net/ipv4/ip_gre.c

   1 /*
   2  *      Linux NET3:     GRE over IP protocol decoder.
   3  *
   4  *      Authors: Alexey Kuznetsov (kuznet@ms2.inr.ac.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/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                                 __be32 id, u32 index, bool truncate);
 118
 119 static unsigned int ipgre_net_id __read_mostly;
 120 static unsigned int gre_tap_net_id __read_mostly;
 121 static unsigned int erspan_net_id __read_mostly;
 122
 123 static void ipgre_err(struct sk_buff *skb, u32 info,
 124                       const struct tnl_ptk_info *tpi)
 125 {
 126
 127         /* All the routers (except for Linux) return only
 128            8 bytes of packet payload. It means, that precise relaying of
 129            ICMP in the real Internet is absolutely infeasible.
 130
 131            Moreover, Cisco "wise men" put GRE key to the third word
 132            in GRE header. It makes impossible maintaining even soft
 133            state for keyed GRE tunnels with enabled checksum. Tell
 134            them "thank you".
 135
 136            Well, I wonder, rfc1812 was written by Cisco employee,
 137            what the hell these idiots break standards established
 138            by themselves???
 139            */
 140         struct net *net = dev_net(skb->dev);
 141         struct ip_tunnel_net *itn;
 142         const struct iphdr *iph;
 143         const int type = icmp_hdr(skb)->type;
 144         const int code = icmp_hdr(skb)->code;
 145         unsigned int data_len = 0;
 146         struct ip_tunnel *t;
 147
 148         switch (type) {
 149         default:
 150         case ICMP_PARAMETERPROB:
 151                 return;
 152
 153         case ICMP_DEST_UNREACH:
 154                 switch (code) {
 155                 case ICMP_SR_FAILED:
 156                 case ICMP_PORT_UNREACH:
 157                         /* Impossible event. */
 158                         return;
 159                 default:
 160                         /* All others are translated to HOST_UNREACH.
 161                            rfc2003 contains "deep thoughts" about NET_UNREACH,
 162                            I believe they are just ether pollution. --ANK
 163                          */
 164                         break;
 165                 }
 166                 break;
 167
 168         case ICMP_TIME_EXCEEDED:
 169                 if (code != ICMP_EXC_TTL)
 170                         return;
 171                 data_len = icmp_hdr(skb)->un.reserved[1] * 4; /* RFC 4884 4.1 */
 172                 break;
 173
 174         case ICMP_REDIRECT:
 175                 break;
 176         }
 177
 178         if (tpi->proto == htons(ETH_P_TEB))
 179                 itn = net_generic(net, gre_tap_net_id);
 180         else
 181                 itn = net_generic(net, ipgre_net_id);
 182
 183         iph = (const struct iphdr *)(icmp_hdr(skb) + 1);
 184         t = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
 185                              iph->daddr, iph->saddr, tpi->key);
 186
 187         if (!t)
 188                 return;
 189
 190 #if IS_ENABLED(CONFIG_IPV6)
 191        if (tpi->proto == htons(ETH_P_IPV6) &&
 192            !ip6_err_gen_icmpv6_unreach(skb, iph->ihl * 4 + tpi->hdr_len,
 193                                        type, data_len))
 194                return;
 195 #endif
 196
 197         if (t->parms.iph.daddr == 0 ||
 198             ipv4_is_multicast(t->parms.iph.daddr))
 199                 return;
 200
 201         if (t->parms.iph.ttl == 0 && type == ICMP_TIME_EXCEEDED)
 202                 return;
 203
 204         if (time_before(jiffies, t->err_time + IPTUNNEL_ERR_TIMEO))
 205                 t->err_count++;
 206         else
 207                 t->err_count = 1;
 208         t->err_time = jiffies;
 209 }
 210
 211 static void gre_err(struct sk_buff *skb, u32 info)
 212 {
 213         /* All the routers (except for Linux) return only
 214          * 8 bytes of packet payload. It means, that precise relaying of
 215          * ICMP in the real Internet is absolutely infeasible.
 216          *
 217          * Moreover, Cisco "wise men" put GRE key to the third word
 218          * in GRE header. It makes impossible maintaining even soft
 219          * state for keyed
 220          * GRE tunnels with enabled checksum. Tell them "thank you".
 221          *
 222          * Well, I wonder, rfc1812 was written by Cisco employee,
 223          * what the hell these idiots break standards established
 224          * by themselves???
 225          */
 226
 227         const struct iphdr *iph = (struct iphdr *)skb->data;
 228         const int type = icmp_hdr(skb)->type;
 229         const int code = icmp_hdr(skb)->code;
 230         struct tnl_ptk_info tpi;
 231         bool csum_err = false;
 232
 233         if (gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP),
 234                              iph->ihl * 4) < 0) {
 235                 if (!csum_err)          /* ignore csum errors. */
 236                         return;
 237         }
 238
 239         if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED) {
 240                 ipv4_update_pmtu(skb, dev_net(skb->dev), info,
 241                                  skb->dev->ifindex, 0, IPPROTO_GRE, 0);
 242                 return;
 243         }
 244         if (type == ICMP_REDIRECT) {
 245                 ipv4_redirect(skb, dev_net(skb->dev), skb->dev->ifindex, 0,
 246                               IPPROTO_GRE, 0);
 247                 return;
 248         }
 249
 250         ipgre_err(skb, info, &tpi);
 251 }
 252
 253 static int erspan_rcv(struct sk_buff *skb, struct tnl_ptk_info *tpi,
 254                       int gre_hdr_len)
 255 {
 256         struct net *net = dev_net(skb->dev);
 257         struct metadata_dst *tun_dst = NULL;
 258         struct ip_tunnel_net *itn;
 259         struct ip_tunnel *tunnel;
 260         struct erspanhdr *ershdr;
 261         const struct iphdr *iph;
 262         __be32 index;
 263         int len;
 264
 265         itn = net_generic(net, erspan_net_id);
 266         len = gre_hdr_len + sizeof(*ershdr);
 267
 268         if (unlikely(!pskb_may_pull(skb, len)))
 269                 return -ENOMEM;
 270
 271         iph = ip_hdr(skb);
 272         ershdr = (struct erspanhdr *)(skb->data + gre_hdr_len);
 273
 274         /* The original GRE header does not have key field,
 275          * Use ERSPAN 10-bit session ID as key.
 276          */
 277         tpi->key = cpu_to_be32(ntohs(ershdr->session_id) & ID_MASK);
 278         index = ershdr->md.index;
 279         tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex,
 280                                   tpi->flags | TUNNEL_KEY,
 281                                   iph->saddr, iph->daddr, tpi->key);
 282
 283         if (tunnel) {
 284                 if (__iptunnel_pull_header(skb,
 285                                            gre_hdr_len + sizeof(*ershdr),
 286                                            htons(ETH_P_TEB),
 287                                            false, false) < 0)
 288                         goto drop;
 289
 290                 if (tunnel->collect_md) {
 291                         struct ip_tunnel_info *info;
 292                         struct erspan_metadata *md;
 293                         __be64 tun_id;
 294                         __be16 flags;
 295
 296                         tpi->flags |= TUNNEL_KEY;
 297                         flags = tpi->flags;
 298                         tun_id = key32_to_tunnel_id(tpi->key);
 299
 300                         tun_dst = ip_tun_rx_dst(skb, flags,
 301                                                 tun_id, sizeof(*md));
 302                         if (!tun_dst)
 303                                 return PACKET_REJECT;
 304
 305                         md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
 306                         if (!md)
 307                                 return PACKET_REJECT;
 308
 309                         md->index = index;
 310                         info = &tun_dst->u.tun_info;
 311                         info->key.tun_flags |= TUNNEL_ERSPAN_OPT;
 312                         info->options_len = sizeof(*md);
 313                 } else {
 314                         tunnel->index = ntohl(index);
 315                 }
 316
 317                 skb_reset_mac_header(skb);
 318                 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
 319                 return PACKET_RCVD;
 320         }
 321 drop:
 322         kfree_skb(skb);
 323         return PACKET_RCVD;
 324 }
 325
 326 static int __ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
 327                        struct ip_tunnel_net *itn, int hdr_len, bool raw_proto)
 328 {
 329         struct metadata_dst *tun_dst = NULL;
 330         const struct iphdr *iph;
 331         struct ip_tunnel *tunnel;
 332
 333         iph = ip_hdr(skb);
 334         tunnel = ip_tunnel_lookup(itn, skb->dev->ifindex, tpi->flags,
 335                                   iph->saddr, iph->daddr, tpi->key);
 336
 337         if (tunnel) {
 338                 if (__iptunnel_pull_header(skb, hdr_len, tpi->proto,
 339                                            raw_proto, false) < 0)
 340                         goto drop;
 341
 342                 if (tunnel->dev->type != ARPHRD_NONE)
 343                         skb_pop_mac_header(skb);
 344                 else
 345                         skb_reset_mac_header(skb);
 346                 if (tunnel->collect_md) {
 347                         __be16 flags;
 348                         __be64 tun_id;
 349
 350                         flags = tpi->flags & (TUNNEL_CSUM | TUNNEL_KEY);
 351                         tun_id = key32_to_tunnel_id(tpi->key);
 352                         tun_dst = ip_tun_rx_dst(skb, flags, tun_id, 0);
 353                         if (!tun_dst)
 354                                 return PACKET_REJECT;
 355                 }
 356
 357                 ip_tunnel_rcv(tunnel, skb, tpi, tun_dst, log_ecn_error);
 358                 return PACKET_RCVD;
 359         }
 360         return PACKET_NEXT;
 361
 362 drop:
 363         kfree_skb(skb);
 364         return PACKET_RCVD;
 365 }
 366
 367 static int ipgre_rcv(struct sk_buff *skb, const struct tnl_ptk_info *tpi,
 368                      int hdr_len)
 369 {
 370         struct net *net = dev_net(skb->dev);
 371         struct ip_tunnel_net *itn;
 372         int res;
 373
 374         if (tpi->proto == htons(ETH_P_TEB))
 375                 itn = net_generic(net, gre_tap_net_id);
 376         else
 377                 itn = net_generic(net, ipgre_net_id);
 378
 379         res = __ipgre_rcv(skb, tpi, itn, hdr_len, false);
 380         if (res == PACKET_NEXT && tpi->proto == htons(ETH_P_TEB)) {
 381                 /* ipgre tunnels in collect metadata mode should receive
 382                  * also ETH_P_TEB traffic.
 383                  */
 384                 itn = net_generic(net, ipgre_net_id);
 385                 res = __ipgre_rcv(skb, tpi, itn, hdr_len, true);
 386         }
 387         return res;
 388 }
 389
 390 static int gre_rcv(struct sk_buff *skb)
 391 {
 392         struct tnl_ptk_info tpi;
 393         bool csum_err = false;
 394         int hdr_len;
 395
 396 #ifdef CONFIG_NET_IPGRE_BROADCAST
 397         if (ipv4_is_multicast(ip_hdr(skb)->daddr)) {
 398                 /* Looped back packet, drop it! */
 399                 if (rt_is_output_route(skb_rtable(skb)))
 400                         goto drop;
 401         }
 402 #endif
 403
 404         hdr_len = gre_parse_header(skb, &tpi, &csum_err, htons(ETH_P_IP), 0);
 405         if (hdr_len < 0)
 406                 goto drop;
 407
 408         if (unlikely(tpi.proto == htons(ETH_P_ERSPAN))) {
 409                 if (erspan_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
 410                         return 0;
 411         }
 412
 413         if (ipgre_rcv(skb, &tpi, hdr_len) == PACKET_RCVD)
 414                 return 0;
 415
 416         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
 417 drop:
 418         kfree_skb(skb);
 419         return 0;
 420 }
 421
 422 static void __gre_xmit(struct sk_buff *skb, struct net_device *dev,
 423                        const struct iphdr *tnl_params,
 424                        __be16 proto)
 425 {
 426         struct ip_tunnel *tunnel = netdev_priv(dev);
 427
 428         if (tunnel->parms.o_flags & TUNNEL_SEQ)
 429                 tunnel->o_seqno++;
 430
 431         /* Push GRE header. */
 432         gre_build_header(skb, tunnel->tun_hlen,
 433                          tunnel->parms.o_flags, proto, tunnel->parms.o_key,
 434                          htonl(tunnel->o_seqno));
 435
 436         ip_tunnel_xmit(skb, dev, tnl_params, tnl_params->protocol);
 437 }
 438
 439 static int gre_handle_offloads(struct sk_buff *skb, bool csum)
 440 {
 441         return iptunnel_handle_offloads(skb, csum ? SKB_GSO_GRE_CSUM : SKB_GSO_GRE);
 442 }
 443
 444 static struct rtable *gre_get_rt(struct sk_buff *skb,
 445                                  struct net_device *dev,
 446                                  struct flowi4 *fl,
 447                                  const struct ip_tunnel_key *key)
 448 {
 449         struct net *net = dev_net(dev);
 450
 451         memset(fl, 0, sizeof(*fl));
 452         fl->daddr = key->u.ipv4.dst;
 453         fl->saddr = key->u.ipv4.src;
 454         fl->flowi4_tos = RT_TOS(key->tos);
 455         fl->flowi4_mark = skb->mark;
 456         fl->flowi4_proto = IPPROTO_GRE;
 457
 458         return ip_route_output_key(net, fl);
 459 }
 460
 461 static struct rtable *prepare_fb_xmit(struct sk_buff *skb,
 462                                       struct net_device *dev,
 463                                       struct flowi4 *fl,
 464                                       int tunnel_hlen)
 465 {
 466         struct ip_tunnel_info *tun_info;
 467         const struct ip_tunnel_key *key;
 468         struct rtable *rt = NULL;
 469         int min_headroom;
 470         bool use_cache;
 471         int err;
 472
 473         tun_info = skb_tunnel_info(skb);
 474         key = &tun_info->key;
 475         use_cache = ip_tunnel_dst_cache_usable(skb, tun_info);
 476
 477         if (use_cache)
 478                 rt = dst_cache_get_ip4(&tun_info->dst_cache, &fl->saddr);
 479         if (!rt) {
 480                 rt = gre_get_rt(skb, dev, fl, key);
 481                 if (IS_ERR(rt))
 482                         goto err_free_skb;
 483                 if (use_cache)
 484                         dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst,
 485                                           fl->saddr);
 486         }
 487
 488         min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
 489                         + tunnel_hlen + sizeof(struct iphdr);
 490         if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
 491                 int head_delta = SKB_DATA_ALIGN(min_headroom -
 492                                                 skb_headroom(skb) +
 493                                                 16);
 494                 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
 495                                        0, GFP_ATOMIC);
 496                 if (unlikely(err))
 497                         goto err_free_rt;
 498         }
 499         return rt;
 500
 501 err_free_rt:
 502         ip_rt_put(rt);
 503 err_free_skb:
 504         kfree_skb(skb);
 505         dev->stats.tx_dropped++;
 506         return NULL;
 507 }
 508
 509 static void gre_fb_xmit(struct sk_buff *skb, struct net_device *dev,
 510                         __be16 proto)
 511 {
 512         struct ip_tunnel_info *tun_info;
 513         const struct ip_tunnel_key *key;
 514         struct rtable *rt = NULL;
 515         struct flowi4 fl;
 516         int tunnel_hlen;
 517         __be16 df, flags;
 518
 519         tun_info = skb_tunnel_info(skb);
 520         if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
 521                      ip_tunnel_info_af(tun_info) != AF_INET))
 522                 goto err_free_skb;
 523
 524         key = &tun_info->key;
 525         tunnel_hlen = gre_calc_hlen(key->tun_flags);
 526
 527         rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
 528         if (!rt)
 529                 return;
 530
 531         /* Push Tunnel header. */
 532         if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM)))
 533                 goto err_free_rt;
 534
 535         flags = tun_info->key.tun_flags & (TUNNEL_CSUM | TUNNEL_KEY);
 536         gre_build_header(skb, tunnel_hlen, flags, proto,
 537                          tunnel_id_to_key32(tun_info->key.tun_id), 0);
 538
 539         df = key->tun_flags & TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
 540
 541         iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
 542                       key->tos, key->ttl, df, false);
 543         return;
 544
 545 err_free_rt:
 546         ip_rt_put(rt);
 547 err_free_skb:
 548         kfree_skb(skb);
 549         dev->stats.tx_dropped++;
 550 }
 551
 552 static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev,
 553                            __be16 proto)
 554 {
 555         struct ip_tunnel *tunnel = netdev_priv(dev);
 556         struct ip_tunnel_info *tun_info;
 557         const struct ip_tunnel_key *key;
 558         struct erspan_metadata *md;
 559         struct rtable *rt = NULL;
 560         bool truncate = false;
 561         struct flowi4 fl;
 562         int tunnel_hlen;
 563         __be16 df;
 564
 565         tun_info = skb_tunnel_info(skb);
 566         if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
 567                      ip_tunnel_info_af(tun_info) != AF_INET))
 568                 goto err_free_skb;
 569
 570         key = &tun_info->key;
 571
 572         /* ERSPAN has fixed 8 byte GRE header */
 573         tunnel_hlen = 8 + sizeof(struct erspanhdr);
 574
 575         rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
 576         if (!rt)
 577                 return;
 578
 579         if (gre_handle_offloads(skb, false))
 580                 goto err_free_rt;
 581
 582         if (skb->len > dev->mtu) {
 583                 pskb_trim(skb, dev->mtu);
 584                 truncate = true;
 585         }
 586
 587         md = ip_tunnel_info_opts(tun_info);
 588         if (!md)
 589                 goto err_free_rt;
 590
 591         erspan_build_header(skb, tunnel_id_to_key32(key->tun_id),
 592                             ntohl(md->index), truncate);
 593
 594         gre_build_header(skb, 8, TUNNEL_SEQ,
 595                          htons(ETH_P_ERSPAN), 0, htonl(tunnel->o_seqno++));
 596
 597         df = key->tun_flags & TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
 598
 599         iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
 600                       key->tos, key->ttl, df, false);
 601         return;
 602
 603 err_free_rt:
 604         ip_rt_put(rt);
 605 err_free_skb:
 606         kfree_skb(skb);
 607         dev->stats.tx_dropped++;
 608 }
 609
 610 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
 611 {
 612         struct ip_tunnel_info *info = skb_tunnel_info(skb);
 613         struct rtable *rt;
 614         struct flowi4 fl4;
 615
 616         if (ip_tunnel_info_af(info) != AF_INET)
 617                 return -EINVAL;
 618
 619         rt = gre_get_rt(skb, dev, &fl4, &info->key);
 620         if (IS_ERR(rt))
 621                 return PTR_ERR(rt);
 622
 623         ip_rt_put(rt);
 624         info->key.u.ipv4.src = fl4.saddr;
 625         return 0;
 626 }
 627
 628 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
 629                               struct net_device *dev)
 630 {
 631         struct ip_tunnel *tunnel = netdev_priv(dev);
 632         const struct iphdr *tnl_params;
 633
 634         if (tunnel->collect_md) {
 635                 gre_fb_xmit(skb, dev, skb->protocol);
 636                 return NETDEV_TX_OK;
 637         }
 638
 639         if (dev->header_ops) {
 640                 /* Need space for new headers */
 641                 if (skb_cow_head(skb, dev->needed_headroom -
 642                                       (tunnel->hlen + sizeof(struct iphdr))))
 643                         goto free_skb;
 644
 645                 tnl_params = (const struct iphdr *)skb->data;
 646
 647                 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing
 648                  * to gre header.
 649                  */
 650                 skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
 651                 skb_reset_mac_header(skb);
 652         } else {
 653                 if (skb_cow_head(skb, dev->needed_headroom))
 654                         goto free_skb;
 655
 656                 tnl_params = &tunnel->parms.iph;
 657         }
 658
 659         if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
 660                 goto free_skb;
 661
 662         __gre_xmit(skb, dev, tnl_params, skb->protocol);
 663         return NETDEV_TX_OK;
 664
 665 free_skb:
 666         kfree_skb(skb);
 667         dev->stats.tx_dropped++;
 668         return NETDEV_TX_OK;
 669 }
 670
 671 static inline u8 tos_to_cos(u8 tos)
 672 {
 673         u8 dscp, cos;
 674
 675         dscp = tos >> 2;
 676         cos = dscp >> 3;
 677         return cos;
 678 }
 679
 680 static void erspan_build_header(struct sk_buff *skb,
 681                                 __be32 id, u32 index, bool truncate)
 682 {
 683         struct iphdr *iphdr = ip_hdr(skb);
 684         struct ethhdr *eth = eth_hdr(skb);
 685         enum erspan_encap_type enc_type;
 686         struct erspanhdr *ershdr;
 687         struct qtag_prefix {
 688                 __be16 eth_type;
 689                 __be16 tci;
 690         } *qp;
 691         u16 vlan_tci = 0;
 692
 693         enc_type = ERSPAN_ENCAP_NOVLAN;
 694
 695         /* If mirrored packet has vlan tag, extract tci and
 696          *  perserve vlan header in the mirrored frame.
 697          */
 698         if (eth->h_proto == htons(ETH_P_8021Q)) {
 699                 qp = (struct qtag_prefix *)(skb->data + 2 * ETH_ALEN);
 700                 vlan_tci = ntohs(qp->tci);
 701                 enc_type = ERSPAN_ENCAP_INFRAME;
 702         }
 703
 704         skb_push(skb, sizeof(*ershdr));
 705         ershdr = (struct erspanhdr *)skb->data;
 706         memset(ershdr, 0, sizeof(*ershdr));
 707
 708         ershdr->ver_vlan = htons((vlan_tci & VLAN_MASK) |
 709                                  (ERSPAN_VERSION << VER_OFFSET));
 710         ershdr->session_id = htons((u16)(ntohl(id) & ID_MASK) |
 711                            ((tos_to_cos(iphdr->tos) << COS_OFFSET) & COS_MASK) |
 712                            (enc_type << EN_OFFSET & EN_MASK) |
 713                            ((truncate << T_OFFSET) & T_MASK));
 714         ershdr->md.index = htonl(index & INDEX_MASK);
 715 }
 716
 717 static netdev_tx_t erspan_xmit(struct sk_buff *skb,
 718                                struct net_device *dev)
 719 {
 720         struct ip_tunnel *tunnel = netdev_priv(dev);
 721         bool truncate = false;
 722
 723         if (tunnel->collect_md) {
 724                 erspan_fb_xmit(skb, dev, skb->protocol);
 725                 return NETDEV_TX_OK;
 726         }
 727
 728         if (gre_handle_offloads(skb, false))
 729                 goto free_skb;
 730
 731         if (skb_cow_head(skb, dev->needed_headroom))
 732                 goto free_skb;
 733
 734         if (skb->len - dev->hard_header_len > dev->mtu) {
 735                 pskb_trim(skb, dev->mtu);
 736                 truncate = true;
 737         }
 738
 739         /* Push ERSPAN header */
 740         erspan_build_header(skb, tunnel->parms.o_key, tunnel->index, truncate);
 741         tunnel->parms.o_flags &= ~TUNNEL_KEY;
 742         __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_ERSPAN));
 743         return NETDEV_TX_OK;
 744
 745 free_skb:
 746         kfree_skb(skb);
 747         dev->stats.tx_dropped++;
 748         return NETDEV_TX_OK;
 749 }
 750
 751 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
 752                                 struct net_device *dev)
 753 {
 754         struct ip_tunnel *tunnel = netdev_priv(dev);
 755
 756         if (tunnel->collect_md) {
 757                 gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
 758                 return NETDEV_TX_OK;
 759         }
 760
 761         if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
 762                 goto free_skb;
 763
 764         if (skb_cow_head(skb, dev->needed_headroom))
 765                 goto free_skb;
 766
 767         __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
 768         return NETDEV_TX_OK;
 769
 770 free_skb:
 771         kfree_skb(skb);
 772         dev->stats.tx_dropped++;
 773         return NETDEV_TX_OK;
 774 }
 775
 776 static int ipgre_tunnel_ioctl(struct net_device *dev,
 777                               struct ifreq *ifr, int cmd)
 778 {
 779         int err;
 780         struct ip_tunnel_parm p;
 781
 782         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
 783                 return -EFAULT;
 784         if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
 785                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
 786                     p.iph.ihl != 5 || (p.iph.frag_off&htons(~IP_DF)) ||
 787                     ((p.i_flags|p.o_flags)&(GRE_VERSION|GRE_ROUTING)))
 788                         return -EINVAL;
 789         }
 790         p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
 791         p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
 792
 793         err = ip_tunnel_ioctl(dev, &p, cmd);
 794         if (err)
 795                 return err;
 796
 797         p.i_flags = gre_tnl_flags_to_gre_flags(p.i_flags);
 798         p.o_flags = gre_tnl_flags_to_gre_flags(p.o_flags);
 799
 800         if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
 801                 return -EFAULT;
 802         return 0;
 803 }
 804
 805 /* Nice toy. Unfortunately, useless in real life :-)
 806    It allows to construct virtual multiprotocol broadcast "LAN"
 807    over the Internet, provided multicast routing is tuned.
 808
 809
 810    I have no idea was this bicycle invented before me,
 811    so that I had to set ARPHRD_IPGRE to a random value.
 812    I have an impression, that Cisco could make something similar,
 813    but this feature is apparently missing in IOS<=11.2(8).
 814
 815    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
 816    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
 817
 818    ping -t 255 224.66.66.66
 819
 820    If nobody answers, mbone does not work.
 821
 822    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
 823    ip addr add 10.66.66.<somewhat>/24 dev Universe
 824    ifconfig Universe up
 825    ifconfig Universe add fe80::<Your_real_addr>/10
 826    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
 827    ftp 10.66.66.66
 828    ...
 829    ftp fec0:6666:6666::193.233.7.65
 830    ...
 831  */
 832 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
 833                         unsigned short type,
 834                         const void *daddr, const void *saddr, unsigned int len)
 835 {
 836         struct ip_tunnel *t = netdev_priv(dev);
 837         struct iphdr *iph;
 838         struct gre_base_hdr *greh;
 839
 840         iph = skb_push(skb, t->hlen + sizeof(*iph));
 841         greh = (struct gre_base_hdr *)(iph+1);
 842         greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
 843         greh->protocol = htons(type);
 844
 845         memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
 846
 847         /* Set the source hardware address. */
 848         if (saddr)
 849                 memcpy(&iph->saddr, saddr, 4);
 850         if (daddr)
 851                 memcpy(&iph->daddr, daddr, 4);
 852         if (iph->daddr)
 853                 return t->hlen + sizeof(*iph);
 854
 855         return -(t->hlen + sizeof(*iph));
 856 }
 857
 858 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
 859 {
 860         const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
 861         memcpy(haddr, &iph->saddr, 4);
 862         return 4;
 863 }
 864
 865 static const struct header_ops ipgre_header_ops = {
 866         .create = ipgre_header,
 867         .parse  = ipgre_header_parse,
 868 };
 869
 870 #ifdef CONFIG_NET_IPGRE_BROADCAST
 871 static int ipgre_open(struct net_device *dev)
 872 {
 873         struct ip_tunnel *t = netdev_priv(dev);
 874
 875         if (ipv4_is_multicast(t->parms.iph.daddr)) {
 876                 struct flowi4 fl4;
 877                 struct rtable *rt;
 878
 879                 rt = ip_route_output_gre(t->net, &fl4,
 880                                          t->parms.iph.daddr,
 881                                          t->parms.iph.saddr,
 882                                          t->parms.o_key,
 883                                          RT_TOS(t->parms.iph.tos),
 884                                          t->parms.link);
 885                 if (IS_ERR(rt))
 886                         return -EADDRNOTAVAIL;
 887                 dev = rt->dst.dev;
 888                 ip_rt_put(rt);
 889                 if (!__in_dev_get_rtnl(dev))
 890                         return -EADDRNOTAVAIL;
 891                 t->mlink = dev->ifindex;
 892                 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
 893         }
 894         return 0;
 895 }
 896
 897 static int ipgre_close(struct net_device *dev)
 898 {
 899         struct ip_tunnel *t = netdev_priv(dev);
 900
 901         if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
 902                 struct in_device *in_dev;
 903                 in_dev = inetdev_by_index(t->net, t->mlink);
 904                 if (in_dev)
 905                         ip_mc_dec_group(in_dev, t->parms.iph.daddr);
 906         }
 907         return 0;
 908 }
 909 #endif
 910
 911 static const struct net_device_ops ipgre_netdev_ops = {
 912         .ndo_init               = ipgre_tunnel_init,
 913         .ndo_uninit             = ip_tunnel_uninit,
 914 #ifdef CONFIG_NET_IPGRE_BROADCAST
 915         .ndo_open               = ipgre_open,
 916         .ndo_stop               = ipgre_close,
 917 #endif
 918         .ndo_start_xmit         = ipgre_xmit,
 919         .ndo_do_ioctl           = ipgre_tunnel_ioctl,
 920         .ndo_change_mtu         = ip_tunnel_change_mtu,
 921         .ndo_get_stats64        = ip_tunnel_get_stats64,
 922         .ndo_get_iflink         = ip_tunnel_get_iflink,
 923 };
 924
 925 #define GRE_FEATURES (NETIF_F_SG |              \
 926                       NETIF_F_FRAGLIST |        \
 927                       NETIF_F_HIGHDMA |         \
 928                       NETIF_F_HW_CSUM)
 929
 930 static void ipgre_tunnel_setup(struct net_device *dev)
 931 {
 932         dev->netdev_ops         = &ipgre_netdev_ops;
 933         dev->type               = ARPHRD_IPGRE;
 934         ip_tunnel_setup(dev, ipgre_net_id);
 935 }
 936
 937 static void __gre_tunnel_init(struct net_device *dev)
 938 {
 939         struct ip_tunnel *tunnel;
 940         int t_hlen;
 941
 942         tunnel = netdev_priv(dev);
 943         tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
 944         tunnel->parms.iph.protocol = IPPROTO_GRE;
 945
 946         tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
 947
 948         t_hlen = tunnel->hlen + sizeof(struct iphdr);
 949
 950         dev->needed_headroom    = LL_MAX_HEADER + t_hlen + 4;
 951         dev->mtu                = ETH_DATA_LEN - t_hlen - 4;
 952
 953         dev->features           |= GRE_FEATURES;
 954         dev->hw_features        |= GRE_FEATURES;
 955
 956         if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
 957                 /* TCP offload with GRE SEQ is not supported, nor
 958                  * can we support 2 levels of outer headers requiring
 959                  * an update.
 960                  */
 961                 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
 962                     (tunnel->encap.type == TUNNEL_ENCAP_NONE)) {
 963                         dev->features    |= NETIF_F_GSO_SOFTWARE;
 964                         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
 965                 }
 966
 967                 /* Can use a lockless transmit, unless we generate
 968                  * output sequences
 969                  */
 970                 dev->features |= NETIF_F_LLTX;
 971         }
 972 }
 973
 974 static int ipgre_tunnel_init(struct net_device *dev)
 975 {
 976         struct ip_tunnel *tunnel = netdev_priv(dev);
 977         struct iphdr *iph = &tunnel->parms.iph;
 978
 979         __gre_tunnel_init(dev);
 980
 981         memcpy(dev->dev_addr, &iph->saddr, 4);
 982         memcpy(dev->broadcast, &iph->daddr, 4);
 983
 984         dev->flags              = IFF_NOARP;
 985         netif_keep_dst(dev);
 986         dev->addr_len           = 4;
 987
 988         if (iph->daddr && !tunnel->collect_md) {
 989 #ifdef CONFIG_NET_IPGRE_BROADCAST
 990                 if (ipv4_is_multicast(iph->daddr)) {
 991                         if (!iph->saddr)
 992                                 return -EINVAL;
 993                         dev->flags = IFF_BROADCAST;
 994                         dev->header_ops = &ipgre_header_ops;
 995                 }
 996 #endif
 997         } else if (!tunnel->collect_md) {
 998                 dev->header_ops = &ipgre_header_ops;
 999         }
1000
1001         return ip_tunnel_init(dev);
1002 }
1003
1004 static const struct gre_protocol ipgre_protocol = {
1005         .handler     = gre_rcv,
1006         .err_handler = gre_err,
1007 };
1008
1009 static int __net_init ipgre_init_net(struct net *net)
1010 {
1011         return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
1012 }
1013
1014 static void __net_exit ipgre_exit_net(struct net *net)
1015 {
1016         struct ip_tunnel_net *itn = net_generic(net, ipgre_net_id);
1017         ip_tunnel_delete_net(itn, &ipgre_link_ops);
1018 }
1019
1020 static struct pernet_operations ipgre_net_ops = {
1021         .init = ipgre_init_net,
1022         .exit = ipgre_exit_net,
1023         .id   = &ipgre_net_id,
1024         .size = sizeof(struct ip_tunnel_net),
1025 };
1026
1027 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1028                                  struct netlink_ext_ack *extack)
1029 {
1030         __be16 flags;
1031
1032         if (!data)
1033                 return 0;
1034
1035         flags = 0;
1036         if (data[IFLA_GRE_IFLAGS])
1037                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1038         if (data[IFLA_GRE_OFLAGS])
1039                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1040         if (flags & (GRE_VERSION|GRE_ROUTING))
1041                 return -EINVAL;
1042
1043         if (data[IFLA_GRE_COLLECT_METADATA] &&
1044             data[IFLA_GRE_ENCAP_TYPE] &&
1045             nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1046                 return -EINVAL;
1047
1048         return 0;
1049 }
1050
1051 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1052                               struct netlink_ext_ack *extack)
1053 {
1054         __be32 daddr;
1055
1056         if (tb[IFLA_ADDRESS]) {
1057                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1058                         return -EINVAL;
1059                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1060                         return -EADDRNOTAVAIL;
1061         }
1062
1063         if (!data)
1064                 goto out;
1065
1066         if (data[IFLA_GRE_REMOTE]) {
1067                 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1068                 if (!daddr)
1069                         return -EINVAL;
1070         }
1071
1072 out:
1073         return ipgre_tunnel_validate(tb, data, extack);
1074 }
1075
1076 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1077                            struct netlink_ext_ack *extack)
1078 {
1079         __be16 flags = 0;
1080         int ret;
1081
1082         if (!data)
1083                 return 0;
1084
1085         ret = ipgre_tap_validate(tb, data, extack);
1086         if (ret)
1087                 return ret;
1088
1089         /* ERSPAN should only have GRE sequence and key flag */
1090         if (data[IFLA_GRE_OFLAGS])
1091                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1092         if (data[IFLA_GRE_IFLAGS])
1093                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1094         if (!data[IFLA_GRE_COLLECT_METADATA] &&
1095             flags != (GRE_SEQ | GRE_KEY))
1096                 return -EINVAL;
1097
1098         /* ERSPAN Session ID only has 10-bit. Since we reuse
1099          * 32-bit key field as ID, check it's range.
1100          */
1101         if (data[IFLA_GRE_IKEY] &&
1102             (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1103                 return -EINVAL;
1104
1105         if (data[IFLA_GRE_OKEY] &&
1106             (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1107                 return -EINVAL;
1108
1109         return 0;
1110 }
1111
1112 static int ipgre_netlink_parms(struct net_device *dev,
1113                                 struct nlattr *data[],
1114                                 struct nlattr *tb[],
1115                                 struct ip_tunnel_parm *parms,
1116                                 __u32 *fwmark)
1117 {
1118         struct ip_tunnel *t = netdev_priv(dev);
1119
1120         memset(parms, 0, sizeof(*parms));
1121
1122         parms->iph.protocol = IPPROTO_GRE;
1123
1124         if (!data)
1125                 return 0;
1126
1127         if (data[IFLA_GRE_LINK])
1128                 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1129
1130         if (data[IFLA_GRE_IFLAGS])
1131                 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
1132
1133         if (data[IFLA_GRE_OFLAGS])
1134                 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
1135
1136         if (data[IFLA_GRE_IKEY])
1137                 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1138
1139         if (data[IFLA_GRE_OKEY])
1140                 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1141
1142         if (data[IFLA_GRE_LOCAL])
1143                 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1144
1145         if (data[IFLA_GRE_REMOTE])
1146                 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1147
1148         if (data[IFLA_GRE_TTL])
1149                 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1150
1151         if (data[IFLA_GRE_TOS])
1152                 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1153
1154         if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1155                 if (t->ignore_df)
1156                         return -EINVAL;
1157                 parms->iph.frag_off = htons(IP_DF);
1158         }
1159
1160         if (data[IFLA_GRE_COLLECT_METADATA]) {
1161                 t->collect_md = true;
1162                 if (dev->type == ARPHRD_IPGRE)
1163                         dev->type = ARPHRD_NONE;
1164         }
1165
1166         if (data[IFLA_GRE_IGNORE_DF]) {
1167                 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1168                   && (parms->iph.frag_off & htons(IP_DF)))
1169                         return -EINVAL;
1170                 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1171         }
1172
1173         if (data[IFLA_GRE_FWMARK])
1174                 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1175
1176         if (data[IFLA_GRE_ERSPAN_INDEX]) {
1177                 t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1178
1179                 if (t->index & ~INDEX_MASK)
1180                         return -EINVAL;
1181         }
1182
1183         return 0;
1184 }
1185
1186 /* This function returns true when ENCAP attributes are present in the nl msg */
1187 static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1188                                       struct ip_tunnel_encap *ipencap)
1189 {
1190         bool ret = false;
1191
1192         memset(ipencap, 0, sizeof(*ipencap));
1193
1194         if (!data)
1195                 return ret;
1196
1197         if (data[IFLA_GRE_ENCAP_TYPE]) {
1198                 ret = true;
1199                 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1200         }
1201
1202         if (data[IFLA_GRE_ENCAP_FLAGS]) {
1203                 ret = true;
1204                 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1205         }
1206
1207         if (data[IFLA_GRE_ENCAP_SPORT]) {
1208                 ret = true;
1209                 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1210         }
1211
1212         if (data[IFLA_GRE_ENCAP_DPORT]) {
1213                 ret = true;
1214                 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1215         }
1216
1217         return ret;
1218 }
1219
1220 static int gre_tap_init(struct net_device *dev)
1221 {
1222         __gre_tunnel_init(dev);
1223         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1224         netif_keep_dst(dev);
1225
1226         return ip_tunnel_init(dev);
1227 }
1228
1229 static const struct net_device_ops gre_tap_netdev_ops = {
1230         .ndo_init               = gre_tap_init,
1231         .ndo_uninit             = ip_tunnel_uninit,
1232         .ndo_start_xmit         = gre_tap_xmit,
1233         .ndo_set_mac_address    = eth_mac_addr,
1234         .ndo_validate_addr      = eth_validate_addr,
1235         .ndo_change_mtu         = ip_tunnel_change_mtu,
1236         .ndo_get_stats64        = ip_tunnel_get_stats64,
1237         .ndo_get_iflink         = ip_tunnel_get_iflink,
1238         .ndo_fill_metadata_dst  = gre_fill_metadata_dst,
1239 };
1240
1241 static int erspan_tunnel_init(struct net_device *dev)
1242 {
1243         struct ip_tunnel *tunnel = netdev_priv(dev);
1244         int t_hlen;
1245
1246         tunnel->tun_hlen = 8;
1247         tunnel->parms.iph.protocol = IPPROTO_GRE;
1248         tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1249                        sizeof(struct erspanhdr);
1250         t_hlen = tunnel->hlen + sizeof(struct iphdr);
1251
1252         dev->needed_headroom = LL_MAX_HEADER + t_hlen + 4;
1253         dev->mtu = ETH_DATA_LEN - t_hlen - 4;
1254         dev->features           |= GRE_FEATURES;
1255         dev->hw_features        |= GRE_FEATURES;
1256         dev->priv_flags         |= IFF_LIVE_ADDR_CHANGE;
1257         netif_keep_dst(dev);
1258
1259         return ip_tunnel_init(dev);
1260 }
1261
1262 static const struct net_device_ops erspan_netdev_ops = {
1263         .ndo_init               = erspan_tunnel_init,
1264         .ndo_uninit             = ip_tunnel_uninit,
1265         .ndo_start_xmit         = erspan_xmit,
1266         .ndo_set_mac_address    = eth_mac_addr,
1267         .ndo_validate_addr      = eth_validate_addr,
1268         .ndo_change_mtu         = ip_tunnel_change_mtu,
1269         .ndo_get_stats64        = ip_tunnel_get_stats64,
1270         .ndo_get_iflink         = ip_tunnel_get_iflink,
1271         .ndo_fill_metadata_dst  = gre_fill_metadata_dst,
1272 };
1273
1274 static void ipgre_tap_setup(struct net_device *dev)
1275 {
1276         ether_setup(dev);
1277         dev->netdev_ops = &gre_tap_netdev_ops;
1278         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1279         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1280         ip_tunnel_setup(dev, gre_tap_net_id);
1281 }
1282
1283 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1284                          struct nlattr *tb[], struct nlattr *data[],
1285                          struct netlink_ext_ack *extack)
1286 {
1287         struct ip_tunnel_parm p;
1288         struct ip_tunnel_encap ipencap;
1289         __u32 fwmark = 0;
1290         int err;
1291
1292         if (ipgre_netlink_encap_parms(data, &ipencap)) {
1293                 struct ip_tunnel *t = netdev_priv(dev);
1294                 err = ip_tunnel_encap_setup(t, &ipencap);
1295
1296                 if (err < 0)
1297                         return err;
1298         }
1299
1300         err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1301         if (err < 0)
1302                 return err;
1303         return ip_tunnel_newlink(dev, tb, &p, fwmark);
1304 }
1305
1306 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1307                             struct nlattr *data[],
1308                             struct netlink_ext_ack *extack)
1309 {
1310         struct ip_tunnel *t = netdev_priv(dev);
1311         struct ip_tunnel_parm p;
1312         struct ip_tunnel_encap ipencap;
1313         __u32 fwmark = t->fwmark;
1314         int err;
1315
1316         if (ipgre_netlink_encap_parms(data, &ipencap)) {
1317                 err = ip_tunnel_encap_setup(t, &ipencap);
1318
1319                 if (err < 0)
1320                         return err;
1321         }
1322
1323         err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1324         if (err < 0)
1325                 return err;
1326         return ip_tunnel_changelink(dev, tb, &p, fwmark);
1327 }
1328
1329 static size_t ipgre_get_size(const struct net_device *dev)
1330 {
1331         return
1332                 /* IFLA_GRE_LINK */
1333                 nla_total_size(4) +
1334                 /* IFLA_GRE_IFLAGS */
1335                 nla_total_size(2) +
1336                 /* IFLA_GRE_OFLAGS */
1337                 nla_total_size(2) +
1338                 /* IFLA_GRE_IKEY */
1339                 nla_total_size(4) +
1340                 /* IFLA_GRE_OKEY */
1341                 nla_total_size(4) +
1342                 /* IFLA_GRE_LOCAL */
1343                 nla_total_size(4) +
1344                 /* IFLA_GRE_REMOTE */
1345                 nla_total_size(4) +
1346                 /* IFLA_GRE_TTL */
1347                 nla_total_size(1) +
1348                 /* IFLA_GRE_TOS */
1349                 nla_total_size(1) +
1350                 /* IFLA_GRE_PMTUDISC */
1351                 nla_total_size(1) +
1352                 /* IFLA_GRE_ENCAP_TYPE */
1353                 nla_total_size(2) +
1354                 /* IFLA_GRE_ENCAP_FLAGS */
1355                 nla_total_size(2) +
1356                 /* IFLA_GRE_ENCAP_SPORT */
1357                 nla_total_size(2) +
1358                 /* IFLA_GRE_ENCAP_DPORT */
1359                 nla_total_size(2) +
1360                 /* IFLA_GRE_COLLECT_METADATA */
1361                 nla_total_size(0) +
1362                 /* IFLA_GRE_IGNORE_DF */
1363                 nla_total_size(1) +
1364                 /* IFLA_GRE_FWMARK */
1365                 nla_total_size(4) +
1366                 /* IFLA_GRE_ERSPAN_INDEX */
1367                 nla_total_size(4) +
1368                 0;
1369 }
1370
1371 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1372 {
1373         struct ip_tunnel *t = netdev_priv(dev);
1374         struct ip_tunnel_parm *p = &t->parms;
1375
1376         if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1377             nla_put_be16(skb, IFLA_GRE_IFLAGS,
1378                          gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1379             nla_put_be16(skb, IFLA_GRE_OFLAGS,
1380                          gre_tnl_flags_to_gre_flags(p->o_flags)) ||
1381             nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1382             nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1383             nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1384             nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1385             nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1386             nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1387             nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1388                        !!(p->iph.frag_off & htons(IP_DF))) ||
1389             nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1390                 goto nla_put_failure;
1391
1392         if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1393                         t->encap.type) ||
1394             nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1395                          t->encap.sport) ||
1396             nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1397                          t->encap.dport) ||
1398             nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1399                         t->encap.flags))
1400                 goto nla_put_failure;
1401
1402         if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1403                 goto nla_put_failure;
1404
1405         if (t->collect_md) {
1406                 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1407                         goto nla_put_failure;
1408         }
1409
1410         if (t->index)
1411                 if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1412                         goto nla_put_failure;
1413
1414         return 0;
1415
1416 nla_put_failure:
1417         return -EMSGSIZE;
1418 }
1419
1420 static void erspan_setup(struct net_device *dev)
1421 {
1422         ether_setup(dev);
1423         dev->netdev_ops = &erspan_netdev_ops;
1424         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1425         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1426         ip_tunnel_setup(dev, erspan_net_id);
1427 }
1428
1429 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1430         [IFLA_GRE_LINK]         = { .type = NLA_U32 },
1431         [IFLA_GRE_IFLAGS]       = { .type = NLA_U16 },
1432         [IFLA_GRE_OFLAGS]       = { .type = NLA_U16 },
1433         [IFLA_GRE_IKEY]         = { .type = NLA_U32 },
1434         [IFLA_GRE_OKEY]         = { .type = NLA_U32 },
1435         [IFLA_GRE_LOCAL]        = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1436         [IFLA_GRE_REMOTE]       = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1437         [IFLA_GRE_TTL]          = { .type = NLA_U8 },
1438         [IFLA_GRE_TOS]          = { .type = NLA_U8 },
1439         [IFLA_GRE_PMTUDISC]     = { .type = NLA_U8 },
1440         [IFLA_GRE_ENCAP_TYPE]   = { .type = NLA_U16 },
1441         [IFLA_GRE_ENCAP_FLAGS]  = { .type = NLA_U16 },
1442         [IFLA_GRE_ENCAP_SPORT]  = { .type = NLA_U16 },
1443         [IFLA_GRE_ENCAP_DPORT]  = { .type = NLA_U16 },
1444         [IFLA_GRE_COLLECT_METADATA]     = { .type = NLA_FLAG },
1445         [IFLA_GRE_IGNORE_DF]    = { .type = NLA_U8 },
1446         [IFLA_GRE_FWMARK]       = { .type = NLA_U32 },
1447         [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 },
1448 };
1449
1450 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1451         .kind           = "gre",
1452         .maxtype        = IFLA_GRE_MAX,
1453         .policy         = ipgre_policy,
1454         .priv_size      = sizeof(struct ip_tunnel),
1455         .setup          = ipgre_tunnel_setup,
1456         .validate       = ipgre_tunnel_validate,
1457         .newlink        = ipgre_newlink,
1458         .changelink     = ipgre_changelink,
1459         .dellink        = ip_tunnel_dellink,
1460         .get_size       = ipgre_get_size,
1461         .fill_info      = ipgre_fill_info,
1462         .get_link_net   = ip_tunnel_get_link_net,
1463 };
1464
1465 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1466         .kind           = "gretap",
1467         .maxtype        = IFLA_GRE_MAX,
1468         .policy         = ipgre_policy,
1469         .priv_size      = sizeof(struct ip_tunnel),
1470         .setup          = ipgre_tap_setup,
1471         .validate       = ipgre_tap_validate,
1472         .newlink        = ipgre_newlink,
1473         .changelink     = ipgre_changelink,
1474         .dellink        = ip_tunnel_dellink,
1475         .get_size       = ipgre_get_size,
1476         .fill_info      = ipgre_fill_info,
1477         .get_link_net   = ip_tunnel_get_link_net,
1478 };
1479
1480 static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1481         .kind           = "erspan",
1482         .maxtype        = IFLA_GRE_MAX,
1483         .policy         = ipgre_policy,
1484         .priv_size      = sizeof(struct ip_tunnel),
1485         .setup          = erspan_setup,
1486         .validate       = erspan_validate,
1487         .newlink        = ipgre_newlink,
1488         .changelink     = ipgre_changelink,
1489         .dellink        = ip_tunnel_dellink,
1490         .get_size       = ipgre_get_size,
1491         .fill_info      = ipgre_fill_info,
1492         .get_link_net   = ip_tunnel_get_link_net,
1493 };
1494
1495 struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1496                                         u8 name_assign_type)
1497 {
1498         struct nlattr *tb[IFLA_MAX + 1];
1499         struct net_device *dev;
1500         LIST_HEAD(list_kill);
1501         struct ip_tunnel *t;
1502         int err;
1503
1504         memset(&tb, 0, sizeof(tb));
1505
1506         dev = rtnl_create_link(net, name, name_assign_type,
1507                                &ipgre_tap_ops, tb);
1508         if (IS_ERR(dev))
1509                 return dev;
1510
1511         /* Configure flow based GRE device. */
1512         t = netdev_priv(dev);
1513         t->collect_md = true;
1514
1515         err = ipgre_newlink(net, dev, tb, NULL, NULL);
1516         if (err < 0) {
1517                 free_netdev(dev);
1518                 return ERR_PTR(err);
1519         }
1520
1521         /* openvswitch users expect packet sizes to be unrestricted,
1522          * so set the largest MTU we can.
1523          */
1524         err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1525         if (err)
1526                 goto out;
1527
1528         err = rtnl_configure_link(dev, NULL);
1529         if (err < 0)
1530                 goto out;
1531
1532         return dev;
1533 out:
1534         ip_tunnel_dellink(dev, &list_kill);
1535         unregister_netdevice_many(&list_kill);
1536         return ERR_PTR(err);
1537 }
1538 EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1539
1540 static int __net_init ipgre_tap_init_net(struct net *net)
1541 {
1542         return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1543 }
1544
1545 static void __net_exit ipgre_tap_exit_net(struct net *net)
1546 {
1547         struct ip_tunnel_net *itn = net_generic(net, gre_tap_net_id);
1548         ip_tunnel_delete_net(itn, &ipgre_tap_ops);
1549 }
1550
1551 static struct pernet_operations ipgre_tap_net_ops = {
1552         .init = ipgre_tap_init_net,
1553         .exit = ipgre_tap_exit_net,
1554         .id   = &gre_tap_net_id,
1555         .size = sizeof(struct ip_tunnel_net),
1556 };
1557
1558 static int __net_init erspan_init_net(struct net *net)
1559 {
1560         return ip_tunnel_init_net(net, erspan_net_id,
1561                                   &erspan_link_ops, "erspan0");
1562 }
1563
1564 static void __net_exit erspan_exit_net(struct net *net)
1565 {
1566         struct ip_tunnel_net *itn = net_generic(net, erspan_net_id);
1567
1568         ip_tunnel_delete_net(itn, &erspan_link_ops);
1569 }
1570
1571 static struct pernet_operations erspan_net_ops = {
1572         .init = erspan_init_net,
1573         .exit = erspan_exit_net,
1574         .id   = &erspan_net_id,
1575         .size = sizeof(struct ip_tunnel_net),
1576 };
1577
1578 static int __init ipgre_init(void)
1579 {
1580         int err;
1581
1582         pr_info("GRE over IPv4 tunneling driver\n");
1583
1584         err = register_pernet_device(&ipgre_net_ops);
1585         if (err < 0)
1586                 return err;
1587
1588         err = register_pernet_device(&ipgre_tap_net_ops);
1589         if (err < 0)
1590                 goto pnet_tap_failed;
1591
1592         err = register_pernet_device(&erspan_net_ops);
1593         if (err < 0)
1594                 goto pnet_erspan_failed;
1595
1596         err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1597         if (err < 0) {
1598                 pr_info("%s: can't add protocol\n", __func__);
1599                 goto add_proto_failed;
1600         }
1601
1602         err = rtnl_link_register(&ipgre_link_ops);
1603         if (err < 0)
1604                 goto rtnl_link_failed;
1605
1606         err = rtnl_link_register(&ipgre_tap_ops);
1607         if (err < 0)
1608                 goto tap_ops_failed;
1609
1610         err = rtnl_link_register(&erspan_link_ops);
1611         if (err < 0)
1612                 goto erspan_link_failed;
1613
1614         return 0;
1615
1616 erspan_link_failed:
1617         rtnl_link_unregister(&ipgre_tap_ops);
1618 tap_ops_failed:
1619         rtnl_link_unregister(&ipgre_link_ops);
1620 rtnl_link_failed:
1621         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1622 add_proto_failed:
1623         unregister_pernet_device(&erspan_net_ops);
1624 pnet_erspan_failed:
1625         unregister_pernet_device(&ipgre_tap_net_ops);
1626 pnet_tap_failed:
1627         unregister_pernet_device(&ipgre_net_ops);
1628         return err;
1629 }
1630
1631 static void __exit ipgre_fini(void)
1632 {
1633         rtnl_link_unregister(&ipgre_tap_ops);
1634         rtnl_link_unregister(&ipgre_link_ops);
1635         rtnl_link_unregister(&erspan_link_ops);
1636         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1637         unregister_pernet_device(&ipgre_tap_net_ops);
1638         unregister_pernet_device(&ipgre_net_ops);
1639         unregister_pernet_device(&erspan_net_ops);
1640 }
1641
1642 module_init(ipgre_init);
1643 module_exit(ipgre_fini);
1644 MODULE_LICENSE("GPL");
1645 MODULE_ALIAS_RTNL_LINK("gre");
1646 MODULE_ALIAS_RTNL_LINK("gretap");
1647 MODULE_ALIAS_RTNL_LINK("erspan");
1648 MODULE_ALIAS_NETDEV("gre0");
1649 MODULE_ALIAS_NETDEV("gretap0");
1650 MODULE_ALIAS_NETDEV("erspan0");