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 strscpy(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), dev_net(dev), 298 tunnel->parms.link, 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 -= t_hlen + (dev->type == ARPHRD_ETHER ? dev->hard_header_len : 0); 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 - t_hlen; 351 if (dev->type == ARPHRD_ETHER) 352 dev->max_mtu -= dev->hard_header_len; 353 354 ip_tunnel_add(itn, nt); 355 return nt; 356 357 err_dev_set_mtu: 358 unregister_netdevice(dev); 359 return ERR_PTR(err); 360 } 361 362 int ip_tunnel_rcv(struct ip_tunnel *tunnel, struct sk_buff *skb, 363 const struct tnl_ptk_info *tpi, struct metadata_dst *tun_dst, 364 bool log_ecn_error) 365 { 366 const struct iphdr *iph = ip_hdr(skb); 367 int err; 368 369 #ifdef CONFIG_NET_IPGRE_BROADCAST 370 if (ipv4_is_multicast(iph->daddr)) { 371 tunnel->dev->stats.multicast++; 372 skb->pkt_type = PACKET_BROADCAST; 373 } 374 #endif 375 376 if ((!(tpi->flags&TUNNEL_CSUM) && (tunnel->parms.i_flags&TUNNEL_CSUM)) || 377 ((tpi->flags&TUNNEL_CSUM) && !(tunnel->parms.i_flags&TUNNEL_CSUM))) { 378 tunnel->dev->stats.rx_crc_errors++; 379 tunnel->dev->stats.rx_errors++; 380 goto drop; 381 } 382 383 if (tunnel->parms.i_flags&TUNNEL_SEQ) { 384 if (!(tpi->flags&TUNNEL_SEQ) || 385 (tunnel->i_seqno && (s32)(ntohl(tpi->seq) - tunnel->i_seqno) < 0)) { 386 tunnel->dev->stats.rx_fifo_errors++; 387 tunnel->dev->stats.rx_errors++; 388 goto drop; 389 } 390 tunnel->i_seqno = ntohl(tpi->seq) + 1; 391 } 392 393 skb_set_network_header(skb, (tunnel->dev->type == ARPHRD_ETHER) ? ETH_HLEN : 0); 394 395 err = IP_ECN_decapsulate(iph, skb); 396 if (unlikely(err)) { 397 if (log_ecn_error) 398 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n", 399 &iph->saddr, iph->tos); 400 if (err > 1) { 401 ++tunnel->dev->stats.rx_frame_errors; 402 ++tunnel->dev->stats.rx_errors; 403 goto drop; 404 } 405 } 406 407 dev_sw_netstats_rx_add(tunnel->dev, skb->len); 408 skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(tunnel->dev))); 409 410 if (tunnel->dev->type == ARPHRD_ETHER) { 411 skb->protocol = eth_type_trans(skb, tunnel->dev); 412 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 413 } else { 414 skb->dev = tunnel->dev; 415 } 416 417 if (tun_dst) 418 skb_dst_set(skb, (struct dst_entry *)tun_dst); 419 420 gro_cells_receive(&tunnel->gro_cells, skb); 421 return 0; 422 423 drop: 424 if (tun_dst) 425 dst_release((struct dst_entry *)tun_dst); 426 kfree_skb(skb); 427 return 0; 428 } 429 EXPORT_SYMBOL_GPL(ip_tunnel_rcv); 430 431 int ip_tunnel_encap_add_ops(const struct ip_tunnel_encap_ops *ops, 432 unsigned int num) 433 { 434 if (num >= MAX_IPTUN_ENCAP_OPS) 435 return -ERANGE; 436 437 return !cmpxchg((const struct ip_tunnel_encap_ops **) 438 &iptun_encaps[num], 439 NULL, ops) ? 0 : -1; 440 } 441 EXPORT_SYMBOL(ip_tunnel_encap_add_ops); 442 443 int ip_tunnel_encap_del_ops(const struct ip_tunnel_encap_ops *ops, 444 unsigned int num) 445 { 446 int ret; 447 448 if (num >= MAX_IPTUN_ENCAP_OPS) 449 return -ERANGE; 450 451 ret = (cmpxchg((const struct ip_tunnel_encap_ops **) 452 &iptun_encaps[num], 453 ops, NULL) == ops) ? 0 : -1; 454 455 synchronize_net(); 456 457 return ret; 458 } 459 EXPORT_SYMBOL(ip_tunnel_encap_del_ops); 460 461 int ip_tunnel_encap_setup(struct ip_tunnel *t, 462 struct ip_tunnel_encap *ipencap) 463 { 464 int hlen; 465 466 memset(&t->encap, 0, sizeof(t->encap)); 467 468 hlen = ip_encap_hlen(ipencap); 469 if (hlen < 0) 470 return hlen; 471 472 t->encap.type = ipencap->type; 473 t->encap.sport = ipencap->sport; 474 t->encap.dport = ipencap->dport; 475 t->encap.flags = ipencap->flags; 476 477 t->encap_hlen = hlen; 478 t->hlen = t->encap_hlen + t->tun_hlen; 479 480 return 0; 481 } 482 EXPORT_SYMBOL_GPL(ip_tunnel_encap_setup); 483 484 static int tnl_update_pmtu(struct net_device *dev, struct sk_buff *skb, 485 struct rtable *rt, __be16 df, 486 const struct iphdr *inner_iph, 487 int tunnel_hlen, __be32 dst, bool md) 488 { 489 struct ip_tunnel *tunnel = netdev_priv(dev); 490 int pkt_size; 491 int mtu; 492 493 tunnel_hlen = md ? tunnel_hlen : tunnel->hlen; 494 pkt_size = skb->len - tunnel_hlen; 495 pkt_size -= dev->type == ARPHRD_ETHER ? dev->hard_header_len : 0; 496 497 if (df) { 498 mtu = dst_mtu(&rt->dst) - (sizeof(struct iphdr) + tunnel_hlen); 499 mtu -= dev->type == ARPHRD_ETHER ? dev->hard_header_len : 0; 500 } else { 501 mtu = skb_valid_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu; 502 } 503 504 if (skb_valid_dst(skb)) 505 skb_dst_update_pmtu_no_confirm(skb, mtu); 506 507 if (skb->protocol == htons(ETH_P_IP)) { 508 if (!skb_is_gso(skb) && 509 (inner_iph->frag_off & htons(IP_DF)) && 510 mtu < pkt_size) { 511 icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu)); 512 return -E2BIG; 513 } 514 } 515 #if IS_ENABLED(CONFIG_IPV6) 516 else if (skb->protocol == htons(ETH_P_IPV6)) { 517 struct rt6_info *rt6; 518 __be32 daddr; 519 520 rt6 = skb_valid_dst(skb) ? (struct rt6_info *)skb_dst(skb) : 521 NULL; 522 daddr = md ? dst : tunnel->parms.iph.daddr; 523 524 if (rt6 && mtu < dst_mtu(skb_dst(skb)) && 525 mtu >= IPV6_MIN_MTU) { 526 if ((daddr && !ipv4_is_multicast(daddr)) || 527 rt6->rt6i_dst.plen == 128) { 528 rt6->rt6i_flags |= RTF_MODIFIED; 529 dst_metric_set(skb_dst(skb), RTAX_MTU, mtu); 530 } 531 } 532 533 if (!skb_is_gso(skb) && mtu >= IPV6_MIN_MTU && 534 mtu < pkt_size) { 535 icmpv6_ndo_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); 536 return -E2BIG; 537 } 538 } 539 #endif 540 return 0; 541 } 542 543 void ip_md_tunnel_xmit(struct sk_buff *skb, struct net_device *dev, 544 u8 proto, int tunnel_hlen) 545 { 546 struct ip_tunnel *tunnel = netdev_priv(dev); 547 u32 headroom = sizeof(struct iphdr); 548 struct ip_tunnel_info *tun_info; 549 const struct ip_tunnel_key *key; 550 const struct iphdr *inner_iph; 551 struct rtable *rt = NULL; 552 struct flowi4 fl4; 553 __be16 df = 0; 554 u8 tos, ttl; 555 bool use_cache; 556 557 tun_info = skb_tunnel_info(skb); 558 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) || 559 ip_tunnel_info_af(tun_info) != AF_INET)) 560 goto tx_error; 561 key = &tun_info->key; 562 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); 563 inner_iph = (const struct iphdr *)skb_inner_network_header(skb); 564 tos = key->tos; 565 if (tos == 1) { 566 if (skb->protocol == htons(ETH_P_IP)) 567 tos = inner_iph->tos; 568 else if (skb->protocol == htons(ETH_P_IPV6)) 569 tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph); 570 } 571 ip_tunnel_init_flow(&fl4, proto, key->u.ipv4.dst, key->u.ipv4.src, 572 tunnel_id_to_key32(key->tun_id), RT_TOS(tos), 573 dev_net(dev), 0, skb->mark, skb_get_hash(skb)); 574 if (tunnel->encap.type != TUNNEL_ENCAP_NONE) 575 goto tx_error; 576 577 use_cache = ip_tunnel_dst_cache_usable(skb, tun_info); 578 if (use_cache) 579 rt = dst_cache_get_ip4(&tun_info->dst_cache, &fl4.saddr); 580 if (!rt) { 581 rt = ip_route_output_key(tunnel->net, &fl4); 582 if (IS_ERR(rt)) { 583 dev->stats.tx_carrier_errors++; 584 goto tx_error; 585 } 586 if (use_cache) 587 dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst, 588 fl4.saddr); 589 } 590 if (rt->dst.dev == dev) { 591 ip_rt_put(rt); 592 dev->stats.collisions++; 593 goto tx_error; 594 } 595 596 if (key->tun_flags & TUNNEL_DONT_FRAGMENT) 597 df = htons(IP_DF); 598 if (tnl_update_pmtu(dev, skb, rt, df, inner_iph, tunnel_hlen, 599 key->u.ipv4.dst, true)) { 600 ip_rt_put(rt); 601 goto tx_error; 602 } 603 604 tos = ip_tunnel_ecn_encap(tos, inner_iph, skb); 605 ttl = key->ttl; 606 if (ttl == 0) { 607 if (skb->protocol == htons(ETH_P_IP)) 608 ttl = inner_iph->ttl; 609 else if (skb->protocol == htons(ETH_P_IPV6)) 610 ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit; 611 else 612 ttl = ip4_dst_hoplimit(&rt->dst); 613 } 614 615 headroom += LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len; 616 if (headroom > dev->needed_headroom) 617 dev->needed_headroom = headroom; 618 619 if (skb_cow_head(skb, dev->needed_headroom)) { 620 ip_rt_put(rt); 621 goto tx_dropped; 622 } 623 iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, proto, tos, ttl, 624 df, !net_eq(tunnel->net, dev_net(dev))); 625 return; 626 tx_error: 627 dev->stats.tx_errors++; 628 goto kfree; 629 tx_dropped: 630 dev->stats.tx_dropped++; 631 kfree: 632 kfree_skb(skb); 633 } 634 EXPORT_SYMBOL_GPL(ip_md_tunnel_xmit); 635 636 void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev, 637 const struct iphdr *tnl_params, u8 protocol) 638 { 639 struct ip_tunnel *tunnel = netdev_priv(dev); 640 struct ip_tunnel_info *tun_info = NULL; 641 const struct iphdr *inner_iph; 642 unsigned int max_headroom; /* The extra header space needed */ 643 struct rtable *rt = NULL; /* Route to the other host */ 644 __be16 payload_protocol; 645 bool use_cache = false; 646 struct flowi4 fl4; 647 bool md = false; 648 bool connected; 649 u8 tos, ttl; 650 __be32 dst; 651 __be16 df; 652 653 inner_iph = (const struct iphdr *)skb_inner_network_header(skb); 654 connected = (tunnel->parms.iph.daddr != 0); 655 payload_protocol = skb_protocol(skb, true); 656 657 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); 658 659 dst = tnl_params->daddr; 660 if (dst == 0) { 661 /* NBMA tunnel */ 662 663 if (!skb_dst(skb)) { 664 dev->stats.tx_fifo_errors++; 665 goto tx_error; 666 } 667 668 tun_info = skb_tunnel_info(skb); 669 if (tun_info && (tun_info->mode & IP_TUNNEL_INFO_TX) && 670 ip_tunnel_info_af(tun_info) == AF_INET && 671 tun_info->key.u.ipv4.dst) { 672 dst = tun_info->key.u.ipv4.dst; 673 md = true; 674 connected = true; 675 } else if (payload_protocol == htons(ETH_P_IP)) { 676 rt = skb_rtable(skb); 677 dst = rt_nexthop(rt, inner_iph->daddr); 678 } 679 #if IS_ENABLED(CONFIG_IPV6) 680 else if (payload_protocol == htons(ETH_P_IPV6)) { 681 const struct in6_addr *addr6; 682 struct neighbour *neigh; 683 bool do_tx_error_icmp; 684 int addr_type; 685 686 neigh = dst_neigh_lookup(skb_dst(skb), 687 &ipv6_hdr(skb)->daddr); 688 if (!neigh) 689 goto tx_error; 690 691 addr6 = (const struct in6_addr *)&neigh->primary_key; 692 addr_type = ipv6_addr_type(addr6); 693 694 if (addr_type == IPV6_ADDR_ANY) { 695 addr6 = &ipv6_hdr(skb)->daddr; 696 addr_type = ipv6_addr_type(addr6); 697 } 698 699 if ((addr_type & IPV6_ADDR_COMPATv4) == 0) 700 do_tx_error_icmp = true; 701 else { 702 do_tx_error_icmp = false; 703 dst = addr6->s6_addr32[3]; 704 } 705 neigh_release(neigh); 706 if (do_tx_error_icmp) 707 goto tx_error_icmp; 708 } 709 #endif 710 else 711 goto tx_error; 712 713 if (!md) 714 connected = false; 715 } 716 717 tos = tnl_params->tos; 718 if (tos & 0x1) { 719 tos &= ~0x1; 720 if (payload_protocol == htons(ETH_P_IP)) { 721 tos = inner_iph->tos; 722 connected = false; 723 } else if (payload_protocol == htons(ETH_P_IPV6)) { 724 tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph); 725 connected = false; 726 } 727 } 728 729 ip_tunnel_init_flow(&fl4, protocol, dst, tnl_params->saddr, 730 tunnel->parms.o_key, RT_TOS(tos), 731 dev_net(dev), tunnel->parms.link, 732 tunnel->fwmark, skb_get_hash(skb)); 733 734 if (ip_tunnel_encap(skb, tunnel, &protocol, &fl4) < 0) 735 goto tx_error; 736 737 if (connected && md) { 738 use_cache = ip_tunnel_dst_cache_usable(skb, tun_info); 739 if (use_cache) 740 rt = dst_cache_get_ip4(&tun_info->dst_cache, 741 &fl4.saddr); 742 } else { 743 rt = connected ? dst_cache_get_ip4(&tunnel->dst_cache, 744 &fl4.saddr) : NULL; 745 } 746 747 if (!rt) { 748 rt = ip_route_output_key(tunnel->net, &fl4); 749 750 if (IS_ERR(rt)) { 751 dev->stats.tx_carrier_errors++; 752 goto tx_error; 753 } 754 if (use_cache) 755 dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst, 756 fl4.saddr); 757 else if (!md && connected) 758 dst_cache_set_ip4(&tunnel->dst_cache, &rt->dst, 759 fl4.saddr); 760 } 761 762 if (rt->dst.dev == dev) { 763 ip_rt_put(rt); 764 dev->stats.collisions++; 765 goto tx_error; 766 } 767 768 df = tnl_params->frag_off; 769 if (payload_protocol == htons(ETH_P_IP) && !tunnel->ignore_df) 770 df |= (inner_iph->frag_off & htons(IP_DF)); 771 772 if (tnl_update_pmtu(dev, skb, rt, df, inner_iph, 0, 0, false)) { 773 ip_rt_put(rt); 774 goto tx_error; 775 } 776 777 if (tunnel->err_count > 0) { 778 if (time_before(jiffies, 779 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) { 780 tunnel->err_count--; 781 782 dst_link_failure(skb); 783 } else 784 tunnel->err_count = 0; 785 } 786 787 tos = ip_tunnel_ecn_encap(tos, inner_iph, skb); 788 ttl = tnl_params->ttl; 789 if (ttl == 0) { 790 if (payload_protocol == htons(ETH_P_IP)) 791 ttl = inner_iph->ttl; 792 #if IS_ENABLED(CONFIG_IPV6) 793 else if (payload_protocol == htons(ETH_P_IPV6)) 794 ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit; 795 #endif 796 else 797 ttl = ip4_dst_hoplimit(&rt->dst); 798 } 799 800 max_headroom = LL_RESERVED_SPACE(rt->dst.dev) + sizeof(struct iphdr) 801 + rt->dst.header_len + ip_encap_hlen(&tunnel->encap); 802 if (max_headroom > dev->needed_headroom) 803 dev->needed_headroom = max_headroom; 804 805 if (skb_cow_head(skb, dev->needed_headroom)) { 806 ip_rt_put(rt); 807 dev->stats.tx_dropped++; 808 kfree_skb(skb); 809 return; 810 } 811 812 iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, protocol, tos, ttl, 813 df, !net_eq(tunnel->net, dev_net(dev))); 814 return; 815 816 #if IS_ENABLED(CONFIG_IPV6) 817 tx_error_icmp: 818 dst_link_failure(skb); 819 #endif 820 tx_error: 821 dev->stats.tx_errors++; 822 kfree_skb(skb); 823 } 824 EXPORT_SYMBOL_GPL(ip_tunnel_xmit); 825 826 static void ip_tunnel_update(struct ip_tunnel_net *itn, 827 struct ip_tunnel *t, 828 struct net_device *dev, 829 struct ip_tunnel_parm *p, 830 bool set_mtu, 831 __u32 fwmark) 832 { 833 ip_tunnel_del(itn, t); 834 t->parms.iph.saddr = p->iph.saddr; 835 t->parms.iph.daddr = p->iph.daddr; 836 t->parms.i_key = p->i_key; 837 t->parms.o_key = p->o_key; 838 if (dev->type != ARPHRD_ETHER) { 839 __dev_addr_set(dev, &p->iph.saddr, 4); 840 memcpy(dev->broadcast, &p->iph.daddr, 4); 841 } 842 ip_tunnel_add(itn, t); 843 844 t->parms.iph.ttl = p->iph.ttl; 845 t->parms.iph.tos = p->iph.tos; 846 t->parms.iph.frag_off = p->iph.frag_off; 847 848 if (t->parms.link != p->link || t->fwmark != fwmark) { 849 int mtu; 850 851 t->parms.link = p->link; 852 t->fwmark = fwmark; 853 mtu = ip_tunnel_bind_dev(dev); 854 if (set_mtu) 855 dev->mtu = mtu; 856 } 857 dst_cache_reset(&t->dst_cache); 858 netdev_state_change(dev); 859 } 860 861 int ip_tunnel_ctl(struct net_device *dev, struct ip_tunnel_parm *p, int cmd) 862 { 863 int err = 0; 864 struct ip_tunnel *t = netdev_priv(dev); 865 struct net *net = t->net; 866 struct ip_tunnel_net *itn = net_generic(net, t->ip_tnl_net_id); 867 868 switch (cmd) { 869 case SIOCGETTUNNEL: 870 if (dev == itn->fb_tunnel_dev) { 871 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type); 872 if (!t) 873 t = netdev_priv(dev); 874 } 875 memcpy(p, &t->parms, sizeof(*p)); 876 break; 877 878 case SIOCADDTUNNEL: 879 case SIOCCHGTUNNEL: 880 err = -EPERM; 881 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 882 goto done; 883 if (p->iph.ttl) 884 p->iph.frag_off |= htons(IP_DF); 885 if (!(p->i_flags & VTI_ISVTI)) { 886 if (!(p->i_flags & TUNNEL_KEY)) 887 p->i_key = 0; 888 if (!(p->o_flags & TUNNEL_KEY)) 889 p->o_key = 0; 890 } 891 892 t = ip_tunnel_find(itn, p, itn->type); 893 894 if (cmd == SIOCADDTUNNEL) { 895 if (!t) { 896 t = ip_tunnel_create(net, itn, p); 897 err = PTR_ERR_OR_ZERO(t); 898 break; 899 } 900 901 err = -EEXIST; 902 break; 903 } 904 if (dev != itn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) { 905 if (t) { 906 if (t->dev != dev) { 907 err = -EEXIST; 908 break; 909 } 910 } else { 911 unsigned int nflags = 0; 912 913 if (ipv4_is_multicast(p->iph.daddr)) 914 nflags = IFF_BROADCAST; 915 else if (p->iph.daddr) 916 nflags = IFF_POINTOPOINT; 917 918 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) { 919 err = -EINVAL; 920 break; 921 } 922 923 t = netdev_priv(dev); 924 } 925 } 926 927 if (t) { 928 err = 0; 929 ip_tunnel_update(itn, t, dev, p, true, 0); 930 } else { 931 err = -ENOENT; 932 } 933 break; 934 935 case SIOCDELTUNNEL: 936 err = -EPERM; 937 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 938 goto done; 939 940 if (dev == itn->fb_tunnel_dev) { 941 err = -ENOENT; 942 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type); 943 if (!t) 944 goto done; 945 err = -EPERM; 946 if (t == netdev_priv(itn->fb_tunnel_dev)) 947 goto done; 948 dev = t->dev; 949 } 950 unregister_netdevice(dev); 951 err = 0; 952 break; 953 954 default: 955 err = -EINVAL; 956 } 957 958 done: 959 return err; 960 } 961 EXPORT_SYMBOL_GPL(ip_tunnel_ctl); 962 963 int ip_tunnel_siocdevprivate(struct net_device *dev, struct ifreq *ifr, 964 void __user *data, int cmd) 965 { 966 struct ip_tunnel_parm p; 967 int err; 968 969 if (copy_from_user(&p, data, sizeof(p))) 970 return -EFAULT; 971 err = dev->netdev_ops->ndo_tunnel_ctl(dev, &p, cmd); 972 if (!err && copy_to_user(data, &p, sizeof(p))) 973 return -EFAULT; 974 return err; 975 } 976 EXPORT_SYMBOL_GPL(ip_tunnel_siocdevprivate); 977 978 int __ip_tunnel_change_mtu(struct net_device *dev, int new_mtu, bool strict) 979 { 980 struct ip_tunnel *tunnel = netdev_priv(dev); 981 int t_hlen = tunnel->hlen + sizeof(struct iphdr); 982 int max_mtu = IP_MAX_MTU - t_hlen; 983 984 if (dev->type == ARPHRD_ETHER) 985 max_mtu -= dev->hard_header_len; 986 987 if (new_mtu < ETH_MIN_MTU) 988 return -EINVAL; 989 990 if (new_mtu > max_mtu) { 991 if (strict) 992 return -EINVAL; 993 994 new_mtu = max_mtu; 995 } 996 997 dev->mtu = new_mtu; 998 return 0; 999 } 1000 EXPORT_SYMBOL_GPL(__ip_tunnel_change_mtu); 1001 1002 int ip_tunnel_change_mtu(struct net_device *dev, int new_mtu) 1003 { 1004 return __ip_tunnel_change_mtu(dev, new_mtu, true); 1005 } 1006 EXPORT_SYMBOL_GPL(ip_tunnel_change_mtu); 1007 1008 static void ip_tunnel_dev_free(struct net_device *dev) 1009 { 1010 struct ip_tunnel *tunnel = netdev_priv(dev); 1011 1012 gro_cells_destroy(&tunnel->gro_cells); 1013 dst_cache_destroy(&tunnel->dst_cache); 1014 free_percpu(dev->tstats); 1015 } 1016 1017 void ip_tunnel_dellink(struct net_device *dev, struct list_head *head) 1018 { 1019 struct ip_tunnel *tunnel = netdev_priv(dev); 1020 struct ip_tunnel_net *itn; 1021 1022 itn = net_generic(tunnel->net, tunnel->ip_tnl_net_id); 1023 1024 if (itn->fb_tunnel_dev != dev) { 1025 ip_tunnel_del(itn, netdev_priv(dev)); 1026 unregister_netdevice_queue(dev, head); 1027 } 1028 } 1029 EXPORT_SYMBOL_GPL(ip_tunnel_dellink); 1030 1031 struct net *ip_tunnel_get_link_net(const struct net_device *dev) 1032 { 1033 struct ip_tunnel *tunnel = netdev_priv(dev); 1034 1035 return tunnel->net; 1036 } 1037 EXPORT_SYMBOL(ip_tunnel_get_link_net); 1038 1039 int ip_tunnel_get_iflink(const struct net_device *dev) 1040 { 1041 struct ip_tunnel *tunnel = netdev_priv(dev); 1042 1043 return tunnel->parms.link; 1044 } 1045 EXPORT_SYMBOL(ip_tunnel_get_iflink); 1046 1047 int ip_tunnel_init_net(struct net *net, unsigned int ip_tnl_net_id, 1048 struct rtnl_link_ops *ops, char *devname) 1049 { 1050 struct ip_tunnel_net *itn = net_generic(net, ip_tnl_net_id); 1051 struct ip_tunnel_parm parms; 1052 unsigned int i; 1053 1054 itn->rtnl_link_ops = ops; 1055 for (i = 0; i < IP_TNL_HASH_SIZE; i++) 1056 INIT_HLIST_HEAD(&itn->tunnels[i]); 1057 1058 if (!ops || !net_has_fallback_tunnels(net)) { 1059 struct ip_tunnel_net *it_init_net; 1060 1061 it_init_net = net_generic(&init_net, ip_tnl_net_id); 1062 itn->type = it_init_net->type; 1063 itn->fb_tunnel_dev = NULL; 1064 return 0; 1065 } 1066 1067 memset(&parms, 0, sizeof(parms)); 1068 if (devname) 1069 strscpy(parms.name, devname, IFNAMSIZ); 1070 1071 rtnl_lock(); 1072 itn->fb_tunnel_dev = __ip_tunnel_create(net, ops, &parms); 1073 /* FB netdevice is special: we have one, and only one per netns. 1074 * Allowing to move it to another netns is clearly unsafe. 1075 */ 1076 if (!IS_ERR(itn->fb_tunnel_dev)) { 1077 itn->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL; 1078 itn->fb_tunnel_dev->mtu = ip_tunnel_bind_dev(itn->fb_tunnel_dev); 1079 ip_tunnel_add(itn, netdev_priv(itn->fb_tunnel_dev)); 1080 itn->type = itn->fb_tunnel_dev->type; 1081 } 1082 rtnl_unlock(); 1083 1084 return PTR_ERR_OR_ZERO(itn->fb_tunnel_dev); 1085 } 1086 EXPORT_SYMBOL_GPL(ip_tunnel_init_net); 1087 1088 static void ip_tunnel_destroy(struct net *net, struct ip_tunnel_net *itn, 1089 struct list_head *head, 1090 struct rtnl_link_ops *ops) 1091 { 1092 struct net_device *dev, *aux; 1093 int h; 1094 1095 for_each_netdev_safe(net, dev, aux) 1096 if (dev->rtnl_link_ops == ops) 1097 unregister_netdevice_queue(dev, head); 1098 1099 for (h = 0; h < IP_TNL_HASH_SIZE; h++) { 1100 struct ip_tunnel *t; 1101 struct hlist_node *n; 1102 struct hlist_head *thead = &itn->tunnels[h]; 1103 1104 hlist_for_each_entry_safe(t, n, thead, hash_node) 1105 /* If dev is in the same netns, it has already 1106 * been added to the list by the previous loop. 1107 */ 1108 if (!net_eq(dev_net(t->dev), net)) 1109 unregister_netdevice_queue(t->dev, head); 1110 } 1111 } 1112 1113 void ip_tunnel_delete_nets(struct list_head *net_list, unsigned int id, 1114 struct rtnl_link_ops *ops) 1115 { 1116 struct ip_tunnel_net *itn; 1117 struct net *net; 1118 LIST_HEAD(list); 1119 1120 rtnl_lock(); 1121 list_for_each_entry(net, net_list, exit_list) { 1122 itn = net_generic(net, id); 1123 ip_tunnel_destroy(net, itn, &list, ops); 1124 } 1125 unregister_netdevice_many(&list); 1126 rtnl_unlock(); 1127 } 1128 EXPORT_SYMBOL_GPL(ip_tunnel_delete_nets); 1129 1130 int ip_tunnel_newlink(struct net_device *dev, struct nlattr *tb[], 1131 struct ip_tunnel_parm *p, __u32 fwmark) 1132 { 1133 struct ip_tunnel *nt; 1134 struct net *net = dev_net(dev); 1135 struct ip_tunnel_net *itn; 1136 int mtu; 1137 int err; 1138 1139 nt = netdev_priv(dev); 1140 itn = net_generic(net, nt->ip_tnl_net_id); 1141 1142 if (nt->collect_md) { 1143 if (rtnl_dereference(itn->collect_md_tun)) 1144 return -EEXIST; 1145 } else { 1146 if (ip_tunnel_find(itn, p, dev->type)) 1147 return -EEXIST; 1148 } 1149 1150 nt->net = net; 1151 nt->parms = *p; 1152 nt->fwmark = fwmark; 1153 err = register_netdevice(dev); 1154 if (err) 1155 goto err_register_netdevice; 1156 1157 if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS]) 1158 eth_hw_addr_random(dev); 1159 1160 mtu = ip_tunnel_bind_dev(dev); 1161 if (tb[IFLA_MTU]) { 1162 unsigned int max = IP_MAX_MTU - (nt->hlen + sizeof(struct iphdr)); 1163 1164 if (dev->type == ARPHRD_ETHER) 1165 max -= dev->hard_header_len; 1166 1167 mtu = clamp(dev->mtu, (unsigned int)ETH_MIN_MTU, max); 1168 } 1169 1170 err = dev_set_mtu(dev, mtu); 1171 if (err) 1172 goto err_dev_set_mtu; 1173 1174 ip_tunnel_add(itn, nt); 1175 return 0; 1176 1177 err_dev_set_mtu: 1178 unregister_netdevice(dev); 1179 err_register_netdevice: 1180 return err; 1181 } 1182 EXPORT_SYMBOL_GPL(ip_tunnel_newlink); 1183 1184 int ip_tunnel_changelink(struct net_device *dev, struct nlattr *tb[], 1185 struct ip_tunnel_parm *p, __u32 fwmark) 1186 { 1187 struct ip_tunnel *t; 1188 struct ip_tunnel *tunnel = netdev_priv(dev); 1189 struct net *net = tunnel->net; 1190 struct ip_tunnel_net *itn = net_generic(net, tunnel->ip_tnl_net_id); 1191 1192 if (dev == itn->fb_tunnel_dev) 1193 return -EINVAL; 1194 1195 t = ip_tunnel_find(itn, p, dev->type); 1196 1197 if (t) { 1198 if (t->dev != dev) 1199 return -EEXIST; 1200 } else { 1201 t = tunnel; 1202 1203 if (dev->type != ARPHRD_ETHER) { 1204 unsigned int nflags = 0; 1205 1206 if (ipv4_is_multicast(p->iph.daddr)) 1207 nflags = IFF_BROADCAST; 1208 else if (p->iph.daddr) 1209 nflags = IFF_POINTOPOINT; 1210 1211 if ((dev->flags ^ nflags) & 1212 (IFF_POINTOPOINT | IFF_BROADCAST)) 1213 return -EINVAL; 1214 } 1215 } 1216 1217 ip_tunnel_update(itn, t, dev, p, !tb[IFLA_MTU], fwmark); 1218 return 0; 1219 } 1220 EXPORT_SYMBOL_GPL(ip_tunnel_changelink); 1221 1222 int ip_tunnel_init(struct net_device *dev) 1223 { 1224 struct ip_tunnel *tunnel = netdev_priv(dev); 1225 struct iphdr *iph = &tunnel->parms.iph; 1226 int err; 1227 1228 dev->needs_free_netdev = true; 1229 dev->priv_destructor = ip_tunnel_dev_free; 1230 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 1231 if (!dev->tstats) 1232 return -ENOMEM; 1233 1234 err = dst_cache_init(&tunnel->dst_cache, GFP_KERNEL); 1235 if (err) { 1236 free_percpu(dev->tstats); 1237 return err; 1238 } 1239 1240 err = gro_cells_init(&tunnel->gro_cells, dev); 1241 if (err) { 1242 dst_cache_destroy(&tunnel->dst_cache); 1243 free_percpu(dev->tstats); 1244 return err; 1245 } 1246 1247 tunnel->dev = dev; 1248 tunnel->net = dev_net(dev); 1249 strcpy(tunnel->parms.name, dev->name); 1250 iph->version = 4; 1251 iph->ihl = 5; 1252 1253 if (tunnel->collect_md) 1254 netif_keep_dst(dev); 1255 return 0; 1256 } 1257 EXPORT_SYMBOL_GPL(ip_tunnel_init); 1258 1259 void ip_tunnel_uninit(struct net_device *dev) 1260 { 1261 struct ip_tunnel *tunnel = netdev_priv(dev); 1262 struct net *net = tunnel->net; 1263 struct ip_tunnel_net *itn; 1264 1265 itn = net_generic(net, tunnel->ip_tnl_net_id); 1266 ip_tunnel_del(itn, netdev_priv(dev)); 1267 if (itn->fb_tunnel_dev == dev) 1268 WRITE_ONCE(itn->fb_tunnel_dev, NULL); 1269 1270 dst_cache_reset(&tunnel->dst_cache); 1271 } 1272 EXPORT_SYMBOL_GPL(ip_tunnel_uninit); 1273 1274 /* Do least required initialization, rest of init is done in tunnel_init call */ 1275 void ip_tunnel_setup(struct net_device *dev, unsigned int net_id) 1276 { 1277 struct ip_tunnel *tunnel = netdev_priv(dev); 1278 tunnel->ip_tnl_net_id = net_id; 1279 } 1280 EXPORT_SYMBOL_GPL(ip_tunnel_setup); 1281 1282 MODULE_LICENSE("GPL"); 1283