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                                 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 *tunnel = netdev_priv(dev);
 526         struct ip_tunnel_info *tun_info;
 527         const struct ip_tunnel_key *key;
 528         struct rtable *rt = NULL;
 529         struct flowi4 fl;
 530         int tunnel_hlen;
 531         __be16 df, flags;
 532
 533         tun_info = skb_tunnel_info(skb);
 534         if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
 535                      ip_tunnel_info_af(tun_info) != AF_INET))
 536                 goto err_free_skb;
 537
 538         key = &tun_info->key;
 539         tunnel_hlen = gre_calc_hlen(key->tun_flags);
 540
 541         rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
 542         if (!rt)
 543                 return;
 544
 545         /* Push Tunnel header. */
 546         if (gre_handle_offloads(skb, !!(tun_info->key.tun_flags & TUNNEL_CSUM)))
 547                 goto err_free_rt;
 548
 549         flags = tun_info->key.tun_flags &
 550                 (TUNNEL_CSUM | TUNNEL_KEY | TUNNEL_SEQ);
 551         gre_build_header(skb, tunnel_hlen, flags, proto,
 552                          tunnel_id_to_key32(tun_info->key.tun_id),
 553                          (flags & TUNNEL_SEQ) ? htonl(tunnel->o_seqno++) : 0);
 554
 555         df = key->tun_flags & TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
 556
 557         iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
 558                       key->tos, key->ttl, df, false);
 559         return;
 560
 561 err_free_rt:
 562         ip_rt_put(rt);
 563 err_free_skb:
 564         kfree_skb(skb);
 565         dev->stats.tx_dropped++;
 566 }
 567
 568 static void erspan_fb_xmit(struct sk_buff *skb, struct net_device *dev,
 569                            __be16 proto)
 570 {
 571         struct ip_tunnel *tunnel = netdev_priv(dev);
 572         struct ip_tunnel_info *tun_info;
 573         const struct ip_tunnel_key *key;
 574         struct erspan_metadata *md;
 575         struct rtable *rt = NULL;
 576         bool truncate = false;
 577         struct flowi4 fl;
 578         int tunnel_hlen;
 579         int version;
 580         __be16 df;
 581
 582         tun_info = skb_tunnel_info(skb);
 583         if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
 584                      ip_tunnel_info_af(tun_info) != AF_INET))
 585                 goto err_free_skb;
 586
 587         key = &tun_info->key;
 588         md = ip_tunnel_info_opts(tun_info);
 589         if (!md)
 590                 goto err_free_rt;
 591
 592         /* ERSPAN has fixed 8 byte GRE header */
 593         version = md->version;
 594         tunnel_hlen = 8 + erspan_hdr_len(version);
 595
 596         rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
 597         if (!rt)
 598                 return;
 599
 600         if (gre_handle_offloads(skb, false))
 601                 goto err_free_rt;
 602
 603         if (skb->len > dev->mtu + dev->hard_header_len) {
 604                 pskb_trim(skb, dev->mtu + dev->hard_header_len);
 605                 truncate = true;
 606         }
 607
 608         if (version == 1) {
 609                 erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)),
 610                                     ntohl(md->u.index), truncate, true);
 611         } else if (version == 2) {
 612                 erspan_build_header_v2(skb,
 613                                        ntohl(tunnel_id_to_key32(key->tun_id)),
 614                                        md->u.md2.dir,
 615                                        get_hwid(&md->u.md2),
 616                                        truncate, true);
 617         } else {
 618                 goto err_free_rt;
 619         }
 620
 621         gre_build_header(skb, 8, TUNNEL_SEQ,
 622                          htons(ETH_P_ERSPAN), 0, htonl(tunnel->o_seqno++));
 623
 624         df = key->tun_flags & TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
 625
 626         iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
 627                       key->tos, key->ttl, df, false);
 628         return;
 629
 630 err_free_rt:
 631         ip_rt_put(rt);
 632 err_free_skb:
 633         kfree_skb(skb);
 634         dev->stats.tx_dropped++;
 635 }
 636
 637 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
 638 {
 639         struct ip_tunnel_info *info = skb_tunnel_info(skb);
 640         struct rtable *rt;
 641         struct flowi4 fl4;
 642
 643         if (ip_tunnel_info_af(info) != AF_INET)
 644                 return -EINVAL;
 645
 646         rt = gre_get_rt(skb, dev, &fl4, &info->key);
 647         if (IS_ERR(rt))
 648                 return PTR_ERR(rt);
 649
 650         ip_rt_put(rt);
 651         info->key.u.ipv4.src = fl4.saddr;
 652         return 0;
 653 }
 654
 655 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
 656                               struct net_device *dev)
 657 {
 658         struct ip_tunnel *tunnel = netdev_priv(dev);
 659         const struct iphdr *tnl_params;
 660
 661         if (tunnel->collect_md) {
 662                 gre_fb_xmit(skb, dev, skb->protocol);
 663                 return NETDEV_TX_OK;
 664         }
 665
 666         if (dev->header_ops) {
 667                 /* Need space for new headers */
 668                 if (skb_cow_head(skb, dev->needed_headroom -
 669                                       (tunnel->hlen + sizeof(struct iphdr))))
 670                         goto free_skb;
 671
 672                 tnl_params = (const struct iphdr *)skb->data;
 673
 674                 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing
 675                  * to gre header.
 676                  */
 677                 skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
 678                 skb_reset_mac_header(skb);
 679         } else {
 680                 if (skb_cow_head(skb, dev->needed_headroom))
 681                         goto free_skb;
 682
 683                 tnl_params = &tunnel->parms.iph;
 684         }
 685
 686         if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
 687                 goto free_skb;
 688
 689         __gre_xmit(skb, dev, tnl_params, skb->protocol);
 690         return NETDEV_TX_OK;
 691
 692 free_skb:
 693         kfree_skb(skb);
 694         dev->stats.tx_dropped++;
 695         return NETDEV_TX_OK;
 696 }
 697
 698 static netdev_tx_t erspan_xmit(struct sk_buff *skb,
 699                                struct net_device *dev)
 700 {
 701         struct ip_tunnel *tunnel = netdev_priv(dev);
 702         bool truncate = false;
 703
 704         if (tunnel->collect_md) {
 705                 erspan_fb_xmit(skb, dev, skb->protocol);
 706                 return NETDEV_TX_OK;
 707         }
 708
 709         if (gre_handle_offloads(skb, false))
 710                 goto free_skb;
 711
 712         if (skb_cow_head(skb, dev->needed_headroom))
 713                 goto free_skb;
 714
 715         if (skb->len > dev->mtu + dev->hard_header_len) {
 716                 pskb_trim(skb, dev->mtu + dev->hard_header_len);
 717                 truncate = true;
 718         }
 719
 720         /* Push ERSPAN header */
 721         if (tunnel->erspan_ver == 1)
 722                 erspan_build_header(skb, ntohl(tunnel->parms.o_key),
 723                                     tunnel->index,
 724                                     truncate, true);
 725         else if (tunnel->erspan_ver == 2)
 726                 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key),
 727                                        tunnel->dir, tunnel->hwid,
 728                                        truncate, true);
 729         else
 730                 goto free_skb;
 731
 732         tunnel->parms.o_flags &= ~TUNNEL_KEY;
 733         __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_ERSPAN));
 734         return NETDEV_TX_OK;
 735
 736 free_skb:
 737         kfree_skb(skb);
 738         dev->stats.tx_dropped++;
 739         return NETDEV_TX_OK;
 740 }
 741
 742 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
 743                                 struct net_device *dev)
 744 {
 745         struct ip_tunnel *tunnel = netdev_priv(dev);
 746
 747         if (tunnel->collect_md) {
 748                 gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
 749                 return NETDEV_TX_OK;
 750         }
 751
 752         if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
 753                 goto free_skb;
 754
 755         if (skb_cow_head(skb, dev->needed_headroom))
 756                 goto free_skb;
 757
 758         __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
 759         return NETDEV_TX_OK;
 760
 761 free_skb:
 762         kfree_skb(skb);
 763         dev->stats.tx_dropped++;
 764         return NETDEV_TX_OK;
 765 }
 766
 767 static void ipgre_link_update(struct net_device *dev, bool set_mtu)
 768 {
 769         struct ip_tunnel *tunnel = netdev_priv(dev);
 770         int len;
 771
 772         len = tunnel->tun_hlen;
 773         tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
 774         len = tunnel->tun_hlen - len;
 775         tunnel->hlen = tunnel->hlen + len;
 776
 777         dev->needed_headroom = dev->needed_headroom + len;
 778         if (set_mtu)
 779                 dev->mtu = max_t(int, dev->mtu - len, 68);
 780
 781         if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
 782                 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
 783                     tunnel->encap.type == TUNNEL_ENCAP_NONE) {
 784                         dev->features |= NETIF_F_GSO_SOFTWARE;
 785                         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
 786                 } else {
 787                         dev->features &= ~NETIF_F_GSO_SOFTWARE;
 788                         dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
 789                 }
 790                 dev->features |= NETIF_F_LLTX;
 791         } else {
 792                 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
 793                 dev->features &= ~(NETIF_F_LLTX | NETIF_F_GSO_SOFTWARE);
 794         }
 795 }
 796
 797 static int ipgre_tunnel_ioctl(struct net_device *dev,
 798                               struct ifreq *ifr, int cmd)
 799 {
 800         struct ip_tunnel_parm p;
 801         int err;
 802
 803         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
 804                 return -EFAULT;
 805
 806         if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
 807                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
 808                     p.iph.ihl != 5 || (p.iph.frag_off & htons(~IP_DF)) ||
 809                     ((p.i_flags | p.o_flags) & (GRE_VERSION | GRE_ROUTING)))
 810                         return -EINVAL;
 811         }
 812
 813         p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
 814         p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
 815
 816         err = ip_tunnel_ioctl(dev, &p, cmd);
 817         if (err)
 818                 return err;
 819
 820         if (cmd == SIOCCHGTUNNEL) {
 821                 struct ip_tunnel *t = netdev_priv(dev);
 822
 823                 t->parms.i_flags = p.i_flags;
 824                 t->parms.o_flags = p.o_flags;
 825
 826                 if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
 827                         ipgre_link_update(dev, true);
 828         }
 829
 830         p.i_flags = gre_tnl_flags_to_gre_flags(p.i_flags);
 831         p.o_flags = gre_tnl_flags_to_gre_flags(p.o_flags);
 832
 833         if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
 834                 return -EFAULT;
 835
 836         return 0;
 837 }
 838
 839 /* Nice toy. Unfortunately, useless in real life :-)
 840    It allows to construct virtual multiprotocol broadcast "LAN"
 841    over the Internet, provided multicast routing is tuned.
 842
 843
 844    I have no idea was this bicycle invented before me,
 845    so that I had to set ARPHRD_IPGRE to a random value.
 846    I have an impression, that Cisco could make something similar,
 847    but this feature is apparently missing in IOS<=11.2(8).
 848
 849    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
 850    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
 851
 852    ping -t 255 224.66.66.66
 853
 854    If nobody answers, mbone does not work.
 855
 856    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
 857    ip addr add 10.66.66.<somewhat>/24 dev Universe
 858    ifconfig Universe up
 859    ifconfig Universe add fe80::<Your_real_addr>/10
 860    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
 861    ftp 10.66.66.66
 862    ...
 863    ftp fec0:6666:6666::193.233.7.65
 864    ...
 865  */
 866 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
 867                         unsigned short type,
 868                         const void *daddr, const void *saddr, unsigned int len)
 869 {
 870         struct ip_tunnel *t = netdev_priv(dev);
 871         struct iphdr *iph;
 872         struct gre_base_hdr *greh;
 873
 874         iph = skb_push(skb, t->hlen + sizeof(*iph));
 875         greh = (struct gre_base_hdr *)(iph+1);
 876         greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
 877         greh->protocol = htons(type);
 878
 879         memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
 880
 881         /* Set the source hardware address. */
 882         if (saddr)
 883                 memcpy(&iph->saddr, saddr, 4);
 884         if (daddr)
 885                 memcpy(&iph->daddr, daddr, 4);
 886         if (iph->daddr)
 887                 return t->hlen + sizeof(*iph);
 888
 889         return -(t->hlen + sizeof(*iph));
 890 }
 891
 892 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
 893 {
 894         const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
 895         memcpy(haddr, &iph->saddr, 4);
 896         return 4;
 897 }
 898
 899 static const struct header_ops ipgre_header_ops = {
 900         .create = ipgre_header,
 901         .parse  = ipgre_header_parse,
 902 };
 903
 904 #ifdef CONFIG_NET_IPGRE_BROADCAST
 905 static int ipgre_open(struct net_device *dev)
 906 {
 907         struct ip_tunnel *t = netdev_priv(dev);
 908
 909         if (ipv4_is_multicast(t->parms.iph.daddr)) {
 910                 struct flowi4 fl4;
 911                 struct rtable *rt;
 912
 913                 rt = ip_route_output_gre(t->net, &fl4,
 914                                          t->parms.iph.daddr,
 915                                          t->parms.iph.saddr,
 916                                          t->parms.o_key,
 917                                          RT_TOS(t->parms.iph.tos),
 918                                          t->parms.link);
 919                 if (IS_ERR(rt))
 920                         return -EADDRNOTAVAIL;
 921                 dev = rt->dst.dev;
 922                 ip_rt_put(rt);
 923                 if (!__in_dev_get_rtnl(dev))
 924                         return -EADDRNOTAVAIL;
 925                 t->mlink = dev->ifindex;
 926                 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
 927         }
 928         return 0;
 929 }
 930
 931 static int ipgre_close(struct net_device *dev)
 932 {
 933         struct ip_tunnel *t = netdev_priv(dev);
 934
 935         if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
 936                 struct in_device *in_dev;
 937                 in_dev = inetdev_by_index(t->net, t->mlink);
 938                 if (in_dev)
 939                         ip_mc_dec_group(in_dev, t->parms.iph.daddr);
 940         }
 941         return 0;
 942 }
 943 #endif
 944
 945 static const struct net_device_ops ipgre_netdev_ops = {
 946         .ndo_init               = ipgre_tunnel_init,
 947         .ndo_uninit             = ip_tunnel_uninit,
 948 #ifdef CONFIG_NET_IPGRE_BROADCAST
 949         .ndo_open               = ipgre_open,
 950         .ndo_stop               = ipgre_close,
 951 #endif
 952         .ndo_start_xmit         = ipgre_xmit,
 953         .ndo_do_ioctl           = ipgre_tunnel_ioctl,
 954         .ndo_change_mtu         = ip_tunnel_change_mtu,
 955         .ndo_get_stats64        = ip_tunnel_get_stats64,
 956         .ndo_get_iflink         = ip_tunnel_get_iflink,
 957 };
 958
 959 #define GRE_FEATURES (NETIF_F_SG |              \
 960                       NETIF_F_FRAGLIST |        \
 961                       NETIF_F_HIGHDMA |         \
 962                       NETIF_F_HW_CSUM)
 963
 964 static void ipgre_tunnel_setup(struct net_device *dev)
 965 {
 966         dev->netdev_ops         = &ipgre_netdev_ops;
 967         dev->type               = ARPHRD_IPGRE;
 968         ip_tunnel_setup(dev, ipgre_net_id);
 969 }
 970
 971 static void __gre_tunnel_init(struct net_device *dev)
 972 {
 973         struct ip_tunnel *tunnel;
 974         int t_hlen;
 975
 976         tunnel = netdev_priv(dev);
 977         tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
 978         tunnel->parms.iph.protocol = IPPROTO_GRE;
 979
 980         tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
 981
 982         t_hlen = tunnel->hlen + sizeof(struct iphdr);
 983
 984         dev->features           |= GRE_FEATURES;
 985         dev->hw_features        |= GRE_FEATURES;
 986
 987         if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
 988                 /* TCP offload with GRE SEQ is not supported, nor
 989                  * can we support 2 levels of outer headers requiring
 990                  * an update.
 991                  */
 992                 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
 993                     (tunnel->encap.type == TUNNEL_ENCAP_NONE)) {
 994                         dev->features    |= NETIF_F_GSO_SOFTWARE;
 995                         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
 996                 }
 997
 998                 /* Can use a lockless transmit, unless we generate
 999                  * output sequences
1000                  */
1001                 dev->features |= NETIF_F_LLTX;
1002         }
1003 }
1004
1005 static int ipgre_tunnel_init(struct net_device *dev)
1006 {
1007         struct ip_tunnel *tunnel = netdev_priv(dev);
1008         struct iphdr *iph = &tunnel->parms.iph;
1009
1010         __gre_tunnel_init(dev);
1011
1012         memcpy(dev->dev_addr, &iph->saddr, 4);
1013         memcpy(dev->broadcast, &iph->daddr, 4);
1014
1015         dev->flags              = IFF_NOARP;
1016         netif_keep_dst(dev);
1017         dev->addr_len           = 4;
1018
1019         if (iph->daddr && !tunnel->collect_md) {
1020 #ifdef CONFIG_NET_IPGRE_BROADCAST
1021                 if (ipv4_is_multicast(iph->daddr)) {
1022                         if (!iph->saddr)
1023                                 return -EINVAL;
1024                         dev->flags = IFF_BROADCAST;
1025                         dev->header_ops = &ipgre_header_ops;
1026                 }
1027 #endif
1028         } else if (!tunnel->collect_md) {
1029                 dev->header_ops = &ipgre_header_ops;
1030         }
1031
1032         return ip_tunnel_init(dev);
1033 }
1034
1035 static const struct gre_protocol ipgre_protocol = {
1036         .handler     = gre_rcv,
1037         .err_handler = gre_err,
1038 };
1039
1040 static int __net_init ipgre_init_net(struct net *net)
1041 {
1042         return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
1043 }
1044
1045 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net)
1046 {
1047         ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops);
1048 }
1049
1050 static struct pernet_operations ipgre_net_ops = {
1051         .init = ipgre_init_net,
1052         .exit_batch = ipgre_exit_batch_net,
1053         .id   = &ipgre_net_id,
1054         .size = sizeof(struct ip_tunnel_net),
1055 };
1056
1057 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1058                                  struct netlink_ext_ack *extack)
1059 {
1060         __be16 flags;
1061
1062         if (!data)
1063                 return 0;
1064
1065         flags = 0;
1066         if (data[IFLA_GRE_IFLAGS])
1067                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1068         if (data[IFLA_GRE_OFLAGS])
1069                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1070         if (flags & (GRE_VERSION|GRE_ROUTING))
1071                 return -EINVAL;
1072
1073         if (data[IFLA_GRE_COLLECT_METADATA] &&
1074             data[IFLA_GRE_ENCAP_TYPE] &&
1075             nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1076                 return -EINVAL;
1077
1078         return 0;
1079 }
1080
1081 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1082                               struct netlink_ext_ack *extack)
1083 {
1084         __be32 daddr;
1085
1086         if (tb[IFLA_ADDRESS]) {
1087                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1088                         return -EINVAL;
1089                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1090                         return -EADDRNOTAVAIL;
1091         }
1092
1093         if (!data)
1094                 goto out;
1095
1096         if (data[IFLA_GRE_REMOTE]) {
1097                 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1098                 if (!daddr)
1099                         return -EINVAL;
1100         }
1101
1102 out:
1103         return ipgre_tunnel_validate(tb, data, extack);
1104 }
1105
1106 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1107                            struct netlink_ext_ack *extack)
1108 {
1109         __be16 flags = 0;
1110         int ret;
1111
1112         if (!data)
1113                 return 0;
1114
1115         ret = ipgre_tap_validate(tb, data, extack);
1116         if (ret)
1117                 return ret;
1118
1119         /* ERSPAN should only have GRE sequence and key flag */
1120         if (data[IFLA_GRE_OFLAGS])
1121                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1122         if (data[IFLA_GRE_IFLAGS])
1123                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1124         if (!data[IFLA_GRE_COLLECT_METADATA] &&
1125             flags != (GRE_SEQ | GRE_KEY))
1126                 return -EINVAL;
1127
1128         /* ERSPAN Session ID only has 10-bit. Since we reuse
1129          * 32-bit key field as ID, check it's range.
1130          */
1131         if (data[IFLA_GRE_IKEY] &&
1132             (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1133                 return -EINVAL;
1134
1135         if (data[IFLA_GRE_OKEY] &&
1136             (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1137                 return -EINVAL;
1138
1139         return 0;
1140 }
1141
1142 static int ipgre_netlink_parms(struct net_device *dev,
1143                                 struct nlattr *data[],
1144                                 struct nlattr *tb[],
1145                                 struct ip_tunnel_parm *parms,
1146                                 __u32 *fwmark)
1147 {
1148         struct ip_tunnel *t = netdev_priv(dev);
1149
1150         memset(parms, 0, sizeof(*parms));
1151
1152         parms->iph.protocol = IPPROTO_GRE;
1153
1154         if (!data)
1155                 return 0;
1156
1157         if (data[IFLA_GRE_LINK])
1158                 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1159
1160         if (data[IFLA_GRE_IFLAGS])
1161                 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
1162
1163         if (data[IFLA_GRE_OFLAGS])
1164                 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
1165
1166         if (data[IFLA_GRE_IKEY])
1167                 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1168
1169         if (data[IFLA_GRE_OKEY])
1170                 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1171
1172         if (data[IFLA_GRE_LOCAL])
1173                 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1174
1175         if (data[IFLA_GRE_REMOTE])
1176                 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1177
1178         if (data[IFLA_GRE_TTL])
1179                 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1180
1181         if (data[IFLA_GRE_TOS])
1182                 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1183
1184         if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1185                 if (t->ignore_df)
1186                         return -EINVAL;
1187                 parms->iph.frag_off = htons(IP_DF);
1188         }
1189
1190         if (data[IFLA_GRE_COLLECT_METADATA]) {
1191                 t->collect_md = true;
1192                 if (dev->type == ARPHRD_IPGRE)
1193                         dev->type = ARPHRD_NONE;
1194         }
1195
1196         if (data[IFLA_GRE_IGNORE_DF]) {
1197                 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1198                   && (parms->iph.frag_off & htons(IP_DF)))
1199                         return -EINVAL;
1200                 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1201         }
1202
1203         if (data[IFLA_GRE_FWMARK])
1204                 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1205
1206         if (data[IFLA_GRE_ERSPAN_VER]) {
1207                 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]);
1208
1209                 if (t->erspan_ver != 1 && t->erspan_ver != 2)
1210                         return -EINVAL;
1211         }
1212
1213         if (t->erspan_ver == 1) {
1214                 if (data[IFLA_GRE_ERSPAN_INDEX]) {
1215                         t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1216                         if (t->index & ~INDEX_MASK)
1217                                 return -EINVAL;
1218                 }
1219         } else if (t->erspan_ver == 2) {
1220                 if (data[IFLA_GRE_ERSPAN_DIR]) {
1221                         t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]);
1222                         if (t->dir & ~(DIR_MASK >> DIR_OFFSET))
1223                                 return -EINVAL;
1224                 }
1225                 if (data[IFLA_GRE_ERSPAN_HWID]) {
1226                         t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]);
1227                         if (t->hwid & ~(HWID_MASK >> HWID_OFFSET))
1228                                 return -EINVAL;
1229                 }
1230         }
1231
1232         return 0;
1233 }
1234
1235 /* This function returns true when ENCAP attributes are present in the nl msg */
1236 static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1237                                       struct ip_tunnel_encap *ipencap)
1238 {
1239         bool ret = false;
1240
1241         memset(ipencap, 0, sizeof(*ipencap));
1242
1243         if (!data)
1244                 return ret;
1245
1246         if (data[IFLA_GRE_ENCAP_TYPE]) {
1247                 ret = true;
1248                 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1249         }
1250
1251         if (data[IFLA_GRE_ENCAP_FLAGS]) {
1252                 ret = true;
1253                 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1254         }
1255
1256         if (data[IFLA_GRE_ENCAP_SPORT]) {
1257                 ret = true;
1258                 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1259         }
1260
1261         if (data[IFLA_GRE_ENCAP_DPORT]) {
1262                 ret = true;
1263                 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1264         }
1265
1266         return ret;
1267 }
1268
1269 static int gre_tap_init(struct net_device *dev)
1270 {
1271         __gre_tunnel_init(dev);
1272         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1273         netif_keep_dst(dev);
1274
1275         return ip_tunnel_init(dev);
1276 }
1277
1278 static const struct net_device_ops gre_tap_netdev_ops = {
1279         .ndo_init               = gre_tap_init,
1280         .ndo_uninit             = ip_tunnel_uninit,
1281         .ndo_start_xmit         = gre_tap_xmit,
1282         .ndo_set_mac_address    = eth_mac_addr,
1283         .ndo_validate_addr      = eth_validate_addr,
1284         .ndo_change_mtu         = ip_tunnel_change_mtu,
1285         .ndo_get_stats64        = ip_tunnel_get_stats64,
1286         .ndo_get_iflink         = ip_tunnel_get_iflink,
1287         .ndo_fill_metadata_dst  = gre_fill_metadata_dst,
1288 };
1289
1290 static int erspan_tunnel_init(struct net_device *dev)
1291 {
1292         struct ip_tunnel *tunnel = netdev_priv(dev);
1293         int t_hlen;
1294
1295         tunnel->tun_hlen = 8;
1296         tunnel->parms.iph.protocol = IPPROTO_GRE;
1297         tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1298                        erspan_hdr_len(tunnel->erspan_ver);
1299         t_hlen = tunnel->hlen + sizeof(struct iphdr);
1300
1301         dev->features           |= GRE_FEATURES;
1302         dev->hw_features        |= GRE_FEATURES;
1303         dev->priv_flags         |= IFF_LIVE_ADDR_CHANGE;
1304         netif_keep_dst(dev);
1305
1306         return ip_tunnel_init(dev);
1307 }
1308
1309 static const struct net_device_ops erspan_netdev_ops = {
1310         .ndo_init               = erspan_tunnel_init,
1311         .ndo_uninit             = ip_tunnel_uninit,
1312         .ndo_start_xmit         = erspan_xmit,
1313         .ndo_set_mac_address    = eth_mac_addr,
1314         .ndo_validate_addr      = eth_validate_addr,
1315         .ndo_change_mtu         = ip_tunnel_change_mtu,
1316         .ndo_get_stats64        = ip_tunnel_get_stats64,
1317         .ndo_get_iflink         = ip_tunnel_get_iflink,
1318         .ndo_fill_metadata_dst  = gre_fill_metadata_dst,
1319 };
1320
1321 static void ipgre_tap_setup(struct net_device *dev)
1322 {
1323         ether_setup(dev);
1324         dev->max_mtu = 0;
1325         dev->netdev_ops = &gre_tap_netdev_ops;
1326         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1327         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1328         ip_tunnel_setup(dev, gre_tap_net_id);
1329 }
1330
1331 bool is_gretap_dev(const struct net_device *dev)
1332 {
1333         return dev->netdev_ops == &gre_tap_netdev_ops;
1334 }
1335 EXPORT_SYMBOL_GPL(is_gretap_dev);
1336
1337 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1338                          struct nlattr *tb[], struct nlattr *data[],
1339                          struct netlink_ext_ack *extack)
1340 {
1341         struct ip_tunnel_parm p;
1342         struct ip_tunnel_encap ipencap;
1343         __u32 fwmark = 0;
1344         int err;
1345
1346         if (ipgre_netlink_encap_parms(data, &ipencap)) {
1347                 struct ip_tunnel *t = netdev_priv(dev);
1348                 err = ip_tunnel_encap_setup(t, &ipencap);
1349
1350                 if (err < 0)
1351                         return err;
1352         }
1353
1354         err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1355         if (err < 0)
1356                 return err;
1357         return ip_tunnel_newlink(dev, tb, &p, fwmark);
1358 }
1359
1360 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1361                             struct nlattr *data[],
1362                             struct netlink_ext_ack *extack)
1363 {
1364         struct ip_tunnel *t = netdev_priv(dev);
1365         struct ip_tunnel_encap ipencap;
1366         __u32 fwmark = t->fwmark;
1367         struct ip_tunnel_parm p;
1368         int err;
1369
1370         if (ipgre_netlink_encap_parms(data, &ipencap)) {
1371                 err = ip_tunnel_encap_setup(t, &ipencap);
1372
1373                 if (err < 0)
1374                         return err;
1375         }
1376
1377         err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1378         if (err < 0)
1379                 return err;
1380
1381         err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1382         if (err < 0)
1383                 return err;
1384
1385         t->parms.i_flags = p.i_flags;
1386         t->parms.o_flags = p.o_flags;
1387
1388         if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
1389                 ipgre_link_update(dev, !tb[IFLA_MTU]);
1390
1391         return 0;
1392 }
1393
1394 static size_t ipgre_get_size(const struct net_device *dev)
1395 {
1396         return
1397                 /* IFLA_GRE_LINK */
1398                 nla_total_size(4) +
1399                 /* IFLA_GRE_IFLAGS */
1400                 nla_total_size(2) +
1401                 /* IFLA_GRE_OFLAGS */
1402                 nla_total_size(2) +
1403                 /* IFLA_GRE_IKEY */
1404                 nla_total_size(4) +
1405                 /* IFLA_GRE_OKEY */
1406                 nla_total_size(4) +
1407                 /* IFLA_GRE_LOCAL */
1408                 nla_total_size(4) +
1409                 /* IFLA_GRE_REMOTE */
1410                 nla_total_size(4) +
1411                 /* IFLA_GRE_TTL */
1412                 nla_total_size(1) +
1413                 /* IFLA_GRE_TOS */
1414                 nla_total_size(1) +
1415                 /* IFLA_GRE_PMTUDISC */
1416                 nla_total_size(1) +
1417                 /* IFLA_GRE_ENCAP_TYPE */
1418                 nla_total_size(2) +
1419                 /* IFLA_GRE_ENCAP_FLAGS */
1420                 nla_total_size(2) +
1421                 /* IFLA_GRE_ENCAP_SPORT */
1422                 nla_total_size(2) +
1423                 /* IFLA_GRE_ENCAP_DPORT */
1424                 nla_total_size(2) +
1425                 /* IFLA_GRE_COLLECT_METADATA */
1426                 nla_total_size(0) +
1427                 /* IFLA_GRE_IGNORE_DF */
1428                 nla_total_size(1) +
1429                 /* IFLA_GRE_FWMARK */
1430                 nla_total_size(4) +
1431                 /* IFLA_GRE_ERSPAN_INDEX */
1432                 nla_total_size(4) +
1433                 /* IFLA_GRE_ERSPAN_VER */
1434                 nla_total_size(1) +
1435                 /* IFLA_GRE_ERSPAN_DIR */
1436                 nla_total_size(1) +
1437                 /* IFLA_GRE_ERSPAN_HWID */
1438                 nla_total_size(2) +
1439                 0;
1440 }
1441
1442 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1443 {
1444         struct ip_tunnel *t = netdev_priv(dev);
1445         struct ip_tunnel_parm *p = &t->parms;
1446
1447         if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1448             nla_put_be16(skb, IFLA_GRE_IFLAGS,
1449                          gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1450             nla_put_be16(skb, IFLA_GRE_OFLAGS,
1451                          gre_tnl_flags_to_gre_flags(p->o_flags)) ||
1452             nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1453             nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1454             nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1455             nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1456             nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1457             nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1458             nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1459                        !!(p->iph.frag_off & htons(IP_DF))) ||
1460             nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1461                 goto nla_put_failure;
1462
1463         if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1464                         t->encap.type) ||
1465             nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1466                          t->encap.sport) ||
1467             nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1468                          t->encap.dport) ||
1469             nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1470                         t->encap.flags))
1471                 goto nla_put_failure;
1472
1473         if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1474                 goto nla_put_failure;
1475
1476         if (t->collect_md) {
1477                 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1478                         goto nla_put_failure;
1479         }
1480
1481         if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1482                 goto nla_put_failure;
1483
1484         if (t->erspan_ver == 1) {
1485                 if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1486                         goto nla_put_failure;
1487         } else if (t->erspan_ver == 2) {
1488                 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1489                         goto nla_put_failure;
1490                 if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1491                         goto nla_put_failure;
1492         }
1493
1494         return 0;
1495
1496 nla_put_failure:
1497         return -EMSGSIZE;
1498 }
1499
1500 static void erspan_setup(struct net_device *dev)
1501 {
1502         ether_setup(dev);
1503         dev->netdev_ops = &erspan_netdev_ops;
1504         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1505         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1506         ip_tunnel_setup(dev, erspan_net_id);
1507 }
1508
1509 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1510         [IFLA_GRE_LINK]         = { .type = NLA_U32 },
1511         [IFLA_GRE_IFLAGS]       = { .type = NLA_U16 },
1512         [IFLA_GRE_OFLAGS]       = { .type = NLA_U16 },
1513         [IFLA_GRE_IKEY]         = { .type = NLA_U32 },
1514         [IFLA_GRE_OKEY]         = { .type = NLA_U32 },
1515         [IFLA_GRE_LOCAL]        = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1516         [IFLA_GRE_REMOTE]       = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1517         [IFLA_GRE_TTL]          = { .type = NLA_U8 },
1518         [IFLA_GRE_TOS]          = { .type = NLA_U8 },
1519         [IFLA_GRE_PMTUDISC]     = { .type = NLA_U8 },
1520         [IFLA_GRE_ENCAP_TYPE]   = { .type = NLA_U16 },
1521         [IFLA_GRE_ENCAP_FLAGS]  = { .type = NLA_U16 },
1522         [IFLA_GRE_ENCAP_SPORT]  = { .type = NLA_U16 },
1523         [IFLA_GRE_ENCAP_DPORT]  = { .type = NLA_U16 },
1524         [IFLA_GRE_COLLECT_METADATA]     = { .type = NLA_FLAG },
1525         [IFLA_GRE_IGNORE_DF]    = { .type = NLA_U8 },
1526         [IFLA_GRE_FWMARK]       = { .type = NLA_U32 },
1527         [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 },
1528         [IFLA_GRE_ERSPAN_VER]   = { .type = NLA_U8 },
1529         [IFLA_GRE_ERSPAN_DIR]   = { .type = NLA_U8 },
1530         [IFLA_GRE_ERSPAN_HWID]  = { .type = NLA_U16 },
1531 };
1532
1533 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1534         .kind           = "gre",
1535         .maxtype        = IFLA_GRE_MAX,
1536         .policy         = ipgre_policy,
1537         .priv_size      = sizeof(struct ip_tunnel),
1538         .setup          = ipgre_tunnel_setup,
1539         .validate       = ipgre_tunnel_validate,
1540         .newlink        = ipgre_newlink,
1541         .changelink     = ipgre_changelink,
1542         .dellink        = ip_tunnel_dellink,
1543         .get_size       = ipgre_get_size,
1544         .fill_info      = ipgre_fill_info,
1545         .get_link_net   = ip_tunnel_get_link_net,
1546 };
1547
1548 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1549         .kind           = "gretap",
1550         .maxtype        = IFLA_GRE_MAX,
1551         .policy         = ipgre_policy,
1552         .priv_size      = sizeof(struct ip_tunnel),
1553         .setup          = ipgre_tap_setup,
1554         .validate       = ipgre_tap_validate,
1555         .newlink        = ipgre_newlink,
1556         .changelink     = ipgre_changelink,
1557         .dellink        = ip_tunnel_dellink,
1558         .get_size       = ipgre_get_size,
1559         .fill_info      = ipgre_fill_info,
1560         .get_link_net   = ip_tunnel_get_link_net,
1561 };
1562
1563 static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1564         .kind           = "erspan",
1565         .maxtype        = IFLA_GRE_MAX,
1566         .policy         = ipgre_policy,
1567         .priv_size      = sizeof(struct ip_tunnel),
1568         .setup          = erspan_setup,
1569         .validate       = erspan_validate,
1570         .newlink        = ipgre_newlink,
1571         .changelink     = ipgre_changelink,
1572         .dellink        = ip_tunnel_dellink,
1573         .get_size       = ipgre_get_size,
1574         .fill_info      = ipgre_fill_info,
1575         .get_link_net   = ip_tunnel_get_link_net,
1576 };
1577
1578 struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1579                                         u8 name_assign_type)
1580 {
1581         struct nlattr *tb[IFLA_MAX + 1];
1582         struct net_device *dev;
1583         LIST_HEAD(list_kill);
1584         struct ip_tunnel *t;
1585         int err;
1586
1587         memset(&tb, 0, sizeof(tb));
1588
1589         dev = rtnl_create_link(net, name, name_assign_type,
1590                                &ipgre_tap_ops, tb);
1591         if (IS_ERR(dev))
1592                 return dev;
1593
1594         /* Configure flow based GRE device. */
1595         t = netdev_priv(dev);
1596         t->collect_md = true;
1597
1598         err = ipgre_newlink(net, dev, tb, NULL, NULL);
1599         if (err < 0) {
1600                 free_netdev(dev);
1601                 return ERR_PTR(err);
1602         }
1603
1604         /* openvswitch users expect packet sizes to be unrestricted,
1605          * so set the largest MTU we can.
1606          */
1607         err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1608         if (err)
1609                 goto out;
1610
1611         err = rtnl_configure_link(dev, NULL);
1612         if (err < 0)
1613                 goto out;
1614
1615         return dev;
1616 out:
1617         ip_tunnel_dellink(dev, &list_kill);
1618         unregister_netdevice_many(&list_kill);
1619         return ERR_PTR(err);
1620 }
1621 EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1622
1623 static int __net_init ipgre_tap_init_net(struct net *net)
1624 {
1625         return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1626 }
1627
1628 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net)
1629 {
1630         ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops);
1631 }
1632
1633 static struct pernet_operations ipgre_tap_net_ops = {
1634         .init = ipgre_tap_init_net,
1635         .exit_batch = ipgre_tap_exit_batch_net,
1636         .id   = &gre_tap_net_id,
1637         .size = sizeof(struct ip_tunnel_net),
1638 };
1639
1640 static int __net_init erspan_init_net(struct net *net)
1641 {
1642         return ip_tunnel_init_net(net, erspan_net_id,
1643                                   &erspan_link_ops, "erspan0");
1644 }
1645
1646 static void __net_exit erspan_exit_batch_net(struct list_head *net_list)
1647 {
1648         ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops);
1649 }
1650
1651 static struct pernet_operations erspan_net_ops = {
1652         .init = erspan_init_net,
1653         .exit_batch = erspan_exit_batch_net,
1654         .id   = &erspan_net_id,
1655         .size = sizeof(struct ip_tunnel_net),
1656 };
1657
1658 static int __init ipgre_init(void)
1659 {
1660         int err;
1661
1662         pr_info("GRE over IPv4 tunneling driver\n");
1663
1664         err = register_pernet_device(&ipgre_net_ops);
1665         if (err < 0)
1666                 return err;
1667
1668         err = register_pernet_device(&ipgre_tap_net_ops);
1669         if (err < 0)
1670                 goto pnet_tap_failed;
1671
1672         err = register_pernet_device(&erspan_net_ops);
1673         if (err < 0)
1674                 goto pnet_erspan_failed;
1675
1676         err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1677         if (err < 0) {
1678                 pr_info("%s: can't add protocol\n", __func__);
1679                 goto add_proto_failed;
1680         }
1681
1682         err = rtnl_link_register(&ipgre_link_ops);
1683         if (err < 0)
1684                 goto rtnl_link_failed;
1685
1686         err = rtnl_link_register(&ipgre_tap_ops);
1687         if (err < 0)
1688                 goto tap_ops_failed;
1689
1690         err = rtnl_link_register(&erspan_link_ops);
1691         if (err < 0)
1692                 goto erspan_link_failed;
1693
1694         return 0;
1695
1696 erspan_link_failed:
1697         rtnl_link_unregister(&ipgre_tap_ops);
1698 tap_ops_failed:
1699         rtnl_link_unregister(&ipgre_link_ops);
1700 rtnl_link_failed:
1701         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1702 add_proto_failed:
1703         unregister_pernet_device(&erspan_net_ops);
1704 pnet_erspan_failed:
1705         unregister_pernet_device(&ipgre_tap_net_ops);
1706 pnet_tap_failed:
1707         unregister_pernet_device(&ipgre_net_ops);
1708         return err;
1709 }
1710
1711 static void __exit ipgre_fini(void)
1712 {
1713         rtnl_link_unregister(&ipgre_tap_ops);
1714         rtnl_link_unregister(&ipgre_link_ops);
1715         rtnl_link_unregister(&erspan_link_ops);
1716         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1717         unregister_pernet_device(&ipgre_tap_net_ops);
1718         unregister_pernet_device(&ipgre_net_ops);
1719         unregister_pernet_device(&erspan_net_ops);
1720 }
1721
1722 module_init(ipgre_init);
1723 module_exit(ipgre_fini);
1724 MODULE_LICENSE("GPL");
1725 MODULE_ALIAS_RTNL_LINK("gre");
1726 MODULE_ALIAS_RTNL_LINK("gretap");
1727 MODULE_ALIAS_RTNL_LINK("erspan");
1728 MODULE_ALIAS_NETDEV("gre0");
1729 MODULE_ALIAS_NETDEV("gretap0");
1730 MODULE_ALIAS_NETDEV("erspan0");