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         int nhoff;
 582         int thoff;
 583
 584         tun_info = skb_tunnel_info(skb);
 585         if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) ||
 586                      ip_tunnel_info_af(tun_info) != AF_INET))
 587                 goto err_free_skb;
 588
 589         key = &tun_info->key;
 590         md = ip_tunnel_info_opts(tun_info);
 591         if (!md)
 592                 goto err_free_rt;
 593
 594         /* ERSPAN has fixed 8 byte GRE header */
 595         version = md->version;
 596         tunnel_hlen = 8 + erspan_hdr_len(version);
 597
 598         rt = prepare_fb_xmit(skb, dev, &fl, tunnel_hlen);
 599         if (!rt)
 600                 return;
 601
 602         if (gre_handle_offloads(skb, false))
 603                 goto err_free_rt;
 604
 605         if (skb->len > dev->mtu + dev->hard_header_len) {
 606                 pskb_trim(skb, dev->mtu + dev->hard_header_len);
 607                 truncate = true;
 608         }
 609
 610         nhoff = skb_network_header(skb) - skb_mac_header(skb);
 611         if (skb->protocol == htons(ETH_P_IP) &&
 612             (ntohs(ip_hdr(skb)->tot_len) > skb->len - nhoff))
 613                 truncate = true;
 614
 615         thoff = skb_transport_header(skb) - skb_mac_header(skb);
 616         if (skb->protocol == htons(ETH_P_IPV6) &&
 617             (ntohs(ipv6_hdr(skb)->payload_len) > skb->len - thoff))
 618                 truncate = true;
 619
 620         if (version == 1) {
 621                 erspan_build_header(skb, ntohl(tunnel_id_to_key32(key->tun_id)),
 622                                     ntohl(md->u.index), truncate, true);
 623         } else if (version == 2) {
 624                 erspan_build_header_v2(skb,
 625                                        ntohl(tunnel_id_to_key32(key->tun_id)),
 626                                        md->u.md2.dir,
 627                                        get_hwid(&md->u.md2),
 628                                        truncate, true);
 629         } else {
 630                 goto err_free_rt;
 631         }
 632
 633         gre_build_header(skb, 8, TUNNEL_SEQ,
 634                          htons(ETH_P_ERSPAN), 0, htonl(tunnel->o_seqno++));
 635
 636         df = key->tun_flags & TUNNEL_DONT_FRAGMENT ?  htons(IP_DF) : 0;
 637
 638         iptunnel_xmit(skb->sk, rt, skb, fl.saddr, key->u.ipv4.dst, IPPROTO_GRE,
 639                       key->tos, key->ttl, df, false);
 640         return;
 641
 642 err_free_rt:
 643         ip_rt_put(rt);
 644 err_free_skb:
 645         kfree_skb(skb);
 646         dev->stats.tx_dropped++;
 647 }
 648
 649 static int gre_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
 650 {
 651         struct ip_tunnel_info *info = skb_tunnel_info(skb);
 652         struct rtable *rt;
 653         struct flowi4 fl4;
 654
 655         if (ip_tunnel_info_af(info) != AF_INET)
 656                 return -EINVAL;
 657
 658         rt = gre_get_rt(skb, dev, &fl4, &info->key);
 659         if (IS_ERR(rt))
 660                 return PTR_ERR(rt);
 661
 662         ip_rt_put(rt);
 663         info->key.u.ipv4.src = fl4.saddr;
 664         return 0;
 665 }
 666
 667 static netdev_tx_t ipgre_xmit(struct sk_buff *skb,
 668                               struct net_device *dev)
 669 {
 670         struct ip_tunnel *tunnel = netdev_priv(dev);
 671         const struct iphdr *tnl_params;
 672
 673         if (tunnel->collect_md) {
 674                 gre_fb_xmit(skb, dev, skb->protocol);
 675                 return NETDEV_TX_OK;
 676         }
 677
 678         if (dev->header_ops) {
 679                 /* Need space for new headers */
 680                 if (skb_cow_head(skb, dev->needed_headroom -
 681                                       (tunnel->hlen + sizeof(struct iphdr))))
 682                         goto free_skb;
 683
 684                 tnl_params = (const struct iphdr *)skb->data;
 685
 686                 /* Pull skb since ip_tunnel_xmit() needs skb->data pointing
 687                  * to gre header.
 688                  */
 689                 skb_pull(skb, tunnel->hlen + sizeof(struct iphdr));
 690                 skb_reset_mac_header(skb);
 691         } else {
 692                 if (skb_cow_head(skb, dev->needed_headroom))
 693                         goto free_skb;
 694
 695                 tnl_params = &tunnel->parms.iph;
 696         }
 697
 698         if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
 699                 goto free_skb;
 700
 701         __gre_xmit(skb, dev, tnl_params, skb->protocol);
 702         return NETDEV_TX_OK;
 703
 704 free_skb:
 705         kfree_skb(skb);
 706         dev->stats.tx_dropped++;
 707         return NETDEV_TX_OK;
 708 }
 709
 710 static netdev_tx_t erspan_xmit(struct sk_buff *skb,
 711                                struct net_device *dev)
 712 {
 713         struct ip_tunnel *tunnel = netdev_priv(dev);
 714         bool truncate = false;
 715
 716         if (tunnel->collect_md) {
 717                 erspan_fb_xmit(skb, dev, skb->protocol);
 718                 return NETDEV_TX_OK;
 719         }
 720
 721         if (gre_handle_offloads(skb, false))
 722                 goto free_skb;
 723
 724         if (skb_cow_head(skb, dev->needed_headroom))
 725                 goto free_skb;
 726
 727         if (skb->len > dev->mtu + dev->hard_header_len) {
 728                 pskb_trim(skb, dev->mtu + dev->hard_header_len);
 729                 truncate = true;
 730         }
 731
 732         /* Push ERSPAN header */
 733         if (tunnel->erspan_ver == 1)
 734                 erspan_build_header(skb, ntohl(tunnel->parms.o_key),
 735                                     tunnel->index,
 736                                     truncate, true);
 737         else if (tunnel->erspan_ver == 2)
 738                 erspan_build_header_v2(skb, ntohl(tunnel->parms.o_key),
 739                                        tunnel->dir, tunnel->hwid,
 740                                        truncate, true);
 741         else
 742                 goto free_skb;
 743
 744         tunnel->parms.o_flags &= ~TUNNEL_KEY;
 745         __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_ERSPAN));
 746         return NETDEV_TX_OK;
 747
 748 free_skb:
 749         kfree_skb(skb);
 750         dev->stats.tx_dropped++;
 751         return NETDEV_TX_OK;
 752 }
 753
 754 static netdev_tx_t gre_tap_xmit(struct sk_buff *skb,
 755                                 struct net_device *dev)
 756 {
 757         struct ip_tunnel *tunnel = netdev_priv(dev);
 758
 759         if (tunnel->collect_md) {
 760                 gre_fb_xmit(skb, dev, htons(ETH_P_TEB));
 761                 return NETDEV_TX_OK;
 762         }
 763
 764         if (gre_handle_offloads(skb, !!(tunnel->parms.o_flags & TUNNEL_CSUM)))
 765                 goto free_skb;
 766
 767         if (skb_cow_head(skb, dev->needed_headroom))
 768                 goto free_skb;
 769
 770         __gre_xmit(skb, dev, &tunnel->parms.iph, htons(ETH_P_TEB));
 771         return NETDEV_TX_OK;
 772
 773 free_skb:
 774         kfree_skb(skb);
 775         dev->stats.tx_dropped++;
 776         return NETDEV_TX_OK;
 777 }
 778
 779 static void ipgre_link_update(struct net_device *dev, bool set_mtu)
 780 {
 781         struct ip_tunnel *tunnel = netdev_priv(dev);
 782         int len;
 783
 784         len = tunnel->tun_hlen;
 785         tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
 786         len = tunnel->tun_hlen - len;
 787         tunnel->hlen = tunnel->hlen + len;
 788
 789         dev->needed_headroom = dev->needed_headroom + len;
 790         if (set_mtu)
 791                 dev->mtu = max_t(int, dev->mtu - len, 68);
 792
 793         if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
 794                 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
 795                     tunnel->encap.type == TUNNEL_ENCAP_NONE) {
 796                         dev->features |= NETIF_F_GSO_SOFTWARE;
 797                         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
 798                 } else {
 799                         dev->features &= ~NETIF_F_GSO_SOFTWARE;
 800                         dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
 801                 }
 802                 dev->features |= NETIF_F_LLTX;
 803         } else {
 804                 dev->hw_features &= ~NETIF_F_GSO_SOFTWARE;
 805                 dev->features &= ~(NETIF_F_LLTX | NETIF_F_GSO_SOFTWARE);
 806         }
 807 }
 808
 809 static int ipgre_tunnel_ioctl(struct net_device *dev,
 810                               struct ifreq *ifr, int cmd)
 811 {
 812         struct ip_tunnel_parm p;
 813         int err;
 814
 815         if (copy_from_user(&p, ifr->ifr_ifru.ifru_data, sizeof(p)))
 816                 return -EFAULT;
 817
 818         if (cmd == SIOCADDTUNNEL || cmd == SIOCCHGTUNNEL) {
 819                 if (p.iph.version != 4 || p.iph.protocol != IPPROTO_GRE ||
 820                     p.iph.ihl != 5 || (p.iph.frag_off & htons(~IP_DF)) ||
 821                     ((p.i_flags | p.o_flags) & (GRE_VERSION | GRE_ROUTING)))
 822                         return -EINVAL;
 823         }
 824
 825         p.i_flags = gre_flags_to_tnl_flags(p.i_flags);
 826         p.o_flags = gre_flags_to_tnl_flags(p.o_flags);
 827
 828         err = ip_tunnel_ioctl(dev, &p, cmd);
 829         if (err)
 830                 return err;
 831
 832         if (cmd == SIOCCHGTUNNEL) {
 833                 struct ip_tunnel *t = netdev_priv(dev);
 834
 835                 t->parms.i_flags = p.i_flags;
 836                 t->parms.o_flags = p.o_flags;
 837
 838                 if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
 839                         ipgre_link_update(dev, true);
 840         }
 841
 842         p.i_flags = gre_tnl_flags_to_gre_flags(p.i_flags);
 843         p.o_flags = gre_tnl_flags_to_gre_flags(p.o_flags);
 844
 845         if (copy_to_user(ifr->ifr_ifru.ifru_data, &p, sizeof(p)))
 846                 return -EFAULT;
 847
 848         return 0;
 849 }
 850
 851 /* Nice toy. Unfortunately, useless in real life :-)
 852    It allows to construct virtual multiprotocol broadcast "LAN"
 853    over the Internet, provided multicast routing is tuned.
 854
 855
 856    I have no idea was this bicycle invented before me,
 857    so that I had to set ARPHRD_IPGRE to a random value.
 858    I have an impression, that Cisco could make something similar,
 859    but this feature is apparently missing in IOS<=11.2(8).
 860
 861    I set up 10.66.66/24 and fec0:6666:6666::0/96 as virtual networks
 862    with broadcast 224.66.66.66. If you have access to mbone, play with me :-)
 863
 864    ping -t 255 224.66.66.66
 865
 866    If nobody answers, mbone does not work.
 867
 868    ip tunnel add Universe mode gre remote 224.66.66.66 local <Your_real_addr> ttl 255
 869    ip addr add 10.66.66.<somewhat>/24 dev Universe
 870    ifconfig Universe up
 871    ifconfig Universe add fe80::<Your_real_addr>/10
 872    ifconfig Universe add fec0:6666:6666::<Your_real_addr>/96
 873    ftp 10.66.66.66
 874    ...
 875    ftp fec0:6666:6666::193.233.7.65
 876    ...
 877  */
 878 static int ipgre_header(struct sk_buff *skb, struct net_device *dev,
 879                         unsigned short type,
 880                         const void *daddr, const void *saddr, unsigned int len)
 881 {
 882         struct ip_tunnel *t = netdev_priv(dev);
 883         struct iphdr *iph;
 884         struct gre_base_hdr *greh;
 885
 886         iph = skb_push(skb, t->hlen + sizeof(*iph));
 887         greh = (struct gre_base_hdr *)(iph+1);
 888         greh->flags = gre_tnl_flags_to_gre_flags(t->parms.o_flags);
 889         greh->protocol = htons(type);
 890
 891         memcpy(iph, &t->parms.iph, sizeof(struct iphdr));
 892
 893         /* Set the source hardware address. */
 894         if (saddr)
 895                 memcpy(&iph->saddr, saddr, 4);
 896         if (daddr)
 897                 memcpy(&iph->daddr, daddr, 4);
 898         if (iph->daddr)
 899                 return t->hlen + sizeof(*iph);
 900
 901         return -(t->hlen + sizeof(*iph));
 902 }
 903
 904 static int ipgre_header_parse(const struct sk_buff *skb, unsigned char *haddr)
 905 {
 906         const struct iphdr *iph = (const struct iphdr *) skb_mac_header(skb);
 907         memcpy(haddr, &iph->saddr, 4);
 908         return 4;
 909 }
 910
 911 static const struct header_ops ipgre_header_ops = {
 912         .create = ipgre_header,
 913         .parse  = ipgre_header_parse,
 914 };
 915
 916 #ifdef CONFIG_NET_IPGRE_BROADCAST
 917 static int ipgre_open(struct net_device *dev)
 918 {
 919         struct ip_tunnel *t = netdev_priv(dev);
 920
 921         if (ipv4_is_multicast(t->parms.iph.daddr)) {
 922                 struct flowi4 fl4;
 923                 struct rtable *rt;
 924
 925                 rt = ip_route_output_gre(t->net, &fl4,
 926                                          t->parms.iph.daddr,
 927                                          t->parms.iph.saddr,
 928                                          t->parms.o_key,
 929                                          RT_TOS(t->parms.iph.tos),
 930                                          t->parms.link);
 931                 if (IS_ERR(rt))
 932                         return -EADDRNOTAVAIL;
 933                 dev = rt->dst.dev;
 934                 ip_rt_put(rt);
 935                 if (!__in_dev_get_rtnl(dev))
 936                         return -EADDRNOTAVAIL;
 937                 t->mlink = dev->ifindex;
 938                 ip_mc_inc_group(__in_dev_get_rtnl(dev), t->parms.iph.daddr);
 939         }
 940         return 0;
 941 }
 942
 943 static int ipgre_close(struct net_device *dev)
 944 {
 945         struct ip_tunnel *t = netdev_priv(dev);
 946
 947         if (ipv4_is_multicast(t->parms.iph.daddr) && t->mlink) {
 948                 struct in_device *in_dev;
 949                 in_dev = inetdev_by_index(t->net, t->mlink);
 950                 if (in_dev)
 951                         ip_mc_dec_group(in_dev, t->parms.iph.daddr);
 952         }
 953         return 0;
 954 }
 955 #endif
 956
 957 static const struct net_device_ops ipgre_netdev_ops = {
 958         .ndo_init               = ipgre_tunnel_init,
 959         .ndo_uninit             = ip_tunnel_uninit,
 960 #ifdef CONFIG_NET_IPGRE_BROADCAST
 961         .ndo_open               = ipgre_open,
 962         .ndo_stop               = ipgre_close,
 963 #endif
 964         .ndo_start_xmit         = ipgre_xmit,
 965         .ndo_do_ioctl           = ipgre_tunnel_ioctl,
 966         .ndo_change_mtu         = ip_tunnel_change_mtu,
 967         .ndo_get_stats64        = ip_tunnel_get_stats64,
 968         .ndo_get_iflink         = ip_tunnel_get_iflink,
 969 };
 970
 971 #define GRE_FEATURES (NETIF_F_SG |              \
 972                       NETIF_F_FRAGLIST |        \
 973                       NETIF_F_HIGHDMA |         \
 974                       NETIF_F_HW_CSUM)
 975
 976 static void ipgre_tunnel_setup(struct net_device *dev)
 977 {
 978         dev->netdev_ops         = &ipgre_netdev_ops;
 979         dev->type               = ARPHRD_IPGRE;
 980         ip_tunnel_setup(dev, ipgre_net_id);
 981 }
 982
 983 static void __gre_tunnel_init(struct net_device *dev)
 984 {
 985         struct ip_tunnel *tunnel;
 986         int t_hlen;
 987
 988         tunnel = netdev_priv(dev);
 989         tunnel->tun_hlen = gre_calc_hlen(tunnel->parms.o_flags);
 990         tunnel->parms.iph.protocol = IPPROTO_GRE;
 991
 992         tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen;
 993
 994         t_hlen = tunnel->hlen + sizeof(struct iphdr);
 995
 996         dev->features           |= GRE_FEATURES;
 997         dev->hw_features        |= GRE_FEATURES;
 998
 999         if (!(tunnel->parms.o_flags & TUNNEL_SEQ)) {
1000                 /* TCP offload with GRE SEQ is not supported, nor
1001                  * can we support 2 levels of outer headers requiring
1002                  * an update.
1003                  */
1004                 if (!(tunnel->parms.o_flags & TUNNEL_CSUM) ||
1005                     (tunnel->encap.type == TUNNEL_ENCAP_NONE)) {
1006                         dev->features    |= NETIF_F_GSO_SOFTWARE;
1007                         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
1008                 }
1009
1010                 /* Can use a lockless transmit, unless we generate
1011                  * output sequences
1012                  */
1013                 dev->features |= NETIF_F_LLTX;
1014         }
1015 }
1016
1017 static int ipgre_tunnel_init(struct net_device *dev)
1018 {
1019         struct ip_tunnel *tunnel = netdev_priv(dev);
1020         struct iphdr *iph = &tunnel->parms.iph;
1021
1022         __gre_tunnel_init(dev);
1023
1024         memcpy(dev->dev_addr, &iph->saddr, 4);
1025         memcpy(dev->broadcast, &iph->daddr, 4);
1026
1027         dev->flags              = IFF_NOARP;
1028         netif_keep_dst(dev);
1029         dev->addr_len           = 4;
1030
1031         if (iph->daddr && !tunnel->collect_md) {
1032 #ifdef CONFIG_NET_IPGRE_BROADCAST
1033                 if (ipv4_is_multicast(iph->daddr)) {
1034                         if (!iph->saddr)
1035                                 return -EINVAL;
1036                         dev->flags = IFF_BROADCAST;
1037                         dev->header_ops = &ipgre_header_ops;
1038                 }
1039 #endif
1040         } else if (!tunnel->collect_md) {
1041                 dev->header_ops = &ipgre_header_ops;
1042         }
1043
1044         return ip_tunnel_init(dev);
1045 }
1046
1047 static const struct gre_protocol ipgre_protocol = {
1048         .handler     = gre_rcv,
1049         .err_handler = gre_err,
1050 };
1051
1052 static int __net_init ipgre_init_net(struct net *net)
1053 {
1054         return ip_tunnel_init_net(net, ipgre_net_id, &ipgre_link_ops, NULL);
1055 }
1056
1057 static void __net_exit ipgre_exit_batch_net(struct list_head *list_net)
1058 {
1059         ip_tunnel_delete_nets(list_net, ipgre_net_id, &ipgre_link_ops);
1060 }
1061
1062 static struct pernet_operations ipgre_net_ops = {
1063         .init = ipgre_init_net,
1064         .exit_batch = ipgre_exit_batch_net,
1065         .id   = &ipgre_net_id,
1066         .size = sizeof(struct ip_tunnel_net),
1067 };
1068
1069 static int ipgre_tunnel_validate(struct nlattr *tb[], struct nlattr *data[],
1070                                  struct netlink_ext_ack *extack)
1071 {
1072         __be16 flags;
1073
1074         if (!data)
1075                 return 0;
1076
1077         flags = 0;
1078         if (data[IFLA_GRE_IFLAGS])
1079                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1080         if (data[IFLA_GRE_OFLAGS])
1081                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1082         if (flags & (GRE_VERSION|GRE_ROUTING))
1083                 return -EINVAL;
1084
1085         if (data[IFLA_GRE_COLLECT_METADATA] &&
1086             data[IFLA_GRE_ENCAP_TYPE] &&
1087             nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]) != TUNNEL_ENCAP_NONE)
1088                 return -EINVAL;
1089
1090         return 0;
1091 }
1092
1093 static int ipgre_tap_validate(struct nlattr *tb[], struct nlattr *data[],
1094                               struct netlink_ext_ack *extack)
1095 {
1096         __be32 daddr;
1097
1098         if (tb[IFLA_ADDRESS]) {
1099                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
1100                         return -EINVAL;
1101                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
1102                         return -EADDRNOTAVAIL;
1103         }
1104
1105         if (!data)
1106                 goto out;
1107
1108         if (data[IFLA_GRE_REMOTE]) {
1109                 memcpy(&daddr, nla_data(data[IFLA_GRE_REMOTE]), 4);
1110                 if (!daddr)
1111                         return -EINVAL;
1112         }
1113
1114 out:
1115         return ipgre_tunnel_validate(tb, data, extack);
1116 }
1117
1118 static int erspan_validate(struct nlattr *tb[], struct nlattr *data[],
1119                            struct netlink_ext_ack *extack)
1120 {
1121         __be16 flags = 0;
1122         int ret;
1123
1124         if (!data)
1125                 return 0;
1126
1127         ret = ipgre_tap_validate(tb, data, extack);
1128         if (ret)
1129                 return ret;
1130
1131         /* ERSPAN should only have GRE sequence and key flag */
1132         if (data[IFLA_GRE_OFLAGS])
1133                 flags |= nla_get_be16(data[IFLA_GRE_OFLAGS]);
1134         if (data[IFLA_GRE_IFLAGS])
1135                 flags |= nla_get_be16(data[IFLA_GRE_IFLAGS]);
1136         if (!data[IFLA_GRE_COLLECT_METADATA] &&
1137             flags != (GRE_SEQ | GRE_KEY))
1138                 return -EINVAL;
1139
1140         /* ERSPAN Session ID only has 10-bit. Since we reuse
1141          * 32-bit key field as ID, check it's range.
1142          */
1143         if (data[IFLA_GRE_IKEY] &&
1144             (ntohl(nla_get_be32(data[IFLA_GRE_IKEY])) & ~ID_MASK))
1145                 return -EINVAL;
1146
1147         if (data[IFLA_GRE_OKEY] &&
1148             (ntohl(nla_get_be32(data[IFLA_GRE_OKEY])) & ~ID_MASK))
1149                 return -EINVAL;
1150
1151         return 0;
1152 }
1153
1154 static int ipgre_netlink_parms(struct net_device *dev,
1155                                 struct nlattr *data[],
1156                                 struct nlattr *tb[],
1157                                 struct ip_tunnel_parm *parms,
1158                                 __u32 *fwmark)
1159 {
1160         struct ip_tunnel *t = netdev_priv(dev);
1161
1162         memset(parms, 0, sizeof(*parms));
1163
1164         parms->iph.protocol = IPPROTO_GRE;
1165
1166         if (!data)
1167                 return 0;
1168
1169         if (data[IFLA_GRE_LINK])
1170                 parms->link = nla_get_u32(data[IFLA_GRE_LINK]);
1171
1172         if (data[IFLA_GRE_IFLAGS])
1173                 parms->i_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_IFLAGS]));
1174
1175         if (data[IFLA_GRE_OFLAGS])
1176                 parms->o_flags = gre_flags_to_tnl_flags(nla_get_be16(data[IFLA_GRE_OFLAGS]));
1177
1178         if (data[IFLA_GRE_IKEY])
1179                 parms->i_key = nla_get_be32(data[IFLA_GRE_IKEY]);
1180
1181         if (data[IFLA_GRE_OKEY])
1182                 parms->o_key = nla_get_be32(data[IFLA_GRE_OKEY]);
1183
1184         if (data[IFLA_GRE_LOCAL])
1185                 parms->iph.saddr = nla_get_in_addr(data[IFLA_GRE_LOCAL]);
1186
1187         if (data[IFLA_GRE_REMOTE])
1188                 parms->iph.daddr = nla_get_in_addr(data[IFLA_GRE_REMOTE]);
1189
1190         if (data[IFLA_GRE_TTL])
1191                 parms->iph.ttl = nla_get_u8(data[IFLA_GRE_TTL]);
1192
1193         if (data[IFLA_GRE_TOS])
1194                 parms->iph.tos = nla_get_u8(data[IFLA_GRE_TOS]);
1195
1196         if (!data[IFLA_GRE_PMTUDISC] || nla_get_u8(data[IFLA_GRE_PMTUDISC])) {
1197                 if (t->ignore_df)
1198                         return -EINVAL;
1199                 parms->iph.frag_off = htons(IP_DF);
1200         }
1201
1202         if (data[IFLA_GRE_COLLECT_METADATA]) {
1203                 t->collect_md = true;
1204                 if (dev->type == ARPHRD_IPGRE)
1205                         dev->type = ARPHRD_NONE;
1206         }
1207
1208         if (data[IFLA_GRE_IGNORE_DF]) {
1209                 if (nla_get_u8(data[IFLA_GRE_IGNORE_DF])
1210                   && (parms->iph.frag_off & htons(IP_DF)))
1211                         return -EINVAL;
1212                 t->ignore_df = !!nla_get_u8(data[IFLA_GRE_IGNORE_DF]);
1213         }
1214
1215         if (data[IFLA_GRE_FWMARK])
1216                 *fwmark = nla_get_u32(data[IFLA_GRE_FWMARK]);
1217
1218         if (data[IFLA_GRE_ERSPAN_VER]) {
1219                 t->erspan_ver = nla_get_u8(data[IFLA_GRE_ERSPAN_VER]);
1220
1221                 if (t->erspan_ver != 1 && t->erspan_ver != 2)
1222                         return -EINVAL;
1223         }
1224
1225         if (t->erspan_ver == 1) {
1226                 if (data[IFLA_GRE_ERSPAN_INDEX]) {
1227                         t->index = nla_get_u32(data[IFLA_GRE_ERSPAN_INDEX]);
1228                         if (t->index & ~INDEX_MASK)
1229                                 return -EINVAL;
1230                 }
1231         } else if (t->erspan_ver == 2) {
1232                 if (data[IFLA_GRE_ERSPAN_DIR]) {
1233                         t->dir = nla_get_u8(data[IFLA_GRE_ERSPAN_DIR]);
1234                         if (t->dir & ~(DIR_MASK >> DIR_OFFSET))
1235                                 return -EINVAL;
1236                 }
1237                 if (data[IFLA_GRE_ERSPAN_HWID]) {
1238                         t->hwid = nla_get_u16(data[IFLA_GRE_ERSPAN_HWID]);
1239                         if (t->hwid & ~(HWID_MASK >> HWID_OFFSET))
1240                                 return -EINVAL;
1241                 }
1242         }
1243
1244         return 0;
1245 }
1246
1247 /* This function returns true when ENCAP attributes are present in the nl msg */
1248 static bool ipgre_netlink_encap_parms(struct nlattr *data[],
1249                                       struct ip_tunnel_encap *ipencap)
1250 {
1251         bool ret = false;
1252
1253         memset(ipencap, 0, sizeof(*ipencap));
1254
1255         if (!data)
1256                 return ret;
1257
1258         if (data[IFLA_GRE_ENCAP_TYPE]) {
1259                 ret = true;
1260                 ipencap->type = nla_get_u16(data[IFLA_GRE_ENCAP_TYPE]);
1261         }
1262
1263         if (data[IFLA_GRE_ENCAP_FLAGS]) {
1264                 ret = true;
1265                 ipencap->flags = nla_get_u16(data[IFLA_GRE_ENCAP_FLAGS]);
1266         }
1267
1268         if (data[IFLA_GRE_ENCAP_SPORT]) {
1269                 ret = true;
1270                 ipencap->sport = nla_get_be16(data[IFLA_GRE_ENCAP_SPORT]);
1271         }
1272
1273         if (data[IFLA_GRE_ENCAP_DPORT]) {
1274                 ret = true;
1275                 ipencap->dport = nla_get_be16(data[IFLA_GRE_ENCAP_DPORT]);
1276         }
1277
1278         return ret;
1279 }
1280
1281 static int gre_tap_init(struct net_device *dev)
1282 {
1283         __gre_tunnel_init(dev);
1284         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1285         netif_keep_dst(dev);
1286
1287         return ip_tunnel_init(dev);
1288 }
1289
1290 static const struct net_device_ops gre_tap_netdev_ops = {
1291         .ndo_init               = gre_tap_init,
1292         .ndo_uninit             = ip_tunnel_uninit,
1293         .ndo_start_xmit         = gre_tap_xmit,
1294         .ndo_set_mac_address    = eth_mac_addr,
1295         .ndo_validate_addr      = eth_validate_addr,
1296         .ndo_change_mtu         = ip_tunnel_change_mtu,
1297         .ndo_get_stats64        = ip_tunnel_get_stats64,
1298         .ndo_get_iflink         = ip_tunnel_get_iflink,
1299         .ndo_fill_metadata_dst  = gre_fill_metadata_dst,
1300 };
1301
1302 static int erspan_tunnel_init(struct net_device *dev)
1303 {
1304         struct ip_tunnel *tunnel = netdev_priv(dev);
1305         int t_hlen;
1306
1307         tunnel->tun_hlen = 8;
1308         tunnel->parms.iph.protocol = IPPROTO_GRE;
1309         tunnel->hlen = tunnel->tun_hlen + tunnel->encap_hlen +
1310                        erspan_hdr_len(tunnel->erspan_ver);
1311         t_hlen = tunnel->hlen + sizeof(struct iphdr);
1312
1313         dev->features           |= GRE_FEATURES;
1314         dev->hw_features        |= GRE_FEATURES;
1315         dev->priv_flags         |= IFF_LIVE_ADDR_CHANGE;
1316         netif_keep_dst(dev);
1317
1318         return ip_tunnel_init(dev);
1319 }
1320
1321 static const struct net_device_ops erspan_netdev_ops = {
1322         .ndo_init               = erspan_tunnel_init,
1323         .ndo_uninit             = ip_tunnel_uninit,
1324         .ndo_start_xmit         = erspan_xmit,
1325         .ndo_set_mac_address    = eth_mac_addr,
1326         .ndo_validate_addr      = eth_validate_addr,
1327         .ndo_change_mtu         = ip_tunnel_change_mtu,
1328         .ndo_get_stats64        = ip_tunnel_get_stats64,
1329         .ndo_get_iflink         = ip_tunnel_get_iflink,
1330         .ndo_fill_metadata_dst  = gre_fill_metadata_dst,
1331 };
1332
1333 static void ipgre_tap_setup(struct net_device *dev)
1334 {
1335         ether_setup(dev);
1336         dev->max_mtu = 0;
1337         dev->netdev_ops = &gre_tap_netdev_ops;
1338         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1339         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1340         ip_tunnel_setup(dev, gre_tap_net_id);
1341 }
1342
1343 bool is_gretap_dev(const struct net_device *dev)
1344 {
1345         return dev->netdev_ops == &gre_tap_netdev_ops;
1346 }
1347 EXPORT_SYMBOL_GPL(is_gretap_dev);
1348
1349 static int ipgre_newlink(struct net *src_net, struct net_device *dev,
1350                          struct nlattr *tb[], struct nlattr *data[],
1351                          struct netlink_ext_ack *extack)
1352 {
1353         struct ip_tunnel_parm p;
1354         struct ip_tunnel_encap ipencap;
1355         __u32 fwmark = 0;
1356         int err;
1357
1358         if (ipgre_netlink_encap_parms(data, &ipencap)) {
1359                 struct ip_tunnel *t = netdev_priv(dev);
1360                 err = ip_tunnel_encap_setup(t, &ipencap);
1361
1362                 if (err < 0)
1363                         return err;
1364         }
1365
1366         err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1367         if (err < 0)
1368                 return err;
1369         return ip_tunnel_newlink(dev, tb, &p, fwmark);
1370 }
1371
1372 static int ipgre_changelink(struct net_device *dev, struct nlattr *tb[],
1373                             struct nlattr *data[],
1374                             struct netlink_ext_ack *extack)
1375 {
1376         struct ip_tunnel *t = netdev_priv(dev);
1377         struct ip_tunnel_encap ipencap;
1378         __u32 fwmark = t->fwmark;
1379         struct ip_tunnel_parm p;
1380         int err;
1381
1382         if (ipgre_netlink_encap_parms(data, &ipencap)) {
1383                 err = ip_tunnel_encap_setup(t, &ipencap);
1384
1385                 if (err < 0)
1386                         return err;
1387         }
1388
1389         err = ipgre_netlink_parms(dev, data, tb, &p, &fwmark);
1390         if (err < 0)
1391                 return err;
1392
1393         err = ip_tunnel_changelink(dev, tb, &p, fwmark);
1394         if (err < 0)
1395                 return err;
1396
1397         t->parms.i_flags = p.i_flags;
1398         t->parms.o_flags = p.o_flags;
1399
1400         if (strcmp(dev->rtnl_link_ops->kind, "erspan"))
1401                 ipgre_link_update(dev, !tb[IFLA_MTU]);
1402
1403         return 0;
1404 }
1405
1406 static size_t ipgre_get_size(const struct net_device *dev)
1407 {
1408         return
1409                 /* IFLA_GRE_LINK */
1410                 nla_total_size(4) +
1411                 /* IFLA_GRE_IFLAGS */
1412                 nla_total_size(2) +
1413                 /* IFLA_GRE_OFLAGS */
1414                 nla_total_size(2) +
1415                 /* IFLA_GRE_IKEY */
1416                 nla_total_size(4) +
1417                 /* IFLA_GRE_OKEY */
1418                 nla_total_size(4) +
1419                 /* IFLA_GRE_LOCAL */
1420                 nla_total_size(4) +
1421                 /* IFLA_GRE_REMOTE */
1422                 nla_total_size(4) +
1423                 /* IFLA_GRE_TTL */
1424                 nla_total_size(1) +
1425                 /* IFLA_GRE_TOS */
1426                 nla_total_size(1) +
1427                 /* IFLA_GRE_PMTUDISC */
1428                 nla_total_size(1) +
1429                 /* IFLA_GRE_ENCAP_TYPE */
1430                 nla_total_size(2) +
1431                 /* IFLA_GRE_ENCAP_FLAGS */
1432                 nla_total_size(2) +
1433                 /* IFLA_GRE_ENCAP_SPORT */
1434                 nla_total_size(2) +
1435                 /* IFLA_GRE_ENCAP_DPORT */
1436                 nla_total_size(2) +
1437                 /* IFLA_GRE_COLLECT_METADATA */
1438                 nla_total_size(0) +
1439                 /* IFLA_GRE_IGNORE_DF */
1440                 nla_total_size(1) +
1441                 /* IFLA_GRE_FWMARK */
1442                 nla_total_size(4) +
1443                 /* IFLA_GRE_ERSPAN_INDEX */
1444                 nla_total_size(4) +
1445                 /* IFLA_GRE_ERSPAN_VER */
1446                 nla_total_size(1) +
1447                 /* IFLA_GRE_ERSPAN_DIR */
1448                 nla_total_size(1) +
1449                 /* IFLA_GRE_ERSPAN_HWID */
1450                 nla_total_size(2) +
1451                 0;
1452 }
1453
1454 static int ipgre_fill_info(struct sk_buff *skb, const struct net_device *dev)
1455 {
1456         struct ip_tunnel *t = netdev_priv(dev);
1457         struct ip_tunnel_parm *p = &t->parms;
1458
1459         if (nla_put_u32(skb, IFLA_GRE_LINK, p->link) ||
1460             nla_put_be16(skb, IFLA_GRE_IFLAGS,
1461                          gre_tnl_flags_to_gre_flags(p->i_flags)) ||
1462             nla_put_be16(skb, IFLA_GRE_OFLAGS,
1463                          gre_tnl_flags_to_gre_flags(p->o_flags)) ||
1464             nla_put_be32(skb, IFLA_GRE_IKEY, p->i_key) ||
1465             nla_put_be32(skb, IFLA_GRE_OKEY, p->o_key) ||
1466             nla_put_in_addr(skb, IFLA_GRE_LOCAL, p->iph.saddr) ||
1467             nla_put_in_addr(skb, IFLA_GRE_REMOTE, p->iph.daddr) ||
1468             nla_put_u8(skb, IFLA_GRE_TTL, p->iph.ttl) ||
1469             nla_put_u8(skb, IFLA_GRE_TOS, p->iph.tos) ||
1470             nla_put_u8(skb, IFLA_GRE_PMTUDISC,
1471                        !!(p->iph.frag_off & htons(IP_DF))) ||
1472             nla_put_u32(skb, IFLA_GRE_FWMARK, t->fwmark))
1473                 goto nla_put_failure;
1474
1475         if (nla_put_u16(skb, IFLA_GRE_ENCAP_TYPE,
1476                         t->encap.type) ||
1477             nla_put_be16(skb, IFLA_GRE_ENCAP_SPORT,
1478                          t->encap.sport) ||
1479             nla_put_be16(skb, IFLA_GRE_ENCAP_DPORT,
1480                          t->encap.dport) ||
1481             nla_put_u16(skb, IFLA_GRE_ENCAP_FLAGS,
1482                         t->encap.flags))
1483                 goto nla_put_failure;
1484
1485         if (nla_put_u8(skb, IFLA_GRE_IGNORE_DF, t->ignore_df))
1486                 goto nla_put_failure;
1487
1488         if (t->collect_md) {
1489                 if (nla_put_flag(skb, IFLA_GRE_COLLECT_METADATA))
1490                         goto nla_put_failure;
1491         }
1492
1493         if (nla_put_u8(skb, IFLA_GRE_ERSPAN_VER, t->erspan_ver))
1494                 goto nla_put_failure;
1495
1496         if (t->erspan_ver == 1) {
1497                 if (nla_put_u32(skb, IFLA_GRE_ERSPAN_INDEX, t->index))
1498                         goto nla_put_failure;
1499         } else if (t->erspan_ver == 2) {
1500                 if (nla_put_u8(skb, IFLA_GRE_ERSPAN_DIR, t->dir))
1501                         goto nla_put_failure;
1502                 if (nla_put_u16(skb, IFLA_GRE_ERSPAN_HWID, t->hwid))
1503                         goto nla_put_failure;
1504         }
1505
1506         return 0;
1507
1508 nla_put_failure:
1509         return -EMSGSIZE;
1510 }
1511
1512 static void erspan_setup(struct net_device *dev)
1513 {
1514         ether_setup(dev);
1515         dev->netdev_ops = &erspan_netdev_ops;
1516         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1517         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
1518         ip_tunnel_setup(dev, erspan_net_id);
1519 }
1520
1521 static const struct nla_policy ipgre_policy[IFLA_GRE_MAX + 1] = {
1522         [IFLA_GRE_LINK]         = { .type = NLA_U32 },
1523         [IFLA_GRE_IFLAGS]       = { .type = NLA_U16 },
1524         [IFLA_GRE_OFLAGS]       = { .type = NLA_U16 },
1525         [IFLA_GRE_IKEY]         = { .type = NLA_U32 },
1526         [IFLA_GRE_OKEY]         = { .type = NLA_U32 },
1527         [IFLA_GRE_LOCAL]        = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
1528         [IFLA_GRE_REMOTE]       = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
1529         [IFLA_GRE_TTL]          = { .type = NLA_U8 },
1530         [IFLA_GRE_TOS]          = { .type = NLA_U8 },
1531         [IFLA_GRE_PMTUDISC]     = { .type = NLA_U8 },
1532         [IFLA_GRE_ENCAP_TYPE]   = { .type = NLA_U16 },
1533         [IFLA_GRE_ENCAP_FLAGS]  = { .type = NLA_U16 },
1534         [IFLA_GRE_ENCAP_SPORT]  = { .type = NLA_U16 },
1535         [IFLA_GRE_ENCAP_DPORT]  = { .type = NLA_U16 },
1536         [IFLA_GRE_COLLECT_METADATA]     = { .type = NLA_FLAG },
1537         [IFLA_GRE_IGNORE_DF]    = { .type = NLA_U8 },
1538         [IFLA_GRE_FWMARK]       = { .type = NLA_U32 },
1539         [IFLA_GRE_ERSPAN_INDEX] = { .type = NLA_U32 },
1540         [IFLA_GRE_ERSPAN_VER]   = { .type = NLA_U8 },
1541         [IFLA_GRE_ERSPAN_DIR]   = { .type = NLA_U8 },
1542         [IFLA_GRE_ERSPAN_HWID]  = { .type = NLA_U16 },
1543 };
1544
1545 static struct rtnl_link_ops ipgre_link_ops __read_mostly = {
1546         .kind           = "gre",
1547         .maxtype        = IFLA_GRE_MAX,
1548         .policy         = ipgre_policy,
1549         .priv_size      = sizeof(struct ip_tunnel),
1550         .setup          = ipgre_tunnel_setup,
1551         .validate       = ipgre_tunnel_validate,
1552         .newlink        = ipgre_newlink,
1553         .changelink     = ipgre_changelink,
1554         .dellink        = ip_tunnel_dellink,
1555         .get_size       = ipgre_get_size,
1556         .fill_info      = ipgre_fill_info,
1557         .get_link_net   = ip_tunnel_get_link_net,
1558 };
1559
1560 static struct rtnl_link_ops ipgre_tap_ops __read_mostly = {
1561         .kind           = "gretap",
1562         .maxtype        = IFLA_GRE_MAX,
1563         .policy         = ipgre_policy,
1564         .priv_size      = sizeof(struct ip_tunnel),
1565         .setup          = ipgre_tap_setup,
1566         .validate       = ipgre_tap_validate,
1567         .newlink        = ipgre_newlink,
1568         .changelink     = ipgre_changelink,
1569         .dellink        = ip_tunnel_dellink,
1570         .get_size       = ipgre_get_size,
1571         .fill_info      = ipgre_fill_info,
1572         .get_link_net   = ip_tunnel_get_link_net,
1573 };
1574
1575 static struct rtnl_link_ops erspan_link_ops __read_mostly = {
1576         .kind           = "erspan",
1577         .maxtype        = IFLA_GRE_MAX,
1578         .policy         = ipgre_policy,
1579         .priv_size      = sizeof(struct ip_tunnel),
1580         .setup          = erspan_setup,
1581         .validate       = erspan_validate,
1582         .newlink        = ipgre_newlink,
1583         .changelink     = ipgre_changelink,
1584         .dellink        = ip_tunnel_dellink,
1585         .get_size       = ipgre_get_size,
1586         .fill_info      = ipgre_fill_info,
1587         .get_link_net   = ip_tunnel_get_link_net,
1588 };
1589
1590 struct net_device *gretap_fb_dev_create(struct net *net, const char *name,
1591                                         u8 name_assign_type)
1592 {
1593         struct nlattr *tb[IFLA_MAX + 1];
1594         struct net_device *dev;
1595         LIST_HEAD(list_kill);
1596         struct ip_tunnel *t;
1597         int err;
1598
1599         memset(&tb, 0, sizeof(tb));
1600
1601         dev = rtnl_create_link(net, name, name_assign_type,
1602                                &ipgre_tap_ops, tb);
1603         if (IS_ERR(dev))
1604                 return dev;
1605
1606         /* Configure flow based GRE device. */
1607         t = netdev_priv(dev);
1608         t->collect_md = true;
1609
1610         err = ipgre_newlink(net, dev, tb, NULL, NULL);
1611         if (err < 0) {
1612                 free_netdev(dev);
1613                 return ERR_PTR(err);
1614         }
1615
1616         /* openvswitch users expect packet sizes to be unrestricted,
1617          * so set the largest MTU we can.
1618          */
1619         err = __ip_tunnel_change_mtu(dev, IP_MAX_MTU, false);
1620         if (err)
1621                 goto out;
1622
1623         err = rtnl_configure_link(dev, NULL);
1624         if (err < 0)
1625                 goto out;
1626
1627         return dev;
1628 out:
1629         ip_tunnel_dellink(dev, &list_kill);
1630         unregister_netdevice_many(&list_kill);
1631         return ERR_PTR(err);
1632 }
1633 EXPORT_SYMBOL_GPL(gretap_fb_dev_create);
1634
1635 static int __net_init ipgre_tap_init_net(struct net *net)
1636 {
1637         return ip_tunnel_init_net(net, gre_tap_net_id, &ipgre_tap_ops, "gretap0");
1638 }
1639
1640 static void __net_exit ipgre_tap_exit_batch_net(struct list_head *list_net)
1641 {
1642         ip_tunnel_delete_nets(list_net, gre_tap_net_id, &ipgre_tap_ops);
1643 }
1644
1645 static struct pernet_operations ipgre_tap_net_ops = {
1646         .init = ipgre_tap_init_net,
1647         .exit_batch = ipgre_tap_exit_batch_net,
1648         .id   = &gre_tap_net_id,
1649         .size = sizeof(struct ip_tunnel_net),
1650 };
1651
1652 static int __net_init erspan_init_net(struct net *net)
1653 {
1654         return ip_tunnel_init_net(net, erspan_net_id,
1655                                   &erspan_link_ops, "erspan0");
1656 }
1657
1658 static void __net_exit erspan_exit_batch_net(struct list_head *net_list)
1659 {
1660         ip_tunnel_delete_nets(net_list, erspan_net_id, &erspan_link_ops);
1661 }
1662
1663 static struct pernet_operations erspan_net_ops = {
1664         .init = erspan_init_net,
1665         .exit_batch = erspan_exit_batch_net,
1666         .id   = &erspan_net_id,
1667         .size = sizeof(struct ip_tunnel_net),
1668 };
1669
1670 static int __init ipgre_init(void)
1671 {
1672         int err;
1673
1674         pr_info("GRE over IPv4 tunneling driver\n");
1675
1676         err = register_pernet_device(&ipgre_net_ops);
1677         if (err < 0)
1678                 return err;
1679
1680         err = register_pernet_device(&ipgre_tap_net_ops);
1681         if (err < 0)
1682                 goto pnet_tap_failed;
1683
1684         err = register_pernet_device(&erspan_net_ops);
1685         if (err < 0)
1686                 goto pnet_erspan_failed;
1687
1688         err = gre_add_protocol(&ipgre_protocol, GREPROTO_CISCO);
1689         if (err < 0) {
1690                 pr_info("%s: can't add protocol\n", __func__);
1691                 goto add_proto_failed;
1692         }
1693
1694         err = rtnl_link_register(&ipgre_link_ops);
1695         if (err < 0)
1696                 goto rtnl_link_failed;
1697
1698         err = rtnl_link_register(&ipgre_tap_ops);
1699         if (err < 0)
1700                 goto tap_ops_failed;
1701
1702         err = rtnl_link_register(&erspan_link_ops);
1703         if (err < 0)
1704                 goto erspan_link_failed;
1705
1706         return 0;
1707
1708 erspan_link_failed:
1709         rtnl_link_unregister(&ipgre_tap_ops);
1710 tap_ops_failed:
1711         rtnl_link_unregister(&ipgre_link_ops);
1712 rtnl_link_failed:
1713         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1714 add_proto_failed:
1715         unregister_pernet_device(&erspan_net_ops);
1716 pnet_erspan_failed:
1717         unregister_pernet_device(&ipgre_tap_net_ops);
1718 pnet_tap_failed:
1719         unregister_pernet_device(&ipgre_net_ops);
1720         return err;
1721 }
1722
1723 static void __exit ipgre_fini(void)
1724 {
1725         rtnl_link_unregister(&ipgre_tap_ops);
1726         rtnl_link_unregister(&ipgre_link_ops);
1727         rtnl_link_unregister(&erspan_link_ops);
1728         gre_del_protocol(&ipgre_protocol, GREPROTO_CISCO);
1729         unregister_pernet_device(&ipgre_tap_net_ops);
1730         unregister_pernet_device(&ipgre_net_ops);
1731         unregister_pernet_device(&erspan_net_ops);
1732 }
1733
1734 module_init(ipgre_init);
1735 module_exit(ipgre_fini);
1736 MODULE_LICENSE("GPL");
1737 MODULE_ALIAS_RTNL_LINK("gre");
1738 MODULE_ALIAS_RTNL_LINK("gretap");
1739 MODULE_ALIAS_RTNL_LINK("erspan");
1740 MODULE_ALIAS_NETDEV("gre0");
1741 MODULE_ALIAS_NETDEV("gretap0");
1742 MODULE_ALIAS_NETDEV("erspan0");