net/ipv4/ip_gre.c

   1 /*
   2  *      Linux NET3:     GRE over IP protocol decoder.
   3  *
   4  *      Authors: Alexey Kuznetsov ([email protected])
   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/if_vlan.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
  36 #include <net/sock.h>
  37 #include <net/ip.h>
  38 #include <net/icmp.h>
  39 #include <net/protocol.h>
  40 #include <net/ip_tunnels.h>
  41 #include <net/arp.h>
  42 #include <net/checksum.h>
  43 #include <net/dsfield.h>
  44 #include <net/inet_ecn.h>
  45 #include <net/xfrm.h>
  46 #include <net/net_namespace.h>
  47 #include <net/netns/generic.h>
  48 #include <net/rtnetlink.h>
  49 #include <net/gre.h>
  50 #include <net/dst_metadata.h>
  51 #include <net/erspan.h>
  52
  53 /*
  54    Problems & solutions
  55    --------------------
  56
  57    1. The most important issue is detecting local dead loops.
  58    They would cause complete host lockup in transmit, which
  59    would be "resolved" by stack overflow or, if queueing is enabled,
  60    with infinite looping in net_bh.
  61
  62    We cannot track such dead loops during route installation,
  63    it is infeasible task. The most general solutions would be
  64    to keep skb->encapsulation counter (sort of local ttl),
  65    and silently drop packet when it expires. It is a good
  66    solution, but it supposes maintaining new variable in ALL
  67    skb, even if no tunneling is used.
  68
  69    Current solution: xmit_recursion breaks dead loops. This is a percpu
  70    counter, since when we enter the first ndo_xmit(), cpu migration is
  71    forbidden. We force an exit if this counter reaches RECURSION_LIMIT
  72
  73    2. Networking dead loops would not kill routers, but would really
  74    kill network. IP hop limit plays role of "t->recursion" in this case,
  75    if we copy it from packet being encapsulated to upper header.
  76    It is very good solution, but it introduces two problems:
  77
  78    - Routing protocols, using packets with ttl=1 (OSPF, RIP2),
  79      do not work over tunnels.
  80    - traceroute does not work. I planned to relay ICMP from tunnel,
  81      so that this problem would be solved and traceroute output
  82      would even more informative. This idea appeared to be wrong:
  83      only Linux complies to rfc1812 now (yes, guys, Linux is the only
  84      true router now :-)), all routers (at least, in neighbourhood of mine)
  85      return only 8 bytes of payload. It is the end.
  86
  87    Hence, if we want that OSPF worked or traceroute said something reasonable,
  88    we should search for another solution.
  89
  90    One of them is to parse packet trying to detect inner encapsulation
  91    made by our node. It is difficult or even impossible, especially,
  92    taking into account fragmentation. TO be short, ttl is not solution at all.
  93
  94    Current solution: The solution was UNEXPECTEDLY SIMPLE.
  95    We force DF flag on tunnels with preconfigured hop limit,
  96    that is ALL. :-) Well, it does not remove the problem completely,
  97    but exponential growth of network traffic is changed to linear
  98    (branches, that exceed pmtu are pruned) and tunnel mtu
  99    rapidly degrades to value <68, where looping stops.
 100    Yes, it is not good if there exists a router in the loop,
 101    which does not force DF, even when encapsulating packets have DF set.
 102    But it is not our problem! Nobody could accuse us, we made
 103    all that we could make. Even if it is your gated who injected
 104    fatal route to network, even if it were you who configured
 105    fatal static route: you are innocent. :-)
 106
 107    Alexey Kuznetsov.
 108  */
 109
 110 static bool log_ecn_error = true;
 111 module_param(log_ecn_error, bool, 0644);
 112 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
 113
 114 static struct rtnl_link_ops ipgre_link_ops __read_mostly;
 115 static int ipgre_tunnel_init(struct net_device *dev);
 116 static void erspan_build_header(struct sk_buff *skb,
 117                                 u32 id, u32 index,
 118                                 bool truncate, bool is_ipv4);
 119
 120 static unsigned int ipgre_net_id __read_mostly;
 121 static unsigned int gre_tap_net_id __read_mostly;
 122 static unsigned int erspan_net_id __read_mostly;
 123
 124 static void ipgre_err(struct sk_buff *skb, u32 info,
 125                       const struct tnl_ptk_info *tpi)
 126 {
 127
 128         /* All the routers (except for Linux) return only
 129            8 bytes of packet payload. It means, that precise relaying of
 130            ICMP in the real Internet is absolutely infeasible.
 131
 132            Moreover, Cisco "wise men" put GRE key to the third word
 133            in GRE header. It makes impossible maintaining even soft
 134            state for keyed GRE tunnels with enabled checksum. Tell
 135            them "thank you".
 136
 137            Well, I wonder, rfc1812 was written by Cisco employee,
 138            what the hell these idiots break standards established
 139            by themselves???
 140            */
 141         struct net *net = dev_net(skb->dev);
 142         struct ip_tunnel_net *itn;
 143         const struct iphdr *iph;
 144         const int type = icmp_hdr(skb)->type;
 145         const int code = icmp_hdr(skb)->code;
 146         unsigned int data_len = 0;
 147         struct ip_tunnel *t;
 148
 149         switch (type) {
 150         default:
 151         case ICMP_PARAMETERPROB:
 152                 return;
 153
 154         case ICMP_DEST_UNREACH:
 155                 switch (code) {
 156                 case ICMP_SR_FAILED:
 157                 case ICMP_PORT_UNREACH:
 158                         /* Impossible event. */
 159                         return;
 160                 default:
 161                         /* All others are translated to HOST_UNREACH.
 162                            rfc2003 contains "deep thoughts" about NET_UNREACH,
 163                            I believe they are just ether pollution. --ANK
 164                          */
 165                         break;
 166                 }
 167                 break;
 168
 169         case ICMP_TIME_EXCEEDED:
 170                 if (code != ICMP_EXC_TTL)
 171                         return;
 172                 data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */
 173                 break;
 174
 175         case ICMP_REDIRECT:
 176                 break;
 177         }
 178
 179         if (tpi->proto == htons(ETH_P_TEB))
 180                 itn = net_generic(net, gre_tap_net_id);
 181         else
 182                 itn = net_generic(net, ipgre_net_id);
 183
 184         iph = (const struct iphdr *)(icmp_hdr(skb) + 1);
 185         t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
 186                              iph->daddr, iph->saddr, tpi->key);
 187
 188         if (!t)
 189                 return;
 190
 191 #if IS_ENABLED(CONFIG_IPV6)
 192        if (tpi->proto == htons(ETH_P_IPV6) &&
 193            !ip6_err_gen_icmpv6_unreach(skb, iph->ihl * 4 + tpi->hdr_len,
 194                                        type, data_len))
 195                return;
 196 #endif
 197
 198         if (t->parms.iph.daddr == 0 ||
 199             ipv4_is_multicast(t->parms.iph.daddr))
 200                 return;
 201
 202         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
 203                 return;
 204
 205         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
 206                 t->err_count++;
 207         else
 208                 t->err_count = 1;
 209         t->err_time = jiffies;
 210 }
 211
 212 static void gre_err(struct sk_buff *skb, u32 info)
 213 {
 214         /* All the routers (except for Linux) return only
 215          * 8 bytes of packet payload. It means, that precise relaying of
 216          * ICMP in the real Internet is absolutely infeasible.
 217          *
 218          * Moreover, Cisco "wise men" put GRE key to the third word
 219          * in GRE header. It makes impossible maintaining even soft
 220          * state for keyed
 221          * GRE tunnels with enabled checksum. Tell them "thank you".
 222          *
 223          * Well, I wonder, rfc1812 was written by Cisco employee,
 224          * what the hell these idiots break standards established
 225          * by themselves???
 226          */
 227
 228         const struct iphdr *iph = (struct iphdr *)skb->data;
 229         const int type = icmp_hdr(skb)->type;
 230         const int code = icmp_hdr(skb)->code;
 231         struct tnl_ptk_info tpi;
 232         bool csum_err = false;
 233
 234         if (gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP),
 235                              iph->ihl * 4) < 0) {
 236                 if (!csum_err)          /* ignore csum errors. */
 237                         return;
 238         }
 239
 240         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
 241                 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
 242                                  skb->dev->ifindex, 0, IPPROTO_GRE, 0);
 243                 return;
 244         }
 245         if (type == ICMP_REDIRECT) {
 246                 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex, 0,
 247                               IPPROTO_GRE, 0);
 248                 return;
 249         }
 250
 251         ipgre_err(skb, info, &tpi);
 252 }
 253
 254 static int erspan_rcv(struct sk_buff *skb, struct tnl_ptk_info *tpi,
 255                       int gre_hdr_len)
 256 {
 257         struct net *net = dev_net(skb->dev);
 258         struct metadata_dst *tun_dst = NULL;
 259         struct erspan_base_hdr *ershdr;
 260         struct erspan_metadata *pkt_md;
 261         struct ip_tunnel_net *itn;
 262         struct ip_tunnel *tunnel;
 263         const struct iphdr *iph;
 264         struct erspan_md2 *md2;
 265         int ver;
 266         int len;
 267
 268         itn = net_generic(net, erspan_net_id);
 269         len = gre_hdr_len + sizeof(*ershdr);
 270
 271         /* Check based hdr len */
 272         if (unlikely(!pskb_may_pull(skb, len)))
 273                 return PACKET_REJECT;
 274
 275         iph = ip_hdr(skb);
 276         ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
 277         ver = ershdr->ver;
 278
 279         /* The original GRE header does not have key field,
 280          * Use ERSPAN 10-bit session ID as key.
 281          */
 282         tpi->key = cpu_to_be32(get_session_id(ershdr));
 283         tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex,
 284                                   tpi->flags | TUNNEL_KEY,
 285                                   iph->saddr, iph->daddr, tpi->key);
 286
 287         if (tunnel) {
 288                 len = gre_hdr_len + erspan_hdr_len(ver);
 289                 if (unlikely(!pskb_may_pull(skb, len)))
 290                         return PACKET_REJECT;
 291
 292                 ershdr = (struct erspan_base_hdr *)(skb->data + gre_hdr_len);
 293                 pkt_md = (struct erspan_metadata *)(ershdr + 1);
 294
 295                 if (__iptunnel_pull_header(skb,
 296                                            len,
 297                                            htons(ETH_P_TEB),
 298                                            false, false) < 0)
 299                         goto drop;
 300
 301                 if (tunnel->collect_md) {
 302                         struct ip_tunnel_info *info;
 303                         struct erspan_metadata *md;
 304                         __be64 tun_id;
 305                         __be16 flags;
 306
 307                         tpi->flags |= TUNNEL_KEY;
 308                         flags = tpi->flags;
 309                         tun_id = key32_to_tunnel_id(tpi->key);
 310
 311                         tun_dst = ip_tun_rx_dst(skb, flags,
 312                                                 tun_id, sizeof(*md));
 313                         if (!tun_dst)
 314                                 return PACKET_REJECT;
 315
 316                         md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
 317                         md->version = ver;
 318                         md2 = &md->u.md2;
 319                         memcpy(md2, pkt_md, ver == 1 ? ERSPAN_V1_MDSIZE :
 320                                                        ERSPAN_V2_MDSIZE);
 321
 322                         info = &tun_dst->u.tun_info;
 323                         info->key.tun_flags |= TUNNEL_ERSPAN_OPT;
 324                         info->options_len = sizeof(*md);
 325                 }
 326
 327                 skb_reset_mac_header(skb);
 328                 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
 329                 return PACKET_RCVD;
 330         }
 331 drop:
 332         kfree_skb(skb);
 333         return PACKET_RCVD;
 334 }
 335
 336 static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
 337                        struct ip_tunnel_net *itn, int hdr_len, bool raw_proto)
 338 {
 339         struct metadata_dst *tun_dst = NULL;
 340         const struct iphdr *iph;
 341         struct ip_tunnel *tunnel;
 342
 343         iph = ip_hdr(skb);
 344         tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
 345                                   iph->saddr, iph->daddr, tpi->key);
 346
 347         if (tunnel) {
 348                 if (__iptunnel_pull_header(skb, hdr_len, tpi->proto,
 349                                            raw_proto, false) < 0)
 350                         goto drop;
 351
 352                 if (tunnel->dev->type != ARPHRD_NONE)
 353                         skb_pop_mac_header(skb);
 354                 else
 355                         skb_reset_mac_header(skb);
 356                 if (tunnel->collect_md) {
 357                         __be16 flags;
 358                         __be64 tun_id;
 359
 360                         flags = tpi->flags & (TUNNEL_CSUM | TUNNEL_KEY);
 361                         tun_id = key32_to_tunnel_id(tpi->key);
 362                         tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
 363                         if (!tun_dst)
 364                                 return PACKET_REJECT;
 365                 }
 366
 367                 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
 368                 return PACKET_RCVD;
 369         }
 370         return PACKET_NEXT;
 371
 372 drop:
 373         kfree_skb(skb);
 374         return PACKET_RCVD;
 375 }
 376
 377 static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
 378                      int hdr_len)
 379 {
 380         struct net *net = dev_net(skb->dev);
 381         struct ip_tunnel_net *itn;
 382         int res;
 383
 384         if (tpi->proto == htons(ETH_P_TEB))
 385                 itn = net_generic(net, gre_tap_net_id);
 386         else
 387                 itn = net_generic(net, ipgre_net_id);
 388
 389         res = __ipgre_rcv(skb, tpi, itn, hdr_len, false);
 390         if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) {
 391                 /* ipgre tunnels in collect metadata mode should receive
 392                  * also ETH_P_TEB traffic.
 393                  */
 394                 itn = net_generic(net, ipgre_net_id);
 395                 res = __ipgre_rcv(skb, tpi, itn, hdr_len, true);
 396         }
 397         return res;
 398 }
 399
 400 static int gre_rcv(struct sk_buff *skb)
 401 {
 402         struct tnl_ptk_info tpi;
 403         bool csum_err = false;
 404         int hdr_len;
 405
 406 #ifdef CONFIG_NET_IPGRE_BROADCAST
 407         if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
 408                 /* Looped back packet, drop it! */
 409                 if (rt_is_output_route(skb_rtable(skb)))
 410                         goto drop;
 411         }
 412 #endif
 413
 414         hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0);
 415         if (hdr_len < 0)
 416                 goto drop;
 417
 418         if (unlikely(tpi.proto == htons(ETH_P_ERSPAN) ||
 419                      tpi.proto == htons(ETH_P_ERSPAN2))) {
 420                 if (erspan_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
 421                         return 0;
 422                 goto out;
 423         }
 424
 425         if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
 426                 return 0;
 427
 428 out:
 429         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 430 drop:
 431         kfree_skb(skb);
 432         return 0;
 433 }
 434
 435 static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
 436                        const struct iphdr *tnl_params,
 437                        __be16 proto)
 438 {
 439         struct ip_tunnel *tunnel = netdev_priv(dev);
 440
 441         if (tunnel->parms.o_flags & TUNNEL_SEQ)
 442                 tunnel->o_seqno++;
 443
 444         /* Push GRE header. */
 445         gre_build_header(skb, tunnel->tun_hlen,
 446                          tunnel->parms.o_flags, proto, tunnel->parms.o_key,
 447                          htonl(tunnel->o_seqno));
 448
 449         ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
 450 }
 451
 452 static int gre_handle_offloads(struct sk_buff *skb, bool csum)
 453 {
 454         return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
 455 }
 456
 457 static struct rtable *gre_get_rt(struct sk_buff *skb,
 458                                  struct net_device *dev,
 459                                  struct flowi4 *fl,
 460                                  const struct ip_tunnel_key *key)
 461 {
 462         struct net *net = dev_net(dev);
 463
 464         memset(fl, 0, sizeof(*fl));
 465         fl->daddr = key->u.ipv4.dst;
 466         fl->saddr = key->u.ipv4.src;
 467         fl->flowi4_tos = RT_TOS(key->tos);
 468         fl->flowi4_mark = skb->mark;
 469         fl->flowi4_proto = IPPROTO_GRE;
 470
 471         return ip_route_output_key(net, fl);
 472 }
 473
 474 static struct rtable *prepare_fb_xmit(struct sk_buff *skb,
 475                                       struct net_device *dev,
 476                                       struct flowi4 *fl,
 477                                       int tunnel_hlen)
 478 {
 479         struct ip_tunnel_info *tun_info;
 480         const struct ip_tunnel_key *key;
 481         struct rtable *rt = NULL;
 482         int min_headroom;
 483         bool use_cache;
 484         int err;
 485
 486         tun_info = skb_tunnel_info(skb);
 487         key = &tun_info->key;
 488         use_cache = ip_tunnel_dst_cache_usable(skb, tun_info);
 489
 490         if (use_cache)
 491                 rt = dst_cache_get_ip4(&tun_info->dst_cache, &fl->saddr);
 492         if (!rt) {
 493                 rt = gre_get_rt(skb, dev, fl, key);
 494                 if (IS_ERR(rt))
 495                         goto err_free_skb;
 496                 if (use_cache)
 497                         dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst,
 498                                           fl->saddr);
 499         }
 500
 501         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 502                         + tunnel_hlen + sizeof(struct iphdr);
 503         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
 504                 int head_delta = SKB_DATA_ALIGN(min_headroom -
 505                                                 skb_headroom(skb) +
 506                                                 16);
 507                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
 508                                        0, GFP_ATOMIC);
 509                 if (unlikely(err))
 510                         goto err_free_rt;
 511         }
 512         return rt;
 513
 514 err_free_rt:
 515         ip_rt_put(rt);
 516 err_free_skb:
 517         kfree_skb(skb);
 518         dev->stats.tx_dropped++;
 519         return NULL;
 520 }
 521
 522 static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev,
 523                         __be16 proto)
 524 {
 525         struct ip_tunnel_info *tun_info;
 526         const struct ip_tunnel_key *key;
 527         struct rtable *rt = NULL;
 528         struct flowi4 fl;
 529         int tunnel_hlen;
 530         __be16 df, flags;
 531
 532         tun_info = skb_tunnel_info(skb);
 533         if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
 534                      ip_tunnel_info_af(tun_info) != AF_INET))
 535                 goto err_free_skb;
 536
 537         key = &tun_info->key;
 538         tunnel_hlen = gre_calc_hlen(key->tun_flags);
 539
 540         rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
 541         if (!rt)
 542                 return;
 543
 544         /* Push Tunnel header. */
 545         if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM)))
 546                 goto err_free_rt;
 547
 548         flags = tun_info->key.tun_flags & (TUNNEL_CSUM | TUNNEL_KEY);
 549         gre_build_header(skb, tunnel_hlen, flags, proto,
 550                          tunnel_id_to_key32(tun_info->key.tun_id), 0);
 551
 552         df = key->tun_flags & TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
 553
 554         iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
 555                       key->tos, key->ttl, df, false);
 556         return;
 557
 558 err_free_rt:
 559         ip_rt_put(rt);
 560 err_free_skb:
 561         kfree_skb(skb);
 562         dev->stats.tx_dropped++;
 563 }
 564
 565 static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev,
 566                            __be16 proto)
 567 {
 568         struct ip_tunnel *tunnel = netdev_priv(dev);
 569         struct ip_tunnel_info *tun_info;
 570         const struct ip_tunnel_key *key;
 571         struct erspan_metadata *md;
 572         struct rtable *rt = NULL;
 573         bool truncate = false;
 574         struct flowi4 fl;
 575         int tunnel_hlen;
 576         int version;
 577         __be16 df;
 578
 579         tun_info = skb_tunnel_info(skb);
 580         if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
 581                      ip_tunnel_info_af(tun_info) != AF_INET))
 582                 goto err_free_skb;
 583
 584         key = &tun_info->key;
 585         md = ip_tunnel_info_opts(tun_info);
 586         if (!md)
 587                 goto err_free_rt;
 588
 589         /* ERSPAN has fixed 8 byte GRE header */
 590         version = md->version;
 591         tunnel_hlen = 8 + erspan_hdr_len(version);
 592
 593         rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
 594         if (!rt)
 595                 return;
 596
 597         if (gre_handle_offloads(skb, false))
 598                 goto err_free_rt;
 599
 600         if (skb->len > dev->mtu + dev->hard_header_len) {
 601                 pskb_trim(skb, dev->mtu + dev->hard_header_len);
 602                 truncate = true;
 603         }
 604
 605         if (version == 1) {
 606                 erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)),
 607                                     ntohl(md->u.index), truncate, true);
 608         } else if (version == 2) {
 609                 erspan_build_header_v2(skb,
 610                                        ntohl(tunnel_id_to_key32(key->tun_id)),
 611                                        md->u.md2.dir,
 612                                        get_hwid(&md->u.md2),
 613                                        truncate, true);
 614         } else {
 615                 goto err_free_rt;
 616         }
 617
 618         gre_build_header(skb, 8, TUNNEL_SEQ,
 619                          htons(ETH_P_ERSPAN), 0, htonl(tunnel->o_seqno++));
 620
 621         df = key->tun_flags & TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
 622
 623         iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
 624                       key->tos, key->ttl, df, false);
 625         return;
 626
 627 err_free_rt:
 628         ip_rt_put(rt);
 629 err_free_skb:
 630         kfree_skb(skb);
 631         dev->stats.tx_dropped++;
 632 }
 633
 634 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
 635 {
 636         struct ip_tunnel_info *info = skb_tunnel_info(skb);
 637         struct rtable *rt;
 638         struct flowi4 fl4;
 639
 640         if (ip_tunnel_info_af(info) != AF_INET)
 641                 return -EINVAL;
 642
 643         rt = gre_get_rt(skb, dev, &fl4, &info->key);
 644         if (IS_ERR(rt))
 645                 return PTR_ERR(rt);
 646
 647         ip_rt_put(rt);
 648         info->key.u.ipv4.src = fl4.saddr;
 649         return 0;
 650 }
 651
 652 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
 653                               struct net_device *dev)
 654 {
 655         struct ip_tunnel *tunnel = netdev_priv(dev);
 656         const struct iphdr *tnl_params;
 657
 658         if (tunnel->collect_md) {
 659                 gre_fb_xmit(skb, dev, skb->protocol);
 660                 return NETDEV_TX_OK;
 661         }
 662
 663         if (dev->header_ops) {
 664                 /* Need space for new headers */
 665                 if (skb_cow_head(skb, dev->needed_headroom -
 666                                       (tunnel->hlen + sizeof(struct iphdr))))
 667                         goto free_skb;
 668
 669                 tnl_params = (const struct iphdr *)skb->data;
 670
 671                 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing
 672                  * to gre header.
 673                  */
 674                 skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
 675                 skb_reset_mac_header(skb);
 676         } else {
 677                 if (skb_cow_head(skb, dev->needed_headroom))
 678                         goto free_skb;
 679
 680                 tnl_params = &tunnel->parms.iph;
 681         }
 682
 683         if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
 684                 goto free_skb;
 685
 686         __gre_xmit(skb, dev, tnl_params, skb->protocol);
 687         return NETDEV_TX_OK;
 688
 689 free_skb:
 690         kfree_skb(skb);
 691         dev->stats.tx_dropped++;
 692         return NETDEV_TX_OK;
 693 }
 694
 695 static netdev_tx_t erspan_xmit(struct sk_buff *skb,
 696                                struct net_device *dev)
 697 {
 698         struct ip_tunnel *tunnel = netdev_priv(dev);
 699         bool truncate = false;
 700
 701         if (tunnel->collect_md) {
 702                 erspan_fb_xmit(skb, dev, skb->protocol);
 703                 return NETDEV_TX_OK;
 704         }
 705
 706         if (gre_handle_offloads(skb, false))
 707                 goto free_skb;
 708
 709         if (skb_cow_head(skb, dev->needed_headroom))
 710                 goto free_skb;
 711
 712         if (skb->len > dev->mtu + dev->hard_header_len) {
 713                 pskb_trim(skb, dev->mtu + dev->hard_header_len);
 714                 truncate = true;
 715         }
 716
 717         /* Push ERSPAN header */
 718         if (tunnel->erspan_ver == 1)
 719                 erspan_build_header(skb, ntohl(tunnel->parms.o_key),
 720                                     tunnel->index,
 721                                     truncate, true);
 722         else
 723                 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key),
 724                                        tunnel->dir, tunnel->hwid,
 725                                        truncate, true);
 726
 727         tunnel->parms.o_flags &= ~TUNNEL_KEY;
 728         __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_ERSPAN));
 729         return NETDEV_TX_OK;
 730
 731 free_skb:
 732         kfree_skb(skb);
 733         dev->stats.tx_dropped++;
 734         return NETDEV_TX_OK;
 735 }
 736
 737 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
 738                                 struct net_device *dev)
 739 {
 740         struct ip_tunnel *tunnel = netdev_priv(dev);
 741
 742         if (tunnel->collect_md) {
 743                 gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
 744                 return NETDEV_TX_OK;
 745         }
 746
 747         if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
 748                 goto free_skb;
 749
 750         if (skb_cow_head(skb, dev->needed_headroom))
 751                 goto free_skb;
 752
 753         __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
 754         return NETDEV_TX_OK;
 755
 756 free_skb:
 757         kfree_skb(skb);
 758         dev->stats.tx_dropped++;
 759         return NETDEV_TX_OK;
 760 }
 761
 762 static void ipgre_link_update(struct net_device *dev, bool set_mtu)
 763 {
 764         struct ip_tunnel *tunnel = netdev_priv(dev);
 765         int len;
 766
 767         len = tunnel->tun_hlen;
 768         tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
 769         len = tunnel->tun_hlen - len;
 770         tunnel->hlen = tunnel->hlen + len;
 771
 772         dev->needed_headroom = dev->needed_headroom + len;
 773         if (set_mtu)
 774                 dev->mtu = max_t(int, dev->mtu - len, 68);
 775
 776         if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
 777                 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
 778                     tunnel->encap.type == TUNNEL_ENCAP_NONE) {
 779                         dev->features |= NETIF_F_GSO_SOFTWARE;
 780                         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
 781                 }
 782                 dev->features |= NETIF_F_LLTX;
 783         }
 784 }
 785
 786 static int ipgre_tunnel_ioctl(struct net_device *dev,
 787                               struct ifreq *ifr, int cmd)
 788 {
 789         struct ip_tunnel_parm p;
 790         int err;
 791
 792         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
 793                 return -EFAULT;
 794
 795         if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
 796                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
 797                     p.iph.ihl != 5 || (p.iph.frag_off & htons(~IP_DF)) ||
 798                     ((p.i_flags | p.o_flags) & (GRE_VERSION | GRE_ROUTING)))
 799                         return -EINVAL;
 800         }
 801
 802         p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
 803         p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
 804
 805         err = ip_tunnel_ioctl(dev, &p, cmd);
 806         if (err)
 807                 return err;
 808
 809         if (cmd == SIOCCHGTUNNEL) {
 810                 struct ip_tunnel *t = netdev_priv(dev);
 811
 812                 t->parms.i_flags = p.i_flags;
 813                 t->parms.o_flags = p.o_flags;
 814
 815                 if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
 816                         ipgre_link_update(dev, true);
 817         }
 818
 819         p.i_flags = gre_tnl_flags_to_gre_flags(p.i_flags);
 820         p.o_flags = gre_tnl_flags_to_gre_flags(p.o_flags);
 821
 822         if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
 823                 return -EFAULT;
 824
 825         return 0;
 826 }
 827
 828 /* Nice toy. Unfortunately, useless in real life :-)
 829    It allows to construct virtual multiprotocol broadcast "LAN"
 830    over the Internet, provided multicast routing is tuned.
 831
 832
 833    I have no idea was this bicycle invented before me,
 834    so that I had to set ARPHRD_IPGRE to a random value.
 835    I have an impression, that Cisco could make something similar,
 836    but this feature is apparently missing in IOS<=11.2(8).
 837
 838    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
 839    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
 840
 841    ping -t 255 224.66.66.66
 842
 843    If nobody answers, mbone does not work.
 844
 845    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
 846    ip addr add 10.66.66.<somewhat>/24 dev Universe
 847    ifconfig Universe up
 848    ifconfig Universe add fe80::<Your_real_addr>/10
 849    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
 850    ftp 10.66.66.66
 851    ...
 852    ftp fec0:6666:6666::193.233.7.65
 853    ...
 854  */
 855 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
 856                         unsigned short type,
 857                         const void *daddr, const void *saddr, unsigned int len)
 858 {
 859         struct ip_tunnel *t = netdev_priv(dev);
 860         struct iphdr *iph;
 861         struct gre_base_hdr *greh;
 862
 863         iph = skb_push(skb, t->hlen + sizeof(*iph));
 864         greh = (struct gre_base_hdr *)(iph+1);
 865         greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
 866         greh->protocol = htons(type);
 867
 868         memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
 869
 870         /* Set the source hardware address. */
 871         if (saddr)
 872                 memcpy(&iph->saddr, saddr, 4);
 873         if (daddr)
 874                 memcpy(&iph->daddr, daddr, 4);
 875         if (iph->daddr)
 876                 return t->hlen + sizeof(*iph);
 877
 878         return -(t->hlen + sizeof(*iph));
 879 }
 880
 881 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
 882 {
 883         const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
 884         memcpy(haddr, &iph->saddr, 4);
 885         return 4;
 886 }
 887
 888 static const struct header_ops ipgre_header_ops = {
 889         .create = ipgre_header,
 890         .parse  = ipgre_header_parse,
 891 };
 892
 893 #ifdef CONFIG_NET_IPGRE_BROADCAST
 894 static int ipgre_open(struct net_device *dev)
 895 {
 896         struct ip_tunnel *t = netdev_priv(dev);
 897
 898         if (ipv4_is_multicast(t->parms.iph.daddr)) {
 899                 struct flowi4 fl4;
 900                 struct rtable *rt;
 901
 902                 rt = ip_route_output_gre(t->net, &fl4,
 903                                          t->parms.iph.daddr,
 904                                          t->parms.iph.saddr,
 905                                          t->parms.o_key,
 906                                          RT_TOS(t->parms.iph.tos),
 907                                          t->parms.link);
 908                 if (IS_ERR(rt))
 909                         return -EADDRNOTAVAIL;
 910                 dev = rt->dst.dev;
 911                 ip_rt_put(rt);
 912                 if (!__in_dev_get_rtnl(dev))
 913                         return -EADDRNOTAVAIL;
 914                 t->mlink = dev->ifindex;
 915                 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
 916         }
 917         return 0;
 918 }
 919
 920 static int ipgre_close(struct net_device *dev)
 921 {
 922         struct ip_tunnel *t = netdev_priv(dev);
 923
 924         if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
 925                 struct in_device *in_dev;
 926                 in_dev = inetdev_by_index(t->net, t->mlink);
 927                 if (in_dev)
 928                         ip_mc_dec_group(in_dev, t->parms.iph.daddr);
 929         }
 930         return 0;
 931 }
 932 #endif
 933
 934 static const struct net_device_ops ipgre_netdev_ops = {
 935         .ndo_init               = ipgre_tunnel_init,
 936         .ndo_uninit             = ip_tunnel_uninit,
 937 #ifdef CONFIG_NET_IPGRE_BROADCAST
 938         .ndo_open               = ipgre_open,
 939         .ndo_stop               = ipgre_close,
 940 #endif
 941         .ndo_start_xmit         = ipgre_xmit,
 942         .ndo_do_ioctl           = ipgre_tunnel_ioctl,
 943         .ndo_change_mtu         = ip_tunnel_change_mtu,
 944         .ndo_get_stats64        = ip_tunnel_get_stats64,
 945         .ndo_get_iflink         = ip_tunnel_get_iflink,
 946 };
 947
 948 #define GRE_FEATURES (NETIF_F_SG |              \
 949                       NETIF_F_FRAGLIST |        \
 950                       NETIF_F_HIGHDMA |         \
 951                       NETIF_F_HW_CSUM)
 952
 953 static void ipgre_tunnel_setup(struct net_device *dev)
 954 {
 955         dev->netdev_ops         = &ipgre_netdev_ops;
 956         dev->type               = ARPHRD_IPGRE;
 957         ip_tunnel_setup(dev, ipgre_net_id);
 958 }
 959
 960 static void __gre_tunnel_init(struct net_device *dev)
 961 {
 962         struct ip_tunnel *tunnel;
 963         int t_hlen;
 964
 965         tunnel = netdev_priv(dev);
 966         tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
 967         tunnel->parms.iph.protocol = IPPROTO_GRE;
 968
 969         tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
 970
 971         t_hlen = tunnel->hlen + sizeof(struct iphdr);
 972
 973         dev->needed_headroom    = LL_MAX_HEADER + t_hlen + 4;
 974         dev->mtu                = ETH_DATA_LEN - t_hlen - 4;
 975
 976         dev->features           |= GRE_FEATURES;
 977         dev->hw_features        |= GRE_FEATURES;
 978
 979         if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
 980                 /* TCP offload with GRE SEQ is not supported, nor
 981                  * can we support 2 levels of outer headers requiring
 982                  * an update.
 983                  */
 984                 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
 985                     (tunnel->encap.type == TUNNEL_ENCAP_NONE)) {
 986                         dev->features    |= NETIF_F_GSO_SOFTWARE;
 987                         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
 988                 }
 989
 990                 /* Can use a lockless transmit, unless we generate
 991                  * output sequences
 992                  */
 993                 dev->features |= NETIF_F_LLTX;
 994         }
 995 }
 996
 997 static int ipgre_tunnel_init(struct net_device *dev)
 998 {
 999         struct ip_tunnel *tunnel = netdev_priv(dev);
1000         struct iphdr *iph = &tunnel->parms.iph;
1001
1002         __gre_tunnel_init(dev);
1003
1004         memcpy(dev->dev_addr, &iph->saddr, 4);
1005         memcpy(dev->broadcast, &iph->daddr, 4);
1006
1007         dev->flags              = IFF_NOARP;
1008         netif_keep_dst(dev);
1009         dev->addr_len           = 4;
1010
1011         if (iph->daddr && !tunnel->collect_md) {
1012 #ifdef CONFIG_NET_IPGRE_BROADCAST
1013                 if (ipv4_is_multicast(iph->daddr)) {
1014                         if (!iph->saddr)
1015                                 return -EINVAL;
1016                         dev->flags = IFF_BROADCAST;
1017                         dev->header_ops = &ipgre_header_ops;
1018                 }
1019 #endif
1020         } else if (!tunnel->collect_md) {
1021                 dev->header_ops = &ipgre_header_ops;
1022         }
1023
1024         return ip_tunnel_init(dev);
1025 }
1026
1027 static const struct gre_protocol ipgre_protocol = {
1028         .handler     = gre_rcv,
1029         .err_handler = gre_err,
1030 };
1031
1032 static int __net_init ipgre_init_net(struct net *net)
1033 {
1034         return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
1035 }
1036
1037 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net)
1038 {
1039         ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops);
1040 }
1041
1042 static struct pernet_operations ipgre_net_ops = {
1043         .init = ipgre_init_net,
1044         .exit_batch = ipgre_exit_batch_net,
1045         .id   = &ipgre_net_id,
1046         .size = sizeof(struct ip_tunnel_net),
1047 };
1048
1049 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1050                                  struct netlink_ext_ack *extack)
1051 {
1052         __be16 flags;
1053
1054         if (!data)
1055                 return 0;
1056
1057         flags = 0;
1058         if (data[IFLA_GRE_IFLAGS])
1059                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1060         if (data[IFLA_GRE_OFLAGS])
1061                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1062         if (flags & (GRE_VERSION|GRE_ROUTING))
1063                 return -EINVAL;
1064
1065         if (data[IFLA_GRE_COLLECT_METADATA] &&
1066             data[IFLA_GRE_ENCAP_TYPE] &&
1067             nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1068                 return -EINVAL;
1069
1070         return 0;
1071 }
1072
1073 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1074                               struct netlink_ext_ack *extack)
1075 {
1076         __be32 daddr;
1077
1078         if (tb[IFLA_ADDRESS]) {
1079                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1080                         return -EINVAL;
1081                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1082                         return -EADDRNOTAVAIL;
1083         }
1084
1085         if (!data)
1086                 goto out;
1087
1088         if (data[IFLA_GRE_REMOTE]) {
1089                 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1090                 if (!daddr)
1091                         return -EINVAL;
1092         }
1093
1094 out:
1095         return ipgre_tunnel_validate(tb, data, extack);
1096 }
1097
1098 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1099                            struct netlink_ext_ack *extack)
1100 {
1101         __be16 flags = 0;
1102         int ret;
1103
1104         if (!data)
1105                 return 0;
1106
1107         ret = ipgre_tap_validate(tb, data, extack);
1108         if (ret)
1109                 return ret;
1110
1111         /* ERSPAN should only have GRE sequence and key flag */
1112         if (data[IFLA_GRE_OFLAGS])
1113                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1114         if (data[IFLA_GRE_IFLAGS])
1115                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1116         if (!data[IFLA_GRE_COLLECT_METADATA] &&
1117             flags != (GRE_SEQ | GRE_KEY))
1118                 return -EINVAL;
1119
1120         /* ERSPAN Session ID only has 10-bit. Since we reuse
1121          * 32-bit key field as ID, check it's range.
1122          */
1123         if (data[IFLA_GRE_IKEY] &&
1124             (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1125                 return -EINVAL;
1126
1127         if (data[IFLA_GRE_OKEY] &&
1128             (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1129                 return -EINVAL;
1130
1131         return 0;
1132 }
1133
1134 static int ipgre_netlink_parms(struct net_device *dev,
1135                                 struct nlattr *data[],
1136                                 struct nlattr *tb[],
1137                                 struct ip_tunnel_parm *parms,
1138                                 __u32 *fwmark)
1139 {
1140         struct ip_tunnel *t = netdev_priv(dev);
1141
1142         memset(parms, 0, sizeof(*parms));
1143
1144         parms->iph.protocol = IPPROTO_GRE;
1145
1146         if (!data)
1147                 return 0;
1148
1149         if (data[IFLA_GRE_LINK])
1150                 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1151
1152         if (data[IFLA_GRE_IFLAGS])
1153                 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
1154
1155         if (data[IFLA_GRE_OFLAGS])
1156                 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
1157
1158         if (data[IFLA_GRE_IKEY])
1159                 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1160
1161         if (data[IFLA_GRE_OKEY])
1162                 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1163
1164         if (data[IFLA_GRE_LOCAL])
1165                 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1166
1167         if (data[IFLA_GRE_REMOTE])
1168                 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1169
1170         if (data[IFLA_GRE_TTL])
1171                 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1172
1173         if (data[IFLA_GRE_TOS])
1174                 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1175
1176         if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1177                 if (t->ignore_df)
1178                         return -EINVAL;
1179                 parms->iph.frag_off = htons(IP_DF);
1180         }
1181
1182         if (data[IFLA_GRE_COLLECT_METADATA]) {
1183                 t->collect_md = true;
1184                 if (dev->type == ARPHRD_IPGRE)
1185                         dev->type = ARPHRD_NONE;
1186         }
1187
1188         if (data[IFLA_GRE_IGNORE_DF]) {
1189                 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1190                   && (parms->iph.frag_off & htons(IP_DF)))
1191                         return -EINVAL;
1192                 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1193         }
1194
1195         if (data[IFLA_GRE_FWMARK])
1196                 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1197
1198         if (data[IFLA_GRE_ERSPAN_VER]) {
1199                 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]);
1200
1201                 if (t->erspan_ver != 1 && t->erspan_ver != 2)
1202                         return -EINVAL;
1203         }
1204
1205         if (t->erspan_ver == 1) {
1206                 if (data[IFLA_GRE_ERSPAN_INDEX]) {
1207                         t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1208                         if (t->index & ~INDEX_MASK)
1209                                 return -EINVAL;
1210                 }
1211         } else if (t->erspan_ver == 2) {
1212                 if (data[IFLA_GRE_ERSPAN_DIR]) {
1213                         t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]);
1214                         if (t->dir & ~(DIR_MASK >> DIR_OFFSET))
1215                                 return -EINVAL;
1216                 }
1217                 if (data[IFLA_GRE_ERSPAN_HWID]) {
1218                         t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]);
1219                         if (t->hwid & ~(HWID_MASK >> HWID_OFFSET))
1220                                 return -EINVAL;
1221                 }
1222         }
1223
1224         return 0;
1225 }
1226
1227 /* This function returns true when ENCAP attributes are present in the nl msg */
1228 static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1229                                       struct ip_tunnel_encap *ipencap)
1230 {
1231         bool ret = false;
1232
1233         memset(ipencap, 0, sizeof(*ipencap));
1234
1235         if (!data)
1236                 return ret;
1237
1238         if (data[IFLA_GRE_ENCAP_TYPE]) {
1239                 ret = true;
1240                 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1241         }
1242
1243         if (data[IFLA_GRE_ENCAP_FLAGS]) {
1244                 ret = true;
1245                 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1246         }
1247
1248         if (data[IFLA_GRE_ENCAP_SPORT]) {
1249                 ret = true;
1250                 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1251         }
1252
1253         if (data[IFLA_GRE_ENCAP_DPORT]) {
1254                 ret = true;
1255                 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1256         }
1257
1258         return ret;
1259 }
1260
1261 static int gre_tap_init(struct net_device *dev)
1262 {
1263         __gre_tunnel_init(dev);
1264         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1265         netif_keep_dst(dev);
1266
1267         return ip_tunnel_init(dev);
1268 }
1269
1270 static const struct net_device_ops gre_tap_netdev_ops = {
1271         .ndo_init               = gre_tap_init,
1272         .ndo_uninit             = ip_tunnel_uninit,
1273         .ndo_start_xmit         = gre_tap_xmit,
1274         .ndo_set_mac_address    = eth_mac_addr,
1275         .ndo_validate_addr      = eth_validate_addr,
1276         .ndo_change_mtu         = ip_tunnel_change_mtu,
1277         .ndo_get_stats64        = ip_tunnel_get_stats64,
1278         .ndo_get_iflink         = ip_tunnel_get_iflink,
1279         .ndo_fill_metadata_dst  = gre_fill_metadata_dst,
1280 };
1281
1282 static int erspan_tunnel_init(struct net_device *dev)
1283 {
1284         struct ip_tunnel *tunnel = netdev_priv(dev);
1285         int t_hlen;
1286
1287         tunnel->tun_hlen = 8;
1288         tunnel->parms.iph.protocol = IPPROTO_GRE;
1289         tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1290                        erspan_hdr_len(tunnel->erspan_ver);
1291         t_hlen = tunnel->hlen + sizeof(struct iphdr);
1292
1293         dev->needed_headroom = LL_MAX_HEADER + t_hlen + 4;
1294         dev->mtu = ETH_DATA_LEN - t_hlen - 4;
1295         dev->features           |= GRE_FEATURES;
1296         dev->hw_features        |= GRE_FEATURES;
1297         dev->priv_flags         |= IFF_LIVE_ADDR_CHANGE;
1298         netif_keep_dst(dev);
1299
1300         return ip_tunnel_init(dev);
1301 }
1302
1303 static const struct net_device_ops erspan_netdev_ops = {
1304         .ndo_init               = erspan_tunnel_init,
1305         .ndo_uninit             = ip_tunnel_uninit,
1306         .ndo_start_xmit         = erspan_xmit,
1307         .ndo_set_mac_address    = eth_mac_addr,
1308         .ndo_validate_addr      = eth_validate_addr,
1309         .ndo_change_mtu         = ip_tunnel_change_mtu,
1310         .ndo_get_stats64        = ip_tunnel_get_stats64,
1311         .ndo_get_iflink         = ip_tunnel_get_iflink,
1312         .ndo_fill_metadata_dst  = gre_fill_metadata_dst,
1313 };
1314
1315 static void ipgre_tap_setup(struct net_device *dev)
1316 {
1317         ether_setup(dev);
1318         dev->max_mtu = 0;
1319         dev->netdev_ops = &gre_tap_netdev_ops;
1320         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1321         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1322         ip_tunnel_setup(dev, gre_tap_net_id);
1323 }
1324
1325 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1326                          struct nlattr *tb[], struct nlattr *data[],
1327                          struct netlink_ext_ack *extack)
1328 {
1329         struct ip_tunnel_parm p;
1330         struct ip_tunnel_encap ipencap;
1331         __u32 fwmark = 0;
1332         int err;
1333
1334         if (ipgre_netlink_encap_parms(data, &ipencap)) {
1335                 struct ip_tunnel *t = netdev_priv(dev);
1336                 err = ip_tunnel_encap_setup(t, &ipencap);
1337
1338                 if (err < 0)
1339                         return err;
1340         }
1341
1342         err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1343         if (err < 0)
1344                 return err;
1345         return ip_tunnel_newlink(dev, tb, &p, fwmark);
1346 }
1347
1348 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1349                             struct nlattr *data[],
1350                             struct netlink_ext_ack *extack)
1351 {
1352         struct ip_tunnel *t = netdev_priv(dev);
1353         struct ip_tunnel_encap ipencap;
1354         __u32 fwmark = t->fwmark;
1355         struct ip_tunnel_parm p;
1356         int err;
1357
1358         if (ipgre_netlink_encap_parms(data, &ipencap)) {
1359                 err = ip_tunnel_encap_setup(t, &ipencap);
1360
1361                 if (err < 0)
1362                         return err;
1363         }
1364
1365         err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1366         if (err < 0)
1367                 return err;
1368
1369         err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1370         if (err < 0)
1371                 return err;
1372
1373         t->parms.i_flags = p.i_flags;
1374         t->parms.o_flags = p.o_flags;
1375
1376         if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
1377                 ipgre_link_update(dev, !tb[IFLA_MTU]);
1378
1379         return 0;
1380 }
1381
1382 static size_t ipgre_get_size(const struct net_device *dev)
1383 {
1384         return
1385                 /* IFLA_GRE_LINK */
1386                 nla_total_size(4) +
1387                 /* IFLA_GRE_IFLAGS */
1388                 nla_total_size(2) +
1389                 /* IFLA_GRE_OFLAGS */
1390                 nla_total_size(2) +
1391                 /* IFLA_GRE_IKEY */
1392                 nla_total_size(4) +
1393                 /* IFLA_GRE_OKEY */
1394                 nla_total_size(4) +
1395                 /* IFLA_GRE_LOCAL */
1396                 nla_total_size(4) +
1397                 /* IFLA_GRE_REMOTE */
1398                 nla_total_size(4) +
1399                 /* IFLA_GRE_TTL */
1400                 nla_total_size(1) +
1401                 /* IFLA_GRE_TOS */
1402                 nla_total_size(1) +
1403                 /* IFLA_GRE_PMTUDISC */
1404                 nla_total_size(1) +
1405                 /* IFLA_GRE_ENCAP_TYPE */
1406                 nla_total_size(2) +
1407                 /* IFLA_GRE_ENCAP_FLAGS */
1408                 nla_total_size(2) +
1409                 /* IFLA_GRE_ENCAP_SPORT */
1410                 nla_total_size(2) +
1411                 /* IFLA_GRE_ENCAP_DPORT */
1412                 nla_total_size(2) +
1413                 /* IFLA_GRE_COLLECT_METADATA */
1414                 nla_total_size(0) +
1415                 /* IFLA_GRE_IGNORE_DF */
1416                 nla_total_size(1) +
1417                 /* IFLA_GRE_FWMARK */
1418                 nla_total_size(4) +
1419                 /* IFLA_GRE_ERSPAN_INDEX */
1420                 nla_total_size(4) +
1421                 /* IFLA_GRE_ERSPAN_VER */
1422                 nla_total_size(1) +
1423                 /* IFLA_GRE_ERSPAN_DIR */
1424                 nla_total_size(1) +
1425                 /* IFLA_GRE_ERSPAN_HWID */
1426                 nla_total_size(2) +
1427                 0;
1428 }
1429
1430 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1431 {
1432         struct ip_tunnel *t = netdev_priv(dev);
1433         struct ip_tunnel_parm *p = &t->parms;
1434
1435         if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1436             nla_put_be16(skb, IFLA_GRE_IFLAGS,
1437                          gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1438             nla_put_be16(skb, IFLA_GRE_OFLAGS,
1439                          gre_tnl_flags_to_gre_flags(p->o_flags)) ||
1440             nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1441             nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1442             nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1443             nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1444             nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1445             nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1446             nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1447                        !!(p->iph.frag_off & htons(IP_DF))) ||
1448             nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1449                 goto nla_put_failure;
1450
1451         if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1452                         t->encap.type) ||
1453             nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1454                          t->encap.sport) ||
1455             nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1456                          t->encap.dport) ||
1457             nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1458                         t->encap.flags))
1459                 goto nla_put_failure;
1460
1461         if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1462                 goto nla_put_failure;
1463
1464         if (t->collect_md) {
1465                 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1466                         goto nla_put_failure;
1467         }
1468
1469         if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1470                 goto nla_put_failure;
1471
1472         if (t->erspan_ver == 1) {
1473                 if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1474                         goto nla_put_failure;
1475         } else if (t->erspan_ver == 2) {
1476                 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1477                         goto nla_put_failure;
1478                 if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1479                         goto nla_put_failure;
1480         }
1481
1482         return 0;
1483
1484 nla_put_failure:
1485         return -EMSGSIZE;
1486 }
1487
1488 static void erspan_setup(struct net_device *dev)
1489 {
1490         ether_setup(dev);
1491         dev->netdev_ops = &erspan_netdev_ops;
1492         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1493         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1494         ip_tunnel_setup(dev, erspan_net_id);
1495 }
1496
1497 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1498         [IFLA_GRE_LINK]         = { .type = NLA_U32 },
1499         [IFLA_GRE_IFLAGS]       = { .type = NLA_U16 },
1500         [IFLA_GRE_OFLAGS]       = { .type = NLA_U16 },
1501         [IFLA_GRE_IKEY]         = { .type = NLA_U32 },
1502         [IFLA_GRE_OKEY]         = { .type = NLA_U32 },
1503         [IFLA_GRE_LOCAL]        = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1504         [IFLA_GRE_REMOTE]       = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1505         [IFLA_GRE_TTL]          = { .type = NLA_U8 },
1506         [IFLA_GRE_TOS]          = { .type = NLA_U8 },
1507         [IFLA_GRE_PMTUDISC]     = { .type = NLA_U8 },
1508         [IFLA_GRE_ENCAP_TYPE]   = { .type = NLA_U16 },
1509         [IFLA_GRE_ENCAP_FLAGS]  = { .type = NLA_U16 },
1510         [IFLA_GRE_ENCAP_SPORT]  = { .type = NLA_U16 },
1511         [IFLA_GRE_ENCAP_DPORT]  = { .type = NLA_U16 },
1512         [IFLA_GRE_COLLECT_METADATA]     = { .type = NLA_FLAG },
1513         [IFLA_GRE_IGNORE_DF]    = { .type = NLA_U8 },
1514         [IFLA_GRE_FWMARK]       = { .type = NLA_U32 },
1515         [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 },
1516         [IFLA_GRE_ERSPAN_VER]   = { .type = NLA_U8 },
1517         [IFLA_GRE_ERSPAN_DIR]   = { .type = NLA_U8 },
1518         [IFLA_GRE_ERSPAN_HWID]  = { .type = NLA_U16 },
1519 };
1520
1521 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1522         .kind           = "gre",
1523         .maxtype        = IFLA_GRE_MAX,
1524         .policy         = ipgre_policy,
1525         .priv_size      = sizeof(struct ip_tunnel),
1526         .setup          = ipgre_tunnel_setup,
1527         .validate       = ipgre_tunnel_validate,
1528         .newlink        = ipgre_newlink,
1529         .changelink     = ipgre_changelink,
1530         .dellink        = ip_tunnel_dellink,
1531         .get_size       = ipgre_get_size,
1532         .fill_info      = ipgre_fill_info,
1533         .get_link_net   = ip_tunnel_get_link_net,
1534 };
1535
1536 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1537         .kind           = "gretap",
1538         .maxtype        = IFLA_GRE_MAX,
1539         .policy         = ipgre_policy,
1540         .priv_size      = sizeof(struct ip_tunnel),
1541         .setup          = ipgre_tap_setup,
1542         .validate       = ipgre_tap_validate,
1543         .newlink        = ipgre_newlink,
1544         .changelink     = ipgre_changelink,
1545         .dellink        = ip_tunnel_dellink,
1546         .get_size       = ipgre_get_size,
1547         .fill_info      = ipgre_fill_info,
1548         .get_link_net   = ip_tunnel_get_link_net,
1549 };
1550
1551 static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1552         .kind           = "erspan",
1553         .maxtype        = IFLA_GRE_MAX,
1554         .policy         = ipgre_policy,
1555         .priv_size      = sizeof(struct ip_tunnel),
1556         .setup          = erspan_setup,
1557         .validate       = erspan_validate,
1558         .newlink        = ipgre_newlink,
1559         .changelink     = ipgre_changelink,
1560         .dellink        = ip_tunnel_dellink,
1561         .get_size       = ipgre_get_size,
1562         .fill_info      = ipgre_fill_info,
1563         .get_link_net   = ip_tunnel_get_link_net,
1564 };
1565
1566 struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1567                                         u8 name_assign_type)
1568 {
1569         struct nlattr *tb[IFLA_MAX + 1];
1570         struct net_device *dev;
1571         LIST_HEAD(list_kill);
1572         struct ip_tunnel *t;
1573         int err;
1574
1575         memset(&tb, 0, sizeof(tb));
1576
1577         dev = rtnl_create_link(net, name, name_assign_type,
1578                                &ipgre_tap_ops, tb);
1579         if (IS_ERR(dev))
1580                 return dev;
1581
1582         /* Configure flow based GRE device. */
1583         t = netdev_priv(dev);
1584         t->collect_md = true;
1585
1586         err = ipgre_newlink(net, dev, tb, NULL, NULL);
1587         if (err < 0) {
1588                 free_netdev(dev);
1589                 return ERR_PTR(err);
1590         }
1591
1592         /* openvswitch users expect packet sizes to be unrestricted,
1593          * so set the largest MTU we can.
1594          */
1595         err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1596         if (err)
1597                 goto out;
1598
1599         err = rtnl_configure_link(dev, NULL);
1600         if (err < 0)
1601                 goto out;
1602
1603         return dev;
1604 out:
1605         ip_tunnel_dellink(dev, &list_kill);
1606         unregister_netdevice_many(&list_kill);
1607         return ERR_PTR(err);
1608 }
1609 EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1610
1611 static int __net_init ipgre_tap_init_net(struct net *net)
1612 {
1613         return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1614 }
1615
1616 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net)
1617 {
1618         ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops);
1619 }
1620
1621 static struct pernet_operations ipgre_tap_net_ops = {
1622         .init = ipgre_tap_init_net,
1623         .exit_batch = ipgre_tap_exit_batch_net,
1624         .id   = &gre_tap_net_id,
1625         .size = sizeof(struct ip_tunnel_net),
1626 };
1627
1628 static int __net_init erspan_init_net(struct net *net)
1629 {
1630         return ip_tunnel_init_net(net, erspan_net_id,
1631                                   &erspan_link_ops, "erspan0");
1632 }
1633
1634 static void __net_exit erspan_exit_batch_net(struct list_head *net_list)
1635 {
1636         ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops);
1637 }
1638
1639 static struct pernet_operations erspan_net_ops = {
1640         .init = erspan_init_net,
1641         .exit_batch = erspan_exit_batch_net,
1642         .id   = &erspan_net_id,
1643         .size = sizeof(struct ip_tunnel_net),
1644 };
1645
1646 static int __init ipgre_init(void)
1647 {
1648         int err;
1649
1650         pr_info("GRE over IPv4 tunneling driver\n");
1651
1652         err = register_pernet_device(&ipgre_net_ops);
1653         if (err < 0)
1654                 return err;
1655
1656         err = register_pernet_device(&ipgre_tap_net_ops);
1657         if (err < 0)
1658                 goto pnet_tap_failed;
1659
1660         err = register_pernet_device(&erspan_net_ops);
1661         if (err < 0)
1662                 goto pnet_erspan_failed;
1663
1664         err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1665         if (err < 0) {
1666                 pr_info("%s: can't add protocol\n", __func__);
1667                 goto add_proto_failed;
1668         }
1669
1670         err = rtnl_link_register(&ipgre_link_ops);
1671         if (err < 0)
1672                 goto rtnl_link_failed;
1673
1674         err = rtnl_link_register(&ipgre_tap_ops);
1675         if (err < 0)
1676                 goto tap_ops_failed;
1677
1678         err = rtnl_link_register(&erspan_link_ops);
1679         if (err < 0)
1680                 goto erspan_link_failed;
1681
1682         return 0;
1683
1684 erspan_link_failed:
1685         rtnl_link_unregister(&ipgre_tap_ops);
1686 tap_ops_failed:
1687         rtnl_link_unregister(&ipgre_link_ops);
1688 rtnl_link_failed:
1689         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1690 add_proto_failed:
1691         unregister_pernet_device(&erspan_net_ops);
1692 pnet_erspan_failed:
1693         unregister_pernet_device(&ipgre_tap_net_ops);
1694 pnet_tap_failed:
1695         unregister_pernet_device(&ipgre_net_ops);
1696         return err;
1697 }
1698
1699 static void __exit ipgre_fini(void)
1700 {
1701         rtnl_link_unregister(&ipgre_tap_ops);
1702         rtnl_link_unregister(&ipgre_link_ops);
1703         rtnl_link_unregister(&erspan_link_ops);
1704         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1705         unregister_pernet_device(&ipgre_tap_net_ops);
1706         unregister_pernet_device(&ipgre_net_ops);
1707         unregister_pernet_device(&erspan_net_ops);
1708 }
1709
1710 module_init(ipgre_init);
1711 module_exit(ipgre_fini);
1712 MODULE_LICENSE("GPL");
1713 MODULE_ALIAS_RTNL_LINK("gre");
1714 MODULE_ALIAS_RTNL_LINK("gretap");
1715 MODULE_ALIAS_RTNL_LINK("erspan");
1716 MODULE_ALIAS_NETDEV("gre0");
1717 MODULE_ALIAS_NETDEV("gretap0");
1718 MODULE_ALIAS_NETDEV("erspan0");