net/ipv6/ip6_gre.c

   1 /*
   2  *      GRE over IPv6 protocol decoder.
   3  *
   4  *      Authors: Dmitry Kozlov (xeb@mail.ru)
   5  *
   6  *      This program is free software; you can redistribute it and/or
   7  *      modify it under the terms of the GNU General Public License
   8  *      as published by the Free Software Foundation; either version
   9  *      2 of the License, or (at your option) any later version.
  10  *
  11  */
  12
  13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  14
  15 #include <linux/capability.h>
  16 #include <linux/module.h>
  17 #include <linux/types.h>
  18 #include <linux/kernel.h>
  19 #include <linux/slab.h>
  20 #include <linux/uaccess.h>
  21 #include <linux/skbuff.h>
  22 #include <linux/netdevice.h>
  23 #include <linux/in.h>
  24 #include <linux/tcp.h>
  25 #include <linux/udp.h>
  26 #include <linux/if_arp.h>
  27 #include <linux/mroute.h>
  28 #include <linux/init.h>
  29 #include <linux/in6.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/igmp.h>
  32 #include <linux/netfilter_ipv4.h>
  33 #include <linux/etherdevice.h>
  34 #include <linux/if_ether.h>
  35 #include <linux/hash.h>
  36 #include <linux/if_tunnel.h>
  37 #include <linux/ip6_tunnel.h>
  38
  39 #include <net/sock.h>
  40 #include <net/ip.h>
  41 #include <net/ip_tunnels.h>
  42 #include <net/icmp.h>
  43 #include <net/protocol.h>
  44 #include <net/addrconf.h>
  45 #include <net/arp.h>
  46 #include <net/checksum.h>
  47 #include <net/dsfield.h>
  48 #include <net/inet_ecn.h>
  49 #include <net/xfrm.h>
  50 #include <net/net_namespace.h>
  51 #include <net/netns/generic.h>
  52 #include <net/rtnetlink.h>
  53
  54 #include <net/ipv6.h>
  55 #include <net/ip6_fib.h>
  56 #include <net/ip6_route.h>
  57 #include <net/ip6_tunnel.h>
  58 #include <net/gre.h>
  59
  60
  61 static bool log_ecn_error = true;
  62 module_param(log_ecn_error, bool, 0644);
  63 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  64
  65 #define HASH_SIZE_SHIFT  5
  66 #define HASH_SIZE (1 << HASH_SIZE_SHIFT)
  67
  68 static int ip6gre_net_id __read_mostly;
  69 struct ip6gre_net {
  70         struct ip6_tnl __rcu *tunnels[4][HASH_SIZE];
  71
  72         struct net_device *fb_tunnel_dev;
  73 };
  74
  75 static struct rtnl_link_ops ip6gre_link_ops __read_mostly;
  76 static struct rtnl_link_ops ip6gre_tap_ops __read_mostly;
  77 static int ip6gre_tunnel_init(struct net_device *dev);
  78 static void ip6gre_tunnel_setup(struct net_device *dev);
  79 static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t);
  80 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu);
  81
  82 /* Tunnel hash table */
  83
  84 /*
  85    4 hash tables:
  86
  87    3: (remote,local)
  88    2: (remote,*)
  89    1: (*,local)
  90    0: (*,*)
  91
  92    We require exact key match i.e. if a key is present in packet
  93    it will match only tunnel with the same key; if it is not present,
  94    it will match only keyless tunnel.
  95
  96    All keysless packets, if not matched configured keyless tunnels
  97    will match fallback tunnel.
  98  */
  99
 100 #define HASH_KEY(key) (((__force u32)key^((__force u32)key>>4))&(HASH_SIZE - 1))
 101 static u32 HASH_ADDR(const struct in6_addr *addr)
 102 {
 103         u32 hash = ipv6_addr_hash(addr);
 104
 105         return hash_32(hash, HASH_SIZE_SHIFT);
 106 }
 107
 108 #define tunnels_r_l     tunnels[3]
 109 #define tunnels_r       tunnels[2]
 110 #define tunnels_l       tunnels[1]
 111 #define tunnels_wc      tunnels[0]
 112
 113 /* Given src, dst and key, find appropriate for input tunnel. */
 114
 115 static struct ip6_tnl *ip6gre_tunnel_lookup(struct net_device *dev,
 116                 const struct in6_addr *remote, const struct in6_addr *local,
 117                 __be32 key, __be16 gre_proto)
 118 {
 119         struct net *net = dev_net(dev);
 120         int link = dev->ifindex;
 121         unsigned int h0 = HASH_ADDR(remote);
 122         unsigned int h1 = HASH_KEY(key);
 123         struct ip6_tnl *t, *cand = NULL;
 124         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
 125         int dev_type = (gre_proto == htons(ETH_P_TEB)) ?
 126                        ARPHRD_ETHER : ARPHRD_IP6GRE;
 127         int score, cand_score = 4;
 128         struct net_device *ndev;
 129
 130         for_each_ip_tunnel_rcu(t, ign->tunnels_r_l[h0 ^ h1]) {
 131                 if (!ipv6_addr_equal(local, &t->parms.laddr) ||
 132                     !ipv6_addr_equal(remote, &t->parms.raddr) ||
 133                     key != t->parms.i_key ||
 134                     !(t->dev->flags & IFF_UP))
 135                         continue;
 136
 137                 if (t->dev->type != ARPHRD_IP6GRE &&
 138                     t->dev->type != dev_type)
 139                         continue;
 140
 141                 score = 0;
 142                 if (t->parms.link != link)
 143                         score |= 1;
 144                 if (t->dev->type != dev_type)
 145                         score |= 2;
 146                 if (score == 0)
 147                         return t;
 148
 149                 if (score < cand_score) {
 150                         cand = t;
 151                         cand_score = score;
 152                 }
 153         }
 154
 155         for_each_ip_tunnel_rcu(t, ign->tunnels_r[h0 ^ h1]) {
 156                 if (!ipv6_addr_equal(remote, &t->parms.raddr) ||
 157                     key != t->parms.i_key ||
 158                     !(t->dev->flags & IFF_UP))
 159                         continue;
 160
 161                 if (t->dev->type != ARPHRD_IP6GRE &&
 162                     t->dev->type != dev_type)
 163                         continue;
 164
 165                 score = 0;
 166                 if (t->parms.link != link)
 167                         score |= 1;
 168                 if (t->dev->type != dev_type)
 169                         score |= 2;
 170                 if (score == 0)
 171                         return t;
 172
 173                 if (score < cand_score) {
 174                         cand = t;
 175                         cand_score = score;
 176                 }
 177         }
 178
 179         for_each_ip_tunnel_rcu(t, ign->tunnels_l[h1]) {
 180                 if ((!ipv6_addr_equal(local, &t->parms.laddr) &&
 181                           (!ipv6_addr_equal(local, &t->parms.raddr) ||
 182                                  !ipv6_addr_is_multicast(local))) ||
 183                     key != t->parms.i_key ||
 184                     !(t->dev->flags & IFF_UP))
 185                         continue;
 186
 187                 if (t->dev->type != ARPHRD_IP6GRE &&
 188                     t->dev->type != dev_type)
 189                         continue;
 190
 191                 score = 0;
 192                 if (t->parms.link != link)
 193                         score |= 1;
 194                 if (t->dev->type != dev_type)
 195                         score |= 2;
 196                 if (score == 0)
 197                         return t;
 198
 199                 if (score < cand_score) {
 200                         cand = t;
 201                         cand_score = score;
 202                 }
 203         }
 204
 205         for_each_ip_tunnel_rcu(t, ign->tunnels_wc[h1]) {
 206                 if (t->parms.i_key != key ||
 207                     !(t->dev->flags & IFF_UP))
 208                         continue;
 209
 210                 if (t->dev->type != ARPHRD_IP6GRE &&
 211                     t->dev->type != dev_type)
 212                         continue;
 213
 214                 score = 0;
 215                 if (t->parms.link != link)
 216                         score |= 1;
 217                 if (t->dev->type != dev_type)
 218                         score |= 2;
 219                 if (score == 0)
 220                         return t;
 221
 222                 if (score < cand_score) {
 223                         cand = t;
 224                         cand_score = score;
 225                 }
 226         }
 227
 228         if (cand)
 229                 return cand;
 230
 231         ndev = READ_ONCE(ign->fb_tunnel_dev);
 232         if (ndev && ndev->flags & IFF_UP)
 233                 return netdev_priv(ndev);
 234
 235         return NULL;
 236 }
 237
 238 static struct ip6_tnl __rcu **__ip6gre_bucket(struct ip6gre_net *ign,
 239                 const struct __ip6_tnl_parm *p)
 240 {
 241         const struct in6_addr *remote = &p->raddr;
 242         const struct in6_addr *local = &p->laddr;
 243         unsigned int h = HASH_KEY(p->i_key);
 244         int prio = 0;
 245
 246         if (!ipv6_addr_any(local))
 247                 prio |= 1;
 248         if (!ipv6_addr_any(remote) && !ipv6_addr_is_multicast(remote)) {
 249                 prio |= 2;
 250                 h ^= HASH_ADDR(remote);
 251         }
 252
 253         return &ign->tunnels[prio][h];
 254 }
 255
 256 static inline struct ip6_tnl __rcu **ip6gre_bucket(struct ip6gre_net *ign,
 257                 const struct ip6_tnl *t)
 258 {
 259         return __ip6gre_bucket(ign, &t->parms);
 260 }
 261
 262 static void ip6gre_tunnel_link(struct ip6gre_net *ign, struct ip6_tnl *t)
 263 {
 264         struct ip6_tnl __rcu **tp = ip6gre_bucket(ign, t);
 265
 266         rcu_assign_pointer(t->next, rtnl_dereference(*tp));
 267         rcu_assign_pointer(*tp, t);
 268 }
 269
 270 static void ip6gre_tunnel_unlink(struct ip6gre_net *ign, struct ip6_tnl *t)
 271 {
 272         struct ip6_tnl __rcu **tp;
 273         struct ip6_tnl *iter;
 274
 275         for (tp = ip6gre_bucket(ign, t);
 276              (iter = rtnl_dereference(*tp)) != NULL;
 277              tp = &iter->next) {
 278                 if (t == iter) {
 279                         rcu_assign_pointer(*tp, t->next);
 280                         break;
 281                 }
 282         }
 283 }
 284
 285 static struct ip6_tnl *ip6gre_tunnel_find(struct net *net,
 286                                            const struct __ip6_tnl_parm *parms,
 287                                            int type)
 288 {
 289         const struct in6_addr *remote = &parms->raddr;
 290         const struct in6_addr *local = &parms->laddr;
 291         __be32 key = parms->i_key;
 292         int link = parms->link;
 293         struct ip6_tnl *t;
 294         struct ip6_tnl __rcu **tp;
 295         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
 296
 297         for (tp = __ip6gre_bucket(ign, parms);
 298              (t = rtnl_dereference(*tp)) != NULL;
 299              tp = &t->next)
 300                 if (ipv6_addr_equal(local, &t->parms.laddr) &&
 301                     ipv6_addr_equal(remote, &t->parms.raddr) &&
 302                     key == t->parms.i_key &&
 303                     link == t->parms.link &&
 304                     type == t->dev->type)
 305                         break;
 306
 307         return t;
 308 }
 309
 310 static struct ip6_tnl *ip6gre_tunnel_locate(struct net *net,
 311                 const struct __ip6_tnl_parm *parms, int create)
 312 {
 313         struct ip6_tnl *t, *nt;
 314         struct net_device *dev;
 315         char name[IFNAMSIZ];
 316         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
 317
 318         t = ip6gre_tunnel_find(net, parms, ARPHRD_IP6GRE);
 319         if (t && create)
 320                 return NULL;
 321         if (t || !create)
 322                 return t;
 323
 324         if (parms->name[0]) {
 325                 if (!dev_valid_name(parms->name))
 326                         return NULL;
 327                 strlcpy(name, parms->name, IFNAMSIZ);
 328         } else {
 329                 strcpy(name, "ip6gre%d");
 330         }
 331         dev = alloc_netdev(sizeof(*t), name, NET_NAME_UNKNOWN,
 332                            ip6gre_tunnel_setup);
 333         if (!dev)
 334                 return NULL;
 335
 336         dev_net_set(dev, net);
 337
 338         nt = netdev_priv(dev);
 339         nt->parms = *parms;
 340         dev->rtnl_link_ops = &ip6gre_link_ops;
 341
 342         nt->dev = dev;
 343         nt->net = dev_net(dev);
 344         ip6gre_tnl_link_config(nt, 1);
 345
 346         if (register_netdevice(dev) < 0)
 347                 goto failed_free;
 348
 349         /* Can use a lockless transmit, unless we generate output sequences */
 350         if (!(nt->parms.o_flags & GRE_SEQ))
 351                 dev->features |= NETIF_F_LLTX;
 352
 353         ip6gre_tunnel_link(ign, nt);
 354         return nt;
 355
 356 failed_free:
 357         free_netdev(dev);
 358         return NULL;
 359 }
 360
 361 static void ip6gre_tunnel_uninit(struct net_device *dev)
 362 {
 363         struct ip6_tnl *t = netdev_priv(dev);
 364         struct ip6gre_net *ign = net_generic(t->net, ip6gre_net_id);
 365
 366         ip6gre_tunnel_unlink(ign, t);
 367         if (ign->fb_tunnel_dev == dev)
 368                 WRITE_ONCE(ign->fb_tunnel_dev, NULL);
 369         dst_cache_reset(&t->dst_cache);
 370         dev_put(dev);
 371 }
 372
 373
 374 static void ip6gre_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
 375                        u8 type, u8 code, int offset, __be32 info)
 376 {
 377         const struct gre_base_hdr *greh;
 378         const struct ipv6hdr *ipv6h;
 379         int grehlen = sizeof(*greh);
 380         struct ip6_tnl *t;
 381         int key_off = 0;
 382         __be16 flags;
 383         __be32 key;
 384
 385         if (!pskb_may_pull(skb, offset + grehlen))
 386                 return;
 387         greh = (const struct gre_base_hdr *)(skb->data + offset);
 388         flags = greh->flags;
 389         if (flags & (GRE_VERSION | GRE_ROUTING))
 390                 return;
 391         if (flags & GRE_CSUM)
 392                 grehlen += 4;
 393         if (flags & GRE_KEY) {
 394                 key_off = grehlen + offset;
 395                 grehlen += 4;
 396         }
 397
 398         if (!pskb_may_pull(skb, offset + grehlen))
 399                 return;
 400         ipv6h = (const struct ipv6hdr *)skb->data;
 401         greh = (const struct gre_base_hdr *)(skb->data + offset);
 402         key = key_off ? *(__be32 *)(skb->data + key_off) : 0;
 403
 404         t = ip6gre_tunnel_lookup(skb->dev, &ipv6h->daddr, &ipv6h->saddr,
 405                                  key, greh->protocol);
 406         if (!t)
 407                 return;
 408
 409         switch (type) {
 410                 __u32 teli;
 411                 struct ipv6_tlv_tnl_enc_lim *tel;
 412                 __u32 mtu;
 413         case ICMPV6_DEST_UNREACH:
 414                 net_dbg_ratelimited("%s: Path to destination invalid or inactive!\n",
 415                                     t->parms.name);
 416                 if (code != ICMPV6_PORT_UNREACH)
 417                         break;
 418                 return;
 419         case ICMPV6_TIME_EXCEED:
 420                 if (code == ICMPV6_EXC_HOPLIMIT) {
 421                         net_dbg_ratelimited("%s: Too small hop limit or routing loop in tunnel!\n",
 422                                             t->parms.name);
 423                         break;
 424                 }
 425                 return;
 426         case ICMPV6_PARAMPROB:
 427                 teli = 0;
 428                 if (code == ICMPV6_HDR_FIELD)
 429                         teli = ip6_tnl_parse_tlv_enc_lim(skb, skb->data);
 430
 431                 if (teli && teli == be32_to_cpu(info) - 2) {
 432                         tel = (struct ipv6_tlv_tnl_enc_lim *) &skb->data[teli];
 433                         if (tel->encap_limit == 0) {
 434                                 net_dbg_ratelimited("%s: Too small encapsulation limit or routing loop in tunnel!\n",
 435                                                     t->parms.name);
 436                         }
 437                 } else {
 438                         net_dbg_ratelimited("%s: Recipient unable to parse tunneled packet!\n",
 439                                             t->parms.name);
 440                 }
 441                 return;
 442         case ICMPV6_PKT_TOOBIG:
 443                 mtu = be32_to_cpu(info) - offset;
 444                 if (mtu < IPV6_MIN_MTU)
 445                         mtu = IPV6_MIN_MTU;
 446                 t->dev->mtu = mtu;
 447                 return;
 448         }
 449
 450         if (time_before(jiffies, t->err_time + IP6TUNNEL_ERR_TIMEO))
 451                 t->err_count++;
 452         else
 453                 t->err_count = 1;
 454         t->err_time = jiffies;
 455 }
 456
 457 static int ip6gre_rcv(struct sk_buff *skb)
 458 {
 459         const struct ipv6hdr *ipv6h;
 460         u8     *h;
 461         __be16    flags;
 462         __sum16   csum = 0;
 463         __be32 key = 0;
 464         u32    seqno = 0;
 465         struct ip6_tnl *tunnel;
 466         int    offset = 4;
 467         __be16 gre_proto;
 468         int err;
 469
 470         if (!pskb_may_pull(skb, sizeof(struct in6_addr)))
 471                 goto drop;
 472
 473         ipv6h = ipv6_hdr(skb);
 474         h = skb->data;
 475         flags = *(__be16 *)h;
 476
 477         if (flags&(GRE_CSUM|GRE_KEY|GRE_ROUTING|GRE_SEQ|GRE_VERSION)) {
 478                 /* - Version must be 0.
 479                    - We do not support routing headers.
 480                  */
 481                 if (flags&(GRE_VERSION|GRE_ROUTING))
 482                         goto drop;
 483
 484                 if (flags&GRE_CSUM) {
 485                         csum = skb_checksum_simple_validate(skb);
 486                         offset += 4;
 487                 }
 488                 if (flags&GRE_KEY) {
 489                         key = *(__be32 *)(h + offset);
 490                         offset += 4;
 491                 }
 492                 if (flags&GRE_SEQ) {
 493                         seqno = ntohl(*(__be32 *)(h + offset));
 494                         offset += 4;
 495                 }
 496         }
 497
 498         gre_proto = *(__be16 *)(h + 2);
 499
 500         tunnel = ip6gre_tunnel_lookup(skb->dev,
 501                                           &ipv6h->saddr, &ipv6h->daddr, key,
 502                                           gre_proto);
 503         if (tunnel) {
 504                 struct pcpu_sw_netstats *tstats;
 505
 506                 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
 507                         goto drop;
 508
 509                 if (!ip6_tnl_rcv_ctl(tunnel, &ipv6h->daddr, &ipv6h->saddr)) {
 510                         tunnel->dev->stats.rx_dropped++;
 511                         goto drop;
 512                 }
 513
 514                 skb->protocol = gre_proto;
 515                 /* WCCP version 1 and 2 protocol decoding.
 516                  * - Change protocol to IPv6
 517                  * - When dealing with WCCPv2, Skip extra 4 bytes in GRE header
 518                  */
 519                 if (flags == 0 && gre_proto == htons(ETH_P_WCCP)) {
 520                         skb->protocol = htons(ETH_P_IPV6);
 521                         if ((*(h + offset) & 0xF0) != 0x40)
 522                                 offset += 4;
 523                 }
 524
 525                 skb->mac_header = skb->network_header;
 526                 __pskb_pull(skb, offset);
 527                 skb_postpull_rcsum(skb, skb_transport_header(skb), offset);
 528
 529                 if (((flags&GRE_CSUM) && csum) ||
 530                     (!(flags&GRE_CSUM) && tunnel->parms.i_flags&GRE_CSUM)) {
 531                         tunnel->dev->stats.rx_crc_errors++;
 532                         tunnel->dev->stats.rx_errors++;
 533                         goto drop;
 534                 }
 535                 if (tunnel->parms.i_flags&GRE_SEQ) {
 536                         if (!(flags&GRE_SEQ) ||
 537                             (tunnel->i_seqno &&
 538                                         (s32)(seqno - tunnel->i_seqno) < 0)) {
 539                                 tunnel->dev->stats.rx_fifo_errors++;
 540                                 tunnel->dev->stats.rx_errors++;
 541                                 goto drop;
 542                         }
 543                         tunnel->i_seqno = seqno + 1;
 544                 }
 545
 546                 /* Warning: All skb pointers will be invalidated! */
 547                 if (tunnel->dev->type == ARPHRD_ETHER) {
 548                         if (!pskb_may_pull(skb, ETH_HLEN)) {
 549                                 tunnel->dev->stats.rx_length_errors++;
 550                                 tunnel->dev->stats.rx_errors++;
 551                                 goto drop;
 552                         }
 553
 554                         ipv6h = ipv6_hdr(skb);
 555                         skb->protocol = eth_type_trans(skb, tunnel->dev);
 556                         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 557                 }
 558
 559                 __skb_tunnel_rx(skb, tunnel->dev, tunnel->net);
 560
 561                 skb_reset_network_header(skb);
 562
 563                 err = IP6_ECN_decapsulate(ipv6h, skb);
 564                 if (unlikely(err)) {
 565                         if (log_ecn_error)
 566                                 net_info_ratelimited("non-ECT from %pI6 with dsfield=%#x\n",
 567                                                      &ipv6h->saddr,
 568                                                      ipv6_get_dsfield(ipv6h));
 569                         if (err > 1) {
 570                                 ++tunnel->dev->stats.rx_frame_errors;
 571                                 ++tunnel->dev->stats.rx_errors;
 572                                 goto drop;
 573                         }
 574                 }
 575
 576                 tstats = this_cpu_ptr(tunnel->dev->tstats);
 577                 u64_stats_update_begin(&tstats->syncp);
 578                 tstats->rx_packets++;
 579                 tstats->rx_bytes += skb->len;
 580                 u64_stats_update_end(&tstats->syncp);
 581
 582                 netif_rx(skb);
 583
 584                 return 0;
 585         }
 586         icmpv6_send(skb, ICMPV6_DEST_UNREACH, ICMPV6_PORT_UNREACH, 0);
 587
 588 drop:
 589         kfree_skb(skb);
 590         return 0;
 591 }
 592
 593 struct ipv6_tel_txoption {
 594         struct ipv6_txoptions ops;
 595         __u8 dst_opt[8];
 596 };
 597
 598 static void init_tel_txopt(struct ipv6_tel_txoption *opt, __u8 encap_limit)
 599 {
 600         memset(opt, 0, sizeof(struct ipv6_tel_txoption));
 601
 602         opt->dst_opt[2] = IPV6_TLV_TNL_ENCAP_LIMIT;
 603         opt->dst_opt[3] = 1;
 604         opt->dst_opt[4] = encap_limit;
 605         opt->dst_opt[5] = IPV6_TLV_PADN;
 606         opt->dst_opt[6] = 1;
 607
 608         opt->ops.dst0opt = (struct ipv6_opt_hdr *) opt->dst_opt;
 609         opt->ops.opt_nflen = 8;
 610 }
 611
 612 static netdev_tx_t ip6gre_xmit2(struct sk_buff *skb,
 613                          struct net_device *dev,
 614                          __u8 dsfield,
 615                          struct flowi6 *fl6,
 616                          int encap_limit,
 617                          __u32 *pmtu)
 618 {
 619         struct ip6_tnl *tunnel = netdev_priv(dev);
 620         struct net *net = tunnel->net;
 621         struct net_device *tdev;    /* Device to other host */
 622         struct ipv6hdr  *ipv6h;     /* Our new IP header */
 623         unsigned int max_headroom = 0; /* The extra header space needed */
 624         int    gre_hlen;
 625         struct ipv6_tel_txoption opt;
 626         int    mtu;
 627         struct dst_entry *dst = NULL, *ndst = NULL;
 628         struct net_device_stats *stats = &tunnel->dev->stats;
 629         int err = -1;
 630         u8 proto;
 631         struct sk_buff *new_skb;
 632         __be16 protocol;
 633
 634         if (dev->type == ARPHRD_ETHER)
 635                 IPCB(skb)->flags = 0;
 636
 637         if (dev->header_ops && dev->type == ARPHRD_IP6GRE) {
 638                 gre_hlen = 0;
 639                 ipv6h = (struct ipv6hdr *)skb->data;
 640                 fl6->daddr = ipv6h->daddr;
 641         } else {
 642                 gre_hlen = tunnel->hlen;
 643                 fl6->daddr = tunnel->parms.raddr;
 644         }
 645
 646         if (!fl6->flowi6_mark)
 647                 dst = dst_cache_get(&tunnel->dst_cache);
 648
 649         if (!dst) {
 650                 dst = ip6_route_output(net, NULL, fl6);
 651
 652                 if (dst->error)
 653                         goto tx_err_link_failure;
 654                 dst = xfrm_lookup(net, dst, flowi6_to_flowi(fl6), NULL, 0);
 655                 if (IS_ERR(dst)) {
 656                         err = PTR_ERR(dst);
 657                         dst = NULL;
 658                         goto tx_err_link_failure;
 659                 }
 660                 ndst = dst;
 661         }
 662
 663         tdev = dst->dev;
 664
 665         if (tdev == dev) {
 666                 stats->collisions++;
 667                 net_warn_ratelimited("%s: Local routing loop detected!\n",
 668                                      tunnel->parms.name);
 669                 goto tx_err_dst_release;
 670         }
 671
 672         mtu = dst_mtu(dst) - sizeof(*ipv6h);
 673         if (encap_limit >= 0) {
 674                 max_headroom += 8;
 675                 mtu -= 8;
 676         }
 677         if (mtu < IPV6_MIN_MTU)
 678                 mtu = IPV6_MIN_MTU;
 679         if (skb_dst(skb))
 680                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), NULL, skb, mtu);
 681         if (skb->len > mtu) {
 682                 *pmtu = mtu;
 683                 err = -EMSGSIZE;
 684                 goto tx_err_dst_release;
 685         }
 686
 687         if (tunnel->err_count > 0) {
 688                 if (time_before(jiffies,
 689                                 tunnel->err_time + IP6TUNNEL_ERR_TIMEO)) {
 690                         tunnel->err_count--;
 691
 692                         dst_link_failure(skb);
 693                 } else
 694                         tunnel->err_count = 0;
 695         }
 696
 697         skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(dev)));
 698
 699         max_headroom += LL_RESERVED_SPACE(tdev) + gre_hlen + dst->header_len;
 700
 701         if (skb_headroom(skb) < max_headroom || skb_shared(skb) ||
 702             (skb_cloned(skb) && !skb_clone_writable(skb, 0))) {
 703                 new_skb = skb_realloc_headroom(skb, max_headroom);
 704                 if (max_headroom > dev->needed_headroom)
 705                         dev->needed_headroom = max_headroom;
 706                 if (!new_skb)
 707                         goto tx_err_dst_release;
 708
 709                 if (skb->sk)
 710                         skb_set_owner_w(new_skb, skb->sk);
 711                 consume_skb(skb);
 712                 skb = new_skb;
 713         }
 714
 715         if (!fl6->flowi6_mark && ndst)
 716                 dst_cache_set_ip6(&tunnel->dst_cache, ndst, &fl6->saddr);
 717         skb_dst_set(skb, dst);
 718
 719         proto = NEXTHDR_GRE;
 720         if (encap_limit >= 0) {
 721                 init_tel_txopt(&opt, encap_limit);
 722                 ipv6_push_nfrag_opts(skb, &opt.ops, &proto, NULL);
 723         }
 724
 725         if (likely(!skb->encapsulation)) {
 726                 skb_reset_inner_headers(skb);
 727                 skb->encapsulation = 1;
 728         }
 729
 730         skb_push(skb, gre_hlen);
 731         skb_reset_network_header(skb);
 732         skb_set_transport_header(skb, sizeof(*ipv6h));
 733
 734         /*
 735          *      Push down and install the IP header.
 736          */
 737         ipv6h = ipv6_hdr(skb);
 738         ip6_flow_hdr(ipv6h, INET_ECN_encapsulate(0, dsfield),
 739                      ip6_make_flowlabel(net, skb, fl6->flowlabel, true, fl6));
 740         ipv6h->hop_limit = tunnel->parms.hop_limit;
 741         ipv6h->nexthdr = proto;
 742         ipv6h->saddr = fl6->saddr;
 743         ipv6h->daddr = fl6->daddr;
 744
 745         ((__be16 *)(ipv6h + 1))[0] = tunnel->parms.o_flags;
 746         protocol = (dev->type == ARPHRD_ETHER) ?
 747                     htons(ETH_P_TEB) : skb->protocol;
 748         ((__be16 *)(ipv6h + 1))[1] = protocol;
 749
 750         if (tunnel->parms.o_flags&(GRE_KEY|GRE_CSUM|GRE_SEQ)) {
 751                 __be32 *ptr = (__be32 *)(((u8 *)ipv6h) + tunnel->hlen - 4);
 752
 753                 if (tunnel->parms.o_flags&GRE_SEQ) {
 754                         ++tunnel->o_seqno;
 755                         *ptr = htonl(tunnel->o_seqno);
 756                         ptr--;
 757                 }
 758                 if (tunnel->parms.o_flags&GRE_KEY) {
 759                         *ptr = tunnel->parms.o_key;
 760                         ptr--;
 761                 }
 762                 if (tunnel->parms.o_flags&GRE_CSUM) {
 763                         *ptr = 0;
 764                         *(__sum16 *)ptr = ip_compute_csum((void *)(ipv6h+1),
 765                                 skb->len - sizeof(struct ipv6hdr));
 766                 }
 767         }
 768
 769         skb_set_inner_protocol(skb, protocol);
 770
 771         ip6tunnel_xmit(NULL, skb, dev);
 772         return 0;
 773 tx_err_link_failure:
 774         stats->tx_carrier_errors++;
 775         dst_link_failure(skb);
 776 tx_err_dst_release:
 777         dst_release(dst);
 778         return err;
 779 }
 780
 781 static inline int ip6gre_xmit_ipv4(struct sk_buff *skb, struct net_device *dev)
 782 {
 783         struct ip6_tnl *t = netdev_priv(dev);
 784         const struct iphdr  *iph = ip_hdr(skb);
 785         int encap_limit = -1;
 786         struct flowi6 fl6;
 787         __u8 dsfield;
 788         __u32 mtu;
 789         int err;
 790
 791         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 792
 793         if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
 794                 encap_limit = t->parms.encap_limit;
 795
 796         memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
 797         fl6.flowi6_proto = IPPROTO_GRE;
 798
 799         dsfield = ipv4_get_dsfield(iph);
 800
 801         if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
 802                 fl6.flowlabel |= htonl((__u32)iph->tos << IPV6_TCLASS_SHIFT)
 803                                           & IPV6_TCLASS_MASK;
 804         if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
 805                 fl6.flowi6_mark = skb->mark;
 806
 807         err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
 808         if (err != 0) {
 809                 /* XXX: send ICMP error even if DF is not set. */
 810                 if (err == -EMSGSIZE)
 811                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
 812                                   htonl(mtu));
 813                 return -1;
 814         }
 815
 816         return 0;
 817 }
 818
 819 static inline int ip6gre_xmit_ipv6(struct sk_buff *skb, struct net_device *dev)
 820 {
 821         struct ip6_tnl *t = netdev_priv(dev);
 822         struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 823         int encap_limit = -1;
 824         __u16 offset;
 825         struct flowi6 fl6;
 826         __u8 dsfield;
 827         __u32 mtu;
 828         int err;
 829
 830         if (ipv6_addr_equal(&t->parms.raddr, &ipv6h->saddr))
 831                 return -1;
 832
 833         offset = ip6_tnl_parse_tlv_enc_lim(skb, skb_network_header(skb));
 834         if (offset > 0) {
 835                 struct ipv6_tlv_tnl_enc_lim *tel;
 836                 tel = (struct ipv6_tlv_tnl_enc_lim *)&skb_network_header(skb)[offset];
 837                 if (tel->encap_limit == 0) {
 838                         icmpv6_send(skb, ICMPV6_PARAMPROB,
 839                                     ICMPV6_HDR_FIELD, offset + 2);
 840                         return -1;
 841                 }
 842                 encap_limit = tel->encap_limit - 1;
 843         } else if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
 844                 encap_limit = t->parms.encap_limit;
 845
 846         memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
 847         fl6.flowi6_proto = IPPROTO_GRE;
 848
 849         dsfield = ipv6_get_dsfield(ipv6h);
 850         if (t->parms.flags & IP6_TNL_F_USE_ORIG_TCLASS)
 851                 fl6.flowlabel |= (*(__be32 *) ipv6h & IPV6_TCLASS_MASK);
 852         if (t->parms.flags & IP6_TNL_F_USE_ORIG_FLOWLABEL)
 853                 fl6.flowlabel |= ip6_flowlabel(ipv6h);
 854         if (t->parms.flags & IP6_TNL_F_USE_ORIG_FWMARK)
 855                 fl6.flowi6_mark = skb->mark;
 856
 857         err = ip6gre_xmit2(skb, dev, dsfield, &fl6, encap_limit, &mtu);
 858         if (err != 0) {
 859                 if (err == -EMSGSIZE)
 860                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 861                 return -1;
 862         }
 863
 864         return 0;
 865 }
 866
 867 /**
 868  * ip6_tnl_addr_conflict - compare packet addresses to tunnel's own
 869  *   @t: the outgoing tunnel device
 870  *   @hdr: IPv6 header from the incoming packet
 871  *
 872  * Description:
 873  *   Avoid trivial tunneling loop by checking that tunnel exit-point
 874  *   doesn't match source of incoming packet.
 875  *
 876  * Return:
 877  *   1 if conflict,
 878  *   0 else
 879  **/
 880
 881 static inline bool ip6gre_tnl_addr_conflict(const struct ip6_tnl *t,
 882         const struct ipv6hdr *hdr)
 883 {
 884         return ipv6_addr_equal(&t->parms.raddr, &hdr->saddr);
 885 }
 886
 887 static int ip6gre_xmit_other(struct sk_buff *skb, struct net_device *dev)
 888 {
 889         struct ip6_tnl *t = netdev_priv(dev);
 890         int encap_limit = -1;
 891         struct flowi6 fl6;
 892         __u32 mtu;
 893         int err;
 894
 895         if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
 896                 encap_limit = t->parms.encap_limit;
 897
 898         memcpy(&fl6, &t->fl.u.ip6, sizeof(fl6));
 899
 900         err = ip6gre_xmit2(skb, dev, 0, &fl6, encap_limit, &mtu);
 901
 902         return err;
 903 }
 904
 905 static netdev_tx_t ip6gre_tunnel_xmit(struct sk_buff *skb,
 906         struct net_device *dev)
 907 {
 908         struct ip6_tnl *t = netdev_priv(dev);
 909         struct net_device_stats *stats = &t->dev->stats;
 910         int ret;
 911
 912         if (!ip6_tnl_xmit_ctl(t, &t->parms.laddr, &t->parms.raddr))
 913                 goto tx_err;
 914
 915         switch (skb->protocol) {
 916         case htons(ETH_P_IP):
 917                 ret = ip6gre_xmit_ipv4(skb, dev);
 918                 break;
 919         case htons(ETH_P_IPV6):
 920                 ret = ip6gre_xmit_ipv6(skb, dev);
 921                 break;
 922         default:
 923                 ret = ip6gre_xmit_other(skb, dev);
 924                 break;
 925         }
 926
 927         if (ret < 0)
 928                 goto tx_err;
 929
 930         return NETDEV_TX_OK;
 931
 932 tx_err:
 933         stats->tx_errors++;
 934         stats->tx_dropped++;
 935         kfree_skb(skb);
 936         return NETDEV_TX_OK;
 937 }
 938
 939 static void ip6gre_tnl_link_config(struct ip6_tnl *t, int set_mtu)
 940 {
 941         struct net_device *dev = t->dev;
 942         struct __ip6_tnl_parm *p = &t->parms;
 943         struct flowi6 *fl6 = &t->fl.u.ip6;
 944         int addend = sizeof(struct ipv6hdr) + 4;
 945
 946         if (dev->type != ARPHRD_ETHER) {
 947                 memcpy(dev->dev_addr, &p->laddr, sizeof(struct in6_addr));
 948                 memcpy(dev->broadcast, &p->raddr, sizeof(struct in6_addr));
 949         }
 950
 951         /* Set up flowi template */
 952         fl6->saddr = p->laddr;
 953         fl6->daddr = p->raddr;
 954         fl6->flowi6_oif = p->link;
 955         fl6->flowlabel = 0;
 956
 957         if (!(p->flags&IP6_TNL_F_USE_ORIG_TCLASS))
 958                 fl6->flowlabel |= IPV6_TCLASS_MASK & p->flowinfo;
 959         if (!(p->flags&IP6_TNL_F_USE_ORIG_FLOWLABEL))
 960                 fl6->flowlabel |= IPV6_FLOWLABEL_MASK & p->flowinfo;
 961
 962         p->flags &= ~(IP6_TNL_F_CAP_XMIT|IP6_TNL_F_CAP_RCV|IP6_TNL_F_CAP_PER_PACKET);
 963         p->flags |= ip6_tnl_get_cap(t, &p->laddr, &p->raddr);
 964
 965         if (p->flags&IP6_TNL_F_CAP_XMIT &&
 966                         p->flags&IP6_TNL_F_CAP_RCV && dev->type != ARPHRD_ETHER)
 967                 dev->flags |= IFF_POINTOPOINT;
 968         else
 969                 dev->flags &= ~IFF_POINTOPOINT;
 970
 971         /* Precalculate GRE options length */
 972         if (t->parms.o_flags&(GRE_CSUM|GRE_KEY|GRE_SEQ)) {
 973                 if (t->parms.o_flags&GRE_CSUM)
 974                         addend += 4;
 975                 if (t->parms.o_flags&GRE_KEY)
 976                         addend += 4;
 977                 if (t->parms.o_flags&GRE_SEQ)
 978                         addend += 4;
 979         }
 980         t->hlen = addend;
 981
 982         if (p->flags & IP6_TNL_F_CAP_XMIT) {
 983                 int strict = (ipv6_addr_type(&p->raddr) &
 984                               (IPV6_ADDR_MULTICAST|IPV6_ADDR_LINKLOCAL));
 985
 986                 struct rt6_info *rt = rt6_lookup(t->net,
 987                                                  &p->raddr, &p->laddr,
 988                                                  p->link, strict);
 989
 990                 if (!rt)
 991                         return;
 992
 993                 if (rt->dst.dev) {
 994                         dev->hard_header_len = rt->dst.dev->hard_header_len + addend;
 995
 996                         if (set_mtu) {
 997                                 dev->mtu = rt->dst.dev->mtu - addend;
 998                                 if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
 999                                         dev->mtu -= 8;
1000
1001                                 if (dev->mtu < IPV6_MIN_MTU)
1002                                         dev->mtu = IPV6_MIN_MTU;
1003                         }
1004                 }
1005                 ip6_rt_put(rt);
1006         }
1007 }
1008
1009 static int ip6gre_tnl_change(struct ip6_tnl *t,
1010         const struct __ip6_tnl_parm *p, int set_mtu)
1011 {
1012         t->parms.laddr = p->laddr;
1013         t->parms.raddr = p->raddr;
1014         t->parms.flags = p->flags;
1015         t->parms.hop_limit = p->hop_limit;
1016         t->parms.encap_limit = p->encap_limit;
1017         t->parms.flowinfo = p->flowinfo;
1018         t->parms.link = p->link;
1019         t->parms.proto = p->proto;
1020         t->parms.i_key = p->i_key;
1021         t->parms.o_key = p->o_key;
1022         t->parms.i_flags = p->i_flags;
1023         t->parms.o_flags = p->o_flags;
1024         dst_cache_reset(&t->dst_cache);
1025         ip6gre_tnl_link_config(t, set_mtu);
1026         return 0;
1027 }
1028
1029 static void ip6gre_tnl_parm_from_user(struct __ip6_tnl_parm *p,
1030         const struct ip6_tnl_parm2 *u)
1031 {
1032         p->laddr = u->laddr;
1033         p->raddr = u->raddr;
1034         p->flags = u->flags;
1035         p->hop_limit = u->hop_limit;
1036         p->encap_limit = u->encap_limit;
1037         p->flowinfo = u->flowinfo;
1038         p->link = u->link;
1039         p->i_key = u->i_key;
1040         p->o_key = u->o_key;
1041         p->i_flags = u->i_flags;
1042         p->o_flags = u->o_flags;
1043         memcpy(p->name, u->name, sizeof(u->name));
1044 }
1045
1046 static void ip6gre_tnl_parm_to_user(struct ip6_tnl_parm2 *u,
1047         const struct __ip6_tnl_parm *p)
1048 {
1049         u->proto = IPPROTO_GRE;
1050         u->laddr = p->laddr;
1051         u->raddr = p->raddr;
1052         u->flags = p->flags;
1053         u->hop_limit = p->hop_limit;
1054         u->encap_limit = p->encap_limit;
1055         u->flowinfo = p->flowinfo;
1056         u->link = p->link;
1057         u->i_key = p->i_key;
1058         u->o_key = p->o_key;
1059         u->i_flags = p->i_flags;
1060         u->o_flags = p->o_flags;
1061         memcpy(u->name, p->name, sizeof(u->name));
1062 }
1063
1064 static int ip6gre_tunnel_ioctl(struct net_device *dev,
1065         struct ifreq *ifr, int cmd)
1066 {
1067         int err = 0;
1068         struct ip6_tnl_parm2 p;
1069         struct __ip6_tnl_parm p1;
1070         struct ip6_tnl *t = netdev_priv(dev);
1071         struct net *net = t->net;
1072         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1073
1074         switch (cmd) {
1075         case SIOCGETTUNNEL:
1076                 if (dev == ign->fb_tunnel_dev) {
1077                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p))) {
1078                                 err = -EFAULT;
1079                                 break;
1080                         }
1081                         ip6gre_tnl_parm_from_user(&p1, &p);
1082                         t = ip6gre_tunnel_locate(net, &p1, 0);
1083                         if (!t)
1084                                 t = netdev_priv(dev);
1085                 }
1086                 memset(&p, 0, sizeof(p));
1087                 ip6gre_tnl_parm_to_user(&p, &t->parms);
1088                 if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1089                         err = -EFAULT;
1090                 break;
1091
1092         case SIOCADDTUNNEL:
1093         case SIOCCHGTUNNEL:
1094                 err = -EPERM;
1095                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1096                         goto done;
1097
1098                 err = -EFAULT;
1099                 if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1100                         goto done;
1101
1102                 err = -EINVAL;
1103                 if ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING))
1104                         goto done;
1105
1106                 if (!(p.i_flags&GRE_KEY))
1107                         p.i_key = 0;
1108                 if (!(p.o_flags&GRE_KEY))
1109                         p.o_key = 0;
1110
1111                 ip6gre_tnl_parm_from_user(&p1, &p);
1112                 t = ip6gre_tunnel_locate(net, &p1, cmd == SIOCADDTUNNEL);
1113
1114                 if (dev != ign->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
1115                         if (t) {
1116                                 if (t->dev != dev) {
1117                                         err = -EEXIST;
1118                                         break;
1119                                 }
1120                         } else {
1121                                 t = netdev_priv(dev);
1122
1123                                 ip6gre_tunnel_unlink(ign, t);
1124                                 synchronize_net();
1125                                 ip6gre_tnl_change(t, &p1, 1);
1126                                 ip6gre_tunnel_link(ign, t);
1127                                 netdev_state_change(dev);
1128                         }
1129                 }
1130
1131                 if (t) {
1132                         err = 0;
1133
1134                         memset(&p, 0, sizeof(p));
1135                         ip6gre_tnl_parm_to_user(&p, &t->parms);
1136                         if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
1137                                 err = -EFAULT;
1138                 } else
1139                         err = (cmd == SIOCADDTUNNEL ? -ENOBUFS : -ENOENT);
1140                 break;
1141
1142         case SIOCDELTUNNEL:
1143                 err = -EPERM;
1144                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1145                         goto done;
1146
1147                 if (dev == ign->fb_tunnel_dev) {
1148                         err = -EFAULT;
1149                         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
1150                                 goto done;
1151                         err = -ENOENT;
1152                         ip6gre_tnl_parm_from_user(&p1, &p);
1153                         t = ip6gre_tunnel_locate(net, &p1, 0);
1154                         if (!t)
1155                                 goto done;
1156                         err = -EPERM;
1157                         if (t == netdev_priv(ign->fb_tunnel_dev))
1158                                 goto done;
1159                         dev = t->dev;
1160                 }
1161                 unregister_netdevice(dev);
1162                 err = 0;
1163                 break;
1164
1165         default:
1166                 err = -EINVAL;
1167         }
1168
1169 done:
1170         return err;
1171 }
1172
1173 static int ip6gre_tunnel_change_mtu(struct net_device *dev, int new_mtu)
1174 {
1175         if (new_mtu < 68 ||
1176             new_mtu > 0xFFF8 - dev->hard_header_len)
1177                 return -EINVAL;
1178         dev->mtu = new_mtu;
1179         return 0;
1180 }
1181
1182 static int ip6gre_header(struct sk_buff *skb, struct net_device *dev,
1183                          unsigned short type, const void *daddr,
1184                          const void *saddr, unsigned int len)
1185 {
1186         struct ip6_tnl *t = netdev_priv(dev);
1187         struct ipv6hdr *ipv6h;
1188         __be16 *p;
1189
1190         ipv6h = (struct ipv6hdr *)skb_push(skb, t->hlen + sizeof(*ipv6h));
1191         ip6_flow_hdr(ipv6h, 0, ip6_make_flowlabel(dev_net(dev), skb,
1192                                                   t->fl.u.ip6.flowlabel,
1193                                                   true, &t->fl.u.ip6));
1194         ipv6h->hop_limit = t->parms.hop_limit;
1195         ipv6h->nexthdr = NEXTHDR_GRE;
1196         ipv6h->saddr = t->parms.laddr;
1197         ipv6h->daddr = t->parms.raddr;
1198
1199         p = (__be16 *)(ipv6h + 1);
1200         p[0] = t->parms.o_flags;
1201         p[1] = htons(type);
1202
1203         /*
1204          *      Set the source hardware address.
1205          */
1206
1207         if (saddr)
1208                 memcpy(&ipv6h->saddr, saddr, sizeof(struct in6_addr));
1209         if (daddr)
1210                 memcpy(&ipv6h->daddr, daddr, sizeof(struct in6_addr));
1211         if (!ipv6_addr_any(&ipv6h->daddr))
1212                 return t->hlen;
1213
1214         return -t->hlen;
1215 }
1216
1217 static const struct header_ops ip6gre_header_ops = {
1218         .create = ip6gre_header,
1219 };
1220
1221 static const struct net_device_ops ip6gre_netdev_ops = {
1222         .ndo_init               = ip6gre_tunnel_init,
1223         .ndo_uninit             = ip6gre_tunnel_uninit,
1224         .ndo_start_xmit         = ip6gre_tunnel_xmit,
1225         .ndo_do_ioctl           = ip6gre_tunnel_ioctl,
1226         .ndo_change_mtu         = ip6gre_tunnel_change_mtu,
1227         .ndo_get_stats64        = ip_tunnel_get_stats64,
1228         .ndo_get_iflink         = ip6_tnl_get_iflink,
1229 };
1230
1231 static void ip6gre_dev_free(struct net_device *dev)
1232 {
1233         struct ip6_tnl *t = netdev_priv(dev);
1234
1235         dst_cache_destroy(&t->dst_cache);
1236         free_percpu(dev->tstats);
1237         free_netdev(dev);
1238 }
1239
1240 static void ip6gre_tunnel_setup(struct net_device *dev)
1241 {
1242         struct ip6_tnl *t;
1243
1244         dev->netdev_ops = &ip6gre_netdev_ops;
1245         dev->destructor = ip6gre_dev_free;
1246
1247         dev->type = ARPHRD_IP6GRE;
1248         dev->hard_header_len = LL_MAX_HEADER + sizeof(struct ipv6hdr) + 4;
1249         dev->mtu = ETH_DATA_LEN - sizeof(struct ipv6hdr) - 4;
1250         t = netdev_priv(dev);
1251         if (!(t->parms.flags & IP6_TNL_F_IGN_ENCAP_LIMIT))
1252                 dev->mtu -= 8;
1253         dev->flags |= IFF_NOARP;
1254         dev->addr_len = sizeof(struct in6_addr);
1255         netif_keep_dst(dev);
1256 }
1257
1258 static int ip6gre_tunnel_init_common(struct net_device *dev)
1259 {
1260         struct ip6_tnl *tunnel;
1261         int ret;
1262
1263         tunnel = netdev_priv(dev);
1264
1265         tunnel->dev = dev;
1266         tunnel->net = dev_net(dev);
1267         strcpy(tunnel->parms.name, dev->name);
1268
1269         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1270         if (!dev->tstats)
1271                 return -ENOMEM;
1272
1273         ret = dst_cache_init(&tunnel->dst_cache, GFP_KERNEL);
1274         if (ret) {
1275                 free_percpu(dev->tstats);
1276                 dev->tstats = NULL;
1277                 return ret;
1278         }
1279
1280         return 0;
1281 }
1282
1283 static int ip6gre_tunnel_init(struct net_device *dev)
1284 {
1285         struct ip6_tnl *tunnel;
1286         int ret;
1287
1288         ret = ip6gre_tunnel_init_common(dev);
1289         if (ret)
1290                 return ret;
1291
1292         tunnel = netdev_priv(dev);
1293
1294         memcpy(dev->dev_addr, &tunnel->parms.laddr, sizeof(struct in6_addr));
1295         memcpy(dev->broadcast, &tunnel->parms.raddr, sizeof(struct in6_addr));
1296
1297         if (ipv6_addr_any(&tunnel->parms.raddr))
1298                 dev->header_ops = &ip6gre_header_ops;
1299
1300         return 0;
1301 }
1302
1303 static void ip6gre_fb_tunnel_init(struct net_device *dev)
1304 {
1305         struct ip6_tnl *tunnel = netdev_priv(dev);
1306
1307         tunnel->dev = dev;
1308         tunnel->net = dev_net(dev);
1309         strcpy(tunnel->parms.name, dev->name);
1310
1311         tunnel->hlen            = sizeof(struct ipv6hdr) + 4;
1312 }
1313
1314
1315 static struct inet6_protocol ip6gre_protocol __read_mostly = {
1316         .handler     = ip6gre_rcv,
1317         .err_handler = ip6gre_err,
1318         .flags       = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
1319 };
1320
1321 static void ip6gre_destroy_tunnels(struct net *net, struct list_head *head)
1322 {
1323         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1324         struct net_device *dev, *aux;
1325         int prio;
1326
1327         for_each_netdev_safe(net, dev, aux)
1328                 if (dev->rtnl_link_ops == &ip6gre_link_ops ||
1329                     dev->rtnl_link_ops == &ip6gre_tap_ops)
1330                         unregister_netdevice_queue(dev, head);
1331
1332         for (prio = 0; prio < 4; prio++) {
1333                 int h;
1334                 for (h = 0; h < HASH_SIZE; h++) {
1335                         struct ip6_tnl *t;
1336
1337                         t = rtnl_dereference(ign->tunnels[prio][h]);
1338
1339                         while (t) {
1340                                 /* If dev is in the same netns, it has already
1341                                  * been added to the list by the previous loop.
1342                                  */
1343                                 if (!net_eq(dev_net(t->dev), net))
1344                                         unregister_netdevice_queue(t->dev,
1345                                                                    head);
1346                                 t = rtnl_dereference(t->next);
1347                         }
1348                 }
1349         }
1350 }
1351
1352 static int __net_init ip6gre_init_net(struct net *net)
1353 {
1354         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1355         struct net_device *ndev;
1356         int err;
1357
1358         ndev = alloc_netdev(sizeof(struct ip6_tnl), "ip6gre0",
1359                             NET_NAME_UNKNOWN, ip6gre_tunnel_setup);
1360         if (!ndev) {
1361                 err = -ENOMEM;
1362                 goto err_alloc_dev;
1363         }
1364         ign->fb_tunnel_dev = ndev;
1365         dev_net_set(ign->fb_tunnel_dev, net);
1366         /* FB netdevice is special: we have one, and only one per netns.
1367          * Allowing to move it to another netns is clearly unsafe.
1368          */
1369         ign->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL;
1370
1371
1372         ip6gre_fb_tunnel_init(ign->fb_tunnel_dev);
1373         ign->fb_tunnel_dev->rtnl_link_ops = &ip6gre_link_ops;
1374
1375         err = register_netdev(ign->fb_tunnel_dev);
1376         if (err)
1377                 goto err_reg_dev;
1378
1379         rcu_assign_pointer(ign->tunnels_wc[0],
1380                            netdev_priv(ign->fb_tunnel_dev));
1381         return 0;
1382
1383 err_reg_dev:
1384         ip6gre_dev_free(ndev);
1385 err_alloc_dev:
1386         return err;
1387 }
1388
1389 static void __net_exit ip6gre_exit_net(struct net *net)
1390 {
1391         LIST_HEAD(list);
1392
1393         rtnl_lock();
1394         ip6gre_destroy_tunnels(net, &list);
1395         unregister_netdevice_many(&list);
1396         rtnl_unlock();
1397 }
1398
1399 static struct pernet_operations ip6gre_net_ops = {
1400         .init = ip6gre_init_net,
1401         .exit = ip6gre_exit_net,
1402         .id   = &ip6gre_net_id,
1403         .size = sizeof(struct ip6gre_net),
1404 };
1405
1406 static int ip6gre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[])
1407 {
1408         __be16 flags;
1409
1410         if (!data)
1411                 return 0;
1412
1413         flags = 0;
1414         if (data[IFLA_GRE_IFLAGS])
1415                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1416         if (data[IFLA_GRE_OFLAGS])
1417                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1418         if (flags & (GRE_VERSION|GRE_ROUTING))
1419                 return -EINVAL;
1420
1421         return 0;
1422 }
1423
1424 static int ip6gre_tap_validate(struct nlattr *tb[], struct nlattr *data[])
1425 {
1426         struct in6_addr daddr;
1427
1428         if (tb[IFLA_ADDRESS]) {
1429                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1430                         return -EINVAL;
1431                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1432                         return -EADDRNOTAVAIL;
1433         }
1434
1435         if (!data)
1436                 goto out;
1437
1438         if (data[IFLA_GRE_REMOTE]) {
1439                 daddr = nla_get_in6_addr(data[IFLA_GRE_REMOTE]);
1440                 if (ipv6_addr_any(&daddr))
1441                         return -EINVAL;
1442         }
1443
1444 out:
1445         return ip6gre_tunnel_validate(tb, data);
1446 }
1447
1448
1449 static void ip6gre_netlink_parms(struct nlattr *data[],
1450                                 struct __ip6_tnl_parm *parms)
1451 {
1452         memset(parms, 0, sizeof(*parms));
1453
1454         if (!data)
1455                 return;
1456
1457         if (data[IFLA_GRE_LINK])
1458                 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1459
1460         if (data[IFLA_GRE_IFLAGS])
1461                 parms->i_flags = nla_get_be16(data[IFLA_GRE_IFLAGS]);
1462
1463         if (data[IFLA_GRE_OFLAGS])
1464                 parms->o_flags = nla_get_be16(data[IFLA_GRE_OFLAGS]);
1465
1466         if (data[IFLA_GRE_IKEY])
1467                 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1468
1469         if (data[IFLA_GRE_OKEY])
1470                 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1471
1472         if (data[IFLA_GRE_LOCAL])
1473                 parms->laddr = nla_get_in6_addr(data[IFLA_GRE_LOCAL]);
1474
1475         if (data[IFLA_GRE_REMOTE])
1476                 parms->raddr = nla_get_in6_addr(data[IFLA_GRE_REMOTE]);
1477
1478         if (data[IFLA_GRE_TTL])
1479                 parms->hop_limit = nla_get_u8(data[IFLA_GRE_TTL]);
1480
1481         if (data[IFLA_GRE_ENCAP_LIMIT])
1482                 parms->encap_limit = nla_get_u8(data[IFLA_GRE_ENCAP_LIMIT]);
1483
1484         if (data[IFLA_GRE_FLOWINFO])
1485                 parms->flowinfo = nla_get_u32(data[IFLA_GRE_FLOWINFO]);
1486
1487         if (data[IFLA_GRE_FLAGS])
1488                 parms->flags = nla_get_u32(data[IFLA_GRE_FLAGS]);
1489 }
1490
1491 static int ip6gre_tap_init(struct net_device *dev)
1492 {
1493         struct ip6_tnl *tunnel;
1494         int ret;
1495
1496         ret = ip6gre_tunnel_init_common(dev);
1497         if (ret)
1498                 return ret;
1499
1500         tunnel = netdev_priv(dev);
1501
1502         ip6gre_tnl_link_config(tunnel, 1);
1503
1504         return 0;
1505 }
1506
1507 static const struct net_device_ops ip6gre_tap_netdev_ops = {
1508         .ndo_init = ip6gre_tap_init,
1509         .ndo_uninit = ip6gre_tunnel_uninit,
1510         .ndo_start_xmit = ip6gre_tunnel_xmit,
1511         .ndo_set_mac_address = eth_mac_addr,
1512         .ndo_validate_addr = eth_validate_addr,
1513         .ndo_change_mtu = ip6gre_tunnel_change_mtu,
1514         .ndo_get_stats64 = ip_tunnel_get_stats64,
1515         .ndo_get_iflink = ip6_tnl_get_iflink,
1516 };
1517
1518 static void ip6gre_tap_setup(struct net_device *dev)
1519 {
1520
1521         ether_setup(dev);
1522
1523         dev->netdev_ops = &ip6gre_tap_netdev_ops;
1524         dev->destructor = ip6gre_dev_free;
1525
1526         dev->features |= NETIF_F_NETNS_LOCAL;
1527 }
1528
1529 static int ip6gre_newlink(struct net *src_net, struct net_device *dev,
1530         struct nlattr *tb[], struct nlattr *data[])
1531 {
1532         struct ip6_tnl *nt;
1533         struct net *net = dev_net(dev);
1534         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1535         int err;
1536
1537         nt = netdev_priv(dev);
1538         ip6gre_netlink_parms(data, &nt->parms);
1539
1540         if (ip6gre_tunnel_find(net, &nt->parms, dev->type))
1541                 return -EEXIST;
1542
1543         if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1544                 eth_hw_addr_random(dev);
1545
1546         nt->dev = dev;
1547         nt->net = dev_net(dev);
1548         ip6gre_tnl_link_config(nt, !tb[IFLA_MTU]);
1549
1550         /* Can use a lockless transmit, unless we generate output sequences */
1551         if (!(nt->parms.o_flags & GRE_SEQ))
1552                 dev->features |= NETIF_F_LLTX;
1553
1554         err = register_netdevice(dev);
1555         if (err)
1556                 goto out;
1557
1558         dev_hold(dev);
1559         ip6gre_tunnel_link(ign, nt);
1560
1561 out:
1562         return err;
1563 }
1564
1565 static int ip6gre_changelink(struct net_device *dev, struct nlattr *tb[],
1566                             struct nlattr *data[])
1567 {
1568         struct ip6_tnl *t, *nt = netdev_priv(dev);
1569         struct net *net = nt->net;
1570         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1571         struct __ip6_tnl_parm p;
1572
1573         if (dev == ign->fb_tunnel_dev)
1574                 return -EINVAL;
1575
1576         ip6gre_netlink_parms(data, &p);
1577
1578         t = ip6gre_tunnel_locate(net, &p, 0);
1579
1580         if (t) {
1581                 if (t->dev != dev)
1582                         return -EEXIST;
1583         } else {
1584                 t = nt;
1585         }
1586
1587         ip6gre_tunnel_unlink(ign, t);
1588         ip6gre_tnl_change(t, &p, !tb[IFLA_MTU]);
1589         ip6gre_tunnel_link(ign, t);
1590         return 0;
1591 }
1592
1593 static void ip6gre_dellink(struct net_device *dev, struct list_head *head)
1594 {
1595         struct net *net = dev_net(dev);
1596         struct ip6gre_net *ign = net_generic(net, ip6gre_net_id);
1597
1598         if (dev != ign->fb_tunnel_dev)
1599                 unregister_netdevice_queue(dev, head);
1600 }
1601
1602 static size_t ip6gre_get_size(const struct net_device *dev)
1603 {
1604         return
1605                 /* IFLA_GRE_LINK */
1606                 nla_total_size(4) +
1607                 /* IFLA_GRE_IFLAGS */
1608                 nla_total_size(2) +
1609                 /* IFLA_GRE_OFLAGS */
1610                 nla_total_size(2) +
1611                 /* IFLA_GRE_IKEY */
1612                 nla_total_size(4) +
1613                 /* IFLA_GRE_OKEY */
1614                 nla_total_size(4) +
1615                 /* IFLA_GRE_LOCAL */
1616                 nla_total_size(sizeof(struct in6_addr)) +
1617                 /* IFLA_GRE_REMOTE */
1618                 nla_total_size(sizeof(struct in6_addr)) +
1619                 /* IFLA_GRE_TTL */
1620                 nla_total_size(1) +
1621                 /* IFLA_GRE_TOS */
1622                 nla_total_size(1) +
1623                 /* IFLA_GRE_ENCAP_LIMIT */
1624                 nla_total_size(1) +
1625                 /* IFLA_GRE_FLOWINFO */
1626                 nla_total_size(4) +
1627                 /* IFLA_GRE_FLAGS */
1628                 nla_total_size(4) +
1629                 0;
1630 }
1631
1632 static int ip6gre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1633 {
1634         struct ip6_tnl *t = netdev_priv(dev);
1635         struct __ip6_tnl_parm *p = &t->parms;
1636
1637         if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1638             nla_put_be16(skb, IFLA_GRE_IFLAGS, p->i_flags) ||
1639             nla_put_be16(skb, IFLA_GRE_OFLAGS, p->o_flags) ||
1640             nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1641             nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1642             nla_put_in6_addr(skb, IFLA_GRE_LOCAL, &p->laddr) ||
1643             nla_put_in6_addr(skb, IFLA_GRE_REMOTE, &p->raddr) ||
1644             nla_put_u8(skb, IFLA_GRE_TTL, p->hop_limit) ||
1645             /*nla_put_u8(skb, IFLA_GRE_TOS, t->priority) ||*/
1646             nla_put_u8(skb, IFLA_GRE_ENCAP_LIMIT, p->encap_limit) ||
1647             nla_put_be32(skb, IFLA_GRE_FLOWINFO, p->flowinfo) ||
1648             nla_put_u32(skb, IFLA_GRE_FLAGS, p->flags))
1649                 goto nla_put_failure;
1650         return 0;
1651
1652 nla_put_failure:
1653         return -EMSGSIZE;
1654 }
1655
1656 static const struct nla_policy ip6gre_policy[IFLA_GRE_MAX + 1] = {
1657         [IFLA_GRE_LINK]        = { .type = NLA_U32 },
1658         [IFLA_GRE_IFLAGS]      = { .type = NLA_U16 },
1659         [IFLA_GRE_OFLAGS]      = { .type = NLA_U16 },
1660         [IFLA_GRE_IKEY]        = { .type = NLA_U32 },
1661         [IFLA_GRE_OKEY]        = { .type = NLA_U32 },
1662         [IFLA_GRE_LOCAL]       = { .len = FIELD_SIZEOF(struct ipv6hdr, saddr) },
1663         [IFLA_GRE_REMOTE]      = { .len = FIELD_SIZEOF(struct ipv6hdr, daddr) },
1664         [IFLA_GRE_TTL]         = { .type = NLA_U8 },
1665         [IFLA_GRE_ENCAP_LIMIT] = { .type = NLA_U8 },
1666         [IFLA_GRE_FLOWINFO]    = { .type = NLA_U32 },
1667         [IFLA_GRE_FLAGS]       = { .type = NLA_U32 },
1668 };
1669
1670 static struct rtnl_link_ops ip6gre_link_ops __read_mostly = {
1671         .kind           = "ip6gre",
1672         .maxtype        = IFLA_GRE_MAX,
1673         .policy         = ip6gre_policy,
1674         .priv_size      = sizeof(struct ip6_tnl),
1675         .setup          = ip6gre_tunnel_setup,
1676         .validate       = ip6gre_tunnel_validate,
1677         .newlink        = ip6gre_newlink,
1678         .changelink     = ip6gre_changelink,
1679         .dellink        = ip6gre_dellink,
1680         .get_size       = ip6gre_get_size,
1681         .fill_info      = ip6gre_fill_info,
1682         .get_link_net   = ip6_tnl_get_link_net,
1683 };
1684
1685 static struct rtnl_link_ops ip6gre_tap_ops __read_mostly = {
1686         .kind           = "ip6gretap",
1687         .maxtype        = IFLA_GRE_MAX,
1688         .policy         = ip6gre_policy,
1689         .priv_size      = sizeof(struct ip6_tnl),
1690         .setup          = ip6gre_tap_setup,
1691         .validate       = ip6gre_tap_validate,
1692         .newlink        = ip6gre_newlink,
1693         .changelink     = ip6gre_changelink,
1694         .get_size       = ip6gre_get_size,
1695         .fill_info      = ip6gre_fill_info,
1696         .get_link_net   = ip6_tnl_get_link_net,
1697 };
1698
1699 /*
1700  *      And now the modules code and kernel interface.
1701  */
1702
1703 static int __init ip6gre_init(void)
1704 {
1705         int err;
1706
1707         pr_info("GRE over IPv6 tunneling driver\n");
1708
1709         err = register_pernet_device(&ip6gre_net_ops);
1710         if (err < 0)
1711                 return err;
1712
1713         err = inet6_add_protocol(&ip6gre_protocol, IPPROTO_GRE);
1714         if (err < 0) {
1715                 pr_info("%s: can't add protocol\n", __func__);
1716                 goto add_proto_failed;
1717         }
1718
1719         err = rtnl_link_register(&ip6gre_link_ops);
1720         if (err < 0)
1721                 goto rtnl_link_failed;
1722
1723         err = rtnl_link_register(&ip6gre_tap_ops);
1724         if (err < 0)
1725                 goto tap_ops_failed;
1726
1727 out:
1728         return err;
1729
1730 tap_ops_failed:
1731         rtnl_link_unregister(&ip6gre_link_ops);
1732 rtnl_link_failed:
1733         inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1734 add_proto_failed:
1735         unregister_pernet_device(&ip6gre_net_ops);
1736         goto out;
1737 }
1738
1739 static void __exit ip6gre_fini(void)
1740 {
1741         rtnl_link_unregister(&ip6gre_tap_ops);
1742         rtnl_link_unregister(&ip6gre_link_ops);
1743         inet6_del_protocol(&ip6gre_protocol, IPPROTO_GRE);
1744         unregister_pernet_device(&ip6gre_net_ops);
1745 }
1746
1747 module_init(ip6gre_init);
1748 module_exit(ip6gre_fini);
1749 MODULE_LICENSE("GPL");
1750 MODULE_AUTHOR("D. Kozlov (xeb@mail.ru)");
1751 MODULE_DESCRIPTION("GRE over IPv6 tunneling device");
1752 MODULE_ALIAS_RTNL_LINK("ip6gre");
1753 MODULE_ALIAS_RTNL_LINK("ip6gretap");
1754 MODULE_ALIAS_NETDEV("ip6gre0");