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 != NULL)
 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 t;
 319
 320         if (parms->name[0])
 321                 strlcpy(name, parms->name, IFNAMSIZ);
 322         else
 323                 strcpy(name, "ip6gre%d");
 324
 325         dev = alloc_netdev(sizeof(*t), name, ip6gre_tunnel_setup);
 326         if (!dev)
 327                 return NULL;
 328
 329         dev_net_set(dev, net);
 330
 331         nt = netdev_priv(dev);
 332         nt->parms = *parms;
 333         dev->rtnl_link_ops = &ip6gre_link_ops;
 334
 335         nt->dev = dev;
 336         nt->net = dev_net(dev);
 337         ip6gre_tnl_link_config(nt, 1);
 338
 339         if (register_netdevice(dev) < 0)
 340                 goto failed_free;
 341
 342         /* Can use a lockless transmit, unless we generate output sequences */
 343         if (!(nt->parms.o_flags & GRE_SEQ))
 344                 dev->features |= NETIF_F_LLTX;
 345
 346         dev_hold(dev);
 347         ip6gre_tunnel_link(ign, nt);
 348         return nt;
 349
 350 failed_free:
 351         free_netdev(dev);
 352         return NULL;
 353 }
 354
 355 static void ip6gre_tunnel_uninit(struct net_device *dev)
 356 {
 357         struct ip6_tnl *t = netdev_priv(dev);
 358         struct ip6gre_net *ign = net_generic(t->net, ip6gre_net_id);
 359
 360         ip6gre_tunnel_unlink(ign, t);
 361         dev_put(dev);
 362 }
 363
 364
 365 static void ip6gre_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 366                 u8 type, u8 code, int offset, __be32 info)
 367 {
 368         const struct ipv6hdr *ipv6h = (const struct ipv6hdr *)skb->data;
 369         __be16 *p = (__be16 *)(skb->data + offset);
 370         int grehlen = offset + 4;
 371         struct ip6_tnl *t;
 372         __be16 flags;
 373
 374         flags = p[0];
 375         if (flags&(GRE_CSUM|GRE_KEY|GRE_SEQ|GRE_ROUTING|GRE_VERSION)) {
 376                 if (flags&(GRE_VERSION|GRE_ROUTING))
 377                         return;
 378                 if (flags&GRE_KEY) {
 379                         grehlen += 4;
 380                         if (flags&GRE_CSUM)
 381                                 grehlen += 4;
 382                 }
 383         }
 384
 385         /* If only 8 bytes returned, keyed message will be dropped here */
 386         if (!pskb_may_pull(skb, grehlen))
 387                 return;
 388         ipv6h = (const struct ipv6hdr *)skb->data;
 389         p = (__be16 *)(skb->data + offset);
 390
 391         t = ip6gre_tunnel_lookup(skb->dev, &ipv6h->daddr, &ipv6h->saddr,
 392                                 flags & GRE_KEY ?
 393                                 *(((__be32 *)p) + (grehlen / 4) - 1) : 0,
 394                                 p[1]);
 395         if (t == NULL)
 396                 return;
 397
 398         switch (type) {
 399                 __u32 teli;
 400                 struct ipv6_tlv_tnl_enc_lim *tel;
 401                 __u32 mtu;
 402         case ICMPV6_DEST_UNREACH:
 403                 net_warn_ratelimited("%s: Path to destination invalid or inactive!\n",
 404                                      t->parms.name);
 405                 break;
 406         case ICMPV6_TIME_EXCEED:
 407                 if (code == ICMPV6_EXC_HOPLIMIT) {
 408                         net_warn_ratelimited("%s: Too small hop limit or routing loop in tunnel!\n",
 409                                              t->parms.name);
 410                 }
 411                 break;
 412         case ICMPV6_PARAMPROB:
 413                 teli = 0;
 414                 if (code == ICMPV6_HDR_FIELD)
 415                         teli = ip6_tnl_parse_tlv_enc_lim(skb, skb->data);
 416
 417                 if (teli && teli == info - 2) {
 418                         tel = (struct ipv6_tlv_tnl_enc_lim *) &skb->data[teli];
 419                         if (tel->encap_limit == 0) {
 420                                 net_warn_ratelimited("%s: Too small encapsulation limit or routing loop in tunnel!\n",
 421                                                      t->parms.name);
 422                         }
 423                 } else {
 424                         net_warn_ratelimited("%s: Recipient unable to parse tunneled packet!\n",
 425                                              t->parms.name);
 426                 }
 427                 break;
 428         case ICMPV6_PKT_TOOBIG:
 429                 mtu = info - offset;
 430                 if (mtu < IPV6_MIN_MTU)
 431                         mtu = IPV6_MIN_MTU;
 432                 t->dev->mtu = mtu;
 433                 break;
 434         }
 435
 436         if (time_before(jiffies, t->err_time + IP6TUNNEL_ERR_TIMEO))
 437                 t->err_count++;
 438         else
 439                 t->err_count = 1;
 440         t->err_time = jiffies;
 441 }
 442
 443 static int ip6gre_rcv(struct sk_buff *skb)
 444 {
 445         const struct ipv6hdr *ipv6h;
 446         u8     *h;
 447         __be16    flags;
 448         __sum16   csum = 0;
 449         __be32 key = 0;
 450         u32    seqno = 0;
 451         struct ip6_tnl *tunnel;
 452         int    offset = 4;
 453         __be16 gre_proto;
 454         int err;
 455
 456         if (!pskb_may_pull(skb, sizeof(struct in6_addr)))
 457                 goto drop;
 458
 459         ipv6h = ipv6_hdr(skb);
 460         h = skb->data;
 461         flags = *(__be16 *)h;
 462
 463         if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
 464                 /* - Version must be 0.
 465                    - We do not support routing headers.
 466                  */
 467                 if (flags&(GRE_VERSION|GRE_ROUTING))
 468                         goto drop;
 469
 470                 if (flags&GRE_CSUM) {
 471                         switch (skb->ip_summed) {
 472                         case CHECKSUM_COMPLETE:
 473                                 csum = csum_fold(skb->csum);
 474                                 if (!csum)
 475                                         break;
 476                                 /* fall through */
 477                         case CHECKSUM_NONE:
 478                                 skb->csum = 0;
 479                                 csum = __skb_checksum_complete(skb);
 480                                 skb->ip_summed = CHECKSUM_COMPLETE;
 481                         }
 482                         offset += 4;
 483                 }
 484                 if (flags&GRE_KEY) {
 485                         key = *(__be32 *)(h + offset);
 486                         offset += 4;
 487                 }
 488                 if (flags&GRE_SEQ) {
 489                         seqno = ntohl(*(__be32 *)(h + offset));
 490                         offset += 4;
 491                 }
 492         }
 493
 494         gre_proto = *(__be16 *)(h + 2);
 495
 496         tunnel = ip6gre_tunnel_lookup(skb->dev,
 497                                           &ipv6h->saddr, &ipv6h->daddr, key,
 498                                           gre_proto);
 499         if (tunnel) {
 500                 struct pcpu_sw_netstats *tstats;
 501
 502                 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
 503                         goto drop;
 504
 505                 if (!ip6_tnl_rcv_ctl(tunnel, &ipv6h->daddr, &ipv6h->saddr)) {
 506                         tunnel->dev->stats.rx_dropped++;
 507                         goto drop;
 508                 }
 509
 510                 skb->protocol = gre_proto;
 511                 /* WCCP version 1 and 2 protocol decoding.
 512                  * - Change protocol to IP
 513                  * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
 514                  */
 515                 if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) {
 516                         skb->protocol = htons(ETH_P_IP);
 517                         if ((*(h + offset) & 0xF0) != 0x40)
 518                                 offset += 4;
 519                 }
 520
 521                 skb->mac_header = skb->network_header;
 522                 __pskb_pull(skb, offset);
 523                 skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
 524
 525                 if (((flags&GRE_CSUM) && csum) ||
 526                     (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
 527                         tunnel->dev->stats.rx_crc_errors++;
 528                         tunnel->dev->stats.rx_errors++;
 529                         goto drop;
 530                 }
 531                 if (tunnel->parms.i_flags&GRE_SEQ) {
 532                         if (!(flags&GRE_SEQ) ||
 533                             (tunnel->i_seqno &&
 534                                         (s32)(seqno - tunnel->i_seqno) < 0)) {
 535                                 tunnel->dev->stats.rx_fifo_errors++;
 536                                 tunnel->dev->stats.rx_errors++;
 537                                 goto drop;
 538                         }
 539                         tunnel->i_seqno = seqno + 1;
 540                 }
 541
 542                 /* Warning: All skb pointers will be invalidated! */
 543                 if (tunnel->dev->type == ARPHRD_ETHER) {
 544                         if (!pskb_may_pull(skb, ETH_HLEN)) {
 545                                 tunnel->dev->stats.rx_length_errors++;
 546                                 tunnel->dev->stats.rx_errors++;
 547                                 goto drop;
 548                         }
 549
 550                         ipv6h = ipv6_hdr(skb);
 551                         skb->protocol = eth_type_trans(skb, tunnel->dev);
 552                         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 553                 }
 554
 555                 __skb_tunnel_rx(skb, tunnel->dev, tunnel->net);
 556
 557                 skb_reset_network_header(skb);
 558
 559                 err = IP6_ECN_decapsulate(ipv6h, skb);
 560                 if (unlikely(err)) {
 561                         if (log_ecn_error)
 562                                 net_info_ratelimited("non-ECT from %pI6 with dsfield=%#x\n",
 563                                                      &ipv6h->saddr,
 564                                                      ipv6_get_dsfield(ipv6h));
 565                         if (err > 1) {
 566                                 ++tunnel->dev->stats.rx_frame_errors;
 567                                 ++tunnel->dev->stats.rx_errors;
 568                                 goto drop;
 569                         }
 570                 }
 571
 572                 tstats = this_cpu_ptr(tunnel->dev->tstats);
 573                 u64_stats_update_begin(&tstats->syncp);
 574                 tstats->rx_packets++;
 575                 tstats->rx_bytes += skb->len;
 576                 u64_stats_update_end(&tstats->syncp);
 577
 578                 netif_rx(skb);
 579
 580                 return 0;
 581         }
 582         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 583
 584 drop:
 585         kfree_skb(skb);
 586         return 0;
 587 }
 588
 589 struct ipv6_tel_txoption {
 590         struct ipv6_txoptions ops;
 591         __u8 dst_opt[8];
 592 };
 593
 594 static void init_tel_txopt(struct ipv6_tel_txoption *opt, __u8 encap_limit)
 595 {
 596         memset(opt, 0, sizeof(struct ipv6_tel_txoption));
 597
 598         opt->dst_opt[2] = IPV6_TLV_TNL_ENCAP_LIMIT;
 599         opt->dst_opt[3] = 1;
 600         opt->dst_opt[4] = encap_limit;
 601         opt->dst_opt[5] = IPV6_TLV_PADN;
 602         opt->dst_opt[6] = 1;
 603
 604         opt->ops.dst0opt = (struct ipv6_opt_hdr *) opt->dst_opt;
 605         opt->ops.opt_nflen = 8;
 606 }
 607
 608 static netdev_tx_t ip6gre_xmit2(struct sk_buff *skb,
 609                          struct net_device *dev,
 610                          __u8 dsfield,
 611                          struct flowi6 *fl6,
 612                          int encap_limit,
 613                          __u32 *pmtu)
 614 {
 615         struct ip6_tnl *tunnel = netdev_priv(dev);
 616         struct net *net = tunnel->net;
 617         struct net_device *tdev;    /* Device to other host */
 618         struct ipv6hdr  *ipv6h;     /* Our new IP header */
 619         unsigned int max_headroom = 0; /* The extra header space needed */
 620         int    gre_hlen;
 621         struct ipv6_tel_txoption opt;
 622         int    mtu;
 623         struct dst_entry *dst = NULL, *ndst = NULL;
 624         struct net_device_stats *stats = &tunnel->dev->stats;
 625         int err = -1;
 626         u8 proto;
 627         struct sk_buff *new_skb;
 628
 629         if (dev->type == ARPHRD_ETHER)
 630                 IPCB(skb)->flags = 0;
 631
 632         if (dev->header_ops && dev->type == ARPHRD_IP6GRE) {
 633                 gre_hlen = 0;
 634                 ipv6h = (struct ipv6hdr *)skb->data;
 635                 fl6->daddr = ipv6h->daddr;
 636         } else {
 637                 gre_hlen = tunnel->hlen;
 638                 fl6->daddr = tunnel->parms.raddr;
 639         }
 640
 641         if (!fl6->flowi6_mark)
 642                 dst = ip6_tnl_dst_check(tunnel);
 643
 644         if (!dst) {
 645                 ndst = ip6_route_output(net, NULL, fl6);
 646
 647                 if (ndst->error)
 648                         goto tx_err_link_failure;
 649                 ndst = xfrm_lookup(net, ndst, flowi6_to_flowi(fl6), NULL, 0);
 650                 if (IS_ERR(ndst)) {
 651                         err = PTR_ERR(ndst);
 652                         ndst = NULL;
 653                         goto tx_err_link_failure;
 654                 }
 655                 dst = ndst;
 656         }
 657
 658         tdev = dst->dev;
 659
 660         if (tdev == dev) {
 661                 stats->collisions++;
 662                 net_warn_ratelimited("%s: Local routing loop detected!\n",
 663                                      tunnel->parms.name);
 664                 goto tx_err_dst_release;
 665         }
 666
 667         mtu = dst_mtu(dst) - sizeof(*ipv6h);
 668         if (encap_limit >= 0) {
 669                 max_headroom += 8;
 670                 mtu -= 8;
 671         }
 672         if (mtu < IPV6_MIN_MTU)
 673                 mtu = IPV6_MIN_MTU;
 674         if (skb_dst(skb))
 675                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), NULL, skb, mtu);
 676         if (skb->len > mtu) {
 677                 *pmtu = mtu;
 678                 err = -EMSGSIZE;
 679                 goto tx_err_dst_release;
 680         }
 681
 682         if (tunnel->err_count > 0) {
 683                 if (time_before(jiffies,
 684                                 tunnel->err_time + IP6TUNNEL_ERR_TIMEO)) {
 685                         tunnel->err_count--;
 686
 687                         dst_link_failure(skb);
 688                 } else
 689                         tunnel->err_count = 0;
 690         }
 691
 692         skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(dev)));
 693
 694         max_headroom += LL_RESERVED_SPACE(tdev) + gre_hlen + dst->header_len;
 695
 696         if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
 697             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
 698                 new_skb = skb_realloc_headroom(skb, max_headroom);
 699                 if (max_headroom > dev->needed_headroom)
 700                         dev->needed_headroom = max_headroom;
 701                 if (!new_skb)
 702                         goto tx_err_dst_release;
 703
 704                 if (skb->sk)
 705                         skb_set_owner_w(new_skb, skb->sk);
 706                 consume_skb(skb);
 707                 skb = new_skb;
 708         }
 709
 710         if (fl6->flowi6_mark) {
 711                 skb_dst_set(skb, dst);
 712                 ndst = NULL;
 713         } else {
 714                 skb_dst_set_noref(skb, dst);
 715         }
 716
 717         proto = NEXTHDR_GRE;
 718         if (encap_limit >= 0) {
 719                 init_tel_txopt(&opt, encap_limit);
 720                 ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
 721         }
 722
 723         if (likely(!skb->encapsulation)) {
 724                 skb_reset_inner_headers(skb);
 725                 skb->encapsulation = 1;
 726         }
 727
 728         skb_push(skb, gre_hlen);
 729         skb_reset_network_header(skb);
 730         skb_set_transport_header(skb, sizeof(*ipv6h));
 731
 732         /*
 733          *      Push down and install the IP header.
 734          */
 735         ipv6h = ipv6_hdr(skb);
 736         ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield), fl6->flowlabel);
 737         ipv6h->hop_limit = tunnel->parms.hop_limit;
 738         ipv6h->nexthdr = proto;
 739         ipv6h->saddr = fl6->saddr;
 740         ipv6h->daddr = fl6->daddr;
 741
 742         ((__be16 *)(ipv6h + 1))[0] = tunnel->parms.o_flags;
 743         ((__be16 *)(ipv6h + 1))[1] = (dev->type == ARPHRD_ETHER) ?
 744                                    htons(ETH_P_TEB) : skb->protocol;
 745
 746         if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
 747                 __be32 *ptr = (__be32 *)(((u8 *)ipv6h) + tunnel->hlen - 4);
 748
 749                 if (tunnel->parms.o_flags&GRE_SEQ) {
 750                         ++tunnel->o_seqno;
 751                         *ptr = htonl(tunnel->o_seqno);
 752                         ptr--;
 753                 }
 754                 if (tunnel->parms.o_flags&GRE_KEY) {
 755                         *ptr = tunnel->parms.o_key;
 756                         ptr--;
 757                 }
 758                 if (tunnel->parms.o_flags&GRE_CSUM) {
 759                         *ptr = 0;
 760                         *(__sum16 *)ptr = ip_compute_csum((void *)(ipv6h+1),
 761                                 skb->len - sizeof(struct ipv6hdr));
 762                 }
 763         }
 764
 765         ip6tunnel_xmit(skb, dev);
 766         if (ndst)
 767                 ip6_tnl_dst_store(tunnel, ndst);
 768         return 0;
 769 tx_err_link_failure:
 770         stats->tx_carrier_errors++;
 771         dst_link_failure(skb);
 772 tx_err_dst_release:
 773         dst_release(ndst);
 774         return err;
 775 }
 776
 777 static inline int ip6gre_xmit_ipv4(struct sk_buff *skb, struct net_device *dev)
 778 {
 779         struct ip6_tnl *t = netdev_priv(dev);
 780         const struct iphdr  *iph = ip_hdr(skb);
 781         int encap_limit = -1;
 782         struct flowi6 fl6;
 783         __u8 dsfield;
 784         __u32 mtu;
 785         int err;
 786
 787         if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
 788                 encap_limit = t->parms.encap_limit;
 789
 790         memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
 791         fl6.flowi6_proto = IPPROTO_IPIP;
 792
 793         dsfield = ipv4_get_dsfield(iph);
 794
 795         if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
 796                 fl6.flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
 797                                           & IPV6_TCLASS_MASK;
 798         if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
 799                 fl6.flowi6_mark = skb->mark;
 800
 801         err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
 802         if (err != 0) {
 803                 /* XXX: send ICMP error even if DF is not set. */
 804                 if (err == -EMSGSIZE)
 805                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
 806                                   htonl(mtu));
 807                 return -1;
 808         }
 809
 810         return 0;
 811 }
 812
 813 static inline int ip6gre_xmit_ipv6(struct sk_buff *skb, struct net_device *dev)
 814 {
 815         struct ip6_tnl *t = netdev_priv(dev);
 816         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 817         int encap_limit = -1;
 818         __u16 offset;
 819         struct flowi6 fl6;
 820         __u8 dsfield;
 821         __u32 mtu;
 822         int err;
 823
 824         if (ipv6_addr_equal(&t->parms.raddr, &ipv6h->saddr))
 825                 return -1;
 826
 827         offset = ip6_tnl_parse_tlv_enc_lim(skb, skb_network_header(skb));
 828         if (offset > 0) {
 829                 struct ipv6_tlv_tnl_enc_lim *tel;
 830                 tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
 831                 if (tel->encap_limit == 0) {
 832                         icmpv6_send(skb, ICMPV6_PARAMPROB,
 833                                     ICMPV6_HDR_FIELD, offset + 2);
 834                         return -1;
 835                 }
 836                 encap_limit = tel->encap_limit - 1;
 837         } else if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
 838                 encap_limit = t->parms.encap_limit;
 839
 840         memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
 841         fl6.flowi6_proto = IPPROTO_IPV6;
 842
 843         dsfield = ipv6_get_dsfield(ipv6h);
 844         if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
 845                 fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
 846         if (t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL)
 847                 fl6.flowlabel |= ip6_flowlabel(ipv6h);
 848         if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
 849                 fl6.flowi6_mark = skb->mark;
 850
 851         err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
 852         if (err != 0) {
 853                 if (err == -EMSGSIZE)
 854                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 855                 return -1;
 856         }
 857
 858         return 0;
 859 }
 860
 861 /**
 862  * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
 863  *   @t: the outgoing tunnel device
 864  *   @hdr: IPv6 header from the incoming packet
 865  *
 866  * Description:
 867  *   Avoid trivial tunneling loop by checking that tunnel exit-point
 868  *   doesn't match source of incoming packet.
 869  *
 870  * Return:
 871  *   1 if conflict,
 872  *   0 else
 873  **/
 874
 875 static inline bool ip6gre_tnl_addr_conflict(const struct ip6_tnl *t,
 876         const struct ipv6hdr *hdr)
 877 {
 878         return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
 879 }
 880
 881 static int ip6gre_xmit_other(struct sk_buff *skb, struct net_device *dev)
 882 {
 883         struct ip6_tnl *t = netdev_priv(dev);
 884         int encap_limit = -1;
 885         struct flowi6 fl6;
 886         __u32 mtu;
 887         int err;
 888
 889         if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
 890                 encap_limit = t->parms.encap_limit;
 891
 892         memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
 893         fl6.flowi6_proto = skb->protocol;
 894
 895         err = ip6gre_xmit2(skb, dev, 0, &fl6, encap_limit, &mtu);
 896
 897         return err;
 898 }
 899
 900 static netdev_tx_t ip6gre_tunnel_xmit(struct sk_buff *skb,
 901         struct net_device *dev)
 902 {
 903         struct ip6_tnl *t = netdev_priv(dev);
 904         struct net_device_stats *stats = &t->dev->stats;
 905         int ret;
 906
 907         if (!ip6_tnl_xmit_ctl(t))
 908                 goto tx_err;
 909
 910         switch (skb->protocol) {
 911         case htons(ETH_P_IP):
 912                 ret = ip6gre_xmit_ipv4(skb, dev);
 913                 break;
 914         case htons(ETH_P_IPV6):
 915                 ret = ip6gre_xmit_ipv6(skb, dev);
 916                 break;
 917         default:
 918                 ret = ip6gre_xmit_other(skb, dev);
 919                 break;
 920         }
 921
 922         if (ret < 0)
 923                 goto tx_err;
 924
 925         return NETDEV_TX_OK;
 926
 927 tx_err:
 928         stats->tx_errors++;
 929         stats->tx_dropped++;
 930         kfree_skb(skb);
 931         return NETDEV_TX_OK;
 932 }
 933
 934 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu)
 935 {
 936         struct net_device *dev = t->dev;
 937         struct __ip6_tnl_parm *p = &t->parms;
 938         struct flowi6 *fl6 = &t->fl.u.ip6;
 939         int addend = sizeof(struct ipv6hdr) + 4;
 940
 941         if (dev->type != ARPHRD_ETHER) {
 942                 memcpy(dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
 943                 memcpy(dev->broadcast, &p->raddr, sizeof(struct in6_addr));
 944         }
 945
 946         /* Set up flowi template */
 947         fl6->saddr = p->laddr;
 948         fl6->daddr = p->raddr;
 949         fl6->flowi6_oif = p->link;
 950         fl6->flowlabel = 0;
 951
 952         if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
 953                 fl6->flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
 954         if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
 955                 fl6->flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;
 956
 957         p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV|IP6_TNL_F_CAP_PER_PACKET);
 958         p->flags |= ip6_tnl_get_cap(t, &p->laddr, &p->raddr);
 959
 960         if (p->flags&IP6_TNL_F_CAP_XMIT &&
 961                         p->flags&IP6_TNL_F_CAP_RCV && dev->type != ARPHRD_ETHER)
 962                 dev->flags |= IFF_POINTOPOINT;
 963         else
 964                 dev->flags &= ~IFF_POINTOPOINT;
 965
 966         dev->iflink = p->link;
 967
 968         /* Precalculate GRE options length */
 969         if (t->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
 970                 if (t->parms.o_flags&GRE_CSUM)
 971                         addend += 4;
 972                 if (t->parms.o_flags&GRE_KEY)
 973                         addend += 4;
 974                 if (t->parms.o_flags&GRE_SEQ)
 975                         addend += 4;
 976         }
 977         t->hlen = addend;
 978
 979         if (p->flags & IP6_TNL_F_CAP_XMIT) {
 980                 int strict = (ipv6_addr_type(&p->raddr) &
 981                               (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));
 982
 983                 struct rt6_info *rt = rt6_lookup(t->net,
 984                                                  &p->raddr, &p->laddr,
 985                                                  p->link, strict);
 986
 987                 if (rt == NULL)
 988                         return;
 989
 990                 if (rt->dst.dev) {
 991                         dev->hard_header_len = rt->dst.dev->hard_header_len + addend;
 992
 993                         if (set_mtu) {
 994                                 dev->mtu = rt->dst.dev->mtu - addend;
 995                                 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
 996                                         dev->mtu -= 8;
 997
 998                                 if (dev->mtu < IPV6_MIN_MTU)
 999                                         dev->mtu = IPV6_MIN_MTU;
1000                         }
1001                 }
1002                 ip6_rt_put(rt);
1003         }
1004 }
1005
1006 static int ip6gre_tnl_change(struct ip6_tnl *t,
1007         const struct __ip6_tnl_parm *p, int set_mtu)
1008 {
1009         t->parms.laddr = p->laddr;
1010         t->parms.raddr = p->raddr;
1011         t->parms.flags = p->flags;
1012         t->parms.hop_limit = p->hop_limit;
1013         t->parms.encap_limit = p->encap_limit;
1014         t->parms.flowinfo = p->flowinfo;
1015         t->parms.link = p->link;
1016         t->parms.proto = p->proto;
1017         t->parms.i_key = p->i_key;
1018         t->parms.o_key = p->o_key;
1019         t->parms.i_flags = p->i_flags;
1020         t->parms.o_flags = p->o_flags;
1021         ip6_tnl_dst_reset(t);
1022         ip6gre_tnl_link_config(t, set_mtu);
1023         return 0;
1024 }
1025
1026 static void ip6gre_tnl_parm_from_user(struct __ip6_tnl_parm *p,
1027         const struct ip6_tnl_parm2 *u)
1028 {
1029         p->laddr = u->laddr;
1030         p->raddr = u->raddr;
1031         p->flags = u->flags;
1032         p->hop_limit = u->hop_limit;
1033         p->encap_limit = u->encap_limit;
1034         p->flowinfo = u->flowinfo;
1035         p->link = u->link;
1036         p->i_key = u->i_key;
1037         p->o_key = u->o_key;
1038         p->i_flags = u->i_flags;
1039         p->o_flags = u->o_flags;
1040         memcpy(p->name, u->name, sizeof(u->name));
1041 }
1042
1043 static void ip6gre_tnl_parm_to_user(struct ip6_tnl_parm2 *u,
1044         const struct __ip6_tnl_parm *p)
1045 {
1046         u->proto = IPPROTO_GRE;
1047         u->laddr = p->laddr;
1048         u->raddr = p->raddr;
1049         u->flags = p->flags;
1050         u->hop_limit = p->hop_limit;
1051         u->encap_limit = p->encap_limit;
1052         u->flowinfo = p->flowinfo;
1053         u->link = p->link;
1054         u->i_key = p->i_key;
1055         u->o_key = p->o_key;
1056         u->i_flags = p->i_flags;
1057         u->o_flags = p->o_flags;
1058         memcpy(u->name, p->name, sizeof(u->name));
1059 }
1060
1061 static int ip6gre_tunnel_ioctl(struct net_device *dev,
1062         struct ifreq *ifr, int cmd)
1063 {
1064         int err = 0;
1065         struct ip6_tnl_parm2 p;
1066         struct __ip6_tnl_parm p1;
1067         struct ip6_tnl *t = netdev_priv(dev);
1068         struct net *net = t->net;
1069         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1070
1071         switch (cmd) {
1072         case SIOCGETTUNNEL:
1073                 if (dev == ign->fb_tunnel_dev) {
1074                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1075                                 err = -EFAULT;
1076                                 break;
1077                         }
1078                         ip6gre_tnl_parm_from_user(&p1, &p);
1079                         t = ip6gre_tunnel_locate(net, &p1, 0);
1080                         if (t == NULL)
1081                                 t = netdev_priv(dev);
1082                 }
1083                 memset(&p, 0, sizeof(p));
1084                 ip6gre_tnl_parm_to_user(&p, &t->parms);
1085                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1086                         err = -EFAULT;
1087                 break;
1088
1089         case SIOCADDTUNNEL:
1090         case SIOCCHGTUNNEL:
1091                 err = -EPERM;
1092                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1093                         goto done;
1094
1095                 err = -EFAULT;
1096                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1097                         goto done;
1098
1099                 err = -EINVAL;
1100                 if ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))
1101                         goto done;
1102
1103                 if (!(p.i_flags&GRE_KEY))
1104                         p.i_key = 0;
1105                 if (!(p.o_flags&GRE_KEY))
1106                         p.o_key = 0;
1107
1108                 ip6gre_tnl_parm_from_user(&p1, &p);
1109                 t = ip6gre_tunnel_locate(net, &p1, cmd == SIOCADDTUNNEL);
1110
1111                 if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1112                         if (t != NULL) {
1113                                 if (t->dev != dev) {
1114                                         err = -EEXIST;
1115                                         break;
1116                                 }
1117                         } else {
1118                                 t = netdev_priv(dev);
1119
1120                                 ip6gre_tunnel_unlink(ign, t);
1121                                 synchronize_net();
1122                                 ip6gre_tnl_change(t, &p1, 1);
1123                                 ip6gre_tunnel_link(ign, t);
1124                                 netdev_state_change(dev);
1125                         }
1126                 }
1127
1128                 if (t) {
1129                         err = 0;
1130
1131                         memset(&p, 0, sizeof(p));
1132                         ip6gre_tnl_parm_to_user(&p, &t->parms);
1133                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1134                                 err = -EFAULT;
1135                 } else
1136                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1137                 break;
1138
1139         case SIOCDELTUNNEL:
1140                 err = -EPERM;
1141                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1142                         goto done;
1143
1144                 if (dev == ign->fb_tunnel_dev) {
1145                         err = -EFAULT;
1146                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1147                                 goto done;
1148                         err = -ENOENT;
1149                         ip6gre_tnl_parm_from_user(&p1, &p);
1150                         t = ip6gre_tunnel_locate(net, &p1, 0);
1151                         if (t == NULL)
1152                                 goto done;
1153                         err = -EPERM;
1154                         if (t == netdev_priv(ign->fb_tunnel_dev))
1155                                 goto done;
1156                         dev = t->dev;
1157                 }
1158                 unregister_netdevice(dev);
1159                 err = 0;
1160                 break;
1161
1162         default:
1163                 err = -EINVAL;
1164         }
1165
1166 done:
1167         return err;
1168 }
1169
1170 static int ip6gre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1171 {
1172         if (new_mtu < 68 ||
1173             new_mtu > 0xFFF8 - dev->hard_header_len)
1174                 return -EINVAL;
1175         dev->mtu = new_mtu;
1176         return 0;
1177 }
1178
1179 static int ip6gre_header(struct sk_buff *skb, struct net_device *dev,
1180                         unsigned short type,
1181                         const void *daddr, const void *saddr, unsigned int len)
1182 {
1183         struct ip6_tnl *t = netdev_priv(dev);
1184         struct ipv6hdr *ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen);
1185         __be16 *p = (__be16 *)(ipv6h+1);
1186
1187         ip6_flow_hdr(ipv6h, 0, t->fl.u.ip6.flowlabel);
1188         ipv6h->hop_limit = t->parms.hop_limit;
1189         ipv6h->nexthdr = NEXTHDR_GRE;
1190         ipv6h->saddr = t->parms.laddr;
1191         ipv6h->daddr = t->parms.raddr;
1192
1193         p[0]            = t->parms.o_flags;
1194         p[1]            = htons(type);
1195
1196         /*
1197          *      Set the source hardware address.
1198          */
1199
1200         if (saddr)
1201                 memcpy(&ipv6h->saddr, saddr, sizeof(struct in6_addr));
1202         if (daddr)
1203                 memcpy(&ipv6h->daddr, daddr, sizeof(struct in6_addr));
1204         if (!ipv6_addr_any(&ipv6h->daddr))
1205                 return t->hlen;
1206
1207         return -t->hlen;
1208 }
1209
1210 static const struct header_ops ip6gre_header_ops = {
1211         .create = ip6gre_header,
1212 };
1213
1214 static const struct net_device_ops ip6gre_netdev_ops = {
1215         .ndo_init               = ip6gre_tunnel_init,
1216         .ndo_uninit             = ip6gre_tunnel_uninit,
1217         .ndo_start_xmit         = ip6gre_tunnel_xmit,
1218         .ndo_do_ioctl           = ip6gre_tunnel_ioctl,
1219         .ndo_change_mtu         = ip6gre_tunnel_change_mtu,
1220         .ndo_get_stats64        = ip_tunnel_get_stats64,
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->iflink = 0;
1244         dev->addr_len = sizeof(struct in6_addr);
1245         dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1246 }
1247
1248 static int ip6gre_tunnel_init(struct net_device *dev)
1249 {
1250         struct ip6_tnl *tunnel;
1251         int i;
1252
1253         tunnel = netdev_priv(dev);
1254
1255         tunnel->dev = dev;
1256         tunnel->net = dev_net(dev);
1257         strcpy(tunnel->parms.name, dev->name);
1258
1259         memcpy(dev->dev_addr, &tunnel->parms.laddr, sizeof(struct in6_addr));
1260         memcpy(dev->broadcast, &tunnel->parms.raddr, sizeof(struct in6_addr));
1261
1262         if (ipv6_addr_any(&tunnel->parms.raddr))
1263                 dev->header_ops = &ip6gre_header_ops;
1264
1265         dev->tstats = alloc_percpu(struct pcpu_sw_netstats);
1266         if (!dev->tstats)
1267                 return -ENOMEM;
1268
1269         for_each_possible_cpu(i) {
1270                 struct pcpu_sw_netstats *ip6gre_tunnel_stats;
1271                 ip6gre_tunnel_stats = per_cpu_ptr(dev->tstats, i);
1272                 u64_stats_init(&ip6gre_tunnel_stats->syncp);
1273         }
1274
1275
1276         return 0;
1277 }
1278
1279 static void ip6gre_fb_tunnel_init(struct net_device *dev)
1280 {
1281         struct ip6_tnl *tunnel = netdev_priv(dev);
1282
1283         tunnel->dev = dev;
1284         tunnel->net = dev_net(dev);
1285         strcpy(tunnel->parms.name, dev->name);
1286
1287         tunnel->hlen            = sizeof(struct ipv6hdr) + 4;
1288
1289         dev_hold(dev);
1290 }
1291
1292
1293 static struct inet6_protocol ip6gre_protocol __read_mostly = {
1294         .handler     = ip6gre_rcv,
1295         .err_handler = ip6gre_err,
1296         .flags       = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1297 };
1298
1299 static void ip6gre_destroy_tunnels(struct net *net, struct list_head *head)
1300 {
1301         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1302         struct net_device *dev, *aux;
1303         int prio;
1304
1305         for_each_netdev_safe(net, dev, aux)
1306                 if (dev->rtnl_link_ops == &ip6gre_link_ops ||
1307                     dev->rtnl_link_ops == &ip6gre_tap_ops)
1308                         unregister_netdevice_queue(dev, head);
1309
1310         for (prio = 0; prio < 4; prio++) {
1311                 int h;
1312                 for (h = 0; h < HASH_SIZE; h++) {
1313                         struct ip6_tnl *t;
1314
1315                         t = rtnl_dereference(ign->tunnels[prio][h]);
1316
1317                         while (t != NULL) {
1318                                 /* If dev is in the same netns, it has already
1319                                  * been added to the list by the previous loop.
1320                                  */
1321                                 if (!net_eq(dev_net(t->dev), net))
1322                                         unregister_netdevice_queue(t->dev,
1323                                                                    head);
1324                                 t = rtnl_dereference(t->next);
1325                         }
1326                 }
1327         }
1328 }
1329
1330 static int __net_init ip6gre_init_net(struct net *net)
1331 {
1332         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1333         int err;
1334
1335         ign->fb_tunnel_dev = alloc_netdev(sizeof(struct ip6_tnl), "ip6gre0",
1336                                            ip6gre_tunnel_setup);
1337         if (!ign->fb_tunnel_dev) {
1338                 err = -ENOMEM;
1339                 goto err_alloc_dev;
1340         }
1341         dev_net_set(ign->fb_tunnel_dev, net);
1342         /* FB netdevice is special: we have one, and only one per netns.
1343          * Allowing to move it to another netns is clearly unsafe.
1344          */
1345         ign->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL;
1346
1347
1348         ip6gre_fb_tunnel_init(ign->fb_tunnel_dev);
1349         ign->fb_tunnel_dev->rtnl_link_ops = &ip6gre_link_ops;
1350
1351         err = register_netdev(ign->fb_tunnel_dev);
1352         if (err)
1353                 goto err_reg_dev;
1354
1355         rcu_assign_pointer(ign->tunnels_wc[0],
1356                            netdev_priv(ign->fb_tunnel_dev));
1357         return 0;
1358
1359 err_reg_dev:
1360         ip6gre_dev_free(ign->fb_tunnel_dev);
1361 err_alloc_dev:
1362         return err;
1363 }
1364
1365 static void __net_exit ip6gre_exit_net(struct net *net)
1366 {
1367         LIST_HEAD(list);
1368
1369         rtnl_lock();
1370         ip6gre_destroy_tunnels(net, &list);
1371         unregister_netdevice_many(&list);
1372         rtnl_unlock();
1373 }
1374
1375 static struct pernet_operations ip6gre_net_ops = {
1376         .init = ip6gre_init_net,
1377         .exit = ip6gre_exit_net,
1378         .id   = &ip6gre_net_id,
1379         .size = sizeof(struct ip6gre_net),
1380 };
1381
1382 static int ip6gre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1383 {
1384         __be16 flags;
1385
1386         if (!data)
1387                 return 0;
1388
1389         flags = 0;
1390         if (data[IFLA_GRE_IFLAGS])
1391                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1392         if (data[IFLA_GRE_OFLAGS])
1393                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1394         if (flags & (GRE_VERSION|GRE_ROUTING))
1395                 return -EINVAL;
1396
1397         return 0;
1398 }
1399
1400 static int ip6gre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1401 {
1402         struct in6_addr daddr;
1403
1404         if (tb[IFLA_ADDRESS]) {
1405                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1406                         return -EINVAL;
1407                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1408                         return -EADDRNOTAVAIL;
1409         }
1410
1411         if (!data)
1412                 goto out;
1413
1414         if (data[IFLA_GRE_REMOTE]) {
1415                 nla_memcpy(&daddr, data[IFLA_GRE_REMOTE], sizeof(struct in6_addr));
1416                 if (ipv6_addr_any(&daddr))
1417                         return -EINVAL;
1418         }
1419
1420 out:
1421         return ip6gre_tunnel_validate(tb, data);
1422 }
1423
1424
1425 static void ip6gre_netlink_parms(struct nlattr *data[],
1426                                 struct __ip6_tnl_parm *parms)
1427 {
1428         memset(parms, 0, sizeof(*parms));
1429
1430         if (!data)
1431                 return;
1432
1433         if (data[IFLA_GRE_LINK])
1434                 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1435
1436         if (data[IFLA_GRE_IFLAGS])
1437                 parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1438
1439         if (data[IFLA_GRE_OFLAGS])
1440                 parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1441
1442         if (data[IFLA_GRE_IKEY])
1443                 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1444
1445         if (data[IFLA_GRE_OKEY])
1446                 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1447
1448         if (data[IFLA_GRE_LOCAL])
1449                 nla_memcpy(&parms->laddr, data[IFLA_GRE_LOCAL], sizeof(struct in6_addr));
1450
1451         if (data[IFLA_GRE_REMOTE])
1452                 nla_memcpy(&parms->raddr, data[IFLA_GRE_REMOTE], sizeof(struct in6_addr));
1453
1454         if (data[IFLA_GRE_TTL])
1455                 parms->hop_limit = nla_get_u8(data[IFLA_GRE_TTL]);
1456
1457         if (data[IFLA_GRE_ENCAP_LIMIT])
1458                 parms->encap_limit = nla_get_u8(data[IFLA_GRE_ENCAP_LIMIT]);
1459
1460         if (data[IFLA_GRE_FLOWINFO])
1461                 parms->flowinfo = nla_get_u32(data[IFLA_GRE_FLOWINFO]);
1462
1463         if (data[IFLA_GRE_FLAGS])
1464                 parms->flags = nla_get_u32(data[IFLA_GRE_FLAGS]);
1465 }
1466
1467 static int ip6gre_tap_init(struct net_device *dev)
1468 {
1469         struct ip6_tnl *tunnel;
1470
1471         tunnel = netdev_priv(dev);
1472
1473         tunnel->dev = dev;
1474         tunnel->net = dev_net(dev);
1475         strcpy(tunnel->parms.name, dev->name);
1476
1477         ip6gre_tnl_link_config(tunnel, 1);
1478
1479         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1480         if (!dev->tstats)
1481                 return -ENOMEM;
1482
1483         return 0;
1484 }
1485
1486 static const struct net_device_ops ip6gre_tap_netdev_ops = {
1487         .ndo_init = ip6gre_tap_init,
1488         .ndo_uninit = ip6gre_tunnel_uninit,
1489         .ndo_start_xmit = ip6gre_tunnel_xmit,
1490         .ndo_set_mac_address = eth_mac_addr,
1491         .ndo_validate_addr = eth_validate_addr,
1492         .ndo_change_mtu = ip6gre_tunnel_change_mtu,
1493         .ndo_get_stats64 = ip_tunnel_get_stats64,
1494 };
1495
1496 static void ip6gre_tap_setup(struct net_device *dev)
1497 {
1498
1499         ether_setup(dev);
1500
1501         dev->netdev_ops = &ip6gre_tap_netdev_ops;
1502         dev->destructor = ip6gre_dev_free;
1503
1504         dev->iflink = 0;
1505         dev->features |= NETIF_F_NETNS_LOCAL;
1506 }
1507
1508 static int ip6gre_newlink(struct net *src_net, struct net_device *dev,
1509         struct nlattr *tb[], struct nlattr *data[])
1510 {
1511         struct ip6_tnl *nt;
1512         struct net *net = dev_net(dev);
1513         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1514         int err;
1515
1516         nt = netdev_priv(dev);
1517         ip6gre_netlink_parms(data, &nt->parms);
1518
1519         if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
1520                 return -EEXIST;
1521
1522         if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1523                 eth_hw_addr_random(dev);
1524
1525         nt->dev = dev;
1526         nt->net = dev_net(dev);
1527         ip6gre_tnl_link_config(nt, !tb[IFLA_MTU]);
1528
1529         /* Can use a lockless transmit, unless we generate output sequences */
1530         if (!(nt->parms.o_flags & GRE_SEQ))
1531                 dev->features |= NETIF_F_LLTX;
1532
1533         err = register_netdevice(dev);
1534         if (err)
1535                 goto out;
1536
1537         dev_hold(dev);
1538         ip6gre_tunnel_link(ign, nt);
1539
1540 out:
1541         return err;
1542 }
1543
1544 static int ip6gre_changelink(struct net_device *dev, struct nlattr *tb[],
1545                             struct nlattr *data[])
1546 {
1547         struct ip6_tnl *t, *nt = netdev_priv(dev);
1548         struct net *net = nt->net;
1549         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1550         struct __ip6_tnl_parm p;
1551
1552         if (dev == ign->fb_tunnel_dev)
1553                 return -EINVAL;
1554
1555         ip6gre_netlink_parms(data, &p);
1556
1557         t = ip6gre_tunnel_locate(net, &p, 0);
1558
1559         if (t) {
1560                 if (t->dev != dev)
1561                         return -EEXIST;
1562         } else {
1563                 t = nt;
1564
1565                 ip6gre_tunnel_unlink(ign, t);
1566                 ip6gre_tnl_change(t, &p, !tb[IFLA_MTU]);
1567                 ip6gre_tunnel_link(ign, t);
1568                 netdev_state_change(dev);
1569         }
1570
1571         return 0;
1572 }
1573
1574 static void ip6gre_dellink(struct net_device *dev, struct list_head *head)
1575 {
1576         struct net *net = dev_net(dev);
1577         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1578
1579         if (dev != ign->fb_tunnel_dev)
1580                 unregister_netdevice_queue(dev, head);
1581 }
1582
1583 static size_t ip6gre_get_size(const struct net_device *dev)
1584 {
1585         return
1586                 /* IFLA_GRE_LINK */
1587                 nla_total_size(4) +
1588                 /* IFLA_GRE_IFLAGS */
1589                 nla_total_size(2) +
1590                 /* IFLA_GRE_OFLAGS */
1591                 nla_total_size(2) +
1592                 /* IFLA_GRE_IKEY */
1593                 nla_total_size(4) +
1594                 /* IFLA_GRE_OKEY */
1595                 nla_total_size(4) +
1596                 /* IFLA_GRE_LOCAL */
1597                 nla_total_size(sizeof(struct in6_addr)) +
1598                 /* IFLA_GRE_REMOTE */
1599                 nla_total_size(sizeof(struct in6_addr)) +
1600                 /* IFLA_GRE_TTL */
1601                 nla_total_size(1) +
1602                 /* IFLA_GRE_TOS */
1603                 nla_total_size(1) +
1604                 /* IFLA_GRE_ENCAP_LIMIT */
1605                 nla_total_size(1) +
1606                 /* IFLA_GRE_FLOWINFO */
1607                 nla_total_size(4) +
1608                 /* IFLA_GRE_FLAGS */
1609                 nla_total_size(4) +
1610                 0;
1611 }
1612
1613 static int ip6gre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1614 {
1615         struct ip6_tnl *t = netdev_priv(dev);
1616         struct __ip6_tnl_parm *p = &t->parms;
1617
1618         if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1619             nla_put_be16(skb, IFLA_GRE_IFLAGS, p->i_flags) ||
1620             nla_put_be16(skb, IFLA_GRE_OFLAGS, p->o_flags) ||
1621             nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1622             nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1623             nla_put(skb, IFLA_GRE_LOCAL, sizeof(struct in6_addr), &p->laddr) ||
1624             nla_put(skb, IFLA_GRE_REMOTE, sizeof(struct in6_addr), &p->raddr) ||
1625             nla_put_u8(skb, IFLA_GRE_TTL, p->hop_limit) ||
1626             /*nla_put_u8(skb, IFLA_GRE_TOS, t->priority) ||*/
1627             nla_put_u8(skb, IFLA_GRE_ENCAP_LIMIT, p->encap_limit) ||
1628             nla_put_be32(skb, IFLA_GRE_FLOWINFO, p->flowinfo) ||
1629             nla_put_u32(skb, IFLA_GRE_FLAGS, p->flags))
1630                 goto nla_put_failure;
1631         return 0;
1632
1633 nla_put_failure:
1634         return -EMSGSIZE;
1635 }
1636
1637 static const struct nla_policy ip6gre_policy[IFLA_GRE_MAX + 1] = {
1638         [IFLA_GRE_LINK]        = { .type = NLA_U32 },
1639         [IFLA_GRE_IFLAGS]      = { .type = NLA_U16 },
1640         [IFLA_GRE_OFLAGS]      = { .type = NLA_U16 },
1641         [IFLA_GRE_IKEY]        = { .type = NLA_U32 },
1642         [IFLA_GRE_OKEY]        = { .type = NLA_U32 },
1643         [IFLA_GRE_LOCAL]       = { .len = FIELD_SIZEOF(struct ipv6hdr, saddr) },
1644         [IFLA_GRE_REMOTE]      = { .len = FIELD_SIZEOF(struct ipv6hdr, daddr) },
1645         [IFLA_GRE_TTL]         = { .type = NLA_U8 },
1646         [IFLA_GRE_ENCAP_LIMIT] = { .type = NLA_U8 },
1647         [IFLA_GRE_FLOWINFO]    = { .type = NLA_U32 },
1648         [IFLA_GRE_FLAGS]       = { .type = NLA_U32 },
1649 };
1650
1651 static struct rtnl_link_ops ip6gre_link_ops __read_mostly = {
1652         .kind           = "ip6gre",
1653         .maxtype        = IFLA_GRE_MAX,
1654         .policy         = ip6gre_policy,
1655         .priv_size      = sizeof(struct ip6_tnl),
1656         .setup          = ip6gre_tunnel_setup,
1657         .validate       = ip6gre_tunnel_validate,
1658         .newlink        = ip6gre_newlink,
1659         .changelink     = ip6gre_changelink,
1660         .dellink        = ip6gre_dellink,
1661         .get_size       = ip6gre_get_size,
1662         .fill_info      = ip6gre_fill_info,
1663 };
1664
1665 static struct rtnl_link_ops ip6gre_tap_ops __read_mostly = {
1666         .kind           = "ip6gretap",
1667         .maxtype        = IFLA_GRE_MAX,
1668         .policy         = ip6gre_policy,
1669         .priv_size      = sizeof(struct ip6_tnl),
1670         .setup          = ip6gre_tap_setup,
1671         .validate       = ip6gre_tap_validate,
1672         .newlink        = ip6gre_newlink,
1673         .changelink     = ip6gre_changelink,
1674         .get_size       = ip6gre_get_size,
1675         .fill_info      = ip6gre_fill_info,
1676 };
1677
1678 /*
1679  *      And now the modules code and kernel interface.
1680  */
1681
1682 static int __init ip6gre_init(void)
1683 {
1684         int err;
1685
1686         pr_info("GRE over IPv6 tunneling driver\n");
1687
1688         err = register_pernet_device(&ip6gre_net_ops);
1689         if (err < 0)
1690                 return err;
1691
1692         err = inet6_add_protocol(&ip6gre_protocol, IPPROTO_GRE);
1693         if (err < 0) {
1694                 pr_info("%s: can't add protocol\n", __func__);
1695                 goto add_proto_failed;
1696         }
1697
1698         err = rtnl_link_register(&ip6gre_link_ops);
1699         if (err < 0)
1700                 goto rtnl_link_failed;
1701
1702         err = rtnl_link_register(&ip6gre_tap_ops);
1703         if (err < 0)
1704                 goto tap_ops_failed;
1705
1706 out:
1707         return err;
1708
1709 tap_ops_failed:
1710         rtnl_link_unregister(&ip6gre_link_ops);
1711 rtnl_link_failed:
1712         inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1713 add_proto_failed:
1714         unregister_pernet_device(&ip6gre_net_ops);
1715         goto out;
1716 }
1717
1718 static void __exit ip6gre_fini(void)
1719 {
1720         rtnl_link_unregister(&ip6gre_tap_ops);
1721         rtnl_link_unregister(&ip6gre_link_ops);
1722         inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1723         unregister_pernet_device(&ip6gre_net_ops);
1724 }
1725
1726 module_init(ip6gre_init);
1727 module_exit(ip6gre_fini);
1728 MODULE_LICENSE("GPL");
1729 MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
1730 MODULE_DESCRIPTION("GRE over IPv6 tunneling device");
1731 MODULE_ALIAS_RTNL_LINK("ip6gre");
1732 MODULE_ALIAS_NETDEV("ip6gre0");