnel->o_seqno;
 748                         *ptr = htonl(tunnel->o_seqno);
 749                         ptr--;
 750                 }
 751                 if (tunnel->parms.o_flags&GRE_KEY) {
 752                         *ptr = tunnel->parms.o_key;
 753                         ptr--;
 754                 }
 755                 if (tunnel->parms.o_flags&GRE_CSUM) {
 756                         *ptr = 0;
 757                         *(__sum16 *)ptr = ip_compute_csum((void *)(ipv6h+1),
 758                                 skb->len - sizeof(struct ipv6hdr));
 759                 }
 760         }
 761
 762         skb_set_inner_protocol(skb, protocol);
 763
 764         ip6tunnel_xmit(NULL, skb, dev);
 765         return 0;
 766 tx_err_link_failure:
 767         stats->tx_carrier_errors++;
 768         dst_link_failure(skb);
 769 tx_err_dst_release:
 770         dst_release(dst);
 771         return err;
 772 }
 773
 774 static inline int ip6gre_xmit_ipv4(struct sk_buff *skb, struct net_device *dev)
 775 {
 776         struct ip6_tnl *t = netdev_priv(dev);
 777         const struct iphdr  *iph = ip_hdr(skb);
 778         int encap_limit = -1;
 779         struct flowi6 fl6;
 780         __u8 dsfield;
 781         __u32 mtu;
 782         int err;
 783
 784         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 785
 786         if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
 787                 encap_limit = t->parms.encap_limit;
 788
 789         memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
 790         fl6.flowi6_proto = IPPROTO_GRE;
 791
 792         dsfield = ipv4_get_dsfield(iph);
 793
 794         if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
 795                 fl6.flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
 796                                           & IPV6_TCLASS_MASK;
 797         if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
 798                 fl6.flowi6_mark = skb->mark;
 799
 800         err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
 801         if (err != 0) {
 802                 /* XXX: send ICMP error even if DF is not set. */
 803                 if (err == -EMSGSIZE)
 804                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
 805                                   htonl(mtu));
 806                 return -1;
 807         }
 808
 809         return 0;
 810 }
 811
 812 static inline int ip6gre_xmit_ipv6(struct sk_buff *skb, struct net_device *dev)
 813 {
 814         struct ip6_tnl *t = netdev_priv(dev);
 815         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 816         int encap_limit = -1;
 817         __u16 offset;
 818         struct flowi6 fl6;
 819         __u8 dsfield;
 820         __u32 mtu;
 821         int err;
 822
 823         if (ipv6_addr_equal(&t->parms.raddr, &ipv6h->saddr))
 824                 return -1;
 825
 826         offset = ip6_tnl_parse_tlv_enc_lim(skb, skb_network_header(skb));
 827         if (offset > 0) {
 828                 struct ipv6_tlv_tnl_enc_lim *tel;
 829                 tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
 830                 if (tel->encap_limit == 0) {
 831                         icmpv6_send(skb, ICMPV6_PARAMPROB,
 832                                     ICMPV6_HDR_FIELD, offset + 2);
 833                         return -1;
 834                 }
 835                 encap_limit = tel->encap_limit - 1;
 836         } else if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
 837                 encap_limit = t->parms.encap_limit;
 838
 839         memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
 840         fl6.flowi6_proto = IPPROTO_GRE;
 841
 842         dsfield = ipv6_get_dsfield(ipv6h);
 843         if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
 844                 fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
 845         if (t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL)
 846                 fl6.flowlabel |= ip6_flowlabel(ipv6h);
 847         if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
 848                 fl6.flowi6_mark = skb->mark;
 849
 850         err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
 851         if (err != 0) {
 852                 if (err == -EMSGSIZE)
 853                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 854                 return -1;
 855         }
 856
 857         return 0;
 858 }
 859
 860 /**
 861  * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
 862  *   @t: the outgoing tunnel device
 863  *   @hdr: IPv6 header from the incoming packet
 864  *
 865  * Description:
 866  *   Avoid trivial tunneling loop by checking that tunnel exit-point
 867  *   doesn't match source of incoming packet.
 868  *
 869  * Return:
 870  *   1 if conflict,
 871  *   0 else
 872  **/
 873
 874 static inline bool ip6gre_tnl_addr_conflict(const struct ip6_tnl *t,
 875         const struct ipv6hdr *hdr)
 876 {
 877         return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
 878 }
 879
 880 static int ip6gre_xmit_other(struct sk_buff *skb, struct net_device *dev)
 881 {
 882         struct ip6_tnl *t = netdev_priv(dev);
 883         int encap_limit = -1;
 884         struct flowi6 fl6;
 885         __u32 mtu;
 886         int err;
 887
 888         if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
 889                 encap_limit = t->parms.encap_limit;
 890
 891         memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
 892
 893         err = ip6gre_xmit2(skb, dev, 0, &fl6, encap_limit, &mtu);
 894
 895         return err;
 896 }
 897
 898 static netdev_tx_t ip6gre_tunnel_xmit(struct sk_buff *skb,
 899         struct net_device *dev)
 900 {
 901         struct ip6_tnl *t = netdev_priv(dev);
 902         struct net_device_stats *stats = &t->dev->stats;
 903         int ret;
 904
 905         if (!ip6_tnl_xmit_ctl(t, &t->parms.laddr, &t->parms.raddr))
 906                 goto tx_err;
 907
 908         switch (skb->protocol) {
 909         case htons(ETH_P_IP):
 910                 ret = ip6gre_xmit_ipv4(skb, dev);
 911                 break;
 912         case htons(ETH_P_IPV6):
 913                 ret = ip6gre_xmit_ipv6(skb, dev);
 914                 break;
 915         default:
 916                 ret = ip6gre_xmit_other(skb, dev);
 917                 break;
 918         }
 919
 920         if (ret < 0)
 921                 goto tx_err;
 922
 923         return NETDEV_TX_OK;
 924
 925 tx_err:
 926         stats->tx_errors++;
 927         stats->tx_dropped++;
 928         kfree_skb(skb);
 929         return NETDEV_TX_OK;
 930 }
 931
 932 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu)
 933 {
 934         struct net_device *dev = t->dev;
 935         struct __ip6_tnl_parm *p = &t->parms;
 936         struct flowi6 *fl6 = &t->fl.u.ip6;
 937         int addend = sizeof(struct ipv6hdr) + 4;
 938
 939         if (dev->type != ARPHRD_ETHER) {
 940                 memcpy(dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
 941                 memcpy(dev->broadcast, &p->raddr, sizeof(struct in6_addr));
 942         }
 943
 944         /* Set up flowi template */
 945         fl6->saddr = p->laddr;
 946         fl6->daddr = p->raddr;
 947         fl6->flowi6_oif = p->link;
 948         fl6->flowlabel = 0;
 949
 950         if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
 951                 fl6->flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
 952         if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
 953                 fl6->flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;
 954
 955         p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV|IP6_TNL_F_CAP_PER_PACKET);
 956         p->flags |= ip6_tnl_get_cap(t, &p->laddr, &p->raddr);
 957
 958         if (p->flags&IP6_TNL_F_CAP_XMIT &&
 959                         p->flags&IP6_TNL_F_CAP_RCV && dev->type != ARPHRD_ETHER)
 960                 dev->flags |= IFF_POINTOPOINT;
 961         else
 962                 dev->flags &= ~IFF_POINTOPOINT;
 963
 964         /* Precalculate GRE options length */
 965         if (t->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
 966                 if (t->parms.o_flags&GRE_CSUM)
 967                         addend += 4;
 968                 if (t->parms.o_flags&GRE_KEY)
 969                         addend += 4;
 970                 if (t->parms.o_flags&GRE_SEQ)
 971                         addend += 4;
 972         }
 973         t->hlen = addend;
 974
 975         if (p->flags & IP6_TNL_F_CAP_XMIT) {
 976                 int strict = (ipv6_addr_type(&p->raddr) &
 977                               (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));
 978
 979                 struct rt6_info *rt = rt6_lookup(t->net,
 980                                                  &p->raddr, &p->laddr,
 981                                                  p->link, strict);
 982
 983                 if (!rt)
 984                         return;
 985
 986                 if (rt->dst.dev) {
 987                         dev->hard_header_len = rt->dst.dev->hard_header_len + addend;
 988
 989                         if (set_mtu) {
 990                                 dev->mtu = rt->dst.dev->mtu - addend;
 991                                 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
 992                                         dev->mtu -= 8;
 993
 994                                 if (dev->mtu < IPV6_MIN_MTU)
 995                                         dev->mtu = IPV6_MIN_MTU;
 996                         }
 997                 }
 998                 ip6_rt_put(rt);
 999         }
1000 }
1001
1002 static int ip6gre_tnl_change(struct ip6_tnl *t,
1003         const struct __ip6_tnl_parm *p, int set_mtu)
1004 {
1005         t->parms.laddr = p->laddr;
1006         t->parms.raddr = p->raddr;
1007         t->parms.flags = p->flags;
1008         t->parms.hop_limit = p->hop_limit;
1009         t->parms.encap_limit = p->encap_limit;
1010         t->parms.flowinfo = p->flowinfo;
1011         t->parms.link = p->link;
1012         t->parms.proto = p->proto;
1013         t->parms.i_key = p->i_key;
1014         t->parms.o_key = p->o_key;
1015         t->parms.i_flags = p->i_flags;
1016         t->parms.o_flags = p->o_flags;
1017         ip6_tnl_dst_reset(t);
1018         ip6gre_tnl_link_config(t, set_mtu);
1019         return 0;
1020 }
1021
1022 static void ip6gre_tnl_parm_from_user(struct __ip6_tnl_parm *p,
1023         const struct ip6_tnl_parm2 *u)
1024 {
1025         p->laddr = u->laddr;
1026         p->raddr = u->raddr;
1027         p->flags = u->flags;
1028         p->hop_limit = u->hop_limit;
1029         p->encap_limit = u->encap_limit;
1030         p->flowinfo = u->flowinfo;
1031         p->link = u->link;
1032         p->i_key = u->i_key;
1033         p->o_key = u->o_key;
1034         p->i_flags = u->i_flags;
1035         p->o_flags = u->o_flags;
1036         memcpy(p->name, u->name, sizeof(u->name));
1037 }
1038
1039 static void ip6gre_tnl_parm_to_user(struct ip6_tnl_parm2 *u,
1040         const struct __ip6_tnl_parm *p)
1041 {
1042         u->proto = IPPROTO_GRE;
1043         u->laddr = p->laddr;
1044         u->raddr = p->raddr;
1045         u->flags = p->flags;
1046         u->hop_limit = p->hop_limit;
1047         u->encap_limit = p->encap_limit;
1048         u->flowinfo = p->flowinfo;
1049         u->link = p->link;
1050         u->i_key = p->i_key;
1051         u->o_key = p->o_key;
1052         u->i_flags = p->i_flags;
1053         u->o_flags = p->o_flags;
1054         memcpy(u->name, p->name, sizeof(u->name));
1055 }
1056
1057 static int ip6gre_tunnel_ioctl(struct net_device *dev,
1058         struct ifreq *ifr, int cmd)
1059 {
1060         int err = 0;
1061         struct ip6_tnl_parm2 p;
1062         struct __ip6_tnl_parm p1;
1063         struct ip6_tnl *t = netdev_priv(dev);
1064         struct net *net = t->net;
1065         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1066
1067         switch (cmd) {
1068         case SIOCGETTUNNEL:
1069                 if (dev == ign->fb_tunnel_dev) {
1070                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1071                                 err = -EFAULT;
1072                                 break;
1073                         }
1074                         ip6gre_tnl_parm_from_user(&p1, &p);
1075                         t = ip6gre_tunnel_locate(net, &p1, 0);
1076                         if (!t)
1077                                 t = netdev_priv(dev);
1078                 }
1079                 memset(&p, 0, sizeof(p));
1080                 ip6gre_tnl_parm_to_user(&p, &t->parms);
1081                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1082                         err = -EFAULT;
1083                 break;
1084
1085         case SIOCADDTUNNEL:
1086         case SIOCCHGTUNNEL:
1087                 err = -EPERM;
1088                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1089                         goto done;
1090
1091                 err = -EFAULT;
1092                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1093                         goto done;
1094
1095                 err = -EINVAL;
1096                 if ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))
1097                         goto done;
1098
1099                 if (!(p.i_flags&GRE_KEY))
1100                         p.i_key = 0;
1101                 if (!(p.o_flags&GRE_KEY))
1102                         p.o_key = 0;
1103
1104                 ip6gre_tnl_parm_from_user(&p1, &p);
1105                 t = ip6gre_tunnel_locate(net, &p1, cmd == SIOCADDTUNNEL);
1106
1107                 if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1108                         if (t) {
1109                                 if (t->dev != dev) {
1110                                         err = -EEXIST;
1111                                         break;
1112                                 }
1113                         } else {
1114                                 t = netdev_priv(dev);
1115
1116                                 ip6gre_tunnel_unlink(ign, t);
1117                                 synchronize_net();
1118                                 ip6gre_tnl_change(t, &p1, 1);
1119                                 ip6gre_tunnel_link(ign, t);
1120                                 netdev_state_change(dev);
1121                         }
1122                 }
1123
1124                 if (t) {
1125                         err = 0;
1126
1127                         memset(&p, 0, sizeof(p));
1128                         ip6gre_tnl_parm_to_user(&p, &t->parms);
1129                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1130                                 err = -EFAULT;
1131                 } else
1132                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1133                 break;
1134
1135         case SIOCDELTUNNEL:
1136                 err = -EPERM;
1137                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1138                         goto done;
1139
1140                 if (dev == ign->fb_tunnel_dev) {
1141                         err = -EFAULT;
1142                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1143                                 goto done;
1144                         err = -ENOENT;
1145                         ip6gre_tnl_parm_from_user(&p1, &p);
1146                         t = ip6gre_tunnel_locate(net, &p1, 0);
1147                         if (!t)
1148                                 goto done;
1149                         err = -EPERM;
1150                         if (t == netdev_priv(ign->fb_tunnel_dev))
1151                                 goto done;
1152                         dev = t->dev;
1153                 }
1154                 unregister_netdevice(dev);
1155                 err = 0;
1156                 break;
1157
1158         default:
1159                 err = -EINVAL;
1160         }
1161
1162 done:
1163         return err;
1164 }
1165
1166 static int ip6gre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1167 {
1168         if (new_mtu < 68 ||
1169             new_mtu > 0xFFF8 - dev->hard_header_len)
1170                 return -EINVAL;
1171         dev->mtu = new_mtu;
1172         return 0;
1173 }
1174
1175 static int ip6gre_header(struct sk_buff *skb, struct net_device *dev,
1176                         unsigned short type,
1177                         const void *daddr, const void *saddr, unsigned int len)
1178 {
1179         struct ip6_tnl *t = netdev_priv(dev);
1180         struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen);
1181         __be16 *p = (__be16 *)(ipv6h+1);
1182
1183         ip6_flow_hdr(ipv6h, 0,
1184                      ip6_make_flowlabel(dev_net(dev), skb,
1185                                         t->fl.u.ip6.flowlabel, true,
1186                                         &t->fl.u.ip6));
1187         ipv6h->hop_limit = t->parms.hop_limit;
1188         ipv6h->nexthdr = NEXTHDR_GRE;
1189         ipv6h->saddr = t->parms.laddr;
1190         ipv6h->daddr = t->parms.raddr;
1191
1192         p[0]            = t->parms.o_flags;
1193         p[1]            = htons(type);
1194
1195         /*
1196          *      Set the source hardware address.
1197          */
1198
1199         if (saddr)
1200                 memcpy(&ipv6h->saddr, saddr, sizeof(struct in6_addr));
1201         if (daddr)
1202                 memcpy(&ipv6h->daddr, daddr, sizeof(struct in6_addr));
1203         if (!ipv6_addr_any(&ipv6h->daddr))
1204                 return t->hlen;
1205
1206         return -t->hlen;
1207 }
1208
1209 static const struct header_ops ip6gre_header_ops = {
1210         .create = ip6gre_header,
1211 };
1212
1213 static const struct net_device_ops ip6gre_netdev_ops = {
1214         .ndo_init               = ip6gre_tunnel_init,
1215         .ndo_uninit             = ip6gre_tunnel_uninit,
1216         .ndo_start_xmit         = ip6gre_tunnel_xmit,
1217         .ndo_do_ioctl           = ip6gre_tunnel_ioctl,
1218         .ndo_change_mtu         = ip6gre_tunnel_change_mtu,
1219         .ndo_get_stats64        = ip_tunnel_get_stats64,
1220         .ndo_get_iflink         = ip6_tnl_get_iflink,
1221 };
1222
1223 static void ip6gre_dev_free(struct net_device *dev)
1224 {
1225         struct ip6_tnl *t = netdev_priv(dev);
1226
1227         ip6_tnl_dst_destroy(t);
1228         free_percpu(dev->tstats);
1229         free_netdev(dev);
1230 }
1231
1232 static void ip6gre_tunnel_setup(struct net_device *dev)
1233 {
1234         struct ip6_tnl *t;
1235
1236         dev->netdev_ops = &ip6gre_netdev_ops;
1237         dev->destructor = ip6gre_dev_free;
1238
1239         dev->type = ARPHRD_IP6GRE;
1240         dev->hard_header_len = LL_MAX_HEADER + sizeof(struct ipv6hdr) + 4;
1241         dev->mtu = ETH_DATA_LEN - sizeof(struct ipv6hdr) - 4;
1242         t = netdev_priv(dev);
1243         if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1244                 dev->mtu -= 8;
1245         dev->flags |= IFF_NOARP;
1246         dev->addr_len = sizeof(struct in6_addr);
1247         netif_keep_dst(dev);
1248 }
1249
1250 static int ip6gre_tunnel_init_common(struct net_device *dev)
1251 {
1252         struct ip6_tnl *tunnel;
1253         int ret;
1254
1255         tunnel = netdev_priv(dev);
1256
1257         tunnel->dev = dev;
1258         tunnel->net = dev_net(dev);
1259         strcpy(tunnel->parms.name, dev->name);
1260
1261         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1262         if (!dev->tstats)
1263                 return -ENOMEM;
1264
1265         ret = ip6_tnl_dst_init(tunnel);
1266         if (ret) {
1267                 free_percpu(dev->tstats);
1268                 dev->tstats = NULL;
1269                 return ret;
1270         }
1271
1272         return 0;
1273 }
1274
1275 static int ip6gre_tunnel_init(struct net_device *dev)
1276 {
1277         struct ip6_tnl *tunnel;
1278         int ret;
1279
1280         ret = ip6gre_tunnel_init_common(dev);
1281         if (ret)
1282                 return ret;
1283
1284         tunnel = netdev_priv(dev);
1285
1286         memcpy(dev->dev_addr, &tunnel->parms.laddr, sizeof(struct in6_addr));
1287         memcpy(dev->broadcast, &tunnel->parms.raddr, sizeof(struct in6_addr));
1288
1289         if (ipv6_addr_any(&tunnel->parms.raddr))
1290                 dev->header_ops = &ip6gre_header_ops;
1291
1292         return 0;
1293 }
1294
1295 static void ip6gre_fb_tunnel_init(struct net_device *dev)
1296 {
1297         struct ip6_tnl *tunnel = netdev_priv(dev);
1298
1299         tunnel->dev = dev;
1300         tunnel->net = dev_net(dev);
1301         strcpy(tunnel->parms.name, dev->name);
1302
1303         tunnel->hlen            = sizeof(struct ipv6hdr) + 4;
1304
1305         dev_hold(dev);
1306 }
1307
1308
1309 static struct inet6_protocol ip6gre_protocol __read_mostly = {
1310         .handler     = ip6gre_rcv,
1311         .err_handler = ip6gre_err,
1312         .flags       = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1313 };
1314
1315 static void ip6gre_destroy_tunnels(struct net *net, struct list_head *head)
1316 {
1317         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1318         struct net_device *dev, *aux;
1319         int prio;
1320
1321         for_each_netdev_safe(net, dev, aux)
1322                 if (dev->rtnl_link_ops == &ip6gre_link_ops ||
1323                     dev->rtnl_link_ops == &ip6gre_tap_ops)
1324                         unregister_netdevice_queue(dev, head);
1325
1326         for (prio = 0; prio < 4; prio++) {
1327                 int h;
1328                 for (h = 0; h < HASH_SIZE; h++) {
1329                         struct ip6_tnl *t;
1330
1331                         t = rtnl_dereference(ign->tunnels[prio][h]);
1332
1333                         while (t) {
1334                                 /* If dev is in the same netns, it has already
1335                                  * been added to the list by the previous loop.
1336                                  */
1337                                 if (!net_eq(dev_net(t->dev), net))
1338                                         unregister_netdevice_queue(t->dev,
1339                                                                    head);
1340                                 t = rtnl_dereference(t->next);
1341                         }
1342                 }
1343         }
1344 }
1345
1346 static int __net_init ip6gre_init_net(struct net *net)
1347 {
1348         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1349         int err;
1350
1351         ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6gre0",
1352                                           NET_NAME_UNKNOWN,
1353                                           ip6gre_tunnel_setup);
1354         if (!ign->fb_tunnel_dev) {
1355                 err = -ENOMEM;
1356                 goto err_alloc_dev;
1357         }
1358         dev_net_set(ign->fb_tunnel_dev, net);
1359         /* FB netdevice is special: we have one, and only one per netns.
1360          * Allowing to move it to another netns is clearly unsafe.
1361          */
1362         ign->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL;
1363
1364
1365         ip6gre_fb_tunnel_init(ign->fb_tunnel_dev);
1366         ign->fb_tunnel_dev->rtnl_link_ops = &ip6gre_link_ops;
1367
1368         err = register_netdev(ign->fb_tunnel_dev);
1369         if (err)
1370                 goto err_reg_dev;
1371
1372         rcu_assign_pointer(ign->tunnels_wc[0],
1373                            netdev_priv(ign->fb_tunnel_dev));
1374         return 0;
1375
1376 err_reg_dev:
1377         ip6gre_dev_free(ign->fb_tunnel_dev);
1378 err_alloc_dev:
1379         return err;
1380 }
1381
1382 static void __net_exit ip6gre_exit_net(struct net *net)
1383 {
1384         LIST_HEAD(list);
1385
1386         rtnl_lock();
1387         ip6gre_destroy_tunnels(net, &list);
1388         unregister_netdevice_many(&list);
1389         rtnl_unlock();
1390 }
1391
1392 static struct pernet_operations ip6gre_net_ops = {
1393         .init = ip6gre_init_net,
1394         .exit = ip6gre_exit_net,
1395         .id   = &ip6gre_net_id,
1396         .size = sizeof(struct ip6gre_net),
1397 };
1398
1399 static int ip6gre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1400 {
1401         __be16 flags;
1402
1403         if (!data)
1404                 return 0;
1405
1406         flags = 0;
1407         if (data[IFLA_GRE_IFLAGS])
1408                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1409         if (data[IFLA_GRE_OFLAGS])
1410                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1411         if (flags & (GRE_VERSION|GRE_ROUTING))
1412                 return -EINVAL;
1413
1414         return 0;
1415 }
1416
1417 static int ip6gre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1418 {
1419         struct in6_addr daddr;
1420
1421         if (tb[IFLA_ADDRESS]) {
1422                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1423                         return -EINVAL;
1424                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1425                         return -EADDRNOTAVAIL;
1426         }
1427
1428         if (!data)
1429                 goto out;
1430
1431         if (data[IFLA_GRE_REMOTE]) {
1432                 daddr = nla_get_in6_addr(data[IFLA_GRE_REMOTE]);
1433                 if (ipv6_addr_any(&daddr))
1434                         return -EINVAL;
1435         }
1436
1437 out:
1438         return ip6gre_tunnel_validate(tb, data);
1439 }
1440
1441
1442 static void ip6gre_netlink_parms(struct nlattr *data[],
1443                                 struct __ip6_tnl_parm *parms)
1444 {
1445         memset(parms, 0, sizeof(*parms));
1446
1447         if (!data)
1448                 return;
1449
1450         if (data[IFLA_GRE_LINK])
1451                 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1452
1453         if (data[IFLA_GRE_IFLAGS])
1454                 parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1455
1456         if (data[IFLA_GRE_OFLAGS])
1457                 parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1458
1459         if (data[IFLA_GRE_IKEY])
1460                 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1461
1462         if (data[IFLA_GRE_OKEY])
1463                 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1464
1465         if (data[IFLA_GRE_LOCAL])
1466                 parms->laddr = nla_get_in6_addr(data[IFLA_GRE_LOCAL]);
1467
1468         if (data[IFLA_GRE_REMOTE])
1469                 parms->raddr = nla_get_in6_addr(data[IFLA_GRE_REMOTE]);
1470
1471         if (data[IFLA_GRE_TTL])
1472                 parms->hop_limit = nla_get_u8(data[IFLA_GRE_TTL]);
1473
1474         if (data[IFLA_GRE_ENCAP_LIMIT])
1475                 parms->encap_limit = nla_get_u8(data[IFLA_GRE_ENCAP_LIMIT]);
1476
1477         if (data[IFLA_GRE_FLOWINFO])
1478                 parms->flowinfo = nla_get_u32(data[IFLA_GRE_FLOWINFO]);
1479
1480         if (data[IFLA_GRE_FLAGS])
1481                 parms->flags = nla_get_u32(data[IFLA_GRE_FLAGS]);
1482 }
1483
1484 static int ip6gre_tap_init(struct net_device *dev)
1485 {
1486         struct ip6_tnl *tunnel;
1487         int ret;
1488
1489         ret = ip6gre_tunnel_init_common(dev);
1490         if (ret)
1491                 return ret;
1492
1493         tunnel = netdev_priv(dev);
1494
1495         ip6gre_tnl_link_config(tunnel, 1);
1496
1497         return 0;
1498 }
1499
1500 static const struct net_device_ops ip6gre_tap_netdev_ops = {
1501         .ndo_init = ip6gre_tap_init,
1502         .ndo_uninit = ip6gre_tunnel_uninit,
1503         .ndo_start_xmit = ip6gre_tunnel_xmit,
1504         .ndo_set_mac_address = eth_mac_addr,
1505         .ndo_validate_addr = eth_validate_addr,
1506         .ndo_change_mtu = ip6gre_tunnel_change_mtu,
1507         .ndo_get_stats64 = ip_tunnel_get_stats64,
1508         .ndo_get_iflink = ip6_tnl_get_iflink,
1509 };
1510
1511 static void ip6gre_tap_setup(struct net_device *dev)
1512 {
1513
1514         ether_setup(dev);
1515
1516         dev->netdev_ops = &ip6gre_tap_netdev_ops;
1517         dev->destructor = ip6gre_dev_free;
1518
1519         dev->features |= NETIF_F_NETNS_LOCAL;
1520 }
1521
1522 static int ip6gre_newlink(struct net *src_net, struct net_device *dev,
1523         struct nlattr *tb[], struct nlattr *data[])
1524 {
1525         struct ip6_tnl *nt;
1526         struct net *net = dev_net(dev);
1527         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1528         int err;
1529
1530         nt = netdev_priv(dev);
1531         ip6gre_netlink_parms(data, &nt->parms);
1532
1533         if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
1534                 return -EEXIST;
1535
1536         if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1537                 eth_hw_addr_random(dev);
1538
1539         nt->dev = dev;
1540         nt->net = dev_net(dev);
1541         ip6gre_tnl_link_config(nt, !tb[IFLA_MTU]);
1542
1543         /* Can use a lockless transmit, unless we generate output sequences */
1544         if (!(nt->parms.o_flags & GRE_SEQ))
1545                 dev->features |= NETIF_F_LLTX;
1546
1547         err = register_netdevice(dev);
1548         if (err)
1549                 goto out;
1550
1551         dev_hold(dev);
1552         ip6gre_tunnel_link(ign, nt);
1553
1554 out:
1555         return err;
1556 }
1557
1558 static int ip6gre_changelink(struct net_device *dev, struct nlattr *tb[],
1559                             struct nlattr *data[])
1560 {
1561         struct ip6_tnl *t, *nt = netdev_priv(dev);
1562         struct net *net = nt->net;
1563         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1564         struct __ip6_tnl_parm p;
1565
1566         if (dev == ign->fb_tunnel_dev)
1567                 return -EINVAL;
1568
1569         ip6gre_netlink_parms(data, &p);
1570
1571         t = ip6gre_tunnel_locate(net, &p, 0);
1572
1573         if (t) {
1574                 if (t->dev != dev)
1575                         return -EEXIST;
1576         } else {
1577                 t = nt;
1578         }
1579
1580         ip6gre_tunnel_unlink(ign, t);
1581         ip6gre_tnl_change(t, &p, !tb[IFLA_MTU]);
1582         ip6gre_tunnel_link(ign, t);
1583         return 0;
1584 }
1585
1586 static void ip6gre_dellink(struct net_device *dev, struct list_head *head)
1587 {
1588         struct net *net = dev_net(dev);
1589         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1590
1591         if (dev != ign->fb_tunnel_dev)
1592                 unregister_netdevice_queue(dev, head);
1593 }
1594
1595 static size_t ip6gre_get_size(const struct net_device *dev)
1596 {
1597         return
1598                 /* IFLA_GRE_LINK */
1599                 nla_total_size(4) +
1600                 /* IFLA_GRE_IFLAGS */
1601                 nla_total_size(2) +
1602                 /* IFLA_GRE_OFLAGS */
1603                 nla_total_size(2) +
1604                 /* IFLA_GRE_IKEY */
1605                 nla_total_size(4) +
1606                 /* IFLA_GRE_OKEY */
1607                 nla_total_size(4) +
1608                 /* IFLA_GRE_LOCAL */
1609                 nla_total_size(sizeof(struct in6_addr)) +
1610                 /* IFLA_GRE_REMOTE */
1611                 nla_total_size(sizeof(struct in6_addr)) +
1612                 /* IFLA_GRE_TTL */
1613                 nla_total_size(1) +
1614                 /* IFLA_GRE_TOS */
1615                 nla_total_size(1) +
1616                 /* IFLA_GRE_ENCAP_LIMIT */
1617                 nla_total_size(1) +
1618                 /* IFLA_GRE_FLOWINFO */
1619                 nla_total_size(4) +
1620                 /* IFLA_GRE_FLAGS */
1621                 nla_total_size(4) +
1622                 0;
1623 }
1624
1625 static int ip6gre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1626 {
1627         struct ip6_tnl *t = netdev_priv(dev);
1628         struct __ip6_tnl_parm *p = &t->parms;
1629
1630         if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1631             nla_put_be16(skb, IFLA_GRE_IFLAGS, p->i_flags) ||
1632             nla_put_be16(skb, IFLA_GRE_OFLAGS, p->o_flags) ||
1633             nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1634             nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1635             nla_put_in6_addr(skb, IFLA_GRE_LOCAL, &p->laddr) ||
1636             nla_put_in6_addr(skb, IFLA_GRE_REMOTE, &p->raddr) ||
1637             nla_put_u8(skb, IFLA_GRE_TTL, p->hop_limit) ||
1638             /*nla_put_u8(skb, IFLA_GRE_TOS, t->priority) ||*/
1639             nla_put_u8(skb, IFLA_GRE_ENCAP_LIMIT, p->encap_limit) ||
1640             nla_put_be32(skb, IFLA_GRE_FLOWINFO, p->flowinfo) ||
1641             nla_put_u32(skb, IFLA_GRE_FLAGS, p->flags))
1642                 goto nla_put_failure;
1643         return 0;
1644
1645 nla_put_failure:
1646         return -EMSGSIZE;
1647 }
1648
1649 static const struct nla_policy ip6gre_policy[IFLA_GRE_MAX + 1] = {
1650         [IFLA_GRE_LINK]        = { .type = NLA_U32 },
1651         [IFLA_GRE_IFLAGS]      = { .type = NLA_U16 },
1652         [IFLA_GRE_OFLAGS]      = { .type = NLA_U16 },
1653         [IFLA_GRE_IKEY]        = { .type = NLA_U32 },
1654         [IFLA_GRE_OKEY]        = { .type = NLA_U32 },
1655         [IFLA_GRE_LOCAL]       = { .len = FIELD_SIZEOF(struct ipv6hdr, saddr) },
1656         [IFLA_GRE_REMOTE]      = { .len = FIELD_SIZEOF(struct ipv6hdr, daddr) },
1657         [IFLA_GRE_TTL]         = { .type = NLA_U8 },
1658         [IFLA_GRE_ENCAP_LIMIT] = { .type = NLA_U8 },
1659         [IFLA_GRE_FLOWINFO]    = { .type = NLA_U32 },
1660         [IFLA_GRE_FLAGS]       = { .type = NLA_U32 },
1661 };
1662
1663 static struct rtnl_link_ops ip6gre_link_ops __read_mostly = {
1664         .kind           = "ip6gre",
1665         .maxtype        = IFLA_GRE_MAX,
1666         .policy         = ip6gre_policy,
1667         .priv_size      = sizeof(struct ip6_tnl),
1668         .setup          = ip6gre_tunnel_setup,
1669         .validate       = ip6gre_tunnel_validate,
1670         .newlink        = ip6gre_newlink,
1671         .changelink     = ip6gre_changelink,
1672         .dellink        = ip6gre_dellink,
1673         .get_size       = ip6gre_get_size,
1674         .fill_info      = ip6gre_fill_info,
1675         .get_link_net   = ip6_tnl_get_link_net,
1676 };
1677
1678 static struct rtnl_link_ops ip6gre_tap_ops __read_mostly = {
1679         .kind           = "ip6gretap",
1680         .maxtype        = IFLA_GRE_MAX,
1681         .policy         = ip6gre_policy,
1682         .priv_size      = sizeof(struct ip6_tnl),
1683         .setup          = ip6gre_tap_setup,
1684         .validate       = ip6gre_tap_validate,
1685         .newlink        = ip6gre_newlink,
1686         .changelink     = ip6gre_changelink,
1687         .get_size       = ip6gre_get_size,
1688         .fill_info      = ip6gre_fill_info,
1689         .get_link_net   = ip6_tnl_get_link_net,
1690 };
1691
1692 /*
1693  *      And now the modules code and kernel interface.
1694  */
1695
1696 static int __init ip6gre_init(void)
1697 {
1698         int err;
1699
1700         pr_info("GRE over IPv6 tunneling driver\n");
1701
1702         err = register_pernet_device(&ip6gre_net_ops);
1703         if (err < 0)
1704                 return err;
1705
1706         err = inet6_add_protocol(&ip6gre_protocol, IPPROTO_GRE);
1707         if (err < 0) {
1708                 pr_info("%s: can't add protocol\n", __func__);
1709                 goto add_proto_failed;
1710         }
1711
1712         err = rtnl_link_register(&ip6gre_link_ops);
1713         if (err < 0)
1714                 goto rtnl_link_failed;
1715
1716         err = rtnl_link_register(&ip6gre_tap_ops);
1717         if (err < 0)
1718                 goto tap_ops_failed;
1719
1720 out:
1721         return err;
1722
1723 tap_ops_failed:
1724         rtnl_link_unregister(&ip6gre_link_ops);
1725 rtnl_link_failed:
1726         inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1727 add_proto_failed:
1728         unregister_pernet_device(&ip6gre_net_ops);
1729         goto out;
1730 }
1731
1732 static void __exit ip6gre_fini(void)
1733 {
1734         rtnl_link_unregister(&ip6gre_tap_ops);
1735         rtnl_link_unregister(&ip6gre_link_ops);
1736         inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1737         unregister_pernet_device(&ip6gre_net_ops);
1738 }
1739
1740 module_init(ip6gre_init);
1741 module_exit(ip6gre_fini);
1742 MODULE_LICENSE("GPL");
1743 MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
1744 MODULE_DESCRIPTION("GRE over IPv6 tunneling device");
1745 MODULE_ALIAS_RTNL_LINK("ip6gre");
1746 MODULE_ALIAS_RTNL_LINK("ip6gretap");
1747 MODULE_ALIAS_NETDEV("ip6gre0");