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
  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_warn_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_warn_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_warn_ratelimited("%s: Too small encapsulation limit or routing loop in tunnel!\n",
 425                                                      t->parms.name);
 426                         }
 427                 } else {
 428                         net_warn_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_check(tunnel);
 638
 639         if (!dst) {
 640                 ndst = ip6_route_output(net, NULL, fl6);
 641
 642                 if (ndst->error)
 643                         goto tx_err_link_failure;
 644                 ndst = xfrm_lookup(net, ndst, flowi6_to_flowi(fl6), NULL, 0);
 645                 if (IS_ERR(ndst)) {
 646                         err = PTR_ERR(ndst);
 647                         ndst = NULL;
 648                         goto tx_err_link_failure;
 649                 }
 650                 dst = ndst;
 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) {
 706                 skb_dst_set(skb, dst);
 707                 ndst = NULL;
 708         } else {
 709                 skb_dst_set_noref(skb, dst);
 710         }
 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, false));
 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         if (ndst)
 766                 ip6_tnl_dst_store(tunnel, ndst);
 767         return 0;
 768 tx_err_link_failure:
 769         stats->tx_carrier_errors++;
 770         dst_link_failure(skb);
 771 tx_err_dst_release:
 772         dst_release(ndst);
 773         return err;
 774 }
 775
 776 static inline int ip6gre_xmit_ipv4(struct sk_buff *skb, struct net_device *dev)
 777 {
 778         struct ip6_tnl *t = netdev_priv(dev);
 779         const struct iphdr  *iph = ip_hdr(skb);
 780         int encap_limit = -1;
 781         struct flowi6 fl6;
 782         __u8 dsfield;
 783         __u32 mtu;
 784         int err;
 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         fl6.flowi6_proto = skb->protocol;
 893
 894         err = ip6gre_xmit2(skb, dev, 0, &fl6, encap_limit, &mtu);
 895
 896         return err;
 897 }
 898
 899 static netdev_tx_t ip6gre_tunnel_xmit(struct sk_buff *skb,
 900         struct net_device *dev)
 901 {
 902         struct ip6_tnl *t = netdev_priv(dev);
 903         struct net_device_stats *stats = &t->dev->stats;
 904         int ret;
 905
 906         if (!ip6_tnl_xmit_ctl(t, &t->parms.laddr, &t->parms.raddr))
 907                 goto tx_err;
 908
 909         switch (skb->protocol) {
 910         case htons(ETH_P_IP):
 911                 ret = ip6gre_xmit_ipv4(skb, dev);
 912                 break;
 913         case htons(ETH_P_IPV6):
 914                 ret = ip6gre_xmit_ipv6(skb, dev);
 915                 break;
 916         default:
 917                 ret = ip6gre_xmit_other(skb, dev);
 918                 break;
 919         }
 920
 921         if (ret < 0)
 922                 goto tx_err;
 923
 924         return NETDEV_TX_OK;
 925
 926 tx_err:
 927         stats->tx_errors++;
 928         stats->tx_dropped++;
 929         kfree_skb(skb);
 930         return NETDEV_TX_OK;
 931 }
 932
 933 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu)
 934 {
 935         struct net_device *dev = t->dev;
 936         struct __ip6_tnl_parm *p = &t->parms;
 937         struct flowi6 *fl6 = &t->fl.u.ip6;
 938         int addend = sizeof(struct ipv6hdr) + 4;
 939
 940         if (dev->type != ARPHRD_ETHER) {
 941                 memcpy(dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
 942                 memcpy(dev->broadcast, &p->raddr, sizeof(struct in6_addr));
 943         }
 944
 945         /* Set up flowi template */
 946         fl6->saddr = p->laddr;
 947         fl6->daddr = p->raddr;
 948         fl6->flowi6_oif = p->link;
 949         fl6->flowlabel = 0;
 950
 951         if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
 952                 fl6->flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
 953         if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
 954                 fl6->flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;
 955
 956         p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV|IP6_TNL_F_CAP_PER_PACKET);
 957         p->flags |= ip6_tnl_get_cap(t, &p->laddr, &p->raddr);
 958
 959         if (p->flags&IP6_TNL_F_CAP_XMIT &&
 960                         p->flags&IP6_TNL_F_CAP_RCV && dev->type != ARPHRD_ETHER)
 961                 dev->flags |= IFF_POINTOPOINT;
 962         else
 963                 dev->flags &= ~IFF_POINTOPOINT;
 964
 965         /* Precalculate GRE options length */
 966         if (t->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
 967                 if (t->parms.o_flags&GRE_CSUM)
 968                         addend += 4;
 969                 if (t->parms.o_flags&GRE_KEY)
 970                         addend += 4;
 971                 if (t->parms.o_flags&GRE_SEQ)
 972                         addend += 4;
 973         }
 974         t->hlen = addend;
 975
 976         if (p->flags & IP6_TNL_F_CAP_XMIT) {
 977                 int strict = (ipv6_addr_type(&p->raddr) &
 978                               (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));
 979
 980                 struct rt6_info *rt = rt6_lookup(t->net,
 981                                                  &p->raddr, &p->laddr,
 982                                                  p->link, strict);
 983
 984                 if (!rt)
 985                         return;
 986
 987                 if (rt->dst.dev) {
 988                         dev->hard_header_len = rt->dst.dev->hard_header_len + addend;
 989
 990                         if (set_mtu) {
 991                                 dev->mtu = rt->dst.dev->mtu - addend;
 992                                 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
 993                                         dev->mtu -= 8;
 994
 995                                 if (dev->mtu < IPV6_MIN_MTU)
 996                                         dev->mtu = IPV6_MIN_MTU;
 997                         }
 998                 }
 999                 ip6_rt_put(rt);
1000         }
1001 }
1002
1003 static int ip6gre_tnl_change(struct ip6_tnl *t,
1004         const struct __ip6_tnl_parm *p, int set_mtu)
1005 {
1006         t->parms.laddr = p->laddr;
1007         t->parms.raddr = p->raddr;
1008         t->parms.flags = p->flags;
1009         t->parms.hop_limit = p->hop_limit;
1010         t->parms.encap_limit = p->encap_limit;
1011         t->parms.flowinfo = p->flowinfo;
1012         t->parms.link = p->link;
1013         t->parms.proto = p->proto;
1014         t->parms.i_key = p->i_key;
1015         t->parms.o_key = p->o_key;
1016         t->parms.i_flags = p->i_flags;
1017         t->parms.o_flags = p->o_flags;
1018         ip6_tnl_dst_reset(t);
1019         ip6gre_tnl_link_config(t, set_mtu);
1020         return 0;
1021 }
1022
1023 static void ip6gre_tnl_parm_from_user(struct __ip6_tnl_parm *p,
1024         const struct ip6_tnl_parm2 *u)
1025 {
1026         p->laddr = u->laddr;
1027         p->raddr = u->raddr;
1028         p->flags = u->flags;
1029         p->hop_limit = u->hop_limit;
1030         p->encap_limit = u->encap_limit;
1031         p->flowinfo = u->flowinfo;
1032         p->link = u->link;
1033         p->i_key = u->i_key;
1034         p->o_key = u->o_key;
1035         p->i_flags = u->i_flags;
1036         p->o_flags = u->o_flags;
1037         memcpy(p->name, u->name, sizeof(u->name));
1038 }
1039
1040 static void ip6gre_tnl_parm_to_user(struct ip6_tnl_parm2 *u,
1041         const struct __ip6_tnl_parm *p)
1042 {
1043         u->proto = IPPROTO_GRE;
1044         u->laddr = p->laddr;
1045         u->raddr = p->raddr;
1046         u->flags = p->flags;
1047         u->hop_limit = p->hop_limit;
1048         u->encap_limit = p->encap_limit;
1049         u->flowinfo = p->flowinfo;
1050         u->link = p->link;
1051         u->i_key = p->i_key;
1052         u->o_key = p->o_key;
1053         u->i_flags = p->i_flags;
1054         u->o_flags = p->o_flags;
1055         memcpy(u->name, p->name, sizeof(u->name));
1056 }
1057
1058 static int ip6gre_tunnel_ioctl(struct net_device *dev,
1059         struct ifreq *ifr, int cmd)
1060 {
1061         int err = 0;
1062         struct ip6_tnl_parm2 p;
1063         struct __ip6_tnl_parm p1;
1064         struct ip6_tnl *t = netdev_priv(dev);
1065         struct net *net = t->net;
1066         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1067
1068         switch (cmd) {
1069         case SIOCGETTUNNEL:
1070                 if (dev == ign->fb_tunnel_dev) {
1071                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1072                                 err = -EFAULT;
1073                                 break;
1074                         }
1075                         ip6gre_tnl_parm_from_user(&p1, &p);
1076                         t = ip6gre_tunnel_locate(net, &p1, 0);
1077                         if (!t)
1078                                 t = netdev_priv(dev);
1079                 }
1080                 memset(&p, 0, sizeof(p));
1081                 ip6gre_tnl_parm_to_user(&p, &t->parms);
1082                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1083                         err = -EFAULT;
1084                 break;
1085
1086         case SIOCADDTUNNEL:
1087         case SIOCCHGTUNNEL:
1088                 err = -EPERM;
1089                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1090                         goto done;
1091
1092                 err = -EFAULT;
1093                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1094                         goto done;
1095
1096                 err = -EINVAL;
1097                 if ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))
1098                         goto done;
1099
1100                 if (!(p.i_flags&GRE_KEY))
1101                         p.i_key = 0;
1102                 if (!(p.o_flags&GRE_KEY))
1103                         p.o_key = 0;
1104
1105                 ip6gre_tnl_parm_from_user(&p1, &p);
1106                 t = ip6gre_tunnel_locate(net, &p1, cmd == SIOCADDTUNNEL);
1107
1108                 if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1109                         if (t) {
1110                                 if (t->dev != dev) {
1111                                         err = -EEXIST;
1112                                         break;
1113                                 }
1114                         } else {
1115                                 t = netdev_priv(dev);
1116
1117                                 ip6gre_tunnel_unlink(ign, t);
1118                                 synchronize_net();
1119                                 ip6gre_tnl_change(t, &p1, 1);
1120                                 ip6gre_tunnel_link(ign, t);
1121                                 netdev_state_change(dev);
1122                         }
1123                 }
1124
1125                 if (t) {
1126                         err = 0;
1127
1128                         memset(&p, 0, sizeof(p));
1129                         ip6gre_tnl_parm_to_user(&p, &t->parms);
1130                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1131                                 err = -EFAULT;
1132                 } else
1133                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1134                 break;
1135
1136         case SIOCDELTUNNEL:
1137                 err = -EPERM;
1138                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1139                         goto done;
1140
1141                 if (dev == ign->fb_tunnel_dev) {
1142                         err = -EFAULT;
1143                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1144                                 goto done;
1145                         err = -ENOENT;
1146                         ip6gre_tnl_parm_from_user(&p1, &p);
1147                         t = ip6gre_tunnel_locate(net, &p1, 0);
1148                         if (!t)
1149                                 goto done;
1150                         err = -EPERM;
1151                         if (t == netdev_priv(ign->fb_tunnel_dev))
1152                                 goto done;
1153                         dev = t->dev;
1154                 }
1155                 unregister_netdevice(dev);
1156                 err = 0;
1157                 break;
1158
1159         default:
1160                 err = -EINVAL;
1161         }
1162
1163 done:
1164         return err;
1165 }
1166
1167 static int ip6gre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1168 {
1169         if (new_mtu < 68 ||
1170             new_mtu > 0xFFF8 - dev->hard_header_len)
1171                 return -EINVAL;
1172         dev->mtu = new_mtu;
1173         return 0;
1174 }
1175
1176 static int ip6gre_header(struct sk_buff *skb, struct net_device *dev,
1177                         unsigned short type,
1178                         const void *daddr, const void *saddr, unsigned int len)
1179 {
1180         struct ip6_tnl *t = netdev_priv(dev);
1181         struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen);
1182         __be16 *p = (__be16 *)(ipv6h+1);
1183
1184         ip6_flow_hdr(ipv6h, 0,
1185                      ip6_make_flowlabel(dev_net(dev), skb,
1186                                         t->fl.u.ip6.flowlabel, false));
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         free_percpu(dev->tstats);
1226         free_netdev(dev);
1227 }
1228
1229 static void ip6gre_tunnel_setup(struct net_device *dev)
1230 {
1231         struct ip6_tnl *t;
1232
1233         dev->netdev_ops = &ip6gre_netdev_ops;
1234         dev->destructor = ip6gre_dev_free;
1235
1236         dev->type = ARPHRD_IP6GRE;
1237         dev->hard_header_len = LL_MAX_HEADER + sizeof(struct ipv6hdr) + 4;
1238         dev->mtu = ETH_DATA_LEN - sizeof(struct ipv6hdr) - 4;
1239         t = netdev_priv(dev);
1240         if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1241                 dev->mtu -= 8;
1242         dev->flags |= IFF_NOARP;
1243         dev->addr_len = sizeof(struct in6_addr);
1244         netif_keep_dst(dev);
1245 }
1246
1247 static int ip6gre_tunnel_init(struct net_device *dev)
1248 {
1249         struct ip6_tnl *tunnel;
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         memcpy(dev->dev_addr, &tunnel->parms.laddr, sizeof(struct in6_addr));
1258         memcpy(dev->broadcast, &tunnel->parms.raddr, sizeof(struct in6_addr));
1259
1260         if (ipv6_addr_any(&tunnel->parms.raddr))
1261                 dev->header_ops = &ip6gre_header_ops;
1262
1263         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1264         if (!dev->tstats)
1265                 return -ENOMEM;
1266
1267         return 0;
1268 }
1269
1270 static void ip6gre_fb_tunnel_init(struct net_device *dev)
1271 {
1272         struct ip6_tnl *tunnel = netdev_priv(dev);
1273
1274         tunnel->dev = dev;
1275         tunnel->net = dev_net(dev);
1276         strcpy(tunnel->parms.name, dev->name);
1277
1278         tunnel->hlen            = sizeof(struct ipv6hdr) + 4;
1279
1280         dev_hold(dev);
1281 }
1282
1283
1284 static struct inet6_protocol ip6gre_protocol __read_mostly = {
1285         .handler     = ip6gre_rcv,
1286         .err_handler = ip6gre_err,
1287         .flags       = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1288 };
1289
1290 static void ip6gre_destroy_tunnels(struct net *net, struct list_head *head)
1291 {
1292         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1293         struct net_device *dev, *aux;
1294         int prio;
1295
1296         for_each_netdev_safe(net, dev, aux)
1297                 if (dev->rtnl_link_ops == &ip6gre_link_ops ||
1298                     dev->rtnl_link_ops == &ip6gre_tap_ops)
1299                         unregister_netdevice_queue(dev, head);
1300
1301         for (prio = 0; prio < 4; prio++) {
1302                 int h;
1303                 for (h = 0; h < HASH_SIZE; h++) {
1304                         struct ip6_tnl *t;
1305
1306                         t = rtnl_dereference(ign->tunnels[prio][h]);
1307
1308                         while (t) {
1309                                 /* If dev is in the same netns, it has already
1310                                  * been added to the list by the previous loop.
1311                                  */
1312                                 if (!net_eq(dev_net(t->dev), net))
1313                                         unregister_netdevice_queue(t->dev,
1314                                                                    head);
1315                                 t = rtnl_dereference(t->next);
1316                         }
1317                 }
1318         }
1319 }
1320
1321 static int __net_init ip6gre_init_net(struct net *net)
1322 {
1323         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1324         int err;
1325
1326         ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6gre0",
1327                                           NET_NAME_UNKNOWN,
1328                                           ip6gre_tunnel_setup);
1329         if (!ign->fb_tunnel_dev) {
1330                 err = -ENOMEM;
1331                 goto err_alloc_dev;
1332         }
1333         dev_net_set(ign->fb_tunnel_dev, net);
1334         /* FB netdevice is special: we have one, and only one per netns.
1335          * Allowing to move it to another netns is clearly unsafe.
1336          */
1337         ign->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL;
1338
1339
1340         ip6gre_fb_tunnel_init(ign->fb_tunnel_dev);
1341         ign->fb_tunnel_dev->rtnl_link_ops = &ip6gre_link_ops;
1342
1343         err = register_netdev(ign->fb_tunnel_dev);
1344         if (err)
1345                 goto err_reg_dev;
1346
1347         rcu_assign_pointer(ign->tunnels_wc[0],
1348                            netdev_priv(ign->fb_tunnel_dev));
1349         return 0;
1350
1351 err_reg_dev:
1352         ip6gre_dev_free(ign->fb_tunnel_dev);
1353 err_alloc_dev:
1354         return err;
1355 }
1356
1357 static void __net_exit ip6gre_exit_net(struct net *net)
1358 {
1359         LIST_HEAD(list);
1360
1361         rtnl_lock();
1362         ip6gre_destroy_tunnels(net, &list);
1363         unregister_netdevice_many(&list);
1364         rtnl_unlock();
1365 }
1366
1367 static struct pernet_operations ip6gre_net_ops = {
1368         .init = ip6gre_init_net,
1369         .exit = ip6gre_exit_net,
1370         .id   = &ip6gre_net_id,
1371         .size = sizeof(struct ip6gre_net),
1372 };
1373
1374 static int ip6gre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1375 {
1376         __be16 flags;
1377
1378         if (!data)
1379                 return 0;
1380
1381         flags = 0;
1382         if (data[IFLA_GRE_IFLAGS])
1383                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1384         if (data[IFLA_GRE_OFLAGS])
1385                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1386         if (flags & (GRE_VERSION|GRE_ROUTING))
1387                 return -EINVAL;
1388
1389         return 0;
1390 }
1391
1392 static int ip6gre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1393 {
1394         struct in6_addr daddr;
1395
1396         if (tb[IFLA_ADDRESS]) {
1397                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1398                         return -EINVAL;
1399                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1400                         return -EADDRNOTAVAIL;
1401         }
1402
1403         if (!data)
1404                 goto out;
1405
1406         if (data[IFLA_GRE_REMOTE]) {
1407                 daddr = nla_get_in6_addr(data[IFLA_GRE_REMOTE]);
1408                 if (ipv6_addr_any(&daddr))
1409                         return -EINVAL;
1410         }
1411
1412 out:
1413         return ip6gre_tunnel_validate(tb, data);
1414 }
1415
1416
1417 static void ip6gre_netlink_parms(struct nlattr *data[],
1418                                 struct __ip6_tnl_parm *parms)
1419 {
1420         memset(parms, 0, sizeof(*parms));
1421
1422         if (!data)
1423                 return;
1424
1425         if (data[IFLA_GRE_LINK])
1426                 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1427
1428         if (data[IFLA_GRE_IFLAGS])
1429                 parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1430
1431         if (data[IFLA_GRE_OFLAGS])
1432                 parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1433
1434         if (data[IFLA_GRE_IKEY])
1435                 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1436
1437         if (data[IFLA_GRE_OKEY])
1438                 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1439
1440         if (data[IFLA_GRE_LOCAL])
1441                 parms->laddr = nla_get_in6_addr(data[IFLA_GRE_LOCAL]);
1442
1443         if (data[IFLA_GRE_REMOTE])
1444                 parms->raddr = nla_get_in6_addr(data[IFLA_GRE_REMOTE]);
1445
1446         if (data[IFLA_GRE_TTL])
1447                 parms->hop_limit = nla_get_u8(data[IFLA_GRE_TTL]);
1448
1449         if (data[IFLA_GRE_ENCAP_LIMIT])
1450                 parms->encap_limit = nla_get_u8(data[IFLA_GRE_ENCAP_LIMIT]);
1451
1452         if (data[IFLA_GRE_FLOWINFO])
1453                 parms->flowinfo = nla_get_u32(data[IFLA_GRE_FLOWINFO]);
1454
1455         if (data[IFLA_GRE_FLAGS])
1456                 parms->flags = nla_get_u32(data[IFLA_GRE_FLAGS]);
1457 }
1458
1459 static int ip6gre_tap_init(struct net_device *dev)
1460 {
1461         struct ip6_tnl *tunnel;
1462
1463         tunnel = netdev_priv(dev);
1464
1465         tunnel->dev = dev;
1466         tunnel->net = dev_net(dev);
1467         strcpy(tunnel->parms.name, dev->name);
1468
1469         ip6gre_tnl_link_config(tunnel, 1);
1470
1471         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1472         if (!dev->tstats)
1473                 return -ENOMEM;
1474
1475         return 0;
1476 }
1477
1478 static const struct net_device_ops ip6gre_tap_netdev_ops = {
1479         .ndo_init = ip6gre_tap_init,
1480         .ndo_uninit = ip6gre_tunnel_uninit,
1481         .ndo_start_xmit = ip6gre_tunnel_xmit,
1482         .ndo_set_mac_address = eth_mac_addr,
1483         .ndo_validate_addr = eth_validate_addr,
1484         .ndo_change_mtu = ip6gre_tunnel_change_mtu,
1485         .ndo_get_stats64 = ip_tunnel_get_stats64,
1486         .ndo_get_iflink = ip6_tnl_get_iflink,
1487 };
1488
1489 static void ip6gre_tap_setup(struct net_device *dev)
1490 {
1491
1492         ether_setup(dev);
1493
1494         dev->netdev_ops = &ip6gre_tap_netdev_ops;
1495         dev->destructor = ip6gre_dev_free;
1496
1497         dev->features |= NETIF_F_NETNS_LOCAL;
1498 }
1499
1500 static int ip6gre_newlink(struct net *src_net, struct net_device *dev,
1501         struct nlattr *tb[], struct nlattr *data[])
1502 {
1503         struct ip6_tnl *nt;
1504         struct net *net = dev_net(dev);
1505         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1506         int err;
1507
1508         nt = netdev_priv(dev);
1509         ip6gre_netlink_parms(data, &nt->parms);
1510
1511         if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
1512                 return -EEXIST;
1513
1514         if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1515                 eth_hw_addr_random(dev);
1516
1517         nt->dev = dev;
1518         nt->net = dev_net(dev);
1519         ip6gre_tnl_link_config(nt, !tb[IFLA_MTU]);
1520
1521         /* Can use a lockless transmit, unless we generate output sequences */
1522         if (!(nt->parms.o_flags & GRE_SEQ))
1523                 dev->features |= NETIF_F_LLTX;
1524
1525         err = register_netdevice(dev);
1526         if (err)
1527                 goto out;
1528
1529         dev_hold(dev);
1530         ip6gre_tunnel_link(ign, nt);
1531
1532 out:
1533         return err;
1534 }
1535
1536 static int ip6gre_changelink(struct net_device *dev, struct nlattr *tb[],
1537                             struct nlattr *data[])
1538 {
1539         struct ip6_tnl *t, *nt = netdev_priv(dev);
1540         struct net *net = nt->net;
1541         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1542         struct __ip6_tnl_parm p;
1543
1544         if (dev == ign->fb_tunnel_dev)
1545                 return -EINVAL;
1546
1547         ip6gre_netlink_parms(data, &p);
1548
1549         t = ip6gre_tunnel_locate(net, &p, 0);
1550
1551         if (t) {
1552                 if (t->dev != dev)
1553                         return -EEXIST;
1554         } else {
1555                 t = nt;
1556
1557                 ip6gre_tunnel_unlink(ign, t);
1558                 ip6gre_tnl_change(t, &p, !tb[IFLA_MTU]);
1559                 ip6gre_tunnel_link(ign, t);
1560                 netdev_state_change(dev);
1561         }
1562
1563         return 0;
1564 }
1565
1566 static void ip6gre_dellink(struct net_device *dev, struct list_head *head)
1567 {
1568         struct net *net = dev_net(dev);
1569         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1570
1571         if (dev != ign->fb_tunnel_dev)
1572                 unregister_netdevice_queue(dev, head);
1573 }
1574
1575 static size_t ip6gre_get_size(const struct net_device *dev)
1576 {
1577         return
1578                 /* IFLA_GRE_LINK */
1579                 nla_total_size(4) +
1580                 /* IFLA_GRE_IFLAGS */
1581                 nla_total_size(2) +
1582                 /* IFLA_GRE_OFLAGS */
1583                 nla_total_size(2) +
1584                 /* IFLA_GRE_IKEY */
1585                 nla_total_size(4) +
1586                 /* IFLA_GRE_OKEY */
1587                 nla_total_size(4) +
1588                 /* IFLA_GRE_LOCAL */
1589                 nla_total_size(sizeof(struct in6_addr)) +
1590                 /* IFLA_GRE_REMOTE */
1591                 nla_total_size(sizeof(struct in6_addr)) +
1592                 /* IFLA_GRE_TTL */
1593                 nla_total_size(1) +
1594                 /* IFLA_GRE_TOS */
1595                 nla_total_size(1) +
1596                 /* IFLA_GRE_ENCAP_LIMIT */
1597                 nla_total_size(1) +
1598                 /* IFLA_GRE_FLOWINFO */
1599                 nla_total_size(4) +
1600                 /* IFLA_GRE_FLAGS */
1601                 nla_total_size(4) +
1602                 0;
1603 }
1604
1605 static int ip6gre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1606 {
1607         struct ip6_tnl *t = netdev_priv(dev);
1608         struct __ip6_tnl_parm *p = &t->parms;
1609
1610         if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1611             nla_put_be16(skb, IFLA_GRE_IFLAGS, p->i_flags) ||
1612             nla_put_be16(skb, IFLA_GRE_OFLAGS, p->o_flags) ||
1613             nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1614             nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1615             nla_put_in6_addr(skb, IFLA_GRE_LOCAL, &p->laddr) ||
1616             nla_put_in6_addr(skb, IFLA_GRE_REMOTE, &p->raddr) ||
1617             nla_put_u8(skb, IFLA_GRE_TTL, p->hop_limit) ||
1618             /*nla_put_u8(skb, IFLA_GRE_TOS, t->priority) ||*/
1619             nla_put_u8(skb, IFLA_GRE_ENCAP_LIMIT, p->encap_limit) ||
1620             nla_put_be32(skb, IFLA_GRE_FLOWINFO, p->flowinfo) ||
1621             nla_put_u32(skb, IFLA_GRE_FLAGS, p->flags))
1622                 goto nla_put_failure;
1623         return 0;
1624
1625 nla_put_failure:
1626         return -EMSGSIZE;
1627 }
1628
1629 static const struct nla_policy ip6gre_policy[IFLA_GRE_MAX + 1] = {
1630         [IFLA_GRE_LINK]        = { .type = NLA_U32 },
1631         [IFLA_GRE_IFLAGS]      = { .type = NLA_U16 },
1632         [IFLA_GRE_OFLAGS]      = { .type = NLA_U16 },
1633         [IFLA_GRE_IKEY]        = { .type = NLA_U32 },
1634         [IFLA_GRE_OKEY]        = { .type = NLA_U32 },
1635         [IFLA_GRE_LOCAL]       = { .len = FIELD_SIZEOF(struct ipv6hdr, saddr) },
1636         [IFLA_GRE_REMOTE]      = { .len = FIELD_SIZEOF(struct ipv6hdr, daddr) },
1637         [IFLA_GRE_TTL]         = { .type = NLA_U8 },
1638         [IFLA_GRE_ENCAP_LIMIT] = { .type = NLA_U8 },
1639         [IFLA_GRE_FLOWINFO]    = { .type = NLA_U32 },
1640         [IFLA_GRE_FLAGS]       = { .type = NLA_U32 },
1641 };
1642
1643 static struct rtnl_link_ops ip6gre_link_ops __read_mostly = {
1644         .kind           = "ip6gre",
1645         .maxtype        = IFLA_GRE_MAX,
1646         .policy         = ip6gre_policy,
1647         .priv_size      = sizeof(struct ip6_tnl),
1648         .setup          = ip6gre_tunnel_setup,
1649         .validate       = ip6gre_tunnel_validate,
1650         .newlink        = ip6gre_newlink,
1651         .changelink     = ip6gre_changelink,
1652         .dellink        = ip6gre_dellink,
1653         .get_size       = ip6gre_get_size,
1654         .fill_info      = ip6gre_fill_info,
1655         .get_link_net   = ip6_tnl_get_link_net,
1656 };
1657
1658 static struct rtnl_link_ops ip6gre_tap_ops __read_mostly = {
1659         .kind           = "ip6gretap",
1660         .maxtype        = IFLA_GRE_MAX,
1661         .policy         = ip6gre_policy,
1662         .priv_size      = sizeof(struct ip6_tnl),
1663         .setup          = ip6gre_tap_setup,
1664         .validate       = ip6gre_tap_validate,
1665         .newlink        = ip6gre_newlink,
1666         .changelink     = ip6gre_changelink,
1667         .get_size       = ip6gre_get_size,
1668         .fill_info      = ip6gre_fill_info,
1669         .get_link_net   = ip6_tnl_get_link_net,
1670 };
1671
1672 /*
1673  *      And now the modules code and kernel interface.
1674  */
1675
1676 static int __init ip6gre_init(void)
1677 {
1678         int err;
1679
1680         pr_info("GRE over IPv6 tunneling driver\n");
1681
1682         err = register_pernet_device(&ip6gre_net_ops);
1683         if (err < 0)
1684                 return err;
1685
1686         err = inet6_add_protocol(&ip6gre_protocol, IPPROTO_GRE);
1687         if (err < 0) {
1688                 pr_info("%s: can't add protocol\n", __func__);
1689                 goto add_proto_failed;
1690         }
1691
1692         err = rtnl_link_register(&ip6gre_link_ops);
1693         if (err < 0)
1694                 goto rtnl_link_failed;
1695
1696         err = rtnl_link_register(&ip6gre_tap_ops);
1697         if (err < 0)
1698                 goto tap_ops_failed;
1699
1700 out:
1701         return err;
1702
1703 tap_ops_failed:
1704         rtnl_link_unregister(&ip6gre_link_ops);
1705 rtnl_link_failed:
1706         inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1707 add_proto_failed:
1708         unregister_pernet_device(&ip6gre_net_ops);
1709         goto out;
1710 }
1711
1712 static void __exit ip6gre_fini(void)
1713 {
1714         rtnl_link_unregister(&ip6gre_tap_ops);
1715         rtnl_link_unregister(&ip6gre_link_ops);
1716         inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1717         unregister_pernet_device(&ip6gre_net_ops);
1718 }
1719
1720 module_init(ip6gre_init);
1721 module_exit(ip6gre_fini);
1722 MODULE_LICENSE("GPL");
1723 MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
1724 MODULE_DESCRIPTION("GRE over IPv6 tunneling device");
1725 MODULE_ALIAS_RTNL_LINK("ip6gre");
1726 MODULE_ALIAS_RTNL_LINK("ip6gretap");
1727 MODULE_ALIAS_NETDEV("ip6gre0");