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/init.h> 34 #include <linux/in6.h> 35 #include <linux/inetdevice.h> 36 #include <linux/igmp.h> 37 #include <linux/netfilter_ipv4.h> 38 #include <linux/etherdevice.h> 39 #include <linux/if_ether.h> 40 #include <linux/if_vlan.h> 41 #include <linux/rculist.h> 42 #include <linux/err.h> 43 44 #include <net/sock.h> 45 #include <net/ip.h> 46 #include <net/icmp.h> 47 #include <net/protocol.h> 48 #include <net/ip_tunnels.h> 49 #include <net/arp.h> 50 #include <net/checksum.h> 51 #include <net/dsfield.h> 52 #include <net/inet_ecn.h> 53 #include <net/xfrm.h> 54 #include <net/net_namespace.h> 55 #include <net/netns/generic.h> 56 #include <net/rtnetlink.h> 57 #include <net/udp.h> 58 #include <net/dst_metadata.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 bool ip_tunnel_key_match(const struct ip_tunnel_parm *p, 73 __be16 flags, __be32 key) 74 { 75 if (p->i_flags & TUNNEL_KEY) { 76 if (flags & TUNNEL_KEY) 77 return key == p->i_key; 78 else 79 /* key expected, none present */ 80 return false; 81 } else 82 return !(flags & TUNNEL_KEY); 83 } 84 85 /* Fallback tunnel: no source, no destination, no key, no options 86 87 Tunnel hash table: 88 We require exact key match i.e. if a key is present in packet 89 it will match only tunnel with the same key; if it is not present, 90 it will match only keyless tunnel. 91 92 All keysless packets, if not matched configured keyless tunnels 93 will match fallback tunnel. 94 Given src, dst and key, find appropriate for input tunnel. 95 */ 96 struct ip_tunnel *ip_tunnel_lookup(struct ip_tunnel_net *itn, 97 int link, __be16 flags, 98 __be32 remote, __be32 local, 99 __be32 key) 100 { 101 unsigned int hash; 102 struct ip_tunnel *t, *cand = NULL; 103 struct hlist_head *head; 104 105 hash = ip_tunnel_hash(key, remote); 106 head = &itn->tunnels[hash]; 107 108 hlist_for_each_entry_rcu(t, head, hash_node) { 109 if (local != t->parms.iph.saddr || 110 remote != t->parms.iph.daddr || 111 !(t->dev->flags & IFF_UP)) 112 continue; 113 114 if (!ip_tunnel_key_match(&t->parms, flags, key)) 115 continue; 116 117 if (t->parms.link == link) 118 return t; 119 else 120 cand = t; 121 } 122 123 hlist_for_each_entry_rcu(t, head, hash_node) { 124 if (remote != t->parms.iph.daddr || 125 t->parms.iph.saddr != 0 || 126 !(t->dev->flags & IFF_UP)) 127 continue; 128 129 if (!ip_tunnel_key_match(&t->parms, flags, key)) 130 continue; 131 132 if (t->parms.link == link) 133 return t; 134 else if (!cand) 135 cand = t; 136 } 137 138 hash = ip_tunnel_hash(key, 0); 139 head = &itn->tunnels[hash]; 140 141 hlist_for_each_entry_rcu(t, head, hash_node) { 142 if ((local != t->parms.iph.saddr || t->parms.iph.daddr != 0) && 143 (local != t->parms.iph.daddr || !ipv4_is_multicast(local))) 144 continue; 145 146 if (!(t->dev->flags & IFF_UP)) 147 continue; 148 149 if (!ip_tunnel_key_match(&t->parms, flags, key)) 150 continue; 151 152 if (t->parms.link == link) 153 return t; 154 else if (!cand) 155 cand = t; 156 } 157 158 if (flags & TUNNEL_NO_KEY) 159 goto skip_key_lookup; 160 161 hlist_for_each_entry_rcu(t, head, hash_node) { 162 if (t->parms.i_key != key || 163 t->parms.iph.saddr != 0 || 164 t->parms.iph.daddr != 0 || 165 !(t->dev->flags & IFF_UP)) 166 continue; 167 168 if (t->parms.link == link) 169 return t; 170 else if (!cand) 171 cand = t; 172 } 173 174 skip_key_lookup: 175 if (cand) 176 return cand; 177 178 t = rcu_dereference(itn->collect_md_tun); 179 if (t && t->dev->flags & IFF_UP) 180 return t; 181 182 if (itn->fb_tunnel_dev && itn->fb_tunnel_dev->flags & IFF_UP) 183 return netdev_priv(itn->fb_tunnel_dev); 184 185 return NULL; 186 } 187 EXPORT_SYMBOL_GPL(ip_tunnel_lookup); 188 189 static struct hlist_head *ip_bucket(struct ip_tunnel_net *itn, 190 struct ip_tunnel_parm *parms) 191 { 192 unsigned int h; 193 __be32 remote; 194 __be32 i_key = parms->i_key; 195 196 if (parms->iph.daddr && !ipv4_is_multicast(parms->iph.daddr)) 197 remote = parms->iph.daddr; 198 else 199 remote = 0; 200 201 if (!(parms->i_flags & TUNNEL_KEY) && (parms->i_flags & VTI_ISVTI)) 202 i_key = 0; 203 204 h = ip_tunnel_hash(i_key, remote); 205 return &itn->tunnels[h]; 206 } 207 208 static void ip_tunnel_add(struct ip_tunnel_net *itn, struct ip_tunnel *t) 209 { 210 struct hlist_head *head = ip_bucket(itn, &t->parms); 211 212 if (t->collect_md) 213 rcu_assign_pointer(itn->collect_md_tun, t); 214 hlist_add_head_rcu(&t->hash_node, head); 215 } 216 217 static void ip_tunnel_del(struct ip_tunnel_net *itn, struct ip_tunnel *t) 218 { 219 if (t->collect_md) 220 rcu_assign_pointer(itn->collect_md_tun, NULL); 221 hlist_del_init_rcu(&t->hash_node); 222 } 223 224 static struct ip_tunnel *ip_tunnel_find(struct ip_tunnel_net *itn, 225 struct ip_tunnel_parm *parms, 226 int type) 227 { 228 __be32 remote = parms->iph.daddr; 229 __be32 local = parms->iph.saddr; 230 __be32 key = parms->i_key; 231 __be16 flags = parms->i_flags; 232 int link = parms->link; 233 struct ip_tunnel *t = NULL; 234 struct hlist_head *head = ip_bucket(itn, parms); 235 236 hlist_for_each_entry_rcu(t, head, hash_node) { 237 if (local == t->parms.iph.saddr && 238 remote == t->parms.iph.daddr && 239 link == t->parms.link && 240 type == t->dev->type && 241 ip_tunnel_key_match(&t->parms, flags, key)) 242 break; 243 } 244 return t; 245 } 246 247 static struct net_device *__ip_tunnel_create(struct net *net, 248 const struct rtnl_link_ops *ops, 249 struct ip_tunnel_parm *parms) 250 { 251 int err; 252 struct ip_tunnel *tunnel; 253 struct net_device *dev; 254 char name[IFNAMSIZ]; 255 256 err = -E2BIG; 257 if (parms->name[0]) { 258 if (!dev_valid_name(parms->name)) 259 goto failed; 260 strlcpy(name, parms->name, IFNAMSIZ); 261 } else { 262 if (strlen(ops->kind) > (IFNAMSIZ - 3)) 263 goto failed; 264 strcpy(name, ops->kind); 265 strcat(name, "%d"); 266 } 267 268 ASSERT_RTNL(); 269 dev = alloc_netdev(ops->priv_size, name, NET_NAME_UNKNOWN, ops->setup); 270 if (!dev) { 271 err = -ENOMEM; 272 goto failed; 273 } 274 dev_net_set(dev, net); 275 276 dev->rtnl_link_ops = ops; 277 278 tunnel = netdev_priv(dev); 279 tunnel->parms = *parms; 280 tunnel->net = net; 281 282 err = register_netdevice(dev); 283 if (err) 284 goto failed_free; 285 286 return dev; 287 288 failed_free: 289 free_netdev(dev); 290 failed: 291 return ERR_PTR(err); 292 } 293 294 static int ip_tunnel_bind_dev(struct net_device *dev) 295 { 296 struct net_device *tdev = NULL; 297 struct ip_tunnel *tunnel = netdev_priv(dev); 298 const struct iphdr *iph; 299 int hlen = LL_MAX_HEADER; 300 int mtu = ETH_DATA_LEN; 301 int t_hlen = tunnel->hlen + sizeof(struct iphdr); 302 303 iph = &tunnel->parms.iph; 304 305 /* Guess output device to choose reasonable mtu and needed_headroom */ 306 if (iph->daddr) { 307 struct flowi4 fl4; 308 struct rtable *rt; 309 310 ip_tunnel_init_flow(&fl4, iph->protocol, iph->daddr, 311 iph->saddr, tunnel->parms.o_key, 312 RT_TOS(iph->tos), tunnel->parms.link, 313 tunnel->fwmark, 0); 314 rt = ip_route_output_key(tunnel->net, &fl4); 315 316 if (!IS_ERR(rt)) { 317 tdev = rt->dst.dev; 318 ip_rt_put(rt); 319 } 320 if (dev->type != ARPHRD_ETHER) 321 dev->flags |= IFF_POINTOPOINT; 322 323 dst_cache_reset(&tunnel->dst_cache); 324 } 325 326 if (!tdev && tunnel->parms.link) 327 tdev = __dev_get_by_index(tunnel->net, tunnel->parms.link); 328 329 if (tdev) { 330 hlen = tdev->hard_header_len + tdev->needed_headroom; 331 mtu = min(tdev->mtu, IP_MAX_MTU); 332 } 333 334 dev->needed_headroom = t_hlen + hlen; 335 mtu -= (dev->hard_header_len + t_hlen); 336 337 if (mtu < IPV4_MIN_MTU) 338 mtu = IPV4_MIN_MTU; 339 340 return mtu; 341 } 342 343 static struct ip_tunnel *ip_tunnel_create(struct net *net, 344 struct ip_tunnel_net *itn, 345 struct ip_tunnel_parm *parms) 346 { 347 struct ip_tunnel *nt; 348 struct net_device *dev; 349 int t_hlen; 350 int mtu; 351 int err; 352 353 dev = __ip_tunnel_create(net, itn->rtnl_link_ops, parms); 354 if (IS_ERR(dev)) 355 return ERR_CAST(dev); 356 357 mtu = ip_tunnel_bind_dev(dev); 358 err = dev_set_mtu(dev, mtu); 359 if (err) 360 goto err_dev_set_mtu; 361 362 nt = netdev_priv(dev); 363 t_hlen = nt->hlen + sizeof(struct iphdr); 364 dev->min_mtu = ETH_MIN_MTU; 365 dev->max_mtu = IP_MAX_MTU - dev->hard_header_len - t_hlen; 366 ip_tunnel_add(itn, nt); 367 return nt; 368 369 err_dev_set_mtu: 370 unregister_netdevice(dev); 371 return ERR_PTR(err); 372 } 373 374 int ip_tunnel_rcv(struct ip_tunnel *tunnel, struct sk_buff *skb, 375 const struct tnl_ptk_info *tpi, struct metadata_dst *tun_dst, 376 bool log_ecn_error) 377 { 378 struct pcpu_sw_netstats *tstats; 379 const struct iphdr *iph = ip_hdr(skb); 380 int err; 381 382 #ifdef CONFIG_NET_IPGRE_BROADCAST 383 if (ipv4_is_multicast(iph->daddr)) { 384 tunnel->dev->stats.multicast++; 385 skb->pkt_type = PACKET_BROADCAST; 386 } 387 #endif 388 389 if ((!(tpi->flags&TUNNEL_CSUM) && (tunnel->parms.i_flags&TUNNEL_CSUM)) || 390 ((tpi->flags&TUNNEL_CSUM) && !(tunnel->parms.i_flags&TUNNEL_CSUM))) { 391 tunnel->dev->stats.rx_crc_errors++; 392 tunnel->dev->stats.rx_errors++; 393 goto drop; 394 } 395 396 if (tunnel->parms.i_flags&TUNNEL_SEQ) { 397 if (!(tpi->flags&TUNNEL_SEQ) || 398 (tunnel->i_seqno && (s32)(ntohl(tpi->seq) - tunnel->i_seqno) < 0)) { 399 tunnel->dev->stats.rx_fifo_errors++; 400 tunnel->dev->stats.rx_errors++; 401 goto drop; 402 } 403 tunnel->i_seqno = ntohl(tpi->seq) + 1; 404 } 405 406 skb_reset_network_header(skb); 407 408 err = IP_ECN_decapsulate(iph, skb); 409 if (unlikely(err)) { 410 if (log_ecn_error) 411 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n", 412 &iph->saddr, iph->tos); 413 if (err > 1) { 414 ++tunnel->dev->stats.rx_frame_errors; 415 ++tunnel->dev->stats.rx_errors; 416 goto drop; 417 } 418 } 419 420 tstats = this_cpu_ptr(tunnel->dev->tstats); 421 u64_stats_update_begin(&tstats->syncp); 422 tstats->rx_packets++; 423 tstats->rx_bytes += skb->len; 424 u64_stats_update_end(&tstats->syncp); 425 426 skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(tunnel->dev))); 427 428 if (tunnel->dev->type == ARPHRD_ETHER) { 429 skb->protocol = eth_type_trans(skb, tunnel->dev); 430 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 431 } else { 432 skb->dev = tunnel->dev; 433 } 434 435 if (tun_dst) 436 skb_dst_set(skb, (struct dst_entry *)tun_dst); 437 438 gro_cells_receive(&tunnel->gro_cells, skb); 439 return 0; 440 441 drop: 442 if (tun_dst) 443 dst_release((struct dst_entry *)tun_dst); 444 kfree_skb(skb); 445 return 0; 446 } 447 EXPORT_SYMBOL_GPL(ip_tunnel_rcv); 448 449 int ip_tunnel_encap_add_ops(const struct ip_tunnel_encap_ops *ops, 450 unsigned int num) 451 { 452 if (num >= MAX_IPTUN_ENCAP_OPS) 453 return -ERANGE; 454 455 return !cmpxchg((const struct ip_tunnel_encap_ops **) 456 &iptun_encaps[num], 457 NULL, ops) ? 0 : -1; 458 } 459 EXPORT_SYMBOL(ip_tunnel_encap_add_ops); 460 461 int ip_tunnel_encap_del_ops(const struct ip_tunnel_encap_ops *ops, 462 unsigned int num) 463 { 464 int ret; 465 466 if (num >= MAX_IPTUN_ENCAP_OPS) 467 return -ERANGE; 468 469 ret = (cmpxchg((const struct ip_tunnel_encap_ops **) 470 &iptun_encaps[num], 471 ops, NULL) == ops) ? 0 : -1; 472 473 synchronize_net(); 474 475 return ret; 476 } 477 EXPORT_SYMBOL(ip_tunnel_encap_del_ops); 478 479 int ip_tunnel_encap_setup(struct ip_tunnel *t, 480 struct ip_tunnel_encap *ipencap) 481 { 482 int hlen; 483 484 memset(&t->encap, 0, sizeof(t->encap)); 485 486 hlen = ip_encap_hlen(ipencap); 487 if (hlen < 0) 488 return hlen; 489 490 t->encap.type = ipencap->type; 491 t->encap.sport = ipencap->sport; 492 t->encap.dport = ipencap->dport; 493 t->encap.flags = ipencap->flags; 494 495 t->encap_hlen = hlen; 496 t->hlen = t->encap_hlen + t->tun_hlen; 497 498 return 0; 499 } 500 EXPORT_SYMBOL_GPL(ip_tunnel_encap_setup); 501 502 static int tnl_update_pmtu(struct net_device *dev, struct sk_buff *skb, 503 struct rtable *rt, __be16 df, 504 const struct iphdr *inner_iph, 505 int tunnel_hlen, __be32 dst, bool md) 506 { 507 struct ip_tunnel *tunnel = netdev_priv(dev); 508 int pkt_size; 509 int mtu; 510 511 tunnel_hlen = md ? tunnel_hlen : tunnel->hlen; 512 pkt_size = skb->len - tunnel_hlen - dev->hard_header_len; 513 514 if (df) 515 mtu = dst_mtu(&rt->dst) - dev->hard_header_len 516 - sizeof(struct iphdr) - tunnel_hlen; 517 else 518 mtu = skb_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu; 519 520 skb_dst_update_pmtu(skb, mtu); 521 522 if (skb->protocol == htons(ETH_P_IP)) { 523 if (!skb_is_gso(skb) && 524 (inner_iph->frag_off & htons(IP_DF)) && 525 mtu < pkt_size) { 526 memset(IPCB(skb), 0, sizeof(*IPCB(skb))); 527 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu)); 528 return -E2BIG; 529 } 530 } 531 #if IS_ENABLED(CONFIG_IPV6) 532 else if (skb->protocol == htons(ETH_P_IPV6)) { 533 struct rt6_info *rt6 = (struct rt6_info *)skb_dst(skb); 534 __be32 daddr; 535 536 daddr = md ? dst : tunnel->parms.iph.daddr; 537 538 if (rt6 && mtu < dst_mtu(skb_dst(skb)) && 539 mtu >= IPV6_MIN_MTU) { 540 if ((daddr && !ipv4_is_multicast(daddr)) || 541 rt6->rt6i_dst.plen == 128) { 542 rt6->rt6i_flags |= RTF_MODIFIED; 543 dst_metric_set(skb_dst(skb), RTAX_MTU, mtu); 544 } 545 } 546 547 if (!skb_is_gso(skb) && mtu >= IPV6_MIN_MTU && 548 mtu < pkt_size) { 549 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); 550 return -E2BIG; 551 } 552 } 553 #endif 554 return 0; 555 } 556 557 void ip_md_tunnel_xmit(struct sk_buff *skb, struct net_device *dev, 558 u8 proto, int tunnel_hlen) 559 { 560 struct ip_tunnel *tunnel = netdev_priv(dev); 561 u32 headroom = sizeof(struct iphdr); 562 struct ip_tunnel_info *tun_info; 563 const struct ip_tunnel_key *key; 564 const struct iphdr *inner_iph; 565 struct rtable *rt = NULL; 566 struct flowi4 fl4; 567 __be16 df = 0; 568 u8 tos, ttl; 569 bool use_cache; 570 571 tun_info = skb_tunnel_info(skb); 572 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) || 573 ip_tunnel_info_af(tun_info) != AF_INET)) 574 goto tx_error; 575 key = &tun_info->key; 576 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); 577 inner_iph = (const struct iphdr *)skb_inner_network_header(skb); 578 tos = key->tos; 579 if (tos == 1) { 580 if (skb->protocol == htons(ETH_P_IP)) 581 tos = inner_iph->tos; 582 else if (skb->protocol == htons(ETH_P_IPV6)) 583 tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph); 584 } 585 ip_tunnel_init_flow(&fl4, proto, key->u.ipv4.dst, key->u.ipv4.src, 586 tunnel_id_to_key32(key->tun_id), RT_TOS(tos), 587 0, skb->mark, skb_get_hash(skb)); 588 if (tunnel->encap.type != TUNNEL_ENCAP_NONE) 589 goto tx_error; 590 591 use_cache = ip_tunnel_dst_cache_usable(skb, tun_info); 592 if (use_cache) 593 rt = dst_cache_get_ip4(&tun_info->dst_cache, &fl4.saddr); 594 if (!rt) { 595 rt = ip_route_output_key(tunnel->net, &fl4); 596 if (IS_ERR(rt)) { 597 dev->stats.tx_carrier_errors++; 598 goto tx_error; 599 } 600 if (use_cache) 601 dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst, 602 fl4.saddr); 603 } 604 if (rt->dst.dev == dev) { 605 ip_rt_put(rt); 606 dev->stats.collisions++; 607 goto tx_error; 608 } 609 610 if (key->tun_flags & TUNNEL_DONT_FRAGMENT) 611 df = htons(IP_DF); 612 if (tnl_update_pmtu(dev, skb, rt, df, inner_iph, tunnel_hlen, 613 key->u.ipv4.dst, true)) { 614 ip_rt_put(rt); 615 goto tx_error; 616 } 617 618 tos = ip_tunnel_ecn_encap(tos, inner_iph, skb); 619 ttl = key->ttl; 620 if (ttl == 0) { 621 if (skb->protocol == htons(ETH_P_IP)) 622 ttl = inner_iph->ttl; 623 else if (skb->protocol == htons(ETH_P_IPV6)) 624 ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit; 625 else 626 ttl = ip4_dst_hoplimit(&rt->dst); 627 } 628 629 if (!df && skb->protocol == htons(ETH_P_IP)) 630 df = inner_iph->frag_off & htons(IP_DF); 631 632 headroom += LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len; 633 if (headroom > dev->needed_headroom) 634 dev->needed_headroom = headroom; 635 636 if (skb_cow_head(skb, dev->needed_headroom)) { 637 ip_rt_put(rt); 638 goto tx_dropped; 639 } 640 iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, proto, tos, ttl, 641 df, !net_eq(tunnel->net, dev_net(dev))); 642 return; 643 tx_error: 644 dev->stats.tx_errors++; 645 goto kfree; 646 tx_dropped: 647 dev->stats.tx_dropped++; 648 kfree: 649 kfree_skb(skb); 650 } 651 EXPORT_SYMBOL_GPL(ip_md_tunnel_xmit); 652 653 void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev, 654 const struct iphdr *tnl_params, u8 protocol) 655 { 656 struct ip_tunnel *tunnel = netdev_priv(dev); 657 struct ip_tunnel_info *tun_info = NULL; 658 const struct iphdr *inner_iph; 659 unsigned int max_headroom; /* The extra header space needed */ 660 struct rtable *rt = NULL; /* Route to the other host */ 661 bool use_cache = false; 662 struct flowi4 fl4; 663 bool md = false; 664 bool connected; 665 u8 tos, ttl; 666 __be32 dst; 667 __be16 df; 668 669 inner_iph = (const struct iphdr *)skb_inner_network_header(skb); 670 connected = (tunnel->parms.iph.daddr != 0); 671 672 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); 673 674 dst = tnl_params->daddr; 675 if (dst == 0) { 676 /* NBMA tunnel */ 677 678 if (!skb_dst(skb)) { 679 dev->stats.tx_fifo_errors++; 680 goto tx_error; 681 } 682 683 tun_info = skb_tunnel_info(skb); 684 if (tun_info && (tun_info->mode & IP_TUNNEL_INFO_TX) && 685 ip_tunnel_info_af(tun_info) == AF_INET && 686 tun_info->key.u.ipv4.dst) { 687 dst = tun_info->key.u.ipv4.dst; 688 md = true; 689 connected = true; 690 } 691 else if (skb->protocol == htons(ETH_P_IP)) { 692 rt = skb_rtable(skb); 693 dst = rt_nexthop(rt, inner_iph->daddr); 694 } 695 #if IS_ENABLED(CONFIG_IPV6) 696 else if (skb->protocol == htons(ETH_P_IPV6)) { 697 const struct in6_addr *addr6; 698 struct neighbour *neigh; 699 bool do_tx_error_icmp; 700 int addr_type; 701 702 neigh = dst_neigh_lookup(skb_dst(skb), 703 &ipv6_hdr(skb)->daddr); 704 if (!neigh) 705 goto tx_error; 706 707 addr6 = (const struct in6_addr *)&neigh->primary_key; 708 addr_type = ipv6_addr_type(addr6); 709 710 if (addr_type == IPV6_ADDR_ANY) { 711 addr6 = &ipv6_hdr(skb)->daddr; 712 addr_type = ipv6_addr_type(addr6); 713 } 714 715 if ((addr_type & IPV6_ADDR_COMPATv4) == 0) 716 do_tx_error_icmp = true; 717 else { 718 do_tx_error_icmp = false; 719 dst = addr6->s6_addr32[3]; 720 } 721 neigh_release(neigh); 722 if (do_tx_error_icmp) 723 goto tx_error_icmp; 724 } 725 #endif 726 else 727 goto tx_error; 728 729 if (!md) 730 connected = false; 731 } 732 733 tos = tnl_params->tos; 734 if (tos & 0x1) { 735 tos &= ~0x1; 736 if (skb->protocol == htons(ETH_P_IP)) { 737 tos = inner_iph->tos; 738 connected = false; 739 } else if (skb->protocol == htons(ETH_P_IPV6)) { 740 tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph); 741 connected = false; 742 } 743 } 744 745 ip_tunnel_init_flow(&fl4, protocol, dst, tnl_params->saddr, 746 tunnel->parms.o_key, RT_TOS(tos), tunnel->parms.link, 747 tunnel->fwmark, skb_get_hash(skb)); 748 749 if (ip_tunnel_encap(skb, tunnel, &protocol, &fl4) < 0) 750 goto tx_error; 751 752 if (connected && md) { 753 use_cache = ip_tunnel_dst_cache_usable(skb, tun_info); 754 if (use_cache) 755 rt = dst_cache_get_ip4(&tun_info->dst_cache, 756 &fl4.saddr); 757 } else { 758 rt = connected ? dst_cache_get_ip4(&tunnel->dst_cache, 759 &fl4.saddr) : NULL; 760 } 761 762 if (!rt) { 763 rt = ip_route_output_key(tunnel->net, &fl4); 764 765 if (IS_ERR(rt)) { 766 dev->stats.tx_carrier_errors++; 767 goto tx_error; 768 } 769 if (use_cache) 770 dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst, 771 fl4.saddr); 772 else if (!md && connected) 773 dst_cache_set_ip4(&tunnel->dst_cache, &rt->dst, 774 fl4.saddr); 775 } 776 777 if (rt->dst.dev == dev) { 778 ip_rt_put(rt); 779 dev->stats.collisions++; 780 goto tx_error; 781 } 782 783 if (tnl_update_pmtu(dev, skb, rt, tnl_params->frag_off, inner_iph, 784 0, 0, false)) { 785 ip_rt_put(rt); 786 goto tx_error; 787 } 788 789 if (tunnel->err_count > 0) { 790 if (time_before(jiffies, 791 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) { 792 tunnel->err_count--; 793 794 dst_link_failure(skb); 795 } else 796 tunnel->err_count = 0; 797 } 798 799 tos = ip_tunnel_ecn_encap(tos, inner_iph, skb); 800 ttl = tnl_params->ttl; 801 if (ttl == 0) { 802 if (skb->protocol == htons(ETH_P_IP)) 803 ttl = inner_iph->ttl; 804 #if IS_ENABLED(CONFIG_IPV6) 805 else if (skb->protocol == htons(ETH_P_IPV6)) 806 ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit; 807 #endif 808 else 809 ttl = ip4_dst_hoplimit(&rt->dst); 810 } 811 812 df = tnl_params->frag_off; 813 if (skb->protocol == htons(ETH_P_IP) && !tunnel->ignore_df) 814 df |= (inner_iph->frag_off&htons(IP_DF)); 815 816 max_headroom = LL_RESERVED_SPACE(rt->dst.dev) + sizeof(struct iphdr) 817 + rt->dst.header_len + ip_encap_hlen(&tunnel->encap); 818 if (max_headroom > dev->needed_headroom) 819 dev->needed_headroom = max_headroom; 820 821 if (skb_cow_head(skb, dev->needed_headroom)) { 822 ip_rt_put(rt); 823 dev->stats.tx_dropped++; 824 kfree_skb(skb); 825 return; 826 } 827 828 iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, protocol, tos, ttl, 829 df, !net_eq(tunnel->net, dev_net(dev))); 830 return; 831 832 #if IS_ENABLED(CONFIG_IPV6) 833 tx_error_icmp: 834 dst_link_failure(skb); 835 #endif 836 tx_error: 837 dev->stats.tx_errors++; 838 kfree_skb(skb); 839 } 840 EXPORT_SYMBOL_GPL(ip_tunnel_xmit); 841 842 static void ip_tunnel_update(struct ip_tunnel_net *itn, 843 struct ip_tunnel *t, 844 struct net_device *dev, 845 struct ip_tunnel_parm *p, 846 bool set_mtu, 847 __u32 fwmark) 848 { 849 ip_tunnel_del(itn, t); 850 t->parms.iph.saddr = p->iph.saddr; 851 t->parms.iph.daddr = p->iph.daddr; 852 t->parms.i_key = p->i_key; 853 t->parms.o_key = p->o_key; 854 if (dev->type != ARPHRD_ETHER) { 855 memcpy(dev->dev_addr, &p->iph.saddr, 4); 856 memcpy(dev->broadcast, &p->iph.daddr, 4); 857 } 858 ip_tunnel_add(itn, t); 859 860 t->parms.iph.ttl = p->iph.ttl; 861 t->parms.iph.tos = p->iph.tos; 862 t->parms.iph.frag_off = p->iph.frag_off; 863 864 if (t->parms.link != p->link || t->fwmark != fwmark) { 865 int mtu; 866 867 t->parms.link = p->link; 868 t->fwmark = fwmark; 869 mtu = ip_tunnel_bind_dev(dev); 870 if (set_mtu) 871 dev->mtu = mtu; 872 } 873 dst_cache_reset(&t->dst_cache); 874 netdev_state_change(dev); 875 } 876 877 int ip_tunnel_ioctl(struct net_device *dev, struct ip_tunnel_parm *p, int cmd) 878 { 879 int err = 0; 880 struct ip_tunnel *t = netdev_priv(dev); 881 struct net *net = t->net; 882 struct ip_tunnel_net *itn = net_generic(net, t->ip_tnl_net_id); 883 884 switch (cmd) { 885 case SIOCGETTUNNEL: 886 if (dev == itn->fb_tunnel_dev) { 887 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type); 888 if (!t) 889 t = netdev_priv(dev); 890 } 891 memcpy(p, &t->parms, sizeof(*p)); 892 break; 893 894 case SIOCADDTUNNEL: 895 case SIOCCHGTUNNEL: 896 err = -EPERM; 897 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 898 goto done; 899 if (p->iph.ttl) 900 p->iph.frag_off |= htons(IP_DF); 901 if (!(p->i_flags & VTI_ISVTI)) { 902 if (!(p->i_flags & TUNNEL_KEY)) 903 p->i_key = 0; 904 if (!(p->o_flags & TUNNEL_KEY)) 905 p->o_key = 0; 906 } 907 908 t = ip_tunnel_find(itn, p, itn->type); 909 910 if (cmd == SIOCADDTUNNEL) { 911 if (!t) { 912 t = ip_tunnel_create(net, itn, p); 913 err = PTR_ERR_OR_ZERO(t); 914 break; 915 } 916 917 err = -EEXIST; 918 break; 919 } 920 if (dev != itn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) { 921 if (t) { 922 if (t->dev != dev) { 923 err = -EEXIST; 924 break; 925 } 926 } else { 927 unsigned int nflags = 0; 928 929 if (ipv4_is_multicast(p->iph.daddr)) 930 nflags = IFF_BROADCAST; 931 else if (p->iph.daddr) 932 nflags = IFF_POINTOPOINT; 933 934 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) { 935 err = -EINVAL; 936 break; 937 } 938 939 t = netdev_priv(dev); 940 } 941 } 942 943 if (t) { 944 err = 0; 945 ip_tunnel_update(itn, t, dev, p, true, 0); 946 } else { 947 err = -ENOENT; 948 } 949 break; 950 951 case SIOCDELTUNNEL: 952 err = -EPERM; 953 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 954 goto done; 955 956 if (dev == itn->fb_tunnel_dev) { 957 err = -ENOENT; 958 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type); 959 if (!t) 960 goto done; 961 err = -EPERM; 962 if (t == netdev_priv(itn->fb_tunnel_dev)) 963 goto done; 964 dev = t->dev; 965 } 966 unregister_netdevice(dev); 967 err = 0; 968 break; 969 970 default: 971 err = -EINVAL; 972 } 973 974 done: 975 return err; 976 } 977 EXPORT_SYMBOL_GPL(ip_tunnel_ioctl); 978 979 int __ip_tunnel_change_mtu(struct net_device *dev, int new_mtu, bool strict) 980 { 981 struct ip_tunnel *tunnel = netdev_priv(dev); 982 int t_hlen = tunnel->hlen + sizeof(struct iphdr); 983 int max_mtu = IP_MAX_MTU - dev->hard_header_len - t_hlen; 984 985 if (new_mtu < ETH_MIN_MTU) 986 return -EINVAL; 987 988 if (new_mtu > max_mtu) { 989 if (strict) 990 return -EINVAL; 991 992 new_mtu = max_mtu; 993 } 994 995 dev->mtu = new_mtu; 996 return 0; 997 } 998 EXPORT_SYMBOL_GPL(__ip_tunnel_change_mtu); 999 1000 int ip_tunnel_change_mtu(struct net_device *dev, int new_mtu) 1001 { 1002 return __ip_tunnel_change_mtu(dev, new_mtu, true); 1003 } 1004 EXPORT_SYMBOL_GPL(ip_tunnel_change_mtu); 1005 1006 static void ip_tunnel_dev_free(struct net_device *dev) 1007 { 1008 struct ip_tunnel *tunnel = netdev_priv(dev); 1009 1010 gro_cells_destroy(&tunnel->gro_cells); 1011 dst_cache_destroy(&tunnel->dst_cache); 1012 free_percpu(dev->tstats); 1013 } 1014 1015 void ip_tunnel_dellink(struct net_device *dev, struct list_head *head) 1016 { 1017 struct ip_tunnel *tunnel = netdev_priv(dev); 1018 struct ip_tunnel_net *itn; 1019 1020 itn = net_generic(tunnel->net, tunnel->ip_tnl_net_id); 1021 1022 if (itn->fb_tunnel_dev != dev) { 1023 ip_tunnel_del(itn, netdev_priv(dev)); 1024 unregister_netdevice_queue(dev, head); 1025 } 1026 } 1027 EXPORT_SYMBOL_GPL(ip_tunnel_dellink); 1028 1029 struct net *ip_tunnel_get_link_net(const struct net_device *dev) 1030 { 1031 struct ip_tunnel *tunnel = netdev_priv(dev); 1032 1033 return tunnel->net; 1034 } 1035 EXPORT_SYMBOL(ip_tunnel_get_link_net); 1036 1037 int ip_tunnel_get_iflink(const struct net_device *dev) 1038 { 1039 struct ip_tunnel *tunnel = netdev_priv(dev); 1040 1041 return tunnel->parms.link; 1042 } 1043 EXPORT_SYMBOL(ip_tunnel_get_iflink); 1044 1045 int ip_tunnel_init_net(struct net *net, unsigned int ip_tnl_net_id, 1046 struct rtnl_link_ops *ops, char *devname) 1047 { 1048 struct ip_tunnel_net *itn = net_generic(net, ip_tnl_net_id); 1049 struct ip_tunnel_parm parms; 1050 unsigned int i; 1051 1052 itn->rtnl_link_ops = ops; 1053 for (i = 0; i < IP_TNL_HASH_SIZE; i++) 1054 INIT_HLIST_HEAD(&itn->tunnels[i]); 1055 1056 if (!ops || !net_has_fallback_tunnels(net)) { 1057 struct ip_tunnel_net *it_init_net; 1058 1059 it_init_net = net_generic(&init_net, ip_tnl_net_id); 1060 itn->type = it_init_net->type; 1061 itn->fb_tunnel_dev = NULL; 1062 return 0; 1063 } 1064 1065 memset(&parms, 0, sizeof(parms)); 1066 if (devname) 1067 strlcpy(parms.name, devname, IFNAMSIZ); 1068 1069 rtnl_lock(); 1070 itn->fb_tunnel_dev = __ip_tunnel_create(net, ops, &parms); 1071 /* FB netdevice is special: we have one, and only one per netns. 1072 * Allowing to move it to another netns is clearly unsafe. 1073 */ 1074 if (!IS_ERR(itn->fb_tunnel_dev)) { 1075 itn->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL; 1076 itn->fb_tunnel_dev->mtu = ip_tunnel_bind_dev(itn->fb_tunnel_dev); 1077 ip_tunnel_add(itn, netdev_priv(itn->fb_tunnel_dev)); 1078 itn->type = itn->fb_tunnel_dev->type; 1079 } 1080 rtnl_unlock(); 1081 1082 return PTR_ERR_OR_ZERO(itn->fb_tunnel_dev); 1083 } 1084 EXPORT_SYMBOL_GPL(ip_tunnel_init_net); 1085 1086 static void ip_tunnel_destroy(struct net *net, struct ip_tunnel_net *itn, 1087 struct list_head *head, 1088 struct rtnl_link_ops *ops) 1089 { 1090 struct net_device *dev, *aux; 1091 int h; 1092 1093 for_each_netdev_safe(net, dev, aux) 1094 if (dev->rtnl_link_ops == ops) 1095 unregister_netdevice_queue(dev, head); 1096 1097 for (h = 0; h < IP_TNL_HASH_SIZE; h++) { 1098 struct ip_tunnel *t; 1099 struct hlist_node *n; 1100 struct hlist_head *thead = &itn->tunnels[h]; 1101 1102 hlist_for_each_entry_safe(t, n, thead, hash_node) 1103 /* If dev is in the same netns, it has already 1104 * been added to the list by the previous loop. 1105 */ 1106 if (!net_eq(dev_net(t->dev), net)) 1107 unregister_netdevice_queue(t->dev, head); 1108 } 1109 } 1110 1111 void ip_tunnel_delete_nets(struct list_head *net_list, unsigned int id, 1112 struct rtnl_link_ops *ops) 1113 { 1114 struct ip_tunnel_net *itn; 1115 struct net *net; 1116 LIST_HEAD(list); 1117 1118 rtnl_lock(); 1119 list_for_each_entry(net, net_list, exit_list) { 1120 itn = net_generic(net, id); 1121 ip_tunnel_destroy(net, itn, &list, ops); 1122 } 1123 unregister_netdevice_many(&list); 1124 rtnl_unlock(); 1125 } 1126 EXPORT_SYMBOL_GPL(ip_tunnel_delete_nets); 1127 1128 int ip_tunnel_newlink(struct net_device *dev, struct nlattr *tb[], 1129 struct ip_tunnel_parm *p, __u32 fwmark) 1130 { 1131 struct ip_tunnel *nt; 1132 struct net *net = dev_net(dev); 1133 struct ip_tunnel_net *itn; 1134 int mtu; 1135 int err; 1136 1137 nt = netdev_priv(dev); 1138 itn = net_generic(net, nt->ip_tnl_net_id); 1139 1140 if (nt->collect_md) { 1141 if (rtnl_dereference(itn->collect_md_tun)) 1142 return -EEXIST; 1143 } else { 1144 if (ip_tunnel_find(itn, p, dev->type)) 1145 return -EEXIST; 1146 } 1147 1148 nt->net = net; 1149 nt->parms = *p; 1150 nt->fwmark = fwmark; 1151 err = register_netdevice(dev); 1152 if (err) 1153 goto err_register_netdevice; 1154 1155 if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS]) 1156 eth_hw_addr_random(dev); 1157 1158 mtu = ip_tunnel_bind_dev(dev); 1159 if (tb[IFLA_MTU]) { 1160 unsigned int max = IP_MAX_MTU - dev->hard_header_len - nt->hlen; 1161 1162 mtu = clamp(dev->mtu, (unsigned int)ETH_MIN_MTU, 1163 (unsigned int)(max - sizeof(struct iphdr))); 1164 } 1165 1166 err = dev_set_mtu(dev, mtu); 1167 if (err) 1168 goto err_dev_set_mtu; 1169 1170 ip_tunnel_add(itn, nt); 1171 return 0; 1172 1173 err_dev_set_mtu: 1174 unregister_netdevice(dev); 1175 err_register_netdevice: 1176 return err; 1177 } 1178 EXPORT_SYMBOL_GPL(ip_tunnel_newlink); 1179 1180 int ip_tunnel_changelink(struct net_device *dev, struct nlattr *tb[], 1181 struct ip_tunnel_parm *p, __u32 fwmark) 1182 { 1183 struct ip_tunnel *t; 1184 struct ip_tunnel *tunnel = netdev_priv(dev); 1185 struct net *net = tunnel->net; 1186 struct ip_tunnel_net *itn = net_generic(net, tunnel->ip_tnl_net_id); 1187 1188 if (dev == itn->fb_tunnel_dev) 1189 return -EINVAL; 1190 1191 t = ip_tunnel_find(itn, p, dev->type); 1192 1193 if (t) { 1194 if (t->dev != dev) 1195 return -EEXIST; 1196 } else { 1197 t = tunnel; 1198 1199 if (dev->type != ARPHRD_ETHER) { 1200 unsigned int nflags = 0; 1201 1202 if (ipv4_is_multicast(p->iph.daddr)) 1203 nflags = IFF_BROADCAST; 1204 else if (p->iph.daddr) 1205 nflags = IFF_POINTOPOINT; 1206 1207 if ((dev->flags ^ nflags) & 1208 (IFF_POINTOPOINT | IFF_BROADCAST)) 1209 return -EINVAL; 1210 } 1211 } 1212 1213 ip_tunnel_update(itn, t, dev, p, !tb[IFLA_MTU], fwmark); 1214 return 0; 1215 } 1216 EXPORT_SYMBOL_GPL(ip_tunnel_changelink); 1217 1218 int ip_tunnel_init(struct net_device *dev) 1219 { 1220 struct ip_tunnel *tunnel = netdev_priv(dev); 1221 struct iphdr *iph = &tunnel->parms.iph; 1222 int err; 1223 1224 dev->needs_free_netdev = true; 1225 dev->priv_destructor = ip_tunnel_dev_free; 1226 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 1227 if (!dev->tstats) 1228 return -ENOMEM; 1229 1230 err = dst_cache_init(&tunnel->dst_cache, GFP_KERNEL); 1231 if (err) { 1232 free_percpu(dev->tstats); 1233 return err; 1234 } 1235 1236 err = gro_cells_init(&tunnel->gro_cells, dev); 1237 if (err) { 1238 dst_cache_destroy(&tunnel->dst_cache); 1239 free_percpu(dev->tstats); 1240 return err; 1241 } 1242 1243 tunnel->dev = dev; 1244 tunnel->net = dev_net(dev); 1245 strcpy(tunnel->parms.name, dev->name); 1246 iph->version = 4; 1247 iph->ihl = 5; 1248 1249 if (tunnel->collect_md) { 1250 dev->features |= NETIF_F_NETNS_LOCAL; 1251 netif_keep_dst(dev); 1252 } 1253 return 0; 1254 } 1255 EXPORT_SYMBOL_GPL(ip_tunnel_init); 1256 1257 void ip_tunnel_uninit(struct net_device *dev) 1258 { 1259 struct ip_tunnel *tunnel = netdev_priv(dev); 1260 struct net *net = tunnel->net; 1261 struct ip_tunnel_net *itn; 1262 1263 itn = net_generic(net, tunnel->ip_tnl_net_id); 1264 /* fb_tunnel_dev will be unregisted in net-exit call. */ 1265 if (itn->fb_tunnel_dev != dev) 1266 ip_tunnel_del(itn, netdev_priv(dev)); 1267 1268 dst_cache_reset(&tunnel->dst_cache); 1269 } 1270 EXPORT_SYMBOL_GPL(ip_tunnel_uninit); 1271 1272 /* Do least required initialization, rest of init is done in tunnel_init call */ 1273 void ip_tunnel_setup(struct net_device *dev, unsigned int net_id) 1274 { 1275 struct ip_tunnel *tunnel = netdev_priv(dev); 1276 tunnel->ip_tnl_net_id = net_id; 1277 } 1278 EXPORT_SYMBOL_GPL(ip_tunnel_setup); 1279 1280 MODULE_LICENSE("GPL"); 1281