kernel/net/ipv4/ip_tunnel.c

   1 /*
   2  * Copyright (c) 2013 Nicira, Inc.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of version 2 of the GNU General Public
   6  * License as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program; if not, write to the Free Software
  15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
  16  * 02110-1301, USA
  17  */
  18
  19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  20
  21 #include <linux/capability.h>
  22 #include <linux/module.h>
  23 #include <linux/types.h>
  24 #include <linux/kernel.h>
  25 #include <linux/slab.h>
  26 #include <linux/uaccess.h>
  27 #include <linux/skbuff.h>
  28 #include <linux/netdevice.h>
  29 #include <linux/in.h>
  30 #include <linux/tcp.h>
  31 #include <linux/udp.h>
  32 #include <linux/if_arp.h>
  33 #include <linux/mroute.h>
  34 #include <linux/init.h>
  35 #include <linux/in6.h>
  36 #include <linux/inetdevice.h>
  37 #include <linux/igmp.h>
  38 #include <linux/netfilter_ipv4.h>
  39 #include <linux/etherdevice.h>
  40 #include <linux/if_ether.h>
  41 #include <linux/if_vlan.h>
  42 #include <linux/rculist.h>
  43 #include <linux/err.h>
  44
  45 #include <net/sock.h>
  46 #include <net/ip.h>
  47 #include <net/icmp.h>
  48 #include <net/protocol.h>
  49 #include <net/ip_tunnels.h>
  50 #include <net/arp.h>
  51 #include <net/checksum.h>
  52 #include <net/dsfield.h>
  53 #include <net/inet_ecn.h>
  54 #include <net/xfrm.h>
  55 #include <net/net_namespace.h>
  56 #include <net/netns/generic.h>
  57 #include <net/rtnetlink.h>
  58 #include <net/udp.h>
  59
  60 #if IS_ENABLED(CONFIG_IPV6)
  61 #include <net/ipv6.h>
  62 #include <net/ip6_fib.h>
  63 #include <net/ip6_route.h>
  64 #endif
  65
  66 static unsigned int ip_tunnel_hash(__be32 key, __be32 remote)
  67 {
  68         return hash_32((__force u32)key ^ (__force u32)remote,
  69                          IP_TNL_HASH_BITS);
  70 }
  71
  72 static void __tunnel_dst_set(struct ip_tunnel_dst *idst,
  73                              struct dst_entry *dst, __be32 saddr)
  74 {
  75         struct dst_entry *old_dst;
  76
  77         dst_clone(dst);
  78         old_dst = xchg((__force struct dst_entry **)&idst->dst, dst);
  79         dst_release(old_dst);
  80         idst->saddr = saddr;
  81 }
  82
  83 static noinline void tunnel_dst_set(struct ip_tunnel *t,
  84                            struct dst_entry *dst, __be32 saddr)
  85 {
  86         __tunnel_dst_set(raw_cpu_ptr(t->dst_cache), dst, saddr);
  87 }
  88
  89 static void tunnel_dst_reset(struct ip_tunnel *t)
  90 {
  91         tunnel_dst_set(t, NULL, 0);
  92 }
  93
  94 void ip_tunnel_dst_reset_all(struct ip_tunnel *t)
  95 {
  96         int i;
  97
  98         for_each_possible_cpu(i)
  99                 __tunnel_dst_set(per_cpu_ptr(t->dst_cache, i), NULL, 0);
 100 }
 101 EXPORT_SYMBOL(ip_tunnel_dst_reset_all);
 102
 103 static struct rtable *tunnel_rtable_get(struct ip_tunnel *t,
 104                                         u32 cookie, __be32 *saddr)
 105 {
 106         struct ip_tunnel_dst *idst;
 107         struct dst_entry *dst;
 108
 109         rcu_read_lock();
 110         idst = raw_cpu_ptr(t->dst_cache);
 111         dst = rcu_dereference(idst->dst);
 112         if (dst && !atomic_inc_not_zero(&dst->__refcnt))
 113                 dst = NULL;
 114         if (dst) {
 115                 if (!dst->obsolete || dst->ops->check(dst, cookie)) {
 116                         *saddr = idst->saddr;
 117                 } else {
 118                         tunnel_dst_reset(t);
 119                         dst_release(dst);
 120                         dst = NULL;
 121                 }
 122         }
 123         rcu_read_unlock();
 124         return (struct rtable *)dst;
 125 }
 126
 127 static bool ip_tunnel_key_match(const struct ip_tunnel_parm *p,
 128                                 __be16 flags, __be32 key)
 129 {
 130         if (p->i_flags & TUNNEL_KEY) {
 131                 if (flags & TUNNEL_KEY)
 132                         return key == p->i_key;
 133                 else
 134                         /* key expected, none present */
 135                         return false;
 136         } else
 137                 return !(flags & TUNNEL_KEY);
 138 }
 139
 140 /* Fallback tunnel: no source, no destination, no key, no options
 141
 142    Tunnel hash table:
 143    We require exact key match i.e. if a key is present in packet
 144    it will match only tunnel with the same key; if it is not present,
 145    it will match only keyless tunnel.
 146
 147    All keysless packets, if not matched configured keyless tunnels
 148    will match fallback tunnel.
 149    Given src, dst and key, find appropriate for input tunnel.
 150 */
 151 struct ip_tunnel *ip_tunnel_lookup(struct ip_tunnel_net *itn,
 152                                    int link, __be16 flags,
 153                                    __be32 remote, __be32 local,
 154                                    __be32 key)
 155 {
 156         unsigned int hash;
 157         struct ip_tunnel *t, *cand = NULL;
 158         struct hlist_head *head;
 159
 160         hash = ip_tunnel_hash(key, remote);
 161         head = &itn->tunnels[hash];
 162
 163         hlist_for_each_entry_rcu(t, head, hash_node) {
 164                 if (local != t->parms.iph.saddr ||
 165                     remote != t->parms.iph.daddr ||
 166                     !(t->dev->flags & IFF_UP))
 167                         continue;
 168
 169                 if (!ip_tunnel_key_match(&t->parms, flags, key))
 170                         continue;
 171
 172                 if (t->parms.link == link)
 173                         return t;
 174                 else
 175                         cand = t;
 176         }
 177
 178         hlist_for_each_entry_rcu(t, head, hash_node) {
 179                 if (remote != t->parms.iph.daddr ||
 180                     t->parms.iph.saddr != 0 ||
 181                     !(t->dev->flags & IFF_UP))
 182                         continue;
 183
 184                 if (!ip_tunnel_key_match(&t->parms, flags, key))
 185                         continue;
 186
 187                 if (t->parms.link == link)
 188                         return t;
 189                 else if (!cand)
 190                         cand = t;
 191         }
 192
 193         hash = ip_tunnel_hash(key, 0);
 194         head = &itn->tunnels[hash];
 195
 196         hlist_for_each_entry_rcu(t, head, hash_node) {
 197                 if ((local != t->parms.iph.saddr || t->parms.iph.daddr != 0) &&
 198                     (local != t->parms.iph.daddr || !ipv4_is_multicast(local)))
 199                         continue;
 200
 201                 if (!(t->dev->flags & IFF_UP))
 202                         continue;
 203
 204                 if (!ip_tunnel_key_match(&t->parms, flags, key))
 205                         continue;
 206
 207                 if (t->parms.link == link)
 208                         return t;
 209                 else if (!cand)
 210                         cand = t;
 211         }
 212
 213         if (flags & TUNNEL_NO_KEY)
 214                 goto skip_key_lookup;
 215
 216         hlist_for_each_entry_rcu(t, head, hash_node) {
 217                 if (t->parms.i_key != key ||
 218                     t->parms.iph.saddr != 0 ||
 219                     t->parms.iph.daddr != 0 ||
 220                     !(t->dev->flags & IFF_UP))
 221                         continue;
 222
 223                 if (t->parms.link == link)
 224                         return t;
 225                 else if (!cand)
 226                         cand = t;
 227         }
 228
 229 skip_key_lookup:
 230         if (cand)
 231                 return cand;
 232
 233         t = rcu_dereference(itn->collect_md_tun);
 234         if (t)
 235                 return t;
 236
 237         if (itn->fb_tunnel_dev && itn->fb_tunnel_dev->flags & IFF_UP)
 238                 return netdev_priv(itn->fb_tunnel_dev);
 239
 240         return NULL;
 241 }
 242 EXPORT_SYMBOL_GPL(ip_tunnel_lookup);
 243
 244 static struct hlist_head *ip_bucket(struct ip_tunnel_net *itn,
 245                                     struct ip_tunnel_parm *parms)
 246 {
 247         unsigned int h;
 248         __be32 remote;
 249         __be32 i_key = parms->i_key;
 250
 251         if (parms->iph.daddr && !ipv4_is_multicast(parms->iph.daddr))
 252                 remote = parms->iph.daddr;
 253         else
 254                 remote = 0;
 255
 256         if (!(parms->i_flags & TUNNEL_KEY) && (parms->i_flags & VTI_ISVTI))
 257                 i_key = 0;
 258
 259         h = ip_tunnel_hash(i_key, remote);
 260         return &itn->tunnels[h];
 261 }
 262
 263 static void ip_tunnel_add(struct ip_tunnel_net *itn, struct ip_tunnel *t)
 264 {
 265         struct hlist_head *head = ip_bucket(itn, &t->parms);
 266
 267         if (t->collect_md)
 268                 rcu_assign_pointer(itn->collect_md_tun, t);
 269         hlist_add_head_rcu(&t->hash_node, head);
 270 }
 271
 272 static void ip_tunnel_del(struct ip_tunnel_net *itn, struct ip_tunnel *t)
 273 {
 274         if (t->collect_md)
 275                 rcu_assign_pointer(itn->collect_md_tun, NULL);
 276         hlist_del_init_rcu(&t->hash_node);
 277 }
 278
 279 static struct ip_tunnel *ip_tunnel_find(struct ip_tunnel_net *itn,
 280                                         struct ip_tunnel_parm *parms,
 281                                         int type)
 282 {
 283         __be32 remote = parms->iph.daddr;
 284         __be32 local = parms->iph.saddr;
 285         __be32 key = parms->i_key;
 286         __be16 flags = parms->i_flags;
 287         int link = parms->link;
 288         struct ip_tunnel *t = NULL;
 289         struct hlist_head *head = ip_bucket(itn, parms);
 290
 291         hlist_for_each_entry_rcu(t, head, hash_node) {
 292                 if (local == t->parms.iph.saddr &&
 293                     remote == t->parms.iph.daddr &&
 294                     link == t->parms.link &&
 295                     type == t->dev->type &&
 296                     ip_tunnel_key_match(&t->parms, flags, key))
 297                         break;
 298         }
 299         return t;
 300 }
 301
 302 static struct net_device *__ip_tunnel_create(struct net *net,
 303                                              const struct rtnl_link_ops *ops,
 304                                              struct ip_tunnel_parm *parms)
 305 {
 306         int err;
 307         struct ip_tunnel *tunnel;
 308         struct net_device *dev;
 309         char name[IFNAMSIZ];
 310
 311         if (parms->name[0])
 312                 strlcpy(name, parms->name, IFNAMSIZ);
 313         else {
 314                 if (strlen(ops->kind) > (IFNAMSIZ - 3)) {
 315                         err = -E2BIG;
 316                         goto failed;
 317                 }
 318                 strlcpy(name, ops->kind, IFNAMSIZ);
 319                 strncat(name, "%d", 2);
 320         }
 321
 322         ASSERT_RTNL();
 323         dev = alloc_netdev(ops->priv_size, name, NET_NAME_UNKNOWN, ops->setup);
 324         if (!dev) {
 325                 err = -ENOMEM;
 326                 goto failed;
 327         }
 328         dev_net_set(dev, net);
 329
 330         dev->rtnl_link_ops = ops;
 331
 332         tunnel = netdev_priv(dev);
 333         tunnel->parms = *parms;
 334         tunnel->net = net;
 335
 336         err = register_netdevice(dev);
 337         if (err)
 338                 goto failed_free;
 339
 340         return dev;
 341
 342 failed_free:
 343         free_netdev(dev);
 344 failed:
 345         return ERR_PTR(err);
 346 }
 347
 348 static inline void init_tunnel_flow(struct flowi4 *fl4,
 349                                     int proto,
 350                                     __be32 daddr, __be32 saddr,
 351                                     __be32 key, __u8 tos, int oif)
 352 {
 353         memset(fl4, 0, sizeof(*fl4));
 354         fl4->flowi4_oif = oif;
 355         fl4->daddr = daddr;
 356         fl4->saddr = saddr;
 357         fl4->flowi4_tos = tos;
 358         fl4->flowi4_proto = proto;
 359         fl4->fl4_gre_key = key;
 360 }
 361
 362 static int ip_tunnel_bind_dev(struct net_device *dev)
 363 {
 364         struct net_device *tdev = NULL;
 365         struct ip_tunnel *tunnel = netdev_priv(dev);
 366         const struct iphdr *iph;
 367         int hlen = LL_MAX_HEADER;
 368         int mtu = ETH_DATA_LEN;
 369         int t_hlen = tunnel->hlen + sizeof(struct iphdr);
 370
 371         iph = &tunnel->parms.iph;
 372
 373         /* Guess output device to choose reasonable mtu and needed_headroom */
 374         if (iph->daddr) {
 375                 struct flowi4 fl4;
 376                 struct rtable *rt;
 377
 378                 init_tunnel_flow(&fl4, iph->protocol, iph->daddr,
 379                                  iph->saddr, tunnel->parms.o_key,
 380                                  RT_TOS(iph->tos), tunnel->parms.link);
 381                 rt = ip_route_output_key(tunnel->net, &fl4);
 382
 383                 if (!IS_ERR(rt)) {
 384                         tdev = rt->dst.dev;
 385                         tunnel_dst_set(tunnel, &rt->dst, fl4.saddr);
 386                         ip_rt_put(rt);
 387                 }
 388                 if (dev->type != ARPHRD_ETHER)
 389                         dev->flags |= IFF_POINTOPOINT;
 390         }
 391
 392         if (!tdev && tunnel->parms.link)
 393                 tdev = __dev_get_by_index(tunnel->net, tunnel->parms.link);
 394
 395         if (tdev) {
 396                 hlen = tdev->hard_header_len + tdev->needed_headroom;
 397                 mtu = tdev->mtu;
 398         }
 399
 400         dev->needed_headroom = t_hlen + hlen;
 401         mtu -= (dev->hard_header_len + t_hlen);
 402
 403         if (mtu < 68)
 404                 mtu = 68;
 405
 406         return mtu;
 407 }
 408
 409 static struct ip_tunnel *ip_tunnel_create(struct net *net,
 410                                           struct ip_tunnel_net *itn,
 411                                           struct ip_tunnel_parm *parms)
 412 {
 413         struct ip_tunnel *nt;
 414         struct net_device *dev;
 415
 416         BUG_ON(!itn->fb_tunnel_dev);
 417         dev = __ip_tunnel_create(net, itn->fb_tunnel_dev->rtnl_link_ops, parms);
 418         if (IS_ERR(dev))
 419                 return ERR_CAST(dev);
 420
 421         dev->mtu = ip_tunnel_bind_dev(dev);
 422
 423         nt = netdev_priv(dev);
 424         ip_tunnel_add(itn, nt);
 425         return nt;
 426 }
 427
 428 int ip_tunnel_rcv(struct ip_tunnel *tunnel, struct sk_buff *skb,
 429                   const struct tnl_ptk_info *tpi, struct metadata_dst *tun_dst,
 430                   bool log_ecn_error)
 431 {
 432         struct pcpu_sw_netstats *tstats;
 433         const struct iphdr *iph = ip_hdr(skb);
 434         int err;
 435
 436 #ifdef CONFIG_NET_IPGRE_BROADCAST
 437         if (ipv4_is_multicast(iph->daddr)) {
 438                 tunnel->dev->stats.multicast++;
 439                 skb->pkt_type = PACKET_BROADCAST;
 440         }
 441 #endif
 442
 443         if ((!(tpi->flags&TUNNEL_CSUM) &&  (tunnel->parms.i_flags&TUNNEL_CSUM)) ||
 444              ((tpi->flags&TUNNEL_CSUM) && !(tunnel->parms.i_flags&TUNNEL_CSUM))) {
 445                 tunnel->dev->stats.rx_crc_errors++;
 446                 tunnel->dev->stats.rx_errors++;
 447                 goto drop;
 448         }
 449
 450         if (tunnel->parms.i_flags&TUNNEL_SEQ) {
 451                 if (!(tpi->flags&TUNNEL_SEQ) ||
 452                     (tunnel->i_seqno && (s32)(ntohl(tpi->seq) - tunnel->i_seqno) < 0)) {
 453                         tunnel->dev->stats.rx_fifo_errors++;
 454                         tunnel->dev->stats.rx_errors++;
 455                         goto drop;
 456                 }
 457                 tunnel->i_seqno = ntohl(tpi->seq) + 1;
 458         }
 459
 460         skb_reset_network_header(skb);
 461
 462         err = IP_ECN_decapsulate(iph, skb);
 463         if (unlikely(err)) {
 464                 if (log_ecn_error)
 465                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
 466                                         &iph->saddr, iph->tos);
 467                 if (err > 1) {
 468                         ++tunnel->dev->stats.rx_frame_errors;
 469                         ++tunnel->dev->stats.rx_errors;
 470                         goto drop;
 471                 }
 472         }
 473
 474         tstats = this_cpu_ptr(tunnel->dev->tstats);
 475         u64_stats_update_begin(&tstats->syncp);
 476         tstats->rx_packets++;
 477         tstats->rx_bytes += skb->len;
 478         u64_stats_update_end(&tstats->syncp);
 479
 480         skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(tunnel->dev)));
 481
 482         if (tunnel->dev->type == ARPHRD_ETHER) {
 483                 skb->protocol = eth_type_trans(skb, tunnel->dev);
 484                 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
 485         } else {
 486                 skb->dev = tunnel->dev;
 487         }
 488
 489         if (tun_dst)
 490                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
 491
 492         gro_cells_receive(&tunnel->gro_cells, skb);
 493         return 0;
 494
 495 drop:
 496         kfree_skb(skb);
 497         return 0;
 498 }
 499 EXPORT_SYMBOL_GPL(ip_tunnel_rcv);
 500
 501 static int ip_encap_hlen(struct ip_tunnel_encap *e)
 502 {
 503         const struct ip_tunnel_encap_ops *ops;
 504         int hlen = -EINVAL;
 505
 506         if (e->type == TUNNEL_ENCAP_NONE)
 507                 return 0;
 508
 509         if (e->type >= MAX_IPTUN_ENCAP_OPS)
 510                 return -EINVAL;
 511
 512         rcu_read_lock();
 513         ops = rcu_dereference(iptun_encaps[e->type]);
 514         if (likely(ops && ops->encap_hlen))
 515                 hlen = ops->encap_hlen(e);
 516         rcu_read_unlock();
 517
 518         return hlen;
 519 }
 520
 521 const struct ip_tunnel_encap_ops __rcu *
 522                 iptun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
 523
 524 int ip_tunnel_encap_add_ops(const struct ip_tunnel_encap_ops *ops,
 525                             unsigned int num)
 526 {
 527         if (num >= MAX_IPTUN_ENCAP_OPS)
 528                 return -ERANGE;
 529
 530         return !cmpxchg((const struct ip_tunnel_encap_ops **)
 531                         &iptun_encaps[num],
 532                         NULL, ops) ? 0 : -1;
 533 }
 534 EXPORT_SYMBOL(ip_tunnel_encap_add_ops);
 535
 536 int ip_tunnel_encap_del_ops(const struct ip_tunnel_encap_ops *ops,
 537                             unsigned int num)
 538 {
 539         int ret;
 540
 541         if (num >= MAX_IPTUN_ENCAP_OPS)
 542                 return -ERANGE;
 543
 544         ret = (cmpxchg((const struct ip_tunnel_encap_ops **)
 545                        &iptun_encaps[num],
 546                        ops, NULL) == ops) ? 0 : -1;
 547
 548         synchronize_net();
 549
 550         return ret;
 551 }
 552 EXPORT_SYMBOL(ip_tunnel_encap_del_ops);
 553
 554 int ip_tunnel_encap_setup(struct ip_tunnel *t,
 555                           struct ip_tunnel_encap *ipencap)
 556 {
 557         int hlen;
 558
 559         memset(&t->encap, 0, sizeof(t->encap));
 560
 561         hlen = ip_encap_hlen(ipencap);
 562         if (hlen < 0)
 563                 return hlen;
 564
 565         t->encap.type = ipencap->type;
 566         t->encap.sport = ipencap->sport;
 567         t->encap.dport = ipencap->dport;
 568         t->encap.flags = ipencap->flags;
 569
 570         t->encap_hlen = hlen;
 571         t->hlen = t->encap_hlen + t->tun_hlen;
 572
 573         return 0;
 574 }
 575 EXPORT_SYMBOL_GPL(ip_tunnel_encap_setup);
 576
 577 int ip_tunnel_encap(struct sk_buff *skb, struct ip_tunnel *t,
 578                     u8 *protocol, struct flowi4 *fl4)
 579 {
 580         const struct ip_tunnel_encap_ops *ops;
 581         int ret = -EINVAL;
 582
 583         if (t->encap.type == TUNNEL_ENCAP_NONE)
 584                 return 0;
 585
 586         if (t->encap.type >= MAX_IPTUN_ENCAP_OPS)
 587                 return -EINVAL;
 588
 589         rcu_read_lock();
 590         ops = rcu_dereference(iptun_encaps[t->encap.type]);
 591         if (likely(ops && ops->build_header))
 592                 ret = ops->build_header(skb, &t->encap, protocol, fl4);
 593         rcu_read_unlock();
 594
 595         return ret;
 596 }
 597 EXPORT_SYMBOL(ip_tunnel_encap);
 598
 599 static int tnl_update_pmtu(struct net_device *dev, struct sk_buff *skb,
 600                             struct rtable *rt, __be16 df,
 601                             const struct iphdr *inner_iph)
 602 {
 603         struct ip_tunnel *tunnel = netdev_priv(dev);
 604         int pkt_size = skb->len - tunnel->hlen - dev->hard_header_len;
 605         int mtu;
 606
 607         if (df)
 608                 mtu = dst_mtu(&rt->dst) - dev->hard_header_len
 609                                         - sizeof(struct iphdr) - tunnel->hlen;
 610         else
 611                 mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu;
 612
 613         if (skb_dst(skb))
 614                 skb_dst(skb)->ops->update_pmtu(skb_dst(skb), NULL, skb, mtu);
 615
 616         if (skb->protocol == htons(ETH_P_IP)) {
 617                 if (!skb_is_gso(skb) &&
 618                     (inner_iph->frag_off & htons(IP_DF)) &&
 619                     mtu < pkt_size) {
 620                         memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
 621                         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
 622                         return -E2BIG;
 623                 }
 624         }
 625 #if IS_ENABLED(CONFIG_IPV6)
 626         else if (skb->protocol == htons(ETH_P_IPV6)) {
 627                 struct rt6_info *rt6 = (struct rt6_info *)skb_dst(skb);
 628
 629                 if (rt6 && mtu < dst_mtu(skb_dst(skb)) &&
 630                            mtu >= IPV6_MIN_MTU) {
 631                         if ((tunnel->parms.iph.daddr &&
 632                             !ipv4_is_multicast(tunnel->parms.iph.daddr)) ||
 633                             rt6->rt6i_dst.plen == 128) {
 634                                 rt6->rt6i_flags |= RTF_MODIFIED;
 635                                 dst_metric_set(skb_dst(skb), RTAX_MTU, mtu);
 636                         }
 637                 }
 638
 639                 if (!skb_is_gso(skb) && mtu >= IPV6_MIN_MTU &&
 640                                         mtu < pkt_size) {
 641                         icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
 642                         return -E2BIG;
 643                 }
 644         }
 645 #endif
 646         return 0;
 647 }
 648
 649 void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev,
 650                     const struct iphdr *tnl_params, u8 protocol)
 651 {
 652         struct ip_tunnel *tunnel = netdev_priv(dev);
 653         const struct iphdr *inner_iph;
 654         struct flowi4 fl4;
 655         u8     tos, ttl;
 656         __be16 df;
 657         struct rtable *rt;              /* Route to the other host */
 658         unsigned int max_headroom;      /* The extra header space needed */
 659         __be32 dst;
 660         int err;
 661         bool connected;
 662
 663         inner_iph = (const struct iphdr *)skb_inner_network_header(skb);
 664         connected = (tunnel->parms.iph.daddr != 0);
 665
 666         dst = tnl_params->daddr;
 667         if (dst == 0) {
 668                 /* NBMA tunnel */
 669
 670                 if (!skb_dst(skb)) {
 671                         dev->stats.tx_fifo_errors++;
 672                         goto tx_error;
 673                 }
 674
 675                 if (skb->protocol == htons(ETH_P_IP)) {
 676                         rt = skb_rtable(skb);
 677                         dst = rt_nexthop(rt, inner_iph->daddr);
 678                 }
 679 #if IS_ENABLED(CONFIG_IPV6)
 680                 else if (skb->protocol == htons(ETH_P_IPV6)) {
 681                         const struct in6_addr *addr6;
 682                         struct neighbour *neigh;
 683                         bool do_tx_error_icmp;
 684                         int addr_type;
 685
 686                         neigh = dst_neigh_lookup(skb_dst(skb),
 687                                                  &ipv6_hdr(skb)->daddr);
 688                         if (!neigh)
 689                                 goto tx_error;
 690
 691                         addr6 = (const struct in6_addr *)&neigh->primary_key;
 692                         addr_type = ipv6_addr_type(addr6);
 693
 694                         if (addr_type == IPV6_ADDR_ANY) {
 695                                 addr6 = &ipv6_hdr(skb)->daddr;
 696                                 addr_type = ipv6_addr_type(addr6);
 697                         }
 698
 699                         if ((addr_type & IPV6_ADDR_COMPATv4) == 0)
 700                                 do_tx_error_icmp = true;
 701                         else {
 702                                 do_tx_error_icmp = false;
 703                                 dst = addr6->s6_addr32[3];
 704                         }
 705                         neigh_release(neigh);
 706                         if (do_tx_error_icmp)
 707                                 goto tx_error_icmp;
 708                 }
 709 #endif
 710                 else
 711                         goto tx_error;
 712
 713                 connected = false;
 714         }
 715
 716         tos = tnl_params->tos;
 717         if (tos & 0x1) {
 718                 tos &= ~0x1;
 719                 if (skb->protocol == htons(ETH_P_IP)) {
 720                         tos = inner_iph->tos;
 721                         connected = false;
 722                 } else if (skb->protocol == htons(ETH_P_IPV6)) {
 723                         tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph);
 724                         connected = false;
 725                 }
 726         }
 727
 728         init_tunnel_flow(&fl4, protocol, dst, tnl_params->saddr,
 729                          tunnel->parms.o_key, RT_TOS(tos), tunnel->parms.link);
 730
 731         if (ip_tunnel_encap(skb, tunnel, &protocol, &fl4) < 0)
 732                 goto tx_error;
 733
 734         rt = connected ? tunnel_rtable_get(tunnel, 0, &fl4.saddr) : NULL;
 735
 736         if (!rt) {
 737                 rt = ip_route_output_key(tunnel->net, &fl4);
 738
 739                 if (IS_ERR(rt)) {
 740                         dev->stats.tx_carrier_errors++;
 741                         goto tx_error;
 742                 }
 743                 if (connected)
 744                         tunnel_dst_set(tunnel, &rt->dst, fl4.saddr);
 745         }
 746
 747         if (rt->dst.dev == dev) {
 748                 ip_rt_put(rt);
 749                 dev->stats.collisions++;
 750                 goto tx_error;
 751         }
 752
 753         if (tnl_update_pmtu(dev, skb, rt, tnl_params->frag_off, inner_iph)) {
 754                 ip_rt_put(rt);
 755                 goto tx_error;
 756         }
 757
 758         if (tunnel->err_count > 0) {
 759                 if (time_before(jiffies,
 760                                 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) {
 761                         tunnel->err_count--;
 762
 763                         memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
 764                         dst_link_failure(skb);
 765                 } else
 766                         tunnel->err_count = 0;
 767         }
 768
 769         tos = ip_tunnel_ecn_encap(tos, inner_iph, skb);
 770         ttl = tnl_params->ttl;
 771         if (ttl == 0) {
 772                 if (skb->protocol == htons(ETH_P_IP))
 773                         ttl = inner_iph->ttl;
 774 #if IS_ENABLED(CONFIG_IPV6)
 775                 else if (skb->protocol == htons(ETH_P_IPV6))
 776                         ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit;
 777 #endif
 778                 else
 779                         ttl = ip4_dst_hoplimit(&rt->dst);
 780         }
 781
 782         df = tnl_params->frag_off;
 783         if (skb->protocol == htons(ETH_P_IP))
 784                 df |= (inner_iph->frag_off&htons(IP_DF));
 785
 786         max_headroom = LL_RESERVED_SPACE(rt->dst.dev) + sizeof(struct iphdr)
 787                         + rt->dst.header_len + ip_encap_hlen(&tunnel->encap);
 788         if (max_headroom > dev->needed_headroom)
 789                 dev->needed_headroom = max_headroom;
 790
 791         if (skb_cow_head(skb, dev->needed_headroom)) {
 792                 ip_rt_put(rt);
 793                 dev->stats.tx_dropped++;
 794                 kfree_skb(skb);
 795                 return;
 796         }
 797
 798         err = iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, protocol,
 799                             tos, ttl, df, !net_eq(tunnel->net, dev_net(dev)));
 800         iptunnel_xmit_stats(err, &dev->stats, dev->tstats);
 801
 802         return;
 803
 804 #if IS_ENABLED(CONFIG_IPV6)
 805 tx_error_icmp:
 806         dst_link_failure(skb);
 807 #endif
 808 tx_error:
 809         dev->stats.tx_errors++;
 810         kfree_skb(skb);
 811 }
 812 EXPORT_SYMBOL_GPL(ip_tunnel_xmit);
 813
 814 static void ip_tunnel_update(struct ip_tunnel_net *itn,
 815                              struct ip_tunnel *t,
 816                              struct net_device *dev,
 817                              struct ip_tunnel_parm *p,
 818                              bool set_mtu)
 819 {
 820         ip_tunnel_del(itn, t);
 821         t->parms.iph.saddr = p->iph.saddr;
 822         t->parms.iph.daddr = p->iph.daddr;
 823         t->parms.i_key = p->i_key;
 824         t->parms.o_key = p->o_key;
 825         if (dev->type != ARPHRD_ETHER) {
 826                 memcpy(dev->dev_addr, &p->iph.saddr, 4);
 827                 memcpy(dev->broadcast, &p->iph.daddr, 4);
 828         }
 829         ip_tunnel_add(itn, t);
 830
 831         t->parms.iph.ttl = p->iph.ttl;
 832         t->parms.iph.tos = p->iph.tos;
 833         t->parms.iph.frag_off = p->iph.frag_off;
 834
 835         if (t->parms.link != p->link) {
 836                 int mtu;
 837
 838                 t->parms.link = p->link;
 839                 mtu = ip_tunnel_bind_dev(dev);
 840                 if (set_mtu)
 841                         dev->mtu = mtu;
 842         }
 843         ip_tunnel_dst_reset_all(t);
 844         netdev_state_change(dev);
 845 }
 846
 847 int ip_tunnel_ioctl(struct net_device *dev, struct ip_tunnel_parm *p, int cmd)
 848 {
 849         int err = 0;
 850         struct ip_tunnel *t = netdev_priv(dev);
 851         struct net *net = t->net;
 852         struct ip_tunnel_net *itn = net_generic(net, t->ip_tnl_net_id);
 853
 854         BUG_ON(!itn->fb_tunnel_dev);
 855         switch (cmd) {
 856         case SIOCGETTUNNEL:
 857                 if (dev == itn->fb_tunnel_dev) {
 858                         t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type);
 859                         if (!t)
 860                                 t = netdev_priv(dev);
 861                 }
 862                 memcpy(p, &t->parms, sizeof(*p));
 863                 break;
 864
 865         case SIOCADDTUNNEL:
 866         case SIOCCHGTUNNEL:
 867                 err = -EPERM;
 868                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 869                         goto done;
 870                 if (p->iph.ttl)
 871                         p->iph.frag_off |= htons(IP_DF);
 872                 if (!(p->i_flags & VTI_ISVTI)) {
 873                         if (!(p->i_flags & TUNNEL_KEY))
 874                                 p->i_key = 0;
 875                         if (!(p->o_flags & TUNNEL_KEY))
 876                                 p->o_key = 0;
 877                 }
 878
 879                 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type);
 880
 881                 if (cmd == SIOCADDTUNNEL) {
 882                         if (!t) {
 883                                 t = ip_tunnel_create(net, itn, p);
 884                                 err = PTR_ERR_OR_ZERO(t);
 885                                 break;
 886                         }
 887
 888                         err = -EEXIST;
 889                         break;
 890                 }
 891                 if (dev != itn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) {
 892                         if (t) {
 893                                 if (t->dev != dev) {
 894                                         err = -EEXIST;
 895                                         break;
 896                                 }
 897                         } else {
 898                                 unsigned int nflags = 0;
 899
 900                                 if (ipv4_is_multicast(p->iph.daddr))
 901                                         nflags = IFF_BROADCAST;
 902                                 else if (p->iph.daddr)
 903                                         nflags = IFF_POINTOPOINT;
 904
 905                                 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) {
 906                                         err = -EINVAL;
 907                                         break;
 908                                 }
 909
 910                                 t = netdev_priv(dev);
 911                         }
 912                 }
 913
 914                 if (t) {
 915                         err = 0;
 916                         ip_tunnel_update(itn, t, dev, p, true);
 917                 } else {
 918                         err = -ENOENT;
 919                 }
 920                 break;
 921
 922         case SIOCDELTUNNEL:
 923                 err = -EPERM;
 924                 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
 925                         goto done;
 926
 927                 if (dev == itn->fb_tunnel_dev) {
 928                         err = -ENOENT;
 929                         t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type);
 930                         if (!t)
 931                                 goto done;
 932                         err = -EPERM;
 933                         if (t == netdev_priv(itn->fb_tunnel_dev))
 934                                 goto done;
 935                         dev = t->dev;
 936                 }
 937                 unregister_netdevice(dev);
 938                 err = 0;
 939                 break;
 940
 941         default:
 942                 err = -EINVAL;
 943         }
 944
 945 done:
 946         return err;
 947 }
 948 EXPORT_SYMBOL_GPL(ip_tunnel_ioctl);
 949
 950 int ip_tunnel_change_mtu(struct net_device *dev, int new_mtu)
 951 {
 952         struct ip_tunnel *tunnel = netdev_priv(dev);
 953         int t_hlen = tunnel->hlen + sizeof(struct iphdr);
 954
 955         if (new_mtu < 68 ||
 956             new_mtu > 0xFFF8 - dev->hard_header_len - t_hlen)
 957                 return -EINVAL;
 958         dev->mtu = new_mtu;
 959         return 0;
 960 }
 961 EXPORT_SYMBOL_GPL(ip_tunnel_change_mtu);
 962
 963 static void ip_tunnel_dev_free(struct net_device *dev)
 964 {
 965         struct ip_tunnel *tunnel = netdev_priv(dev);
 966
 967         gro_cells_destroy(&tunnel->gro_cells);
 968         free_percpu(tunnel->dst_cache);
 969         free_percpu(dev->tstats);
 970         free_netdev(dev);
 971 }
 972
 973 void ip_tunnel_dellink(struct net_device *dev, struct list_head *head)
 974 {
 975         struct ip_tunnel *tunnel = netdev_priv(dev);
 976         struct ip_tunnel_net *itn;
 977
 978         itn = net_generic(tunnel->net, tunnel->ip_tnl_net_id);
 979
 980         if (itn->fb_tunnel_dev != dev) {
 981                 ip_tunnel_del(itn, netdev_priv(dev));
 982                 unregister_netdevice_queue(dev, head);
 983         }
 984 }
 985 EXPORT_SYMBOL_GPL(ip_tunnel_dellink);
 986
 987 struct net *ip_tunnel_get_link_net(const struct net_device *dev)
 988 {
 989         struct ip_tunnel *tunnel = netdev_priv(dev);
 990
 991         return tunnel->net;
 992 }
 993 EXPORT_SYMBOL(ip_tunnel_get_link_net);
 994
 995 int ip_tunnel_get_iflink(const struct net_device *dev)
 996 {
 997         struct ip_tunnel *tunnel = netdev_priv(dev);
 998
 999         return tunnel->parms.link;
1000 }
1001 EXPORT_SYMBOL(ip_tunnel_get_iflink);
1002
1003 int ip_tunnel_init_net(struct net *net, int ip_tnl_net_id,
1004                                   struct rtnl_link_ops *ops, char *devname)
1005 {
1006         struct ip_tunnel_net *itn = net_generic(net, ip_tnl_net_id);
1007         struct ip_tunnel_parm parms;
1008         unsigned int i;
1009
1010         for (i = 0; i < IP_TNL_HASH_SIZE; i++)
1011                 INIT_HLIST_HEAD(&itn->tunnels[i]);
1012
1013         if (!ops) {
1014                 itn->fb_tunnel_dev = NULL;
1015                 return 0;
1016         }
1017
1018         memset(&parms, 0, sizeof(parms));
1019         if (devname)
1020                 strlcpy(parms.name, devname, IFNAMSIZ);
1021
1022         rtnl_lock();
1023         itn->fb_tunnel_dev = __ip_tunnel_create(net, ops, &parms);
1024         /* FB netdevice is special: we have one, and only one per netns.
1025          * Allowing to move it to another netns is clearly unsafe.
1026          */
1027         if (!IS_ERR(itn->fb_tunnel_dev)) {
1028                 itn->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL;
1029                 itn->fb_tunnel_dev->mtu = ip_tunnel_bind_dev(itn->fb_tunnel_dev);
1030                 ip_tunnel_add(itn, netdev_priv(itn->fb_tunnel_dev));
1031         }
1032         rtnl_unlock();
1033
1034         return PTR_ERR_OR_ZERO(itn->fb_tunnel_dev);
1035 }
1036 EXPORT_SYMBOL_GPL(ip_tunnel_init_net);
1037
1038 static void ip_tunnel_destroy(struct ip_tunnel_net *itn, struct list_head *head,
1039                               struct rtnl_link_ops *ops)
1040 {
1041         struct net *net = dev_net(itn->fb_tunnel_dev);
1042         struct net_device *dev, *aux;
1043         int h;
1044
1045         for_each_netdev_safe(net, dev, aux)
1046                 if (dev->rtnl_link_ops == ops)
1047                         unregister_netdevice_queue(dev, head);
1048
1049         for (h = 0; h < IP_TNL_HASH_SIZE; h++) {
1050                 struct ip_tunnel *t;
1051                 struct hlist_node *n;
1052                 struct hlist_head *thead = &itn->tunnels[h];
1053
1054                 hlist_for_each_entry_safe(t, n, thead, hash_node)
1055                         /* If dev is in the same netns, it has already
1056                          * been added to the list by the previous loop.
1057                          */
1058                         if (!net_eq(dev_net(t->dev), net))
1059                                 unregister_netdevice_queue(t->dev, head);
1060         }
1061 }
1062
1063 void ip_tunnel_delete_net(struct ip_tunnel_net *itn, struct rtnl_link_ops *ops)
1064 {
1065         LIST_HEAD(list);
1066
1067         rtnl_lock();
1068         ip_tunnel_destroy(itn, &list, ops);
1069         unregister_netdevice_many(&list);
1070         rtnl_unlock();
1071 }
1072 EXPORT_SYMBOL_GPL(ip_tunnel_delete_net);
1073
1074 int ip_tunnel_newlink(struct net_device *dev, struct nlattr *tb[],
1075                       struct ip_tunnel_parm *p)
1076 {
1077         struct ip_tunnel *nt;
1078         struct net *net = dev_net(dev);
1079         struct ip_tunnel_net *itn;
1080         int mtu;
1081         int err;
1082
1083         nt = netdev_priv(dev);
1084         itn = net_generic(net, nt->ip_tnl_net_id);
1085
1086         if (nt->collect_md) {
1087                 if (rtnl_dereference(itn->collect_md_tun))
1088                         return -EEXIST;
1089         } else {
1090                 if (ip_tunnel_find(itn, p, dev->type))
1091                         return -EEXIST;
1092         }
1093
1094         nt->net = net;
1095         nt->parms = *p;
1096         err = register_netdevice(dev);
1097         if (err)
1098                 goto out;
1099
1100         if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS])
1101                 eth_hw_addr_random(dev);
1102
1103         mtu = ip_tunnel_bind_dev(dev);
1104         if (!tb[IFLA_MTU])
1105                 dev->mtu = mtu;
1106
1107         ip_tunnel_add(itn, nt);
1108 out:
1109         return err;
1110 }
1111 EXPORT_SYMBOL_GPL(ip_tunnel_newlink);
1112
1113 int ip_tunnel_changelink(struct net_device *dev, struct nlattr *tb[],
1114                          struct ip_tunnel_parm *p)
1115 {
1116         struct ip_tunnel *t;
1117         struct ip_tunnel *tunnel = netdev_priv(dev);
1118         struct net *net = tunnel->net;
1119         struct ip_tunnel_net *itn = net_generic(net, tunnel->ip_tnl_net_id);
1120
1121         if (dev == itn->fb_tunnel_dev)
1122                 return -EINVAL;
1123
1124         t = ip_tunnel_find(itn, p, dev->type);
1125
1126         if (t) {
1127                 if (t->dev != dev)
1128                         return -EEXIST;
1129         } else {
1130                 t = tunnel;
1131
1132                 if (dev->type != ARPHRD_ETHER) {
1133                         unsigned int nflags = 0;
1134
1135                         if (ipv4_is_multicast(p->iph.daddr))
1136                                 nflags = IFF_BROADCAST;
1137                         else if (p->iph.daddr)
1138                                 nflags = IFF_POINTOPOINT;
1139
1140                         if ((dev->flags ^ nflags) &
1141                             (IFF_POINTOPOINT | IFF_BROADCAST))
1142                                 return -EINVAL;
1143                 }
1144         }
1145
1146         ip_tunnel_update(itn, t, dev, p, !tb[IFLA_MTU]);
1147         return 0;
1148 }
1149 EXPORT_SYMBOL_GPL(ip_tunnel_changelink);
1150
1151 int ip_tunnel_init(struct net_device *dev)
1152 {
1153         struct ip_tunnel *tunnel = netdev_priv(dev);
1154         struct iphdr *iph = &tunnel->parms.iph;
1155         int err;
1156
1157         dev->destructor = ip_tunnel_dev_free;
1158         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
1159         if (!dev->tstats)
1160                 return -ENOMEM;
1161
1162         tunnel->dst_cache = alloc_percpu(struct ip_tunnel_dst);
1163         if (!tunnel->dst_cache) {
1164                 free_percpu(dev->tstats);
1165                 return -ENOMEM;
1166         }
1167
1168         err = gro_cells_init(&tunnel->gro_cells, dev);
1169         if (err) {
1170                 free_percpu(tunnel->dst_cache);
1171                 free_percpu(dev->tstats);
1172                 return err;
1173         }
1174
1175         tunnel->dev = dev;
1176         tunnel->net = dev_net(dev);
1177         strcpy(tunnel->parms.name, dev->name);
1178         iph->version            = 4;
1179         iph->ihl                = 5;
1180
1181         if (tunnel->collect_md) {
1182                 dev->features |= NETIF_F_NETNS_LOCAL;
1183                 netif_keep_dst(dev);
1184         }
1185         return 0;
1186 }
1187 EXPORT_SYMBOL_GPL(ip_tunnel_init);
1188
1189 void ip_tunnel_uninit(struct net_device *dev)
1190 {
1191         struct ip_tunnel *tunnel = netdev_priv(dev);
1192         struct net *net = tunnel->net;
1193         struct ip_tunnel_net *itn;
1194
1195         itn = net_generic(net, tunnel->ip_tnl_net_id);
1196         /* fb_tunnel_dev will be unregisted in net-exit call. */
1197         if (itn->fb_tunnel_dev != dev)
1198                 ip_tunnel_del(itn, netdev_priv(dev));
1199
1200         ip_tunnel_dst_reset_all(tunnel);
1201 }
1202 EXPORT_SYMBOL_GPL(ip_tunnel_uninit);
1203
1204 /* Do least required initialization, rest of init is done in tunnel_init call */
1205 void ip_tunnel_setup(struct net_device *dev, int net_id)
1206 {
1207         struct ip_tunnel *tunnel = netdev_priv(dev);
1208         tunnel->ip_tnl_net_id = net_id;
1209 }
1210 EXPORT_SYMBOL_GPL(ip_tunnel_setup);
1211
1212 MODULE_LICENSE("GPL");