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