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