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