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 53static 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 59static 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*/ 83struct 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} 171EXPORT_SYMBOL_GPL(ip_tunnel_lookup); 172 173static 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 192static 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 201static 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 208static 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 231static 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 272failed_free: 273 free_netdev(dev); 274failed: 275 return ERR_PTR(err); 276} 277 278static 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 327static 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 356err_dev_set_mtu: 357 unregister_netdevice(dev); 358 return ERR_PTR(err); 359} 360 361void 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} 373EXPORT_SYMBOL(ip_tunnel_md_udp_encap); 374 375int 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 nh, 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 /* Save offset of outer header relative to skb->head, 407 * because we are going to reset the network header to the inner header 408 * and might change skb->head. 409 */ 410 nh = skb_network_header(skb) - skb->head; 411 412 skb_set_network_header(skb, (tunnel->dev->type == ARPHRD_ETHER) ? ETH_HLEN : 0); 413 414 if (!pskb_inet_may_pull(skb)) { 415 DEV_STATS_INC(tunnel->dev, rx_length_errors); 416 DEV_STATS_INC(tunnel->dev, rx_errors); 417 goto drop; 418 } 419 iph = (struct iphdr *)(skb->head + nh); 420 421 err = IP_ECN_decapsulate(iph, skb); 422 if (unlikely(err)) { 423 if (log_ecn_error) 424 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n", 425 &iph->saddr, iph->tos); 426 if (err > 1) { 427 DEV_STATS_INC(tunnel->dev, rx_frame_errors); 428 DEV_STATS_INC(tunnel->dev, rx_errors); 429 goto drop; 430 } 431 } 432 433 dev_sw_netstats_rx_add(tunnel->dev, skb->len); 434 skb_scrub_packet(skb, !net_eq(tunnel->net, dev_net(tunnel->dev))); 435 436 if (tunnel->dev->type == ARPHRD_ETHER) { 437 skb->protocol = eth_type_trans(skb, tunnel->dev); 438 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 439 } else { 440 skb->dev = tunnel->dev; 441 } 442 443 if (tun_dst) 444 skb_dst_set(skb, (struct dst_entry *)tun_dst); 445 446 gro_cells_receive(&tunnel->gro_cells, skb); 447 return 0; 448 449drop: 450 if (tun_dst) 451 dst_release((struct dst_entry *)tun_dst); 452 kfree_skb(skb); 453 return 0; 454} 455EXPORT_SYMBOL_GPL(ip_tunnel_rcv); 456 457int ip_tunnel_encap_add_ops(const struct ip_tunnel_encap_ops *ops, 458 unsigned int num) 459{ 460 if (num >= MAX_IPTUN_ENCAP_OPS) 461 return -ERANGE; 462 463 return !cmpxchg((const struct ip_tunnel_encap_ops **) 464 &iptun_encaps[num], 465 NULL, ops) ? 0 : -1; 466} 467EXPORT_SYMBOL(ip_tunnel_encap_add_ops); 468 469int ip_tunnel_encap_del_ops(const struct ip_tunnel_encap_ops *ops, 470 unsigned int num) 471{ 472 int ret; 473 474 if (num >= MAX_IPTUN_ENCAP_OPS) 475 return -ERANGE; 476 477 ret = (cmpxchg((const struct ip_tunnel_encap_ops **) 478 &iptun_encaps[num], 479 ops, NULL) == ops) ? 0 : -1; 480 481 synchronize_net(); 482 483 return ret; 484} 485EXPORT_SYMBOL(ip_tunnel_encap_del_ops); 486 487int ip_tunnel_encap_setup(struct ip_tunnel *t, 488 struct ip_tunnel_encap *ipencap) 489{ 490 int hlen; 491 492 memset(&t->encap, 0, sizeof(t->encap)); 493 494 hlen = ip_encap_hlen(ipencap); 495 if (hlen < 0) 496 return hlen; 497 498 t->encap.type = ipencap->type; 499 t->encap.sport = ipencap->sport; 500 t->encap.dport = ipencap->dport; 501 t->encap.flags = ipencap->flags; 502 503 t->encap_hlen = hlen; 504 t->hlen = t->encap_hlen + t->tun_hlen; 505 506 return 0; 507} 508EXPORT_SYMBOL_GPL(ip_tunnel_encap_setup); 509 510static int tnl_update_pmtu(struct net_device *dev, struct sk_buff *skb, 511 struct rtable *rt, __be16 df, 512 const struct iphdr *inner_iph, 513 int tunnel_hlen, __be32 dst, bool md) 514{ 515 struct ip_tunnel *tunnel = netdev_priv(dev); 516 int pkt_size; 517 int mtu; 518 519 tunnel_hlen = md ? tunnel_hlen : tunnel->hlen; 520 pkt_size = skb->len - tunnel_hlen; 521 pkt_size -= dev->type == ARPHRD_ETHER ? dev->hard_header_len : 0; 522 523 if (df) { 524 mtu = dst_mtu(&rt->dst) - (sizeof(struct iphdr) + tunnel_hlen); 525 mtu -= dev->type == ARPHRD_ETHER ? dev->hard_header_len : 0; 526 } else { 527 mtu = skb_valid_dst(skb) ? dst_mtu(skb_dst(skb)) : dev->mtu; 528 } 529 530 if (skb_valid_dst(skb)) 531 skb_dst_update_pmtu_no_confirm(skb, mtu); 532 533 if (skb->protocol == htons(ETH_P_IP)) { 534 if (!skb_is_gso(skb) && 535 (inner_iph->frag_off & htons(IP_DF)) && 536 mtu < pkt_size) { 537 icmp_ndo_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu)); 538 return -E2BIG; 539 } 540 } 541#if IS_ENABLED(CONFIG_IPV6) 542 else if (skb->protocol == htons(ETH_P_IPV6)) { 543 struct rt6_info *rt6; 544 __be32 daddr; 545 546 rt6 = skb_valid_dst(skb) ? (struct rt6_info *)skb_dst(skb) : 547 NULL; 548 daddr = md ? dst : tunnel->parms.iph.daddr; 549 550 if (rt6 && mtu < dst_mtu(skb_dst(skb)) && 551 mtu >= IPV6_MIN_MTU) { 552 if ((daddr && !ipv4_is_multicast(daddr)) || 553 rt6->rt6i_dst.plen == 128) { 554 rt6->rt6i_flags |= RTF_MODIFIED; 555 dst_metric_set(skb_dst(skb), RTAX_MTU, mtu); 556 } 557 } 558 559 if (!skb_is_gso(skb) && mtu >= IPV6_MIN_MTU && 560 mtu < pkt_size) { 561 icmpv6_ndo_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu); 562 return -E2BIG; 563 } 564 } 565#endif 566 return 0; 567} 568 569static void ip_tunnel_adj_headroom(struct net_device *dev, unsigned int headroom) 570{ 571 /* we must cap headroom to some upperlimit, else pskb_expand_head 572 * will overflow header offsets in skb_headers_offset_update(). 573 */ 574 static const unsigned int max_allowed = 512; 575 576 if (headroom > max_allowed) 577 headroom = max_allowed; 578 579 if (headroom > READ_ONCE(dev->needed_headroom)) 580 WRITE_ONCE(dev->needed_headroom, headroom); 581} 582 583void ip_md_tunnel_xmit(struct sk_buff *skb, struct net_device *dev, 584 u8 proto, int tunnel_hlen) 585{ 586 struct ip_tunnel *tunnel = netdev_priv(dev); 587 u32 headroom = sizeof(struct iphdr); 588 struct ip_tunnel_info *tun_info; 589 const struct ip_tunnel_key *key; 590 const struct iphdr *inner_iph; 591 struct rtable *rt = NULL; 592 struct flowi4 fl4; 593 __be16 df = 0; 594 u8 tos, ttl; 595 bool use_cache; 596 597 tun_info = skb_tunnel_info(skb); 598 if (unlikely(!tun_info || !(tun_info->mode & IP_TUNNEL_INFO_TX) || 599 ip_tunnel_info_af(tun_info) != AF_INET)) 600 goto tx_error; 601 key = &tun_info->key; 602 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); 603 inner_iph = (const struct iphdr *)skb_inner_network_header(skb); 604 tos = key->tos; 605 if (tos == 1) { 606 if (skb->protocol == htons(ETH_P_IP)) 607 tos = inner_iph->tos; 608 else if (skb->protocol == htons(ETH_P_IPV6)) 609 tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph); 610 } 611 ip_tunnel_init_flow(&fl4, proto, key->u.ipv4.dst, key->u.ipv4.src, 612 tunnel_id_to_key32(key->tun_id), RT_TOS(tos), 613 dev_net(dev), 0, skb->mark, skb_get_hash(skb), 614 key->flow_flags); 615 616 if (!tunnel_hlen) 617 tunnel_hlen = ip_encap_hlen(&tun_info->encap); 618 619 if (ip_tunnel_encap(skb, &tun_info->encap, &proto, &fl4) < 0) 620 goto tx_error; 621 622 use_cache = ip_tunnel_dst_cache_usable(skb, tun_info); 623 if (use_cache) 624 rt = dst_cache_get_ip4(&tun_info->dst_cache, &fl4.saddr); 625 if (!rt) { 626 rt = ip_route_output_key(tunnel->net, &fl4); 627 if (IS_ERR(rt)) { 628 DEV_STATS_INC(dev, tx_carrier_errors); 629 goto tx_error; 630 } 631 if (use_cache) 632 dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst, 633 fl4.saddr); 634 } 635 if (rt->dst.dev == dev) { 636 ip_rt_put(rt); 637 DEV_STATS_INC(dev, collisions); 638 goto tx_error; 639 } 640 641 if (key->tun_flags & TUNNEL_DONT_FRAGMENT) 642 df = htons(IP_DF); 643 if (tnl_update_pmtu(dev, skb, rt, df, inner_iph, tunnel_hlen, 644 key->u.ipv4.dst, true)) { 645 ip_rt_put(rt); 646 goto tx_error; 647 } 648 649 tos = ip_tunnel_ecn_encap(tos, inner_iph, skb); 650 ttl = key->ttl; 651 if (ttl == 0) { 652 if (skb->protocol == htons(ETH_P_IP)) 653 ttl = inner_iph->ttl; 654 else if (skb->protocol == htons(ETH_P_IPV6)) 655 ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit; 656 else 657 ttl = ip4_dst_hoplimit(&rt->dst); 658 } 659 660 headroom += LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len; 661 if (skb_cow_head(skb, headroom)) { 662 ip_rt_put(rt); 663 goto tx_dropped; 664 } 665 666 ip_tunnel_adj_headroom(dev, headroom); 667 668 iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, proto, tos, ttl, 669 df, !net_eq(tunnel->net, dev_net(dev))); 670 return; 671tx_error: 672 DEV_STATS_INC(dev, tx_errors); 673 goto kfree; 674tx_dropped: 675 DEV_STATS_INC(dev, tx_dropped); 676kfree: 677 kfree_skb(skb); 678} 679EXPORT_SYMBOL_GPL(ip_md_tunnel_xmit); 680 681void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev, 682 const struct iphdr *tnl_params, u8 protocol) 683{ 684 struct ip_tunnel *tunnel = netdev_priv(dev); 685 struct ip_tunnel_info *tun_info = NULL; 686 const struct iphdr *inner_iph; 687 unsigned int max_headroom; /* The extra header space needed */ 688 struct rtable *rt = NULL; /* Route to the other host */ 689 __be16 payload_protocol; 690 bool use_cache = false; 691 struct flowi4 fl4; 692 bool md = false; 693 bool connected; 694 u8 tos, ttl; 695 __be32 dst; 696 __be16 df; 697 698 inner_iph = (const struct iphdr *)skb_inner_network_header(skb); 699 connected = (tunnel->parms.iph.daddr != 0); 700 payload_protocol = skb_protocol(skb, true); 701 702 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); 703 704 dst = tnl_params->daddr; 705 if (dst == 0) { 706 /* NBMA tunnel */ 707 708 if (!skb_dst(skb)) { 709 DEV_STATS_INC(dev, tx_fifo_errors); 710 goto tx_error; 711 } 712 713 tun_info = skb_tunnel_info(skb); 714 if (tun_info && (tun_info->mode & IP_TUNNEL_INFO_TX) && 715 ip_tunnel_info_af(tun_info) == AF_INET && 716 tun_info->key.u.ipv4.dst) { 717 dst = tun_info->key.u.ipv4.dst; 718 md = true; 719 connected = true; 720 } else if (payload_protocol == htons(ETH_P_IP)) { 721 rt = skb_rtable(skb); 722 dst = rt_nexthop(rt, inner_iph->daddr); 723 } 724#if IS_ENABLED(CONFIG_IPV6) 725 else if (payload_protocol == htons(ETH_P_IPV6)) { 726 const struct in6_addr *addr6; 727 struct neighbour *neigh; 728 bool do_tx_error_icmp; 729 int addr_type; 730 731 neigh = dst_neigh_lookup(skb_dst(skb), 732 &ipv6_hdr(skb)->daddr); 733 if (!neigh) 734 goto tx_error; 735 736 addr6 = (const struct in6_addr *)&neigh->primary_key; 737 addr_type = ipv6_addr_type(addr6); 738 739 if (addr_type == IPV6_ADDR_ANY) { 740 addr6 = &ipv6_hdr(skb)->daddr; 741 addr_type = ipv6_addr_type(addr6); 742 } 743 744 if ((addr_type & IPV6_ADDR_COMPATv4) == 0) 745 do_tx_error_icmp = true; 746 else { 747 do_tx_error_icmp = false; 748 dst = addr6->s6_addr32[3]; 749 } 750 neigh_release(neigh); 751 if (do_tx_error_icmp) 752 goto tx_error_icmp; 753 } 754#endif 755 else 756 goto tx_error; 757 758 if (!md) 759 connected = false; 760 } 761 762 tos = tnl_params->tos; 763 if (tos & 0x1) { 764 tos &= ~0x1; 765 if (payload_protocol == htons(ETH_P_IP)) { 766 tos = inner_iph->tos; 767 connected = false; 768 } else if (payload_protocol == htons(ETH_P_IPV6)) { 769 tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph); 770 connected = false; 771 } 772 } 773 774 ip_tunnel_init_flow(&fl4, protocol, dst, tnl_params->saddr, 775 tunnel->parms.o_key, RT_TOS(tos), 776 dev_net(dev), READ_ONCE(tunnel->parms.link), 777 tunnel->fwmark, skb_get_hash(skb), 0); 778 779 if (ip_tunnel_encap(skb, &tunnel->encap, &protocol, &fl4) < 0) 780 goto tx_error; 781 782 if (connected && md) { 783 use_cache = ip_tunnel_dst_cache_usable(skb, tun_info); 784 if (use_cache) 785 rt = dst_cache_get_ip4(&tun_info->dst_cache, 786 &fl4.saddr); 787 } else { 788 rt = connected ? dst_cache_get_ip4(&tunnel->dst_cache, 789 &fl4.saddr) : NULL; 790 } 791 792 if (!rt) { 793 rt = ip_route_output_key(tunnel->net, &fl4); 794 795 if (IS_ERR(rt)) { 796 DEV_STATS_INC(dev, tx_carrier_errors); 797 goto tx_error; 798 } 799 if (use_cache) 800 dst_cache_set_ip4(&tun_info->dst_cache, &rt->dst, 801 fl4.saddr); 802 else if (!md && connected) 803 dst_cache_set_ip4(&tunnel->dst_cache, &rt->dst, 804 fl4.saddr); 805 } 806 807 if (rt->dst.dev == dev) { 808 ip_rt_put(rt); 809 DEV_STATS_INC(dev, collisions); 810 goto tx_error; 811 } 812 813 df = tnl_params->frag_off; 814 if (payload_protocol == htons(ETH_P_IP) && !tunnel->ignore_df) 815 df |= (inner_iph->frag_off & htons(IP_DF)); 816 817 if (tnl_update_pmtu(dev, skb, rt, df, inner_iph, 0, 0, false)) { 818 ip_rt_put(rt); 819 goto tx_error; 820 } 821 822 if (tunnel->err_count > 0) { 823 if (time_before(jiffies, 824 tunnel->err_time + IPTUNNEL_ERR_TIMEO)) { 825 tunnel->err_count--; 826 827 dst_link_failure(skb); 828 } else 829 tunnel->err_count = 0; 830 } 831 832 tos = ip_tunnel_ecn_encap(tos, inner_iph, skb); 833 ttl = tnl_params->ttl; 834 if (ttl == 0) { 835 if (payload_protocol == htons(ETH_P_IP)) 836 ttl = inner_iph->ttl; 837#if IS_ENABLED(CONFIG_IPV6) 838 else if (payload_protocol == htons(ETH_P_IPV6)) 839 ttl = ((const struct ipv6hdr *)inner_iph)->hop_limit; 840#endif 841 else 842 ttl = ip4_dst_hoplimit(&rt->dst); 843 } 844 845 max_headroom = LL_RESERVED_SPACE(rt->dst.dev) + sizeof(struct iphdr) 846 + rt->dst.header_len + ip_encap_hlen(&tunnel->encap); 847 848 if (skb_cow_head(skb, max_headroom)) { 849 ip_rt_put(rt); 850 DEV_STATS_INC(dev, tx_dropped); 851 kfree_skb(skb); 852 return; 853 } 854 855 ip_tunnel_adj_headroom(dev, max_headroom); 856 857 iptunnel_xmit(NULL, rt, skb, fl4.saddr, fl4.daddr, protocol, tos, ttl, 858 df, !net_eq(tunnel->net, dev_net(dev))); 859 return; 860 861#if IS_ENABLED(CONFIG_IPV6) 862tx_error_icmp: 863 dst_link_failure(skb); 864#endif 865tx_error: 866 DEV_STATS_INC(dev, tx_errors); 867 kfree_skb(skb); 868} 869EXPORT_SYMBOL_GPL(ip_tunnel_xmit); 870 871static void ip_tunnel_update(struct ip_tunnel_net *itn, 872 struct ip_tunnel *t, 873 struct net_device *dev, 874 struct ip_tunnel_parm *p, 875 bool set_mtu, 876 __u32 fwmark) 877{ 878 ip_tunnel_del(itn, t); 879 t->parms.iph.saddr = p->iph.saddr; 880 t->parms.iph.daddr = p->iph.daddr; 881 t->parms.i_key = p->i_key; 882 t->parms.o_key = p->o_key; 883 if (dev->type != ARPHRD_ETHER) { 884 __dev_addr_set(dev, &p->iph.saddr, 4); 885 memcpy(dev->broadcast, &p->iph.daddr, 4); 886 } 887 ip_tunnel_add(itn, t); 888 889 t->parms.iph.ttl = p->iph.ttl; 890 t->parms.iph.tos = p->iph.tos; 891 t->parms.iph.frag_off = p->iph.frag_off; 892 893 if (t->parms.link != p->link || t->fwmark != fwmark) { 894 int mtu; 895 896 WRITE_ONCE(t->parms.link, p->link); 897 t->fwmark = fwmark; 898 mtu = ip_tunnel_bind_dev(dev); 899 if (set_mtu) 900 dev->mtu = mtu; 901 } 902 dst_cache_reset(&t->dst_cache); 903 netdev_state_change(dev); 904} 905 906int ip_tunnel_ctl(struct net_device *dev, struct ip_tunnel_parm *p, int cmd) 907{ 908 int err = 0; 909 struct ip_tunnel *t = netdev_priv(dev); 910 struct net *net = t->net; 911 struct ip_tunnel_net *itn = net_generic(net, t->ip_tnl_net_id); 912 913 switch (cmd) { 914 case SIOCGETTUNNEL: 915 if (dev == itn->fb_tunnel_dev) { 916 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type); 917 if (!t) 918 t = netdev_priv(dev); 919 } 920 memcpy(p, &t->parms, sizeof(*p)); 921 break; 922 923 case SIOCADDTUNNEL: 924 case SIOCCHGTUNNEL: 925 err = -EPERM; 926 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 927 goto done; 928 if (p->iph.ttl) 929 p->iph.frag_off |= htons(IP_DF); 930 if (!(p->i_flags & VTI_ISVTI)) { 931 if (!(p->i_flags & TUNNEL_KEY)) 932 p->i_key = 0; 933 if (!(p->o_flags & TUNNEL_KEY)) 934 p->o_key = 0; 935 } 936 937 t = ip_tunnel_find(itn, p, itn->type); 938 939 if (cmd == SIOCADDTUNNEL) { 940 if (!t) { 941 t = ip_tunnel_create(net, itn, p); 942 err = PTR_ERR_OR_ZERO(t); 943 break; 944 } 945 946 err = -EEXIST; 947 break; 948 } 949 if (dev != itn->fb_tunnel_dev && cmd == SIOCCHGTUNNEL) { 950 if (t) { 951 if (t->dev != dev) { 952 err = -EEXIST; 953 break; 954 } 955 } else { 956 unsigned int nflags = 0; 957 958 if (ipv4_is_multicast(p->iph.daddr)) 959 nflags = IFF_BROADCAST; 960 else if (p->iph.daddr) 961 nflags = IFF_POINTOPOINT; 962 963 if ((dev->flags^nflags)&(IFF_POINTOPOINT|IFF_BROADCAST)) { 964 err = -EINVAL; 965 break; 966 } 967 968 t = netdev_priv(dev); 969 } 970 } 971 972 if (t) { 973 err = 0; 974 ip_tunnel_update(itn, t, dev, p, true, 0); 975 } else { 976 err = -ENOENT; 977 } 978 break; 979 980 case SIOCDELTUNNEL: 981 err = -EPERM; 982 if (!ns_capable(net->user_ns, CAP_NET_ADMIN)) 983 goto done; 984 985 if (dev == itn->fb_tunnel_dev) { 986 err = -ENOENT; 987 t = ip_tunnel_find(itn, p, itn->fb_tunnel_dev->type); 988 if (!t) 989 goto done; 990 err = -EPERM; 991 if (t == netdev_priv(itn->fb_tunnel_dev)) 992 goto done; 993 dev = t->dev; 994 } 995 unregister_netdevice(dev); 996 err = 0; 997 break; 998 999 default: 1000 err = -EINVAL; 1001 } 1002 1003done: 1004 return err; 1005} 1006EXPORT_SYMBOL_GPL(ip_tunnel_ctl); 1007 1008int ip_tunnel_siocdevprivate(struct net_device *dev, struct ifreq *ifr, 1009 void __user *data, int cmd) 1010{ 1011 struct ip_tunnel_parm p; 1012 int err; 1013 1014 if (copy_from_user(&p, data, sizeof(p))) 1015 return -EFAULT; 1016 err = dev->netdev_ops->ndo_tunnel_ctl(dev, &p, cmd); 1017 if (!err && copy_to_user(data, &p, sizeof(p))) 1018 return -EFAULT; 1019 return err; 1020} 1021EXPORT_SYMBOL_GPL(ip_tunnel_siocdevprivate); 1022 1023int __ip_tunnel_change_mtu(struct net_device *dev, int new_mtu, bool strict) 1024{ 1025 struct ip_tunnel *tunnel = netdev_priv(dev); 1026 int t_hlen = tunnel->hlen + sizeof(struct iphdr); 1027 int max_mtu = IP_MAX_MTU - t_hlen; 1028 1029 if (dev->type == ARPHRD_ETHER) 1030 max_mtu -= dev->hard_header_len; 1031 1032 if (new_mtu < ETH_MIN_MTU) 1033 return -EINVAL; 1034 1035 if (new_mtu > max_mtu) { 1036 if (strict) 1037 return -EINVAL; 1038 1039 new_mtu = max_mtu; 1040 } 1041 1042 dev->mtu = new_mtu; 1043 return 0; 1044} 1045EXPORT_SYMBOL_GPL(__ip_tunnel_change_mtu); 1046 1047int ip_tunnel_change_mtu(struct net_device *dev, int new_mtu) 1048{ 1049 return __ip_tunnel_change_mtu(dev, new_mtu, true); 1050} 1051EXPORT_SYMBOL_GPL(ip_tunnel_change_mtu); 1052 1053static void ip_tunnel_dev_free(struct net_device *dev) 1054{ 1055 struct ip_tunnel *tunnel = netdev_priv(dev); 1056 1057 gro_cells_destroy(&tunnel->gro_cells); 1058 dst_cache_destroy(&tunnel->dst_cache); 1059 free_percpu(dev->tstats); 1060} 1061 1062void ip_tunnel_dellink(struct net_device *dev, struct list_head *head) 1063{ 1064 struct ip_tunnel *tunnel = netdev_priv(dev); 1065 struct ip_tunnel_net *itn; 1066 1067 itn = net_generic(tunnel->net, tunnel->ip_tnl_net_id); 1068 1069 if (itn->fb_tunnel_dev != dev) { 1070 ip_tunnel_del(itn, netdev_priv(dev)); 1071 unregister_netdevice_queue(dev, head); 1072 } 1073} 1074EXPORT_SYMBOL_GPL(ip_tunnel_dellink); 1075 1076struct net *ip_tunnel_get_link_net(const struct net_device *dev) 1077{ 1078 struct ip_tunnel *tunnel = netdev_priv(dev); 1079 1080 return tunnel->net; 1081} 1082EXPORT_SYMBOL(ip_tunnel_get_link_net); 1083 1084int ip_tunnel_get_iflink(const struct net_device *dev) 1085{ 1086 const struct ip_tunnel *tunnel = netdev_priv(dev); 1087 1088 return READ_ONCE(tunnel->parms.link); 1089} 1090EXPORT_SYMBOL(ip_tunnel_get_iflink); 1091 1092int ip_tunnel_init_net(struct net *net, unsigned int ip_tnl_net_id, 1093 struct rtnl_link_ops *ops, char *devname) 1094{ 1095 struct ip_tunnel_net *itn = net_generic(net, ip_tnl_net_id); 1096 struct ip_tunnel_parm parms; 1097 unsigned int i; 1098 1099 itn->rtnl_link_ops = ops; 1100 for (i = 0; i < IP_TNL_HASH_SIZE; i++) 1101 INIT_HLIST_HEAD(&itn->tunnels[i]); 1102 1103 if (!ops || !net_has_fallback_tunnels(net)) { 1104 struct ip_tunnel_net *it_init_net; 1105 1106 it_init_net = net_generic(&init_net, ip_tnl_net_id); 1107 itn->type = it_init_net->type; 1108 itn->fb_tunnel_dev = NULL; 1109 return 0; 1110 } 1111 1112 memset(&parms, 0, sizeof(parms)); 1113 if (devname) 1114 strscpy(parms.name, devname, IFNAMSIZ); 1115 1116 rtnl_lock(); 1117 itn->fb_tunnel_dev = __ip_tunnel_create(net, ops, &parms); 1118 /* FB netdevice is special: we have one, and only one per netns. 1119 * Allowing to move it to another netns is clearly unsafe. 1120 */ 1121 if (!IS_ERR(itn->fb_tunnel_dev)) { 1122 itn->fb_tunnel_dev->features |= NETIF_F_NETNS_LOCAL; 1123 itn->fb_tunnel_dev->mtu = ip_tunnel_bind_dev(itn->fb_tunnel_dev); 1124 ip_tunnel_add(itn, netdev_priv(itn->fb_tunnel_dev)); 1125 itn->type = itn->fb_tunnel_dev->type; 1126 } 1127 rtnl_unlock(); 1128 1129 return PTR_ERR_OR_ZERO(itn->fb_tunnel_dev); 1130} 1131EXPORT_SYMBOL_GPL(ip_tunnel_init_net); 1132 1133static void ip_tunnel_destroy(struct net *net, struct ip_tunnel_net *itn, 1134 struct list_head *head, 1135 struct rtnl_link_ops *ops) 1136{ 1137 struct net_device *dev, *aux; 1138 int h; 1139 1140 for_each_netdev_safe(net, dev, aux) 1141 if (dev->rtnl_link_ops == ops) 1142 unregister_netdevice_queue(dev, head); 1143 1144 for (h = 0; h < IP_TNL_HASH_SIZE; h++) { 1145 struct ip_tunnel *t; 1146 struct hlist_node *n; 1147 struct hlist_head *thead = &itn->tunnels[h]; 1148 1149 hlist_for_each_entry_safe(t, n, thead, hash_node) 1150 /* If dev is in the same netns, it has already 1151 * been added to the list by the previous loop. 1152 */ 1153 if (!net_eq(dev_net(t->dev), net)) 1154 unregister_netdevice_queue(t->dev, head); 1155 } 1156} 1157 1158void ip_tunnel_delete_nets(struct list_head *net_list, unsigned int id, 1159 struct rtnl_link_ops *ops, 1160 struct list_head *dev_to_kill) 1161{ 1162 struct ip_tunnel_net *itn; 1163 struct net *net; 1164 1165 ASSERT_RTNL(); 1166 list_for_each_entry(net, net_list, exit_list) { 1167 itn = net_generic(net, id); 1168 ip_tunnel_destroy(net, itn, dev_to_kill, ops); 1169 } 1170} 1171EXPORT_SYMBOL_GPL(ip_tunnel_delete_nets); 1172 1173int ip_tunnel_newlink(struct net_device *dev, struct nlattr *tb[], 1174 struct ip_tunnel_parm *p, __u32 fwmark) 1175{ 1176 struct ip_tunnel *nt; 1177 struct net *net = dev_net(dev); 1178 struct ip_tunnel_net *itn; 1179 int mtu; 1180 int err; 1181 1182 nt = netdev_priv(dev); 1183 itn = net_generic(net, nt->ip_tnl_net_id); 1184 1185 if (nt->collect_md) { 1186 if (rtnl_dereference(itn->collect_md_tun)) 1187 return -EEXIST; 1188 } else { 1189 if (ip_tunnel_find(itn, p, dev->type)) 1190 return -EEXIST; 1191 } 1192 1193 nt->net = net; 1194 nt->parms = *p; 1195 nt->fwmark = fwmark; 1196 err = register_netdevice(dev); 1197 if (err) 1198 goto err_register_netdevice; 1199 1200 if (dev->type == ARPHRD_ETHER && !tb[IFLA_ADDRESS]) 1201 eth_hw_addr_random(dev); 1202 1203 mtu = ip_tunnel_bind_dev(dev); 1204 if (tb[IFLA_MTU]) { 1205 unsigned int max = IP_MAX_MTU - (nt->hlen + sizeof(struct iphdr)); 1206 1207 if (dev->type == ARPHRD_ETHER) 1208 max -= dev->hard_header_len; 1209 1210 mtu = clamp(dev->mtu, (unsigned int)ETH_MIN_MTU, max); 1211 } 1212 1213 err = dev_set_mtu(dev, mtu); 1214 if (err) 1215 goto err_dev_set_mtu; 1216 1217 ip_tunnel_add(itn, nt); 1218 return 0; 1219 1220err_dev_set_mtu: 1221 unregister_netdevice(dev); 1222err_register_netdevice: 1223 return err; 1224} 1225EXPORT_SYMBOL_GPL(ip_tunnel_newlink); 1226 1227int ip_tunnel_changelink(struct net_device *dev, struct nlattr *tb[], 1228 struct ip_tunnel_parm *p, __u32 fwmark) 1229{ 1230 struct ip_tunnel *t; 1231 struct ip_tunnel *tunnel = netdev_priv(dev); 1232 struct net *net = tunnel->net; 1233 struct ip_tunnel_net *itn = net_generic(net, tunnel->ip_tnl_net_id); 1234 1235 if (dev == itn->fb_tunnel_dev) 1236 return -EINVAL; 1237 1238 t = ip_tunnel_find(itn, p, dev->type); 1239 1240 if (t) { 1241 if (t->dev != dev) 1242 return -EEXIST; 1243 } else { 1244 t = tunnel; 1245 1246 if (dev->type != ARPHRD_ETHER) { 1247 unsigned int nflags = 0; 1248 1249 if (ipv4_is_multicast(p->iph.daddr)) 1250 nflags = IFF_BROADCAST; 1251 else if (p->iph.daddr) 1252 nflags = IFF_POINTOPOINT; 1253 1254 if ((dev->flags ^ nflags) & 1255 (IFF_POINTOPOINT | IFF_BROADCAST)) 1256 return -EINVAL; 1257 } 1258 } 1259 1260 ip_tunnel_update(itn, t, dev, p, !tb[IFLA_MTU], fwmark); 1261 return 0; 1262} 1263EXPORT_SYMBOL_GPL(ip_tunnel_changelink); 1264 1265int ip_tunnel_init(struct net_device *dev) 1266{ 1267 struct ip_tunnel *tunnel = netdev_priv(dev); 1268 struct iphdr *iph = &tunnel->parms.iph; 1269 int err; 1270 1271 dev->needs_free_netdev = true; 1272 dev->priv_destructor = ip_tunnel_dev_free; 1273 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 1274 if (!dev->tstats) 1275 return -ENOMEM; 1276 1277 err = dst_cache_init(&tunnel->dst_cache, GFP_KERNEL); 1278 if (err) { 1279 free_percpu(dev->tstats); 1280 return err; 1281 } 1282 1283 err = gro_cells_init(&tunnel->gro_cells, dev); 1284 if (err) { 1285 dst_cache_destroy(&tunnel->dst_cache); 1286 free_percpu(dev->tstats); 1287 return err; 1288 } 1289 1290 tunnel->dev = dev; 1291 tunnel->net = dev_net(dev); 1292 strcpy(tunnel->parms.name, dev->name); 1293 iph->version = 4; 1294 iph->ihl = 5; 1295 1296 if (tunnel->collect_md) 1297 netif_keep_dst(dev); 1298 netdev_lockdep_set_classes(dev); 1299 return 0; 1300} 1301EXPORT_SYMBOL_GPL(ip_tunnel_init); 1302 1303void ip_tunnel_uninit(struct net_device *dev) 1304{ 1305 struct ip_tunnel *tunnel = netdev_priv(dev); 1306 struct net *net = tunnel->net; 1307 struct ip_tunnel_net *itn; 1308 1309 itn = net_generic(net, tunnel->ip_tnl_net_id); 1310 ip_tunnel_del(itn, netdev_priv(dev)); 1311 if (itn->fb_tunnel_dev == dev) 1312 WRITE_ONCE(itn->fb_tunnel_dev, NULL); 1313 1314 dst_cache_reset(&tunnel->dst_cache); 1315} 1316EXPORT_SYMBOL_GPL(ip_tunnel_uninit); 1317 1318/* Do least required initialization, rest of init is done in tunnel_init call */ 1319void ip_tunnel_setup(struct net_device *dev, unsigned int net_id) 1320{ 1321 struct ip_tunnel *tunnel = netdev_priv(dev); 1322 tunnel->ip_tnl_net_id = net_id; 1323} 1324EXPORT_SYMBOL_GPL(ip_tunnel_setup); 1325 1326MODULE_DESCRIPTION("IPv4 tunnel implementation library"); 1327MODULE_LICENSE("GPL"); 1328