1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (c) 2013 Nicira, Inc. 4 */ 5 6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 7 8 #include <linux/capability.h> 9 #include <linux/module.h> 10 #include <linux/types.h> 11 #include <linux/kernel.h> 12 #include <linux/slab.h> 13 #include <linux/uaccess.h> 14 #include <linux/skbuff.h> 15 #include <linux/netdevice.h> 16 #include <linux/in.h> 17 #include <linux/tcp.h> 18 #include <linux/udp.h> 19 #include <linux/if_arp.h> 20 #include <linux/init.h> 21 #include <linux/in6.h> 22 #include <linux/inetdevice.h> 23 #include <linux/igmp.h> 24 #include <linux/netfilter_ipv4.h> 25 #include <linux/etherdevice.h> 26 #include <linux/if_ether.h> 27 #include <linux/if_vlan.h> 28 #include <linux/rculist.h> 29 #include <linux/err.h> 30 31 #include <net/sock.h> 32 #include <net/ip.h> 33 #include <net/icmp.h> 34 #include <net/protocol.h> 35 #include <net/ip_tunnels.h> 36 #include <net/arp.h> 37 #include <net/checksum.h> 38 #include <net/dsfield.h> 39 #include <net/inet_ecn.h> 40 #include <net/xfrm.h> 41 #include <net/net_namespace.h> 42 #include <net/netns/generic.h> 43 #include <net/rtnetlink.h> 44 #include <net/udp.h> 45 #include <net/dst_metadata.h> 46 47 #if IS_ENABLED(CONFIG_IPV6) 48 #include <net/ipv6.h> 49 #include <net/ip6_fib.h> 50 #include <net/ip6_route.h> 51 #endif 52 53 static unsigned int ip_tunnel_hash(__be32 key, __be32 remote) 54 { 55 return hash_32((__force u32)key ^ (__force u32)remote, 56 IP_TNL_HASH_BITS); 57 } 58 59 static bool ip_tunnel_key_match(const struct ip_tunnel_parm *p, 60 __be16 flags, __be32 key) 61 { 62 if (p->i_flags & TUNNEL_KEY) { 63 if (flags & TUNNEL_KEY) 64 return key == p->i_key; 65 else 66 /* key expected, none present */ 67 return false; 68 } else 69 return !(flags & TUNNEL_KEY); 70 } 71 72 /* Fallback tunnel: no source, no destination, no key, no options 73 74 Tunnel hash table: 75 We require exact key match i.e. if a key is present in packet 76 it will match only tunnel with the same key; if it is not present, 77 it will match only keyless tunnel. 78 79 All keysless packets, if not matched configured keyless tunnels 80 will match fallback tunnel. 81 Given src, dst and key, find appropriate for input tunnel. 82 */ 83 struct ip_tunnel *ip_tunnel_lookup(struct ip_tunnel_net *itn, 84 int link, __be16 flags, 85 __be32 remote, __be32 local, 86 __be32 key) 87 { 88 struct ip_tunnel *t, *cand = NULL; 89 struct hlist_head *head; 90 struct net_device *ndev; 91 unsigned int hash; 92 93 hash = ip_tunnel_hash(key, remote); 94 head = &itn->tunnels[hash]; 95 96 hlist_for_each_entry_rcu(t, head, hash_node) { 97 if (local != t->parms.iph.saddr || 98 remote != t->parms.iph.daddr || 99 !(t->dev->flags & IFF_UP)) 100 continue; 101 102 if (!ip_tunnel_key_match(&t->parms, flags, key)) 103 continue; 104 105 if (READ_ONCE(t->parms.link) == link) 106 return t; 107 cand = t; 108 } 109 110 hlist_for_each_entry_rcu(t, head, hash_node) { 111 if (remote != t->parms.iph.daddr || 112 t->parms.iph.saddr != 0 || 113 !(t->dev->flags & IFF_UP)) 114 continue; 115 116 if (!ip_tunnel_key_match(&t->parms, flags, key)) 117 continue; 118 119 if (READ_ONCE(t->parms.link) == link) 120 return t; 121 if (!cand) 122 cand = t; 123 } 124 125 hash = ip_tunnel_hash(key, 0); 126 head = &itn->tunnels[hash]; 127 128 hlist_for_each_entry_rcu(t, head, hash_node) { 129 if ((local != t->parms.iph.saddr || t->parms.iph.daddr != 0) && 130 (local != t->parms.iph.daddr || !ipv4_is_multicast(local))) 131 continue; 132 133 if (!(t->dev->flags & IFF_UP)) 134 continue; 135 136 if (!ip_tunnel_key_match(&t->parms, flags, key)) 137 continue; 138 139 if (READ_ONCE(t->parms.link) == link) 140 return t; 141 if (!cand) 142 cand = t; 143 } 144 145 hlist_for_each_entry_rcu(t, head, hash_node) { 146 if ((!(flags & TUNNEL_NO_KEY) && t->parms.i_key != key) || 147 t->parms.iph.saddr != 0 || 148 t->parms.iph.daddr != 0 || 149 !(t->dev->flags & IFF_UP)) 150 continue; 151 152 if (READ_ONCE(t->parms.link) == link) 153 return t; 154 if (!cand) 155 cand = t; 156 } 157 158 if (cand) 159 return cand; 160 161 t = rcu_dereference(itn->collect_md_tun); 162 if (t && t->dev->flags & IFF_UP) 163 return t; 164 165 ndev = READ_ONCE(itn->fb_tunnel_dev); 166 if (ndev && ndev->flags & IFF_UP) 167 return netdev_priv(ndev); 168 169 return NULL; 170 } 171 EXPORT_SYMBOL_GPL(ip_tunnel_lookup); 172 173 static struct hlist_head *ip_bucket(struct ip_tunnel_net *itn, 174 struct ip_tunnel_parm *parms) 175 { 176 unsigned int h; 177 __be32 remote; 178 __be32 i_key = parms->i_key; 179 180 if (parms->iph.daddr && !ipv4_is_multicast(parms->iph.daddr)) 181 remote = parms->iph.daddr; 182 else 183 remote = 0; 184 185 if (!(parms->i_flags & TUNNEL_KEY) && (parms->i_flags & VTI_ISVTI)) 186 i_key = 0; 187 188 h = ip_tunnel_hash(i_key, remote); 189 return &itn->tunnels[h]; 190 } 191 192 static void ip_tunnel_add(struct ip_tunnel_net *itn, struct ip_tunnel *t) 193 { 194 struct hlist_head *head = ip_bucket(itn, &t->parms); 195 196 if (t->collect_md) 197 rcu_assign_pointer(itn->collect_md_tun, t); 198 hlist_add_head_rcu(&t->hash_node, head); 199 } 200 201 static void ip_tunnel_del(struct ip_tunnel_net *itn, struct ip_tunnel *t) 202 { 203 if (t->collect_md) 204 rcu_assign_pointer(itn->collect_md_tun, NULL); 205 hlist_del_init_rcu(&t->hash_node); 206 } 207 208 static struct ip_tunnel *ip_tunnel_find(struct ip_tunnel_net *itn, 209 struct ip_tunnel_parm *parms, 210 int type) 211 { 212 __be32 remote = parms->iph.daddr; 213 __be32 local = parms->iph.saddr; 214 __be32 key = parms->i_key; 215 __be16 flags = parms->i_flags; 216 int link = parms->link; 217 struct ip_tunnel *t = NULL; 218 struct hlist_head *head = ip_bucket(itn, parms); 219 220 hlist_for_each_entry_rcu(t, head, hash_node) { 221 if (local == t->parms.iph.saddr && 222 remote == t->parms.iph.daddr && 223 link == READ_ONCE(t->parms.link) && 224 type == t->dev->type && 225 ip_tunnel_key_match(&t->parms, flags, key)) 226 break; 227 } 228 return t; 229 } 230 231 static struct net_device *__ip_tunnel_create(struct net *net, 232 const struct rtnl_link_ops *ops, 233 struct ip_tunnel_parm *parms) 234 { 235 int err; 236 struct ip_tunnel *tunnel; 237 struct net_device *dev; 238 char name[IFNAMSIZ]; 239 240 err = -E2BIG; 241 if (parms->name[0]) { 242 if (!dev_valid_name(parms->name)) 243 goto failed; 244 strscpy(name, parms->name, IFNAMSIZ); 245 } else { 246 if (strlen(ops->kind) > (IFNAMSIZ - 3)) 247 goto failed; 248 strcpy(name, ops->kind); 249 strcat(name, "%d"); 250 } 251 252 ASSERT_RTNL(); 253 dev = alloc_netdev(ops->priv_size, name, NET_NAME_UNKNOWN, ops->setup); 254 if (!dev) { 255 err = -ENOMEM; 256 goto failed; 257 } 258 dev_net_set(dev, net); 259 260 dev->rtnl_link_ops = ops; 261 262 tunnel = netdev_priv(dev); 263 tunnel->parms = *parms; 264 tunnel->net = net; 265 266 err = register_netdevice(dev); 267 if (err) 268 goto failed_free; 269 270 return dev; 271 272 failed_free: 273 free_netdev(dev); 274 failed: 275 return ERR_PTR(err); 276 } 277 278 static int ip_tunnel_bind_dev(struct net_device *dev) 279 { 280 struct net_device *tdev = NULL; 281 struct ip_tunnel *tunnel = netdev_priv(dev); 282 const struct iphdr *iph; 283 int hlen = LL_MAX_HEADER; 284 int mtu = ETH_DATA_LEN; 285 int t_hlen = tunnel->hlen + sizeof(struct iphdr); 286 287 iph = &tunnel->parms.iph; 288 289 /* Guess output device to choose reasonable mtu and needed_headroom */ 290 if (iph->daddr) { 291 struct flowi4 fl4; 292 struct rtable *rt; 293 294 ip_tunnel_init_flow(&fl4, iph->protocol, iph->daddr, 295 iph->saddr, tunnel->parms.o_key, 296 RT_TOS(iph->tos), dev_net(dev), 297 tunnel->parms.link, tunnel->fwmark, 0, 0); 298 rt = ip_route_output_key(tunnel->net, &fl4); 299 300 if (!IS_ERR(rt)) { 301 tdev = rt->dst.dev; 302 ip_rt_put(rt); 303 } 304 if (dev->type != ARPHRD_ETHER) 305 dev->flags |= IFF_POINTOPOINT; 306 307 dst_cache_reset(&tunnel->dst_cache); 308 } 309 310 if (!tdev && tunnel->parms.link) 311 tdev = __dev_get_by_index(tunnel->net, tunnel->parms.link); 312 313 if (tdev) { 314 hlen = tdev->hard_header_len + tdev->needed_headroom; 315 mtu = min(tdev->mtu, IP_MAX_MTU); 316 } 317 318 dev->needed_headroom = t_hlen + hlen; 319 mtu -= t_hlen + (dev->type == ARPHRD_ETHER ? dev->hard_header_len : 0); 320 321 if (mtu < IPV4_MIN_MTU) 322 mtu = IPV4_MIN_MTU; 323 324 return mtu; 325 } 326 327 static struct ip_tunnel *ip_tunnel_create(struct net *net, 328 struct ip_tunnel_net *itn, 329 struct ip_tunnel_parm *parms) 330 { 331 struct ip_tunnel *nt; 332 struct net_device *dev; 333 int t_hlen; 334 int mtu; 335 int err; 336 337 dev = __ip_tunnel_create(net, itn->rtnl_link_ops, parms); 338 if (IS_ERR(dev)) 339 return ERR_CAST(dev); 340 341 mtu = ip_tunnel_bind_dev(dev); 342 err = dev_set_mtu(dev, mtu); 343 if (err) 344 goto err_dev_set_mtu; 345 346 nt = netdev_priv(dev); 347 t_hlen = nt->hlen + sizeof(struct iphdr); 348 dev->min_mtu = ETH_MIN_MTU; 349 dev->max_mtu = IP_MAX_MTU - t_hlen; 350 if (dev->type == ARPHRD_ETHER) 351 dev->max_mtu -= dev->hard_header_len; 352 353 ip_tunnel_add(itn, nt); 354 return nt; 355 356 err_dev_set_mtu: 357 unregister_netdevice(dev); 358 return ERR_PTR(err); 359 } 360 361 void ip_tunnel_md_udp_encap(struct sk_buff *skb, struct ip_tunnel_info *info) 362 { 363 const struct iphdr *iph = ip_hdr(skb); 364 const struct udphdr *udph; 365 366 if (iph->protocol != IPPROTO_UDP) 367 return; 368 369 udph = (struct udphdr *)((__u8 *)iph + (iph->ihl << 2)); 370 info->encap.sport = udph->source; 371 info->encap.dport = udph->dest; 372 } 373 EXPORT_SYMBOL(ip_tunnel_md_udp_encap); 374 375 int ip_tunnel_rcv(struct ip_tunnel *tunnel, struct sk_buff *skb, 376 const struct tnl_ptk_info *tpi, struct metadata_dst *tun_dst, 377 bool log_ecn_error) 378 { 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 DEV_STATS_INC(tunnel->dev, 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 DEV_STATS_INC(tunnel->dev, rx_crc_errors); 392 DEV_STATS_INC(tunnel->dev, 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 DEV_STATS_INC(tunnel->dev, rx_fifo_errors); 400 DEV_STATS_INC(tunnel->dev, rx_errors); 401 goto drop; 402 } 403 tunnel->i_seqno = ntohl(tpi->seq) + 1; 404 } 405 406 skb_set_network_header(skb, (tunnel->dev->type == ARPHRD_ETHER) ? ETH_HLEN : 0); 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 DEV_STATS_INC(tunnel->dev, rx_frame_errors); 415 DEV_STATS_INC(tunnel->dev, rx_errors); 416 goto drop; 417 } 418 } 419 420 dev_sw_netstats_rx_add(tunnel->dev, skb->len); 421 skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(tunnel->dev))); 422 423 if (tunnel->dev->type == ARPHRD_ETHER) { 424 skb->protocol = eth_type_trans(skb, tunnel->dev); 425 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 426 } else { 427 skb->dev = tunnel->dev; 428 } 429 430 if (tun_dst) 431 skb_dst_set(skb, (struct dst_entry *)tun_dst); 432 433 gro_cells_receive(&tunnel->gro_cells, skb); 434 return 0; 435 436 drop: 437 if (tun_dst) 438 dst_release((struct dst_entry *)tun_dst); 439 kfree_skb(skb); 440 return 0; 441 } 442 EXPORT_SYMBOL_GPL(ip_tunnel_rcv); 443 444 int ip_tunnel_encap_add_ops(const struct ip_tunnel_encap_ops *ops, 445 unsigned int num) 446 { 447 if (num >= MAX_IPTUN_ENCAP_OPS) 448 return -ERANGE; 449 450 return !cmpxchg((const struct ip_tunnel_encap_ops **) 451 &iptun_encaps[num], 452 NULL, ops) ? 0 : -1; 453 } 454 EXPORT_SYMBOL(ip_tunnel_encap_add_ops); 455 456 int ip_tunnel_encap_del_ops(const struct ip_tunnel_encap_ops *ops, 457 unsigned int num) 458 { 459 int ret; 460 461 if (num >= MAX_IPTUN_ENCAP_OPS) 462 return -ERANGE; 463 464 ret = (cmpxchg((const struct ip_tunnel_encap_ops **) 465 &iptun_encaps[num], 466 ops, NULL) == ops) ? 0 : -1; 467 468 synchronize_net(); 469 470 return ret; 471 } 472 EXPORT_SYMBOL(ip_tunnel_encap_del_ops); 473 474 int ip_tunnel_encap_setup(struct ip_tunnel *t, 475 struct ip_tunnel_encap *ipencap) 476 { 477 int hlen; 478 479 memset(&t->encap, 0, sizeof(t->encap)); 480 481 hlen = ip_encap_hlen(ipencap); 482 if (hlen < 0) 483 return hlen; 484 485 t->encap.type = ipencap->type; 486 t->encap.sport = ipencap->sport; 487 t->encap.dport = ipencap->dport; 488 t->encap.flags = ipencap->flags; 489 490 t->encap_hlen = hlen; 491 t->hlen = t->encap_hlen + t->tun_hlen; 492 493 return 0; 494 } 495 EXPORT_SYMBOL_GPL(ip_tunnel_encap_setup); 496 497 static int tnl_update_pmtu(struct net_device *dev, struct sk_buff *skb, 498 struct rtable *rt, __be16 df, 499 const struct iphdr *inner_iph, 500 int tunnel_hlen, __be32 dst, bool md) 501 { 502 struct ip_tunnel *tunnel = netdev_priv(dev); 503 int pkt_size; 504 int mtu; 505 506 tunnel_hlen = md ? tunnel_hlen : tunnel->hlen; 507 pkt_size = skb->len - tunnel_hlen; 508 pkt_size -= dev->type == ARPHRD_ETHER ? dev->hard_header_len : 0; 509 510 if (df) { 511 mtu = dst_mtu(&rt->dst) - (sizeof(struct iphdr) + tunnel_hlen); 512 mtu -= dev->type == ARPHRD_ETHER ? dev->hard_header_len : 0; 513 } else { 514 mtu = skb_valid_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu; 515 } 516 517 if (skb_valid_dst(skb)) 518 skb_dst_update_pmtu_no_confirm(skb, mtu); 519 520 if (skb->protocol == htons(ETH_P_IP)) { 521 if (!skb_is_gso(skb) && 522 (inner_iph->frag_off & htons(IP_DF)) && 523 mtu < pkt_size) { 524 icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu)); 525 return -E2BIG; 526 } 527 } 528 #if IS_ENABLED(CONFIG_IPV6) 529 else if (skb->protocol == htons(ETH_P_IPV6)) { 530 struct rt6_info *rt6; 531 __be32 daddr; 532 533 rt6 = skb_valid_dst(skb) ? (struct rt6_info *)skb_dst(skb) : 534 NULL; 535 daddr = md ? dst : tunnel->parms.iph.daddr; 536 537 if (rt6 && mtu < dst_mtu(skb_dst(skb)) && 538 mtu >= IPV6_MIN_MTU) { 539 if ((daddr && !ipv4_is_multicast(daddr)) || 540 rt6->rt6i_dst.plen == 128) { 541 rt6->rt6i_flags |= RTF_MODIFIED; 542 dst_metric_set(skb_dst(skb), RTAX_MTU, mtu); 543 } 544 } 545 546 if (!skb_is_gso(skb) && mtu >= IPV6_MIN_MTU && 547 mtu < pkt_size) { 548 icmpv6_ndo_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); 549 return -E2BIG; 550 } 551 } 552 #endif 553 return 0; 554 } 555 556 static void ip_tunnel_adj_headroom(struct net_device *dev, unsigned int headroom) 557 { 558 /* we must cap headroom to some upperlimit, else pskb_expand_head 559 * will overflow header offsets in skb_headers_offset_update(). 560 */ 561 static const unsigned int max_allowed = 512; 562 563 if (headroom > max_allowed) 564 headroom = max_allowed; 565 566 if (headroom > READ_ONCE(dev->needed_headroom)) 567 WRITE_ONCE(dev->needed_headroom, headroom); 568 } 569 570 void ip_md_tunnel_xmit(struct sk_buff *skb, struct net_device *dev, 571 u8 proto, int tunnel_hlen) 572 { 573 struct ip_tunnel *tunnel = netdev_priv(dev); 574 u32 headroom = sizeof(struct iphdr); 575 struct ip_tunnel_info *tun_info; 576 const struct ip_tunnel_key *key; 577 const struct iphdr *inner_iph; 578 struct rtable *rt = NULL; 579 struct flowi4 fl4; 580 __be16 df = 0; 581 u8 tos, ttl; 582 bool use_cache; 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 tx_error; 588 key = &tun_info->key; 589 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); 590 inner_iph = (const struct iphdr *)skb_inner_network_header(skb); 591 tos = key->tos; 592 if (tos == 1) { 593 if (skb->protocol == htons(ETH_P_IP)) 594 tos = inner_iph->tos; 595 else if (skb->protocol == htons(ETH_P_IPV6)) 596 tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph); 597 } 598 ip_tunnel_init_flow(&fl4, proto, key->u.ipv4.dst, key->u.ipv4.src, 599 tunnel_id_to_key32(key->tun_id), RT_TOS(tos), 600 dev_net(dev), 0, skb->mark, skb_get_hash(skb), 601 key->flow_flags); 602 603 if (!tunnel_hlen) 604 tunnel_hlen = ip_encap_hlen(&tun_info->encap); 605 606 if (ip_tunnel_encap(skb, &tun_info->encap, &proto, &fl4) < 0) 607 goto tx_error; 608 609 use_cache = ip_tunnel_dst_cache_usable(skb, tun_info); 610 if (use_cache) 611 rt = dst_cache_get_ip4(&tun_info->dst_cache, &fl4.saddr); 612 if (!rt) { 613 rt = ip_route_output_key(tunnel->net, &fl4); 614 if (IS_ERR(rt)) { 615 DEV_STATS_INC(dev, tx_carrier_errors); 616 goto tx_error; 617 } 618 if (use_cache) 619 dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst, 620 fl4.saddr); 621 } 622 if (rt->dst.dev == dev) { 623 ip_rt_put(rt); 624 DEV_STATS_INC(dev, collisions); 625 goto tx_error; 626 } 627 628 if (key->tun_flags & TUNNEL_DONT_FRAGMENT) 629 df = htons(IP_DF); 630 if (tnl_update_pmtu(dev, skb, rt, df, inner_iph, tunnel_hlen, 631 key->u.ipv4.dst, true)) { 632 ip_rt_put(rt); 633 goto tx_error; 634 } 635 636 tos = ip_tunnel_ecn_encap(tos, inner_iph, skb); 637 ttl = key->ttl; 638 if (ttl == 0) { 639 if (skb->protocol == htons(ETH_P_IP)) 640 ttl = inner_iph->ttl; 641 else if (skb->protocol == htons(ETH_P_IPV6)) 642 ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit; 643 else 644 ttl = ip4_dst_hoplimit(&rt->dst); 645 } 646 647 headroom += LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len; 648 if (skb_cow_head(skb, headroom)) { 649 ip_rt_put(rt); 650 goto tx_dropped; 651 } 652 653 ip_tunnel_adj_headroom(dev, headroom); 654 655 iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, proto, tos, ttl, 656 df, !net_eq(tunnel->net, dev_net(dev))); 657 return; 658 tx_error: 659 DEV_STATS_INC(dev, tx_errors); 660 goto kfree; 661 tx_dropped: 662 DEV_STATS_INC(dev, tx_dropped); 663 kfree: 664 kfree_skb(skb); 665 } 666 EXPORT_SYMBOL_GPL(ip_md_tunnel_xmit); 667 668 void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev, 669 const struct iphdr *tnl_params, u8 protocol) 670 { 671 struct ip_tunnel *tunnel = netdev_priv(dev); 672 struct ip_tunnel_info *tun_info = NULL; 673 const struct iphdr *inner_iph; 674 unsigned int max_headroom; /* The extra header space needed */ 675 struct rtable *rt = NULL; /* Route to the other host */ 676 __be16 payload_protocol; 677 bool use_cache = false; 678 struct flowi4 fl4; 679 bool md = false; 680 bool connected; 681 u8 tos, ttl; 682 __be32 dst; 683 __be16 df; 684 685 inner_iph = (const struct iphdr *)skb_inner_network_header(skb); 686 connected = (tunnel->parms.iph.daddr != 0); 687 payload_protocol = skb_protocol(skb, true); 688 689 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); 690 691 dst = tnl_params->daddr; 692 if (dst == 0) { 693 /* NBMA tunnel */ 694 695 if (!skb_dst(skb)) { 696 DEV_STATS_INC(dev, tx_fifo_errors); 697 goto tx_error; 698 } 699 700 tun_info = skb_tunnel_info(skb); 701 if (tun_info && (tun_info->mode & IP_TUNNEL_INFO_TX) && 702 ip_tunnel_info_af(tun_info) == AF_INET && 703 tun_info->key.u.ipv4.dst) { 704 dst = tun_info->key.u.ipv4.dst; 705 md = true; 706 connected = true; 707 } else if (payload_protocol == htons(ETH_P_IP)) { 708 rt = skb_rtable(skb); 709 dst = rt_nexthop(rt, inner_iph->daddr); 710 } 711 #if IS_ENABLED(CONFIG_IPV6) 712 else if (payload_protocol == htons(ETH_P_IPV6)) { 713 const struct in6_addr *addr6; 714 struct neighbour *neigh; 715 bool do_tx_error_icmp; 716 int addr_type; 717 718 neigh = dst_neigh_lookup(skb_dst(skb), 719 &ipv6_hdr(skb)->daddr); 720 if (!neigh) 721 goto tx_error; 722 723 addr6 = (const struct in6_addr *)&neigh->primary_key; 724 addr_type = ipv6_addr_type(addr6); 725 726 if (addr_type == IPV6_ADDR_ANY) { 727 addr6 = &ipv6_hdr(skb)->daddr; 728 addr_type = ipv6_addr_type(addr6); 729 } 730 731 if ((addr_type & IPV6_ADDR_COMPATv4) == 0) 732 do_tx_error_icmp = true; 733 else { 734 do_tx_error_icmp = false; 735 dst = addr6->s6_addr32[3]; 736 } 737 neigh_release(neigh); 738 if (do_tx_error_icmp) 739 goto tx_error_icmp; 740 } 741 #endif 742 else 743 goto tx_error; 744 745 if (!md) 746 connected = false; 747 } 748 749 tos = tnl_params->tos; 750 if (tos & 0x1) { 751 tos &= ~0x1; 752 if (payload_protocol == htons(ETH_P_IP)) { 753 tos = inner_iph->tos; 754 connected = false; 755 } else if (payload_protocol == htons(ETH_P_IPV6)) { 756 tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph); 757 connected = false; 758 } 759 } 760 761 ip_tunnel_init_flow(&fl4, protocol, dst, tnl_params->saddr, 762 tunnel->parms.o_key, RT_TOS(tos), 763 dev_net(dev), READ_ONCE(tunnel->parms.link), 764 tunnel->fwmark, skb_get_hash(skb), 0); 765 766 if (ip_tunnel_encap(skb, &tunnel->encap, &protocol, &fl4) < 0) 767 goto tx_error; 768 769 if (connected && md) { 770 use_cache = ip_tunnel_dst_cache_usable(skb, tun_info); 771 if (use_cache) 772 rt = dst_cache_get_ip4(&tun_info->dst_cache, 773 &fl4.saddr); 774 } else { 775 rt = connected ? dst_cache_get_ip4(&tunnel->dst_cache, 776 &fl4.saddr) : NULL; 777 } 778 779 if (!rt) { 780 rt = ip_route_output_key(tunnel->net, &fl4); 781 782 if (IS_ERR(rt)) { 783 DEV_STATS_INC(dev, tx_carrier_errors); 784 goto tx_error; 785 } 786 if (use_cache) 787 dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst, 788 fl4.saddr); 789 else if (!md && connected) 790 dst_cache_set_ip4(&tunnel->dst_cache, &rt->dst, 791 fl4.saddr); 792 } 793 794 if (rt->dst.dev == dev) { 795 ip_rt_put(rt); 796 DEV_STATS_INC(dev, collisions); 797 goto tx_error; 798 } 799 800 df = tnl_params->frag_off; 801 if (payload_protocol == htons(ETH_P_IP) && !tunnel->ignore_df) 802 df |= (inner_iph->frag_off & htons(IP_DF)); 803 804 if (tnl_update_pmtu(dev, skb, rt, df, inner_iph, 0, 0, false)) { 805 ip_rt_put(rt); 806 goto tx_error; 807 } 808 809 if (tunnel->err_count > 0) { 810 if (time_before(jiffies, 811 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) { 812 tunnel->err_count--; 813 814 dst_link_failure(skb); 815 } else 816 tunnel->err_count = 0; 817 } 818 819 tos = ip_tunnel_ecn_encap(tos, inner_iph, skb); 820 ttl = tnl_params->ttl; 821 if (ttl == 0) { 822 if (payload_protocol == htons(ETH_P_IP)) 823 ttl = inner_iph->ttl; 824 #if IS_ENABLED(CONFIG_IPV6) 825 else if (payload_protocol == htons(ETH_P_IPV6)) 826 ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit; 827 #endif 828 else 829 ttl = ip4_dst_hoplimit(&rt->dst); 830 } 831 832 max_headroom = LL_RESERVED_SPACE(rt->dst.dev) + sizeof(struct iphdr) 833 + rt->dst.header_len + ip_encap_hlen(&tunnel->encap); 834 835 if (skb_cow_head(skb, max_headroom)) { 836 ip_rt_put(rt); 837 DEV_STATS_INC(dev, tx_dropped); 838 kfree_skb(skb); 839 return; 840 } 841 842 ip_tunnel_adj_headroom(dev, max_headroom); 843 844 iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, protocol, tos, ttl, 845 df, !net_eq(tunnel->net, dev_net(dev))); 846 return; 847 848 #if IS_ENABLED(CONFIG_IPV6) 849 tx_error_icmp: 850 dst_link_failure(skb); 851 #endif 852 tx_error: 853 DEV_STATS_INC(dev, tx_errors); 854 kfree_skb(skb); 855 } 856 EXPORT_SYMBOL_GPL(ip_tunnel_xmit); 857 858 static void ip_tunnel_update(struct ip_tunnel_net *itn, 859 struct ip_tunnel *t, 860 struct net_device *dev, 861 struct ip_tunnel_parm *p, 862 bool set_mtu, 863 __u32 fwmark) 864 { 865 ip_tunnel_del(itn, t); 866 t->parms.iph.saddr = p->iph.saddr; 867 t->parms.iph.daddr = p->iph.daddr; 868 t->parms.i_key = p->i_key; 869 t->parms.o_key = p->o_key; 870 if (dev->type != ARPHRD_ETHER) { 871 __dev_addr_set(dev, &p->iph.saddr, 4); 872 memcpy(dev->broadcast, &p->iph.daddr, 4); 873 } 874 ip_tunnel_add(itn, t); 875 876 t->parms.iph.ttl = p->iph.ttl; 877 t->parms.iph.tos = p->iph.tos; 878 t->parms.iph.frag_off = p->iph.frag_off; 879 880 if (t->parms.link != p->link || t->fwmark != fwmark) { 881 int mtu; 882 883 WRITE_ONCE(t->parms.link, p->link); 884 t->fwmark = fwmark; 885 mtu = ip_tunnel_bind_dev(dev); 886 if (set_mtu) 887 dev->mtu = mtu; 888 } 889 dst_cache_reset(&t->dst_cache); 890 netdev_state_change(dev); 891 } 892 893 int ip_tunnel_ctl(struct net_device *dev, struct ip_tunnel_parm *p, int cmd) 894 { 895 int err = 0; 896 struct ip_tunnel *t = netdev_priv(dev); 897 struct net *net = t->net; 898 struct ip_tunnel_net *itn = net_generic(net, t->ip_tnl_net_id); 899 900 switch (cmd) { 901 case SIOCGETTUNNEL: 902 if (dev == itn->fb_tunnel_dev) { 903 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type); 904 if (!t) 905 t = netdev_priv(dev); 906 } 907 memcpy(p, &t->parms, sizeof(*p)); 908 break; 909 910 case SIOCADDTUNNEL: 911 case SIOCCHGTUNNEL: 912 err = -EPERM; 913 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 914 goto done; 915 if (p->iph.ttl) 916 p->iph.frag_off |= htons(IP_DF); 917 if (!(p->i_flags & VTI_ISVTI)) { 918 if (!(p->i_flags & TUNNEL_KEY)) 919 p->i_key = 0; 920 if (!(p->o_flags & TUNNEL_KEY)) 921 p->o_key = 0; 922 } 923 924 t = ip_tunnel_find(itn, p, itn->type); 925 926 if (cmd == SIOCADDTUNNEL) { 927 if (!t) { 928 t = ip_tunnel_create(net, itn, p); 929 err = PTR_ERR_OR_ZERO(t); 930 break; 931 } 932 933 err = -EEXIST; 934 break; 935 } 936 if (dev != itn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) { 937 if (t) { 938 if (t->dev != dev) { 939 err = -EEXIST; 940 break; 941 } 942 } else { 943 unsigned int nflags = 0; 944 945 if (ipv4_is_multicast(p->iph.daddr)) 946 nflags = IFF_BROADCAST; 947 else if (p->iph.daddr) 948 nflags = IFF_POINTOPOINT; 949 950 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) { 951 err = -EINVAL; 952 break; 953 } 954 955 t = netdev_priv(dev); 956 } 957 } 958 959 if (t) { 960 err = 0; 961 ip_tunnel_update(itn, t, dev, p, true, 0); 962 } else { 963 err = -ENOENT; 964 } 965 break; 966 967 case SIOCDELTUNNEL: 968 err = -EPERM; 969 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 970 goto done; 971 972 if (dev == itn->fb_tunnel_dev) { 973 err = -ENOENT; 974 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type); 975 if (!t) 976 goto done; 977 err = -EPERM; 978 if (t == netdev_priv(itn->fb_tunnel_dev)) 979 goto done; 980 dev = t->dev; 981 } 982 unregister_netdevice(dev); 983 err = 0; 984 break; 985 986 default: 987 err = -EINVAL; 988 } 989 990 done: 991 return err; 992 } 993 EXPORT_SYMBOL_GPL(ip_tunnel_ctl); 994 995 int ip_tunnel_siocdevprivate(struct net_device *dev, struct ifreq *ifr, 996 void __user *data, int cmd) 997 { 998 struct ip_tunnel_parm p; 999 int err; 1000 1001 if (copy_from_user(&p, data, sizeof(p))) 1002 return -EFAULT; 1003 err = dev->netdev_ops->ndo_tunnel_ctl(dev, &p, cmd); 1004 if (!err && copy_to_user(data, &p, sizeof(p))) 1005 return -EFAULT; 1006 return err; 1007 } 1008 EXPORT_SYMBOL_GPL(ip_tunnel_siocdevprivate); 1009 1010 int __ip_tunnel_change_mtu(struct net_device *dev, int new_mtu, bool strict) 1011 { 1012 struct ip_tunnel *tunnel = netdev_priv(dev); 1013 int t_hlen = tunnel->hlen + sizeof(struct iphdr); 1014 int max_mtu = IP_MAX_MTU - t_hlen; 1015 1016 if (dev->type == ARPHRD_ETHER) 1017 max_mtu -= dev->hard_header_len; 1018 1019 if (new_mtu < ETH_MIN_MTU) 1020 return -EINVAL; 1021 1022 if (new_mtu > max_mtu) { 1023 if (strict) 1024 return -EINVAL; 1025 1026 new_mtu = max_mtu; 1027 } 1028 1029 dev->mtu = new_mtu; 1030 return 0; 1031 } 1032 EXPORT_SYMBOL_GPL(__ip_tunnel_change_mtu); 1033 1034 int ip_tunnel_change_mtu(struct net_device *dev, int new_mtu) 1035 { 1036 return __ip_tunnel_change_mtu(dev, new_mtu, true); 1037 } 1038 EXPORT_SYMBOL_GPL(ip_tunnel_change_mtu); 1039 1040 static void ip_tunnel_dev_free(struct net_device *dev) 1041 { 1042 struct ip_tunnel *tunnel = netdev_priv(dev); 1043 1044 gro_cells_destroy(&tunnel->gro_cells); 1045 dst_cache_destroy(&tunnel->dst_cache); 1046 free_percpu(dev->tstats); 1047 } 1048 1049 void ip_tunnel_dellink(struct net_device *dev, struct list_head *head) 1050 { 1051 struct ip_tunnel *tunnel = netdev_priv(dev); 1052 struct ip_tunnel_net *itn; 1053 1054 itn = net_generic(tunnel->net, tunnel->ip_tnl_net_id); 1055 1056 if (itn->fb_tunnel_dev != dev) { 1057 ip_tunnel_del(itn, netdev_priv(dev)); 1058 unregister_netdevice_queue(dev, head); 1059 } 1060 } 1061 EXPORT_SYMBOL_GPL(ip_tunnel_dellink); 1062 1063 struct net *ip_tunnel_get_link_net(const struct net_device *dev) 1064 { 1065 struct ip_tunnel *tunnel = netdev_priv(dev); 1066 1067 return tunnel->net; 1068 } 1069 EXPORT_SYMBOL(ip_tunnel_get_link_net); 1070 1071 int ip_tunnel_get_iflink(const struct net_device *dev) 1072 { 1073 const struct ip_tunnel *tunnel = netdev_priv(dev); 1074 1075 return READ_ONCE(tunnel->parms.link); 1076 } 1077 EXPORT_SYMBOL(ip_tunnel_get_iflink); 1078 1079 int ip_tunnel_init_net(struct net *net, unsigned int ip_tnl_net_id, 1080 struct rtnl_link_ops *ops, char *devname) 1081 { 1082 struct ip_tunnel_net *itn = net_generic(net, ip_tnl_net_id); 1083 struct ip_tunnel_parm parms; 1084 unsigned int i; 1085 1086 itn->rtnl_link_ops = ops; 1087 for (i = 0; i < IP_TNL_HASH_SIZE; i++) 1088 INIT_HLIST_HEAD(&itn->tunnels[i]); 1089 1090 if (!ops || !net_has_fallback_tunnels(net)) { 1091 struct ip_tunnel_net *it_init_net; 1092 1093 it_init_net = net_generic(&init_net, ip_tnl_net_id); 1094 itn->type = it_init_net->type; 1095 itn->fb_tunnel_dev = NULL; 1096 return 0; 1097 } 1098 1099 memset(&parms, 0, sizeof(parms)); 1100 if (devname) 1101 strscpy(parms.name, devname, IFNAMSIZ); 1102 1103 rtnl_lock(); 1104 itn->fb_tunnel_dev = __ip_tunnel_create(net, ops, &parms); 1105 /* FB netdevice is special: we have one, and only one per netns. 1106 * Allowing to move it to another netns is clearly unsafe. 1107 */ 1108 if (!IS_ERR(itn->fb_tunnel_dev)) { 1109 itn->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL; 1110 itn->fb_tunnel_dev->mtu = ip_tunnel_bind_dev(itn->fb_tunnel_dev); 1111 ip_tunnel_add(itn, netdev_priv(itn->fb_tunnel_dev)); 1112 itn->type = itn->fb_tunnel_dev->type; 1113 } 1114 rtnl_unlock(); 1115 1116 return PTR_ERR_OR_ZERO(itn->fb_tunnel_dev); 1117 } 1118 EXPORT_SYMBOL_GPL(ip_tunnel_init_net); 1119 1120 static void ip_tunnel_destroy(struct net *net, struct ip_tunnel_net *itn, 1121 struct list_head *head, 1122 struct rtnl_link_ops *ops) 1123 { 1124 struct net_device *dev, *aux; 1125 int h; 1126 1127 for_each_netdev_safe(net, dev, aux) 1128 if (dev->rtnl_link_ops == ops) 1129 unregister_netdevice_queue(dev, head); 1130 1131 for (h = 0; h < IP_TNL_HASH_SIZE; h++) { 1132 struct ip_tunnel *t; 1133 struct hlist_node *n; 1134 struct hlist_head *thead = &itn->tunnels[h]; 1135 1136 hlist_for_each_entry_safe(t, n, thead, hash_node) 1137 /* If dev is in the same netns, it has already 1138 * been added to the list by the previous loop. 1139 */ 1140 if (!net_eq(dev_net(t->dev), net)) 1141 unregister_netdevice_queue(t->dev, head); 1142 } 1143 } 1144 1145 void ip_tunnel_delete_nets(struct list_head *net_list, unsigned int id, 1146 struct rtnl_link_ops *ops, 1147 struct list_head *dev_to_kill) 1148 { 1149 struct ip_tunnel_net *itn; 1150 struct net *net; 1151 1152 ASSERT_RTNL(); 1153 list_for_each_entry(net, net_list, exit_list) { 1154 itn = net_generic(net, id); 1155 ip_tunnel_destroy(net, itn, dev_to_kill, ops); 1156 } 1157 } 1158 EXPORT_SYMBOL_GPL(ip_tunnel_delete_nets); 1159 1160 int ip_tunnel_newlink(struct net_device *dev, struct nlattr *tb[], 1161 struct ip_tunnel_parm *p, __u32 fwmark) 1162 { 1163 struct ip_tunnel *nt; 1164 struct net *net = dev_net(dev); 1165 struct ip_tunnel_net *itn; 1166 int mtu; 1167 int err; 1168 1169 nt = netdev_priv(dev); 1170 itn = net_generic(net, nt->ip_tnl_net_id); 1171 1172 if (nt->collect_md) { 1173 if (rtnl_dereference(itn->collect_md_tun)) 1174 return -EEXIST; 1175 } else { 1176 if (ip_tunnel_find(itn, p, dev->type)) 1177 return -EEXIST; 1178 } 1179 1180 nt->net = net; 1181 nt->parms = *p; 1182 nt->fwmark = fwmark; 1183 err = register_netdevice(dev); 1184 if (err) 1185 goto err_register_netdevice; 1186 1187 if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS]) 1188 eth_hw_addr_random(dev); 1189 1190 mtu = ip_tunnel_bind_dev(dev); 1191 if (tb[IFLA_MTU]) { 1192 unsigned int max = IP_MAX_MTU - (nt->hlen + sizeof(struct iphdr)); 1193 1194 if (dev->type == ARPHRD_ETHER) 1195 max -= dev->hard_header_len; 1196 1197 mtu = clamp(dev->mtu, (unsigned int)ETH_MIN_MTU, max); 1198 } 1199 1200 err = dev_set_mtu(dev, mtu); 1201 if (err) 1202 goto err_dev_set_mtu; 1203 1204 ip_tunnel_add(itn, nt); 1205 return 0; 1206 1207 err_dev_set_mtu: 1208 unregister_netdevice(dev); 1209 err_register_netdevice: 1210 return err; 1211 } 1212 EXPORT_SYMBOL_GPL(ip_tunnel_newlink); 1213 1214 int ip_tunnel_changelink(struct net_device *dev, struct nlattr *tb[], 1215 struct ip_tunnel_parm *p, __u32 fwmark) 1216 { 1217 struct ip_tunnel *t; 1218 struct ip_tunnel *tunnel = netdev_priv(dev); 1219 struct net *net = tunnel->net; 1220 struct ip_tunnel_net *itn = net_generic(net, tunnel->ip_tnl_net_id); 1221 1222 if (dev == itn->fb_tunnel_dev) 1223 return -EINVAL; 1224 1225 t = ip_tunnel_find(itn, p, dev->type); 1226 1227 if (t) { 1228 if (t->dev != dev) 1229 return -EEXIST; 1230 } else { 1231 t = tunnel; 1232 1233 if (dev->type != ARPHRD_ETHER) { 1234 unsigned int nflags = 0; 1235 1236 if (ipv4_is_multicast(p->iph.daddr)) 1237 nflags = IFF_BROADCAST; 1238 else if (p->iph.daddr) 1239 nflags = IFF_POINTOPOINT; 1240 1241 if ((dev->flags ^ nflags) & 1242 (IFF_POINTOPOINT | IFF_BROADCAST)) 1243 return -EINVAL; 1244 } 1245 } 1246 1247 ip_tunnel_update(itn, t, dev, p, !tb[IFLA_MTU], fwmark); 1248 return 0; 1249 } 1250 EXPORT_SYMBOL_GPL(ip_tunnel_changelink); 1251 1252 int ip_tunnel_init(struct net_device *dev) 1253 { 1254 struct ip_tunnel *tunnel = netdev_priv(dev); 1255 struct iphdr *iph = &tunnel->parms.iph; 1256 int err; 1257 1258 dev->needs_free_netdev = true; 1259 dev->priv_destructor = ip_tunnel_dev_free; 1260 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 1261 if (!dev->tstats) 1262 return -ENOMEM; 1263 1264 err = dst_cache_init(&tunnel->dst_cache, GFP_KERNEL); 1265 if (err) { 1266 free_percpu(dev->tstats); 1267 return err; 1268 } 1269 1270 err = gro_cells_init(&tunnel->gro_cells, dev); 1271 if (err) { 1272 dst_cache_destroy(&tunnel->dst_cache); 1273 free_percpu(dev->tstats); 1274 return err; 1275 } 1276 1277 tunnel->dev = dev; 1278 tunnel->net = dev_net(dev); 1279 strcpy(tunnel->parms.name, dev->name); 1280 iph->version = 4; 1281 iph->ihl = 5; 1282 1283 if (tunnel->collect_md) 1284 netif_keep_dst(dev); 1285 netdev_lockdep_set_classes(dev); 1286 return 0; 1287 } 1288 EXPORT_SYMBOL_GPL(ip_tunnel_init); 1289 1290 void ip_tunnel_uninit(struct net_device *dev) 1291 { 1292 struct ip_tunnel *tunnel = netdev_priv(dev); 1293 struct net *net = tunnel->net; 1294 struct ip_tunnel_net *itn; 1295 1296 itn = net_generic(net, tunnel->ip_tnl_net_id); 1297 ip_tunnel_del(itn, netdev_priv(dev)); 1298 if (itn->fb_tunnel_dev == dev) 1299 WRITE_ONCE(itn->fb_tunnel_dev, NULL); 1300 1301 dst_cache_reset(&tunnel->dst_cache); 1302 } 1303 EXPORT_SYMBOL_GPL(ip_tunnel_uninit); 1304 1305 /* Do least required initialization, rest of init is done in tunnel_init call */ 1306 void ip_tunnel_setup(struct net_device *dev, unsigned int net_id) 1307 { 1308 struct ip_tunnel *tunnel = netdev_priv(dev); 1309 tunnel->ip_tnl_net_id = net_id; 1310 } 1311 EXPORT_SYMBOL_GPL(ip_tunnel_setup); 1312 1313 MODULE_DESCRIPTION("IPv4 tunnel implementation library"); 1314 MODULE_LICENSE("GPL"); 1315