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/types.h> 9 #include <linux/kernel.h> 10 #include <linux/skbuff.h> 11 #include <linux/netdevice.h> 12 #include <linux/in.h> 13 #include <linux/if_arp.h> 14 #include <linux/init.h> 15 #include <linux/in6.h> 16 #include <linux/inetdevice.h> 17 #include <linux/netfilter_ipv4.h> 18 #include <linux/etherdevice.h> 19 #include <linux/if_ether.h> 20 #include <linux/if_vlan.h> 21 #include <linux/static_key.h> 22 23 #include <net/ip.h> 24 #include <net/icmp.h> 25 #include <net/protocol.h> 26 #include <net/ip_tunnels.h> 27 #include <net/ip6_tunnel.h> 28 #include <net/ip6_checksum.h> 29 #include <net/arp.h> 30 #include <net/checksum.h> 31 #include <net/dsfield.h> 32 #include <net/inet_ecn.h> 33 #include <net/xfrm.h> 34 #include <net/net_namespace.h> 35 #include <net/netns/generic.h> 36 #include <net/rtnetlink.h> 37 #include <net/dst_metadata.h> 38 #include <net/geneve.h> 39 #include <net/vxlan.h> 40 #include <net/erspan.h> 41 42 const struct ip_tunnel_encap_ops __rcu * 43 iptun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly; 44 EXPORT_SYMBOL(iptun_encaps); 45 46 const struct ip6_tnl_encap_ops __rcu * 47 ip6tun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly; 48 EXPORT_SYMBOL(ip6tun_encaps); 49 50 void iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb, 51 __be32 src, __be32 dst, __u8 proto, 52 __u8 tos, __u8 ttl, __be16 df, bool xnet) 53 { 54 int pkt_len = skb->len - skb_inner_network_offset(skb); 55 struct net *net = dev_net(rt->dst.dev); 56 struct net_device *dev = skb->dev; 57 struct iphdr *iph; 58 int err; 59 60 skb_scrub_packet(skb, xnet); 61 62 skb_clear_hash_if_not_l4(skb); 63 skb_dst_set(skb, &rt->dst); 64 memset(IPCB(skb), 0, sizeof(*IPCB(skb))); 65 66 /* Push down and install the IP header. */ 67 skb_push(skb, sizeof(struct iphdr)); 68 skb_reset_network_header(skb); 69 70 iph = ip_hdr(skb); 71 72 iph->version = 4; 73 iph->ihl = sizeof(struct iphdr) >> 2; 74 iph->frag_off = ip_mtu_locked(&rt->dst) ? 0 : df; 75 iph->protocol = proto; 76 iph->tos = tos; 77 iph->daddr = dst; 78 iph->saddr = src; 79 iph->ttl = ttl; 80 __ip_select_ident(net, iph, skb_shinfo(skb)->gso_segs ?: 1); 81 82 err = ip_local_out(net, sk, skb); 83 84 if (dev) { 85 if (unlikely(net_xmit_eval(err))) 86 pkt_len = 0; 87 iptunnel_xmit_stats(dev, pkt_len); 88 } 89 } 90 EXPORT_SYMBOL_GPL(iptunnel_xmit); 91 92 int __iptunnel_pull_header(struct sk_buff *skb, int hdr_len, 93 __be16 inner_proto, bool raw_proto, bool xnet) 94 { 95 if (unlikely(!pskb_may_pull(skb, hdr_len))) 96 return -ENOMEM; 97 98 skb_pull_rcsum(skb, hdr_len); 99 100 if (!raw_proto && inner_proto == htons(ETH_P_TEB)) { 101 struct ethhdr *eh; 102 103 if (unlikely(!pskb_may_pull(skb, ETH_HLEN))) 104 return -ENOMEM; 105 106 eh = (struct ethhdr *)skb->data; 107 if (likely(eth_proto_is_802_3(eh->h_proto))) 108 skb->protocol = eh->h_proto; 109 else 110 skb->protocol = htons(ETH_P_802_2); 111 112 } else { 113 skb->protocol = inner_proto; 114 } 115 116 skb_clear_hash_if_not_l4(skb); 117 __vlan_hwaccel_clear_tag(skb); 118 skb_set_queue_mapping(skb, 0); 119 skb_scrub_packet(skb, xnet); 120 121 return iptunnel_pull_offloads(skb); 122 } 123 EXPORT_SYMBOL_GPL(__iptunnel_pull_header); 124 125 struct metadata_dst *iptunnel_metadata_reply(struct metadata_dst *md, 126 gfp_t flags) 127 { 128 struct metadata_dst *res; 129 struct ip_tunnel_info *dst, *src; 130 131 if (!md || md->type != METADATA_IP_TUNNEL || 132 md->u.tun_info.mode & IP_TUNNEL_INFO_TX) 133 return NULL; 134 135 src = &md->u.tun_info; 136 res = metadata_dst_alloc(src->options_len, METADATA_IP_TUNNEL, flags); 137 if (!res) 138 return NULL; 139 140 dst = &res->u.tun_info; 141 dst->key.tun_id = src->key.tun_id; 142 if (src->mode & IP_TUNNEL_INFO_IPV6) 143 memcpy(&dst->key.u.ipv6.dst, &src->key.u.ipv6.src, 144 sizeof(struct in6_addr)); 145 else 146 dst->key.u.ipv4.dst = src->key.u.ipv4.src; 147 dst->key.tun_flags = src->key.tun_flags; 148 dst->mode = src->mode | IP_TUNNEL_INFO_TX; 149 ip_tunnel_info_opts_set(dst, ip_tunnel_info_opts(src), 150 src->options_len, 0); 151 152 return res; 153 } 154 EXPORT_SYMBOL_GPL(iptunnel_metadata_reply); 155 156 int iptunnel_handle_offloads(struct sk_buff *skb, 157 int gso_type_mask) 158 { 159 int err; 160 161 if (likely(!skb->encapsulation)) { 162 skb_reset_inner_headers(skb); 163 skb->encapsulation = 1; 164 } 165 166 if (skb_is_gso(skb)) { 167 err = skb_header_unclone(skb, GFP_ATOMIC); 168 if (unlikely(err)) 169 return err; 170 skb_shinfo(skb)->gso_type |= gso_type_mask; 171 return 0; 172 } 173 174 if (skb->ip_summed != CHECKSUM_PARTIAL) { 175 skb->ip_summed = CHECKSUM_NONE; 176 /* We clear encapsulation here to prevent badly-written 177 * drivers potentially deciding to offload an inner checksum 178 * if we set CHECKSUM_PARTIAL on the outer header. 179 * This should go away when the drivers are all fixed. 180 */ 181 skb->encapsulation = 0; 182 } 183 184 return 0; 185 } 186 EXPORT_SYMBOL_GPL(iptunnel_handle_offloads); 187 188 /** 189 * iptunnel_pmtud_build_icmp() - Build ICMP error message for PMTUD 190 * @skb: Original packet with L2 header 191 * @mtu: MTU value for ICMP error 192 * 193 * Return: length on success, negative error code if message couldn't be built. 194 */ 195 static int iptunnel_pmtud_build_icmp(struct sk_buff *skb, int mtu) 196 { 197 const struct iphdr *iph = ip_hdr(skb); 198 struct icmphdr *icmph; 199 struct iphdr *niph; 200 struct ethhdr eh; 201 int len, err; 202 203 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct iphdr))) 204 return -EINVAL; 205 206 skb_copy_bits(skb, skb_mac_offset(skb), &eh, ETH_HLEN); 207 pskb_pull(skb, ETH_HLEN); 208 skb_reset_network_header(skb); 209 210 err = pskb_trim(skb, 576 - sizeof(*niph) - sizeof(*icmph)); 211 if (err) 212 return err; 213 214 len = skb->len + sizeof(*icmph); 215 err = skb_cow(skb, sizeof(*niph) + sizeof(*icmph) + ETH_HLEN); 216 if (err) 217 return err; 218 219 icmph = skb_push(skb, sizeof(*icmph)); 220 *icmph = (struct icmphdr) { 221 .type = ICMP_DEST_UNREACH, 222 .code = ICMP_FRAG_NEEDED, 223 .checksum = 0, 224 .un.frag.__unused = 0, 225 .un.frag.mtu = ntohs(mtu), 226 }; 227 icmph->checksum = ip_compute_csum(icmph, len); 228 skb_reset_transport_header(skb); 229 230 niph = skb_push(skb, sizeof(*niph)); 231 *niph = (struct iphdr) { 232 .ihl = sizeof(*niph) / 4u, 233 .version = 4, 234 .tos = 0, 235 .tot_len = htons(len + sizeof(*niph)), 236 .id = 0, 237 .frag_off = htons(IP_DF), 238 .ttl = iph->ttl, 239 .protocol = IPPROTO_ICMP, 240 .saddr = iph->daddr, 241 .daddr = iph->saddr, 242 }; 243 ip_send_check(niph); 244 skb_reset_network_header(skb); 245 246 skb->ip_summed = CHECKSUM_NONE; 247 248 eth_header(skb, skb->dev, htons(eh.h_proto), eh.h_source, eh.h_dest, 0); 249 skb_reset_mac_header(skb); 250 251 return skb->len; 252 } 253 254 /** 255 * iptunnel_pmtud_check_icmp() - Trigger ICMP reply if needed and allowed 256 * @skb: Buffer being sent by encapsulation, L2 headers expected 257 * @mtu: Network MTU for path 258 * 259 * Return: 0 for no ICMP reply, length if built, negative value on error. 260 */ 261 static int iptunnel_pmtud_check_icmp(struct sk_buff *skb, int mtu) 262 { 263 const struct icmphdr *icmph = icmp_hdr(skb); 264 const struct iphdr *iph = ip_hdr(skb); 265 266 if (mtu <= 576 || iph->frag_off != htons(IP_DF)) 267 return 0; 268 269 if (ipv4_is_lbcast(iph->daddr) || ipv4_is_multicast(iph->daddr) || 270 ipv4_is_zeronet(iph->saddr) || ipv4_is_loopback(iph->saddr) || 271 ipv4_is_lbcast(iph->saddr) || ipv4_is_multicast(iph->saddr)) 272 return 0; 273 274 if (iph->protocol == IPPROTO_ICMP && icmp_is_err(icmph->type)) 275 return 0; 276 277 return iptunnel_pmtud_build_icmp(skb, mtu); 278 } 279 280 #if IS_ENABLED(CONFIG_IPV6) 281 /** 282 * iptunnel_pmtud_build_icmpv6() - Build ICMPv6 error message for PMTUD 283 * @skb: Original packet with L2 header 284 * @mtu: MTU value for ICMPv6 error 285 * 286 * Return: length on success, negative error code if message couldn't be built. 287 */ 288 static int iptunnel_pmtud_build_icmpv6(struct sk_buff *skb, int mtu) 289 { 290 const struct ipv6hdr *ip6h = ipv6_hdr(skb); 291 struct icmp6hdr *icmp6h; 292 struct ipv6hdr *nip6h; 293 struct ethhdr eh; 294 int len, err; 295 __wsum csum; 296 297 if (!pskb_may_pull(skb, ETH_HLEN + sizeof(struct ipv6hdr))) 298 return -EINVAL; 299 300 skb_copy_bits(skb, skb_mac_offset(skb), &eh, ETH_HLEN); 301 pskb_pull(skb, ETH_HLEN); 302 skb_reset_network_header(skb); 303 304 err = pskb_trim(skb, IPV6_MIN_MTU - sizeof(*nip6h) - sizeof(*icmp6h)); 305 if (err) 306 return err; 307 308 len = skb->len + sizeof(*icmp6h); 309 err = skb_cow(skb, sizeof(*nip6h) + sizeof(*icmp6h) + ETH_HLEN); 310 if (err) 311 return err; 312 313 icmp6h = skb_push(skb, sizeof(*icmp6h)); 314 *icmp6h = (struct icmp6hdr) { 315 .icmp6_type = ICMPV6_PKT_TOOBIG, 316 .icmp6_code = 0, 317 .icmp6_cksum = 0, 318 .icmp6_mtu = htonl(mtu), 319 }; 320 skb_reset_transport_header(skb); 321 322 nip6h = skb_push(skb, sizeof(*nip6h)); 323 *nip6h = (struct ipv6hdr) { 324 .priority = 0, 325 .version = 6, 326 .flow_lbl = { 0 }, 327 .payload_len = htons(len), 328 .nexthdr = IPPROTO_ICMPV6, 329 .hop_limit = ip6h->hop_limit, 330 .saddr = ip6h->daddr, 331 .daddr = ip6h->saddr, 332 }; 333 skb_reset_network_header(skb); 334 335 csum = csum_partial(icmp6h, len, 0); 336 icmp6h->icmp6_cksum = csum_ipv6_magic(&nip6h->saddr, &nip6h->daddr, len, 337 IPPROTO_ICMPV6, csum); 338 339 skb->ip_summed = CHECKSUM_NONE; 340 341 eth_header(skb, skb->dev, htons(eh.h_proto), eh.h_source, eh.h_dest, 0); 342 skb_reset_mac_header(skb); 343 344 return skb->len; 345 } 346 347 /** 348 * iptunnel_pmtud_check_icmpv6() - Trigger ICMPv6 reply if needed and allowed 349 * @skb: Buffer being sent by encapsulation, L2 headers expected 350 * @mtu: Network MTU for path 351 * 352 * Return: 0 for no ICMPv6 reply, length if built, negative value on error. 353 */ 354 static int iptunnel_pmtud_check_icmpv6(struct sk_buff *skb, int mtu) 355 { 356 const struct ipv6hdr *ip6h = ipv6_hdr(skb); 357 int stype = ipv6_addr_type(&ip6h->saddr); 358 u8 proto = ip6h->nexthdr; 359 __be16 frag_off; 360 int offset; 361 362 if (mtu <= IPV6_MIN_MTU) 363 return 0; 364 365 if (stype == IPV6_ADDR_ANY || stype == IPV6_ADDR_MULTICAST || 366 stype == IPV6_ADDR_LOOPBACK) 367 return 0; 368 369 offset = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &proto, 370 &frag_off); 371 if (offset < 0 || (frag_off & htons(~0x7))) 372 return 0; 373 374 if (proto == IPPROTO_ICMPV6) { 375 struct icmp6hdr *icmp6h; 376 377 if (!pskb_may_pull(skb, skb_network_header(skb) + 378 offset + 1 - skb->data)) 379 return 0; 380 381 icmp6h = (struct icmp6hdr *)(skb_network_header(skb) + offset); 382 if (icmpv6_is_err(icmp6h->icmp6_type) || 383 icmp6h->icmp6_type == NDISC_REDIRECT) 384 return 0; 385 } 386 387 return iptunnel_pmtud_build_icmpv6(skb, mtu); 388 } 389 #endif /* IS_ENABLED(CONFIG_IPV6) */ 390 391 /** 392 * skb_tunnel_check_pmtu() - Check, update PMTU and trigger ICMP reply as needed 393 * @skb: Buffer being sent by encapsulation, L2 headers expected 394 * @encap_dst: Destination for tunnel encapsulation (outer IP) 395 * @headroom: Encapsulation header size, bytes 396 * @reply: Build matching ICMP or ICMPv6 message as a result 397 * 398 * L2 tunnel implementations that can carry IP and can be directly bridged 399 * (currently UDP tunnels) can't always rely on IP forwarding paths to handle 400 * PMTU discovery. In the bridged case, ICMP or ICMPv6 messages need to be built 401 * based on payload and sent back by the encapsulation itself. 402 * 403 * For routable interfaces, we just need to update the PMTU for the destination. 404 * 405 * Return: 0 if ICMP error not needed, length if built, negative value on error 406 */ 407 int skb_tunnel_check_pmtu(struct sk_buff *skb, struct dst_entry *encap_dst, 408 int headroom, bool reply) 409 { 410 u32 mtu = dst_mtu(encap_dst) - headroom; 411 412 if ((skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu)) || 413 (!skb_is_gso(skb) && (skb->len - skb_mac_header_len(skb)) <= mtu)) 414 return 0; 415 416 skb_dst_update_pmtu_no_confirm(skb, mtu); 417 418 if (!reply || skb->pkt_type == PACKET_HOST) 419 return 0; 420 421 if (skb->protocol == htons(ETH_P_IP)) 422 return iptunnel_pmtud_check_icmp(skb, mtu); 423 424 #if IS_ENABLED(CONFIG_IPV6) 425 if (skb->protocol == htons(ETH_P_IPV6)) 426 return iptunnel_pmtud_check_icmpv6(skb, mtu); 427 #endif 428 return 0; 429 } 430 EXPORT_SYMBOL(skb_tunnel_check_pmtu); 431 432 /* Often modified stats are per cpu, other are shared (netdev->stats) */ 433 void ip_tunnel_get_stats64(struct net_device *dev, 434 struct rtnl_link_stats64 *tot) 435 { 436 int i; 437 438 netdev_stats_to_stats64(tot, &dev->stats); 439 440 for_each_possible_cpu(i) { 441 const struct pcpu_sw_netstats *tstats = 442 per_cpu_ptr(dev->tstats, i); 443 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 444 unsigned int start; 445 446 do { 447 start = u64_stats_fetch_begin_irq(&tstats->syncp); 448 rx_packets = tstats->rx_packets; 449 tx_packets = tstats->tx_packets; 450 rx_bytes = tstats->rx_bytes; 451 tx_bytes = tstats->tx_bytes; 452 } while (u64_stats_fetch_retry_irq(&tstats->syncp, start)); 453 454 tot->rx_packets += rx_packets; 455 tot->tx_packets += tx_packets; 456 tot->rx_bytes += rx_bytes; 457 tot->tx_bytes += tx_bytes; 458 } 459 } 460 EXPORT_SYMBOL_GPL(ip_tunnel_get_stats64); 461 462 static const struct nla_policy ip_tun_policy[LWTUNNEL_IP_MAX + 1] = { 463 [LWTUNNEL_IP_UNSPEC] = { .strict_start_type = LWTUNNEL_IP_OPTS }, 464 [LWTUNNEL_IP_ID] = { .type = NLA_U64 }, 465 [LWTUNNEL_IP_DST] = { .type = NLA_U32 }, 466 [LWTUNNEL_IP_SRC] = { .type = NLA_U32 }, 467 [LWTUNNEL_IP_TTL] = { .type = NLA_U8 }, 468 [LWTUNNEL_IP_TOS] = { .type = NLA_U8 }, 469 [LWTUNNEL_IP_FLAGS] = { .type = NLA_U16 }, 470 [LWTUNNEL_IP_OPTS] = { .type = NLA_NESTED }, 471 }; 472 473 static const struct nla_policy ip_opts_policy[LWTUNNEL_IP_OPTS_MAX + 1] = { 474 [LWTUNNEL_IP_OPTS_GENEVE] = { .type = NLA_NESTED }, 475 [LWTUNNEL_IP_OPTS_VXLAN] = { .type = NLA_NESTED }, 476 [LWTUNNEL_IP_OPTS_ERSPAN] = { .type = NLA_NESTED }, 477 }; 478 479 static const struct nla_policy 480 geneve_opt_policy[LWTUNNEL_IP_OPT_GENEVE_MAX + 1] = { 481 [LWTUNNEL_IP_OPT_GENEVE_CLASS] = { .type = NLA_U16 }, 482 [LWTUNNEL_IP_OPT_GENEVE_TYPE] = { .type = NLA_U8 }, 483 [LWTUNNEL_IP_OPT_GENEVE_DATA] = { .type = NLA_BINARY, .len = 128 }, 484 }; 485 486 static const struct nla_policy 487 vxlan_opt_policy[LWTUNNEL_IP_OPT_VXLAN_MAX + 1] = { 488 [LWTUNNEL_IP_OPT_VXLAN_GBP] = { .type = NLA_U32 }, 489 }; 490 491 static const struct nla_policy 492 erspan_opt_policy[LWTUNNEL_IP_OPT_ERSPAN_MAX + 1] = { 493 [LWTUNNEL_IP_OPT_ERSPAN_VER] = { .type = NLA_U8 }, 494 [LWTUNNEL_IP_OPT_ERSPAN_INDEX] = { .type = NLA_U32 }, 495 [LWTUNNEL_IP_OPT_ERSPAN_DIR] = { .type = NLA_U8 }, 496 [LWTUNNEL_IP_OPT_ERSPAN_HWID] = { .type = NLA_U8 }, 497 }; 498 499 static int ip_tun_parse_opts_geneve(struct nlattr *attr, 500 struct ip_tunnel_info *info, int opts_len, 501 struct netlink_ext_ack *extack) 502 { 503 struct nlattr *tb[LWTUNNEL_IP_OPT_GENEVE_MAX + 1]; 504 int data_len, err; 505 506 err = nla_parse_nested(tb, LWTUNNEL_IP_OPT_GENEVE_MAX, attr, 507 geneve_opt_policy, extack); 508 if (err) 509 return err; 510 511 if (!tb[LWTUNNEL_IP_OPT_GENEVE_CLASS] || 512 !tb[LWTUNNEL_IP_OPT_GENEVE_TYPE] || 513 !tb[LWTUNNEL_IP_OPT_GENEVE_DATA]) 514 return -EINVAL; 515 516 attr = tb[LWTUNNEL_IP_OPT_GENEVE_DATA]; 517 data_len = nla_len(attr); 518 if (data_len % 4) 519 return -EINVAL; 520 521 if (info) { 522 struct geneve_opt *opt = ip_tunnel_info_opts(info) + opts_len; 523 524 memcpy(opt->opt_data, nla_data(attr), data_len); 525 opt->length = data_len / 4; 526 attr = tb[LWTUNNEL_IP_OPT_GENEVE_CLASS]; 527 opt->opt_class = nla_get_be16(attr); 528 attr = tb[LWTUNNEL_IP_OPT_GENEVE_TYPE]; 529 opt->type = nla_get_u8(attr); 530 info->key.tun_flags |= TUNNEL_GENEVE_OPT; 531 } 532 533 return sizeof(struct geneve_opt) + data_len; 534 } 535 536 static int ip_tun_parse_opts_vxlan(struct nlattr *attr, 537 struct ip_tunnel_info *info, int opts_len, 538 struct netlink_ext_ack *extack) 539 { 540 struct nlattr *tb[LWTUNNEL_IP_OPT_VXLAN_MAX + 1]; 541 int err; 542 543 err = nla_parse_nested(tb, LWTUNNEL_IP_OPT_VXLAN_MAX, attr, 544 vxlan_opt_policy, extack); 545 if (err) 546 return err; 547 548 if (!tb[LWTUNNEL_IP_OPT_VXLAN_GBP]) 549 return -EINVAL; 550 551 if (info) { 552 struct vxlan_metadata *md = 553 ip_tunnel_info_opts(info) + opts_len; 554 555 attr = tb[LWTUNNEL_IP_OPT_VXLAN_GBP]; 556 md->gbp = nla_get_u32(attr); 557 info->key.tun_flags |= TUNNEL_VXLAN_OPT; 558 } 559 560 return sizeof(struct vxlan_metadata); 561 } 562 563 static int ip_tun_parse_opts_erspan(struct nlattr *attr, 564 struct ip_tunnel_info *info, int opts_len, 565 struct netlink_ext_ack *extack) 566 { 567 struct nlattr *tb[LWTUNNEL_IP_OPT_ERSPAN_MAX + 1]; 568 int err; 569 u8 ver; 570 571 err = nla_parse_nested(tb, LWTUNNEL_IP_OPT_ERSPAN_MAX, attr, 572 erspan_opt_policy, extack); 573 if (err) 574 return err; 575 576 if (!tb[LWTUNNEL_IP_OPT_ERSPAN_VER]) 577 return -EINVAL; 578 579 ver = nla_get_u8(tb[LWTUNNEL_IP_OPT_ERSPAN_VER]); 580 if (ver == 1) { 581 if (!tb[LWTUNNEL_IP_OPT_ERSPAN_INDEX]) 582 return -EINVAL; 583 } else if (ver == 2) { 584 if (!tb[LWTUNNEL_IP_OPT_ERSPAN_DIR] || 585 !tb[LWTUNNEL_IP_OPT_ERSPAN_HWID]) 586 return -EINVAL; 587 } else { 588 return -EINVAL; 589 } 590 591 if (info) { 592 struct erspan_metadata *md = 593 ip_tunnel_info_opts(info) + opts_len; 594 595 md->version = ver; 596 if (ver == 1) { 597 attr = tb[LWTUNNEL_IP_OPT_ERSPAN_INDEX]; 598 md->u.index = nla_get_be32(attr); 599 } else { 600 attr = tb[LWTUNNEL_IP_OPT_ERSPAN_DIR]; 601 md->u.md2.dir = nla_get_u8(attr); 602 attr = tb[LWTUNNEL_IP_OPT_ERSPAN_HWID]; 603 set_hwid(&md->u.md2, nla_get_u8(attr)); 604 } 605 606 info->key.tun_flags |= TUNNEL_ERSPAN_OPT; 607 } 608 609 return sizeof(struct erspan_metadata); 610 } 611 612 static int ip_tun_parse_opts(struct nlattr *attr, struct ip_tunnel_info *info, 613 struct netlink_ext_ack *extack) 614 { 615 int err, rem, opt_len, opts_len = 0, type = 0; 616 struct nlattr *nla; 617 618 if (!attr) 619 return 0; 620 621 err = nla_validate(nla_data(attr), nla_len(attr), LWTUNNEL_IP_OPTS_MAX, 622 ip_opts_policy, extack); 623 if (err) 624 return err; 625 626 nla_for_each_attr(nla, nla_data(attr), nla_len(attr), rem) { 627 switch (nla_type(nla)) { 628 case LWTUNNEL_IP_OPTS_GENEVE: 629 if (type && type != TUNNEL_GENEVE_OPT) 630 return -EINVAL; 631 opt_len = ip_tun_parse_opts_geneve(nla, info, opts_len, 632 extack); 633 if (opt_len < 0) 634 return opt_len; 635 opts_len += opt_len; 636 if (opts_len > IP_TUNNEL_OPTS_MAX) 637 return -EINVAL; 638 type = TUNNEL_GENEVE_OPT; 639 break; 640 case LWTUNNEL_IP_OPTS_VXLAN: 641 if (type) 642 return -EINVAL; 643 opt_len = ip_tun_parse_opts_vxlan(nla, info, opts_len, 644 extack); 645 if (opt_len < 0) 646 return opt_len; 647 opts_len += opt_len; 648 type = TUNNEL_VXLAN_OPT; 649 break; 650 case LWTUNNEL_IP_OPTS_ERSPAN: 651 if (type) 652 return -EINVAL; 653 opt_len = ip_tun_parse_opts_erspan(nla, info, opts_len, 654 extack); 655 if (opt_len < 0) 656 return opt_len; 657 opts_len += opt_len; 658 type = TUNNEL_ERSPAN_OPT; 659 break; 660 default: 661 return -EINVAL; 662 } 663 } 664 665 return opts_len; 666 } 667 668 static int ip_tun_get_optlen(struct nlattr *attr, 669 struct netlink_ext_ack *extack) 670 { 671 return ip_tun_parse_opts(attr, NULL, extack); 672 } 673 674 static int ip_tun_set_opts(struct nlattr *attr, struct ip_tunnel_info *info, 675 struct netlink_ext_ack *extack) 676 { 677 return ip_tun_parse_opts(attr, info, extack); 678 } 679 680 static int ip_tun_build_state(struct net *net, struct nlattr *attr, 681 unsigned int family, const void *cfg, 682 struct lwtunnel_state **ts, 683 struct netlink_ext_ack *extack) 684 { 685 struct nlattr *tb[LWTUNNEL_IP_MAX + 1]; 686 struct lwtunnel_state *new_state; 687 struct ip_tunnel_info *tun_info; 688 int err, opt_len; 689 690 err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP_MAX, attr, 691 ip_tun_policy, extack); 692 if (err < 0) 693 return err; 694 695 opt_len = ip_tun_get_optlen(tb[LWTUNNEL_IP_OPTS], extack); 696 if (opt_len < 0) 697 return opt_len; 698 699 new_state = lwtunnel_state_alloc(sizeof(*tun_info) + opt_len); 700 if (!new_state) 701 return -ENOMEM; 702 703 new_state->type = LWTUNNEL_ENCAP_IP; 704 705 tun_info = lwt_tun_info(new_state); 706 707 err = ip_tun_set_opts(tb[LWTUNNEL_IP_OPTS], tun_info, extack); 708 if (err < 0) { 709 lwtstate_free(new_state); 710 return err; 711 } 712 713 #ifdef CONFIG_DST_CACHE 714 err = dst_cache_init(&tun_info->dst_cache, GFP_KERNEL); 715 if (err) { 716 lwtstate_free(new_state); 717 return err; 718 } 719 #endif 720 721 if (tb[LWTUNNEL_IP_ID]) 722 tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP_ID]); 723 724 if (tb[LWTUNNEL_IP_DST]) 725 tun_info->key.u.ipv4.dst = nla_get_in_addr(tb[LWTUNNEL_IP_DST]); 726 727 if (tb[LWTUNNEL_IP_SRC]) 728 tun_info->key.u.ipv4.src = nla_get_in_addr(tb[LWTUNNEL_IP_SRC]); 729 730 if (tb[LWTUNNEL_IP_TTL]) 731 tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP_TTL]); 732 733 if (tb[LWTUNNEL_IP_TOS]) 734 tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP_TOS]); 735 736 if (tb[LWTUNNEL_IP_FLAGS]) 737 tun_info->key.tun_flags |= 738 (nla_get_be16(tb[LWTUNNEL_IP_FLAGS]) & 739 ~TUNNEL_OPTIONS_PRESENT); 740 741 tun_info->mode = IP_TUNNEL_INFO_TX; 742 tun_info->options_len = opt_len; 743 744 *ts = new_state; 745 746 return 0; 747 } 748 749 static void ip_tun_destroy_state(struct lwtunnel_state *lwtstate) 750 { 751 #ifdef CONFIG_DST_CACHE 752 struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate); 753 754 dst_cache_destroy(&tun_info->dst_cache); 755 #endif 756 } 757 758 static int ip_tun_fill_encap_opts_geneve(struct sk_buff *skb, 759 struct ip_tunnel_info *tun_info) 760 { 761 struct geneve_opt *opt; 762 struct nlattr *nest; 763 int offset = 0; 764 765 nest = nla_nest_start_noflag(skb, LWTUNNEL_IP_OPTS_GENEVE); 766 if (!nest) 767 return -ENOMEM; 768 769 while (tun_info->options_len > offset) { 770 opt = ip_tunnel_info_opts(tun_info) + offset; 771 if (nla_put_be16(skb, LWTUNNEL_IP_OPT_GENEVE_CLASS, 772 opt->opt_class) || 773 nla_put_u8(skb, LWTUNNEL_IP_OPT_GENEVE_TYPE, opt->type) || 774 nla_put(skb, LWTUNNEL_IP_OPT_GENEVE_DATA, opt->length * 4, 775 opt->opt_data)) { 776 nla_nest_cancel(skb, nest); 777 return -ENOMEM; 778 } 779 offset += sizeof(*opt) + opt->length * 4; 780 } 781 782 nla_nest_end(skb, nest); 783 return 0; 784 } 785 786 static int ip_tun_fill_encap_opts_vxlan(struct sk_buff *skb, 787 struct ip_tunnel_info *tun_info) 788 { 789 struct vxlan_metadata *md; 790 struct nlattr *nest; 791 792 nest = nla_nest_start_noflag(skb, LWTUNNEL_IP_OPTS_VXLAN); 793 if (!nest) 794 return -ENOMEM; 795 796 md = ip_tunnel_info_opts(tun_info); 797 if (nla_put_u32(skb, LWTUNNEL_IP_OPT_VXLAN_GBP, md->gbp)) { 798 nla_nest_cancel(skb, nest); 799 return -ENOMEM; 800 } 801 802 nla_nest_end(skb, nest); 803 return 0; 804 } 805 806 static int ip_tun_fill_encap_opts_erspan(struct sk_buff *skb, 807 struct ip_tunnel_info *tun_info) 808 { 809 struct erspan_metadata *md; 810 struct nlattr *nest; 811 812 nest = nla_nest_start_noflag(skb, LWTUNNEL_IP_OPTS_ERSPAN); 813 if (!nest) 814 return -ENOMEM; 815 816 md = ip_tunnel_info_opts(tun_info); 817 if (nla_put_u8(skb, LWTUNNEL_IP_OPT_ERSPAN_VER, md->version)) 818 goto err; 819 820 if (md->version == 1 && 821 nla_put_be32(skb, LWTUNNEL_IP_OPT_ERSPAN_INDEX, md->u.index)) 822 goto err; 823 824 if (md->version == 2 && 825 (nla_put_u8(skb, LWTUNNEL_IP_OPT_ERSPAN_DIR, md->u.md2.dir) || 826 nla_put_u8(skb, LWTUNNEL_IP_OPT_ERSPAN_HWID, 827 get_hwid(&md->u.md2)))) 828 goto err; 829 830 nla_nest_end(skb, nest); 831 return 0; 832 err: 833 nla_nest_cancel(skb, nest); 834 return -ENOMEM; 835 } 836 837 static int ip_tun_fill_encap_opts(struct sk_buff *skb, int type, 838 struct ip_tunnel_info *tun_info) 839 { 840 struct nlattr *nest; 841 int err = 0; 842 843 if (!(tun_info->key.tun_flags & TUNNEL_OPTIONS_PRESENT)) 844 return 0; 845 846 nest = nla_nest_start_noflag(skb, type); 847 if (!nest) 848 return -ENOMEM; 849 850 if (tun_info->key.tun_flags & TUNNEL_GENEVE_OPT) 851 err = ip_tun_fill_encap_opts_geneve(skb, tun_info); 852 else if (tun_info->key.tun_flags & TUNNEL_VXLAN_OPT) 853 err = ip_tun_fill_encap_opts_vxlan(skb, tun_info); 854 else if (tun_info->key.tun_flags & TUNNEL_ERSPAN_OPT) 855 err = ip_tun_fill_encap_opts_erspan(skb, tun_info); 856 857 if (err) { 858 nla_nest_cancel(skb, nest); 859 return err; 860 } 861 862 nla_nest_end(skb, nest); 863 return 0; 864 } 865 866 static int ip_tun_fill_encap_info(struct sk_buff *skb, 867 struct lwtunnel_state *lwtstate) 868 { 869 struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate); 870 871 if (nla_put_be64(skb, LWTUNNEL_IP_ID, tun_info->key.tun_id, 872 LWTUNNEL_IP_PAD) || 873 nla_put_in_addr(skb, LWTUNNEL_IP_DST, tun_info->key.u.ipv4.dst) || 874 nla_put_in_addr(skb, LWTUNNEL_IP_SRC, tun_info->key.u.ipv4.src) || 875 nla_put_u8(skb, LWTUNNEL_IP_TOS, tun_info->key.tos) || 876 nla_put_u8(skb, LWTUNNEL_IP_TTL, tun_info->key.ttl) || 877 nla_put_be16(skb, LWTUNNEL_IP_FLAGS, tun_info->key.tun_flags) || 878 ip_tun_fill_encap_opts(skb, LWTUNNEL_IP_OPTS, tun_info)) 879 return -ENOMEM; 880 881 return 0; 882 } 883 884 static int ip_tun_opts_nlsize(struct ip_tunnel_info *info) 885 { 886 int opt_len; 887 888 if (!(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT)) 889 return 0; 890 891 opt_len = nla_total_size(0); /* LWTUNNEL_IP_OPTS */ 892 if (info->key.tun_flags & TUNNEL_GENEVE_OPT) { 893 struct geneve_opt *opt; 894 int offset = 0; 895 896 opt_len += nla_total_size(0); /* LWTUNNEL_IP_OPTS_GENEVE */ 897 while (info->options_len > offset) { 898 opt = ip_tunnel_info_opts(info) + offset; 899 opt_len += nla_total_size(2) /* OPT_GENEVE_CLASS */ 900 + nla_total_size(1) /* OPT_GENEVE_TYPE */ 901 + nla_total_size(opt->length * 4); 902 /* OPT_GENEVE_DATA */ 903 offset += sizeof(*opt) + opt->length * 4; 904 } 905 } else if (info->key.tun_flags & TUNNEL_VXLAN_OPT) { 906 opt_len += nla_total_size(0) /* LWTUNNEL_IP_OPTS_VXLAN */ 907 + nla_total_size(4); /* OPT_VXLAN_GBP */ 908 } else if (info->key.tun_flags & TUNNEL_ERSPAN_OPT) { 909 struct erspan_metadata *md = ip_tunnel_info_opts(info); 910 911 opt_len += nla_total_size(0) /* LWTUNNEL_IP_OPTS_ERSPAN */ 912 + nla_total_size(1) /* OPT_ERSPAN_VER */ 913 + (md->version == 1 ? nla_total_size(4) 914 /* OPT_ERSPAN_INDEX (v1) */ 915 : nla_total_size(1) + 916 nla_total_size(1)); 917 /* OPT_ERSPAN_DIR + HWID (v2) */ 918 } 919 920 return opt_len; 921 } 922 923 static int ip_tun_encap_nlsize(struct lwtunnel_state *lwtstate) 924 { 925 return nla_total_size_64bit(8) /* LWTUNNEL_IP_ID */ 926 + nla_total_size(4) /* LWTUNNEL_IP_DST */ 927 + nla_total_size(4) /* LWTUNNEL_IP_SRC */ 928 + nla_total_size(1) /* LWTUNNEL_IP_TOS */ 929 + nla_total_size(1) /* LWTUNNEL_IP_TTL */ 930 + nla_total_size(2) /* LWTUNNEL_IP_FLAGS */ 931 + ip_tun_opts_nlsize(lwt_tun_info(lwtstate)); 932 /* LWTUNNEL_IP_OPTS */ 933 } 934 935 static int ip_tun_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b) 936 { 937 struct ip_tunnel_info *info_a = lwt_tun_info(a); 938 struct ip_tunnel_info *info_b = lwt_tun_info(b); 939 940 return memcmp(info_a, info_b, sizeof(info_a->key)) || 941 info_a->mode != info_b->mode || 942 info_a->options_len != info_b->options_len || 943 memcmp(ip_tunnel_info_opts(info_a), 944 ip_tunnel_info_opts(info_b), info_a->options_len); 945 } 946 947 static const struct lwtunnel_encap_ops ip_tun_lwt_ops = { 948 .build_state = ip_tun_build_state, 949 .destroy_state = ip_tun_destroy_state, 950 .fill_encap = ip_tun_fill_encap_info, 951 .get_encap_size = ip_tun_encap_nlsize, 952 .cmp_encap = ip_tun_cmp_encap, 953 .owner = THIS_MODULE, 954 }; 955 956 static const struct nla_policy ip6_tun_policy[LWTUNNEL_IP6_MAX + 1] = { 957 [LWTUNNEL_IP6_UNSPEC] = { .strict_start_type = LWTUNNEL_IP6_OPTS }, 958 [LWTUNNEL_IP6_ID] = { .type = NLA_U64 }, 959 [LWTUNNEL_IP6_DST] = { .len = sizeof(struct in6_addr) }, 960 [LWTUNNEL_IP6_SRC] = { .len = sizeof(struct in6_addr) }, 961 [LWTUNNEL_IP6_HOPLIMIT] = { .type = NLA_U8 }, 962 [LWTUNNEL_IP6_TC] = { .type = NLA_U8 }, 963 [LWTUNNEL_IP6_FLAGS] = { .type = NLA_U16 }, 964 [LWTUNNEL_IP6_OPTS] = { .type = NLA_NESTED }, 965 }; 966 967 static int ip6_tun_build_state(struct net *net, struct nlattr *attr, 968 unsigned int family, const void *cfg, 969 struct lwtunnel_state **ts, 970 struct netlink_ext_ack *extack) 971 { 972 struct nlattr *tb[LWTUNNEL_IP6_MAX + 1]; 973 struct lwtunnel_state *new_state; 974 struct ip_tunnel_info *tun_info; 975 int err, opt_len; 976 977 err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP6_MAX, attr, 978 ip6_tun_policy, extack); 979 if (err < 0) 980 return err; 981 982 opt_len = ip_tun_get_optlen(tb[LWTUNNEL_IP6_OPTS], extack); 983 if (opt_len < 0) 984 return opt_len; 985 986 new_state = lwtunnel_state_alloc(sizeof(*tun_info) + opt_len); 987 if (!new_state) 988 return -ENOMEM; 989 990 new_state->type = LWTUNNEL_ENCAP_IP6; 991 992 tun_info = lwt_tun_info(new_state); 993 994 err = ip_tun_set_opts(tb[LWTUNNEL_IP6_OPTS], tun_info, extack); 995 if (err < 0) { 996 lwtstate_free(new_state); 997 return err; 998 } 999 1000 if (tb[LWTUNNEL_IP6_ID]) 1001 tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP6_ID]); 1002 1003 if (tb[LWTUNNEL_IP6_DST]) 1004 tun_info->key.u.ipv6.dst = nla_get_in6_addr(tb[LWTUNNEL_IP6_DST]); 1005 1006 if (tb[LWTUNNEL_IP6_SRC]) 1007 tun_info->key.u.ipv6.src = nla_get_in6_addr(tb[LWTUNNEL_IP6_SRC]); 1008 1009 if (tb[LWTUNNEL_IP6_HOPLIMIT]) 1010 tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP6_HOPLIMIT]); 1011 1012 if (tb[LWTUNNEL_IP6_TC]) 1013 tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP6_TC]); 1014 1015 if (tb[LWTUNNEL_IP6_FLAGS]) 1016 tun_info->key.tun_flags |= 1017 (nla_get_be16(tb[LWTUNNEL_IP6_FLAGS]) & 1018 ~TUNNEL_OPTIONS_PRESENT); 1019 1020 tun_info->mode = IP_TUNNEL_INFO_TX | IP_TUNNEL_INFO_IPV6; 1021 tun_info->options_len = opt_len; 1022 1023 *ts = new_state; 1024 1025 return 0; 1026 } 1027 1028 static int ip6_tun_fill_encap_info(struct sk_buff *skb, 1029 struct lwtunnel_state *lwtstate) 1030 { 1031 struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate); 1032 1033 if (nla_put_be64(skb, LWTUNNEL_IP6_ID, tun_info->key.tun_id, 1034 LWTUNNEL_IP6_PAD) || 1035 nla_put_in6_addr(skb, LWTUNNEL_IP6_DST, &tun_info->key.u.ipv6.dst) || 1036 nla_put_in6_addr(skb, LWTUNNEL_IP6_SRC, &tun_info->key.u.ipv6.src) || 1037 nla_put_u8(skb, LWTUNNEL_IP6_TC, tun_info->key.tos) || 1038 nla_put_u8(skb, LWTUNNEL_IP6_HOPLIMIT, tun_info->key.ttl) || 1039 nla_put_be16(skb, LWTUNNEL_IP6_FLAGS, tun_info->key.tun_flags) || 1040 ip_tun_fill_encap_opts(skb, LWTUNNEL_IP6_OPTS, tun_info)) 1041 return -ENOMEM; 1042 1043 return 0; 1044 } 1045 1046 static int ip6_tun_encap_nlsize(struct lwtunnel_state *lwtstate) 1047 { 1048 return nla_total_size_64bit(8) /* LWTUNNEL_IP6_ID */ 1049 + nla_total_size(16) /* LWTUNNEL_IP6_DST */ 1050 + nla_total_size(16) /* LWTUNNEL_IP6_SRC */ 1051 + nla_total_size(1) /* LWTUNNEL_IP6_HOPLIMIT */ 1052 + nla_total_size(1) /* LWTUNNEL_IP6_TC */ 1053 + nla_total_size(2) /* LWTUNNEL_IP6_FLAGS */ 1054 + ip_tun_opts_nlsize(lwt_tun_info(lwtstate)); 1055 /* LWTUNNEL_IP6_OPTS */ 1056 } 1057 1058 static const struct lwtunnel_encap_ops ip6_tun_lwt_ops = { 1059 .build_state = ip6_tun_build_state, 1060 .fill_encap = ip6_tun_fill_encap_info, 1061 .get_encap_size = ip6_tun_encap_nlsize, 1062 .cmp_encap = ip_tun_cmp_encap, 1063 .owner = THIS_MODULE, 1064 }; 1065 1066 void __init ip_tunnel_core_init(void) 1067 { 1068 /* If you land here, make sure whether increasing ip_tunnel_info's 1069 * options_len is a reasonable choice with its usage in front ends 1070 * (f.e., it's part of flow keys, etc). 1071 */ 1072 BUILD_BUG_ON(IP_TUNNEL_OPTS_MAX != 255); 1073 1074 lwtunnel_encap_add_ops(&ip_tun_lwt_ops, LWTUNNEL_ENCAP_IP); 1075 lwtunnel_encap_add_ops(&ip6_tun_lwt_ops, LWTUNNEL_ENCAP_IP6); 1076 } 1077 1078 DEFINE_STATIC_KEY_FALSE(ip_tunnel_metadata_cnt); 1079 EXPORT_SYMBOL(ip_tunnel_metadata_cnt); 1080 1081 void ip_tunnel_need_metadata(void) 1082 { 1083 static_branch_inc(&ip_tunnel_metadata_cnt); 1084 } 1085 EXPORT_SYMBOL_GPL(ip_tunnel_need_metadata); 1086 1087 void ip_tunnel_unneed_metadata(void) 1088 { 1089 static_branch_dec(&ip_tunnel_metadata_cnt); 1090 } 1091 EXPORT_SYMBOL_GPL(ip_tunnel_unneed_metadata); 1092 1093 /* Returns either the correct skb->protocol value, or 0 if invalid. */ 1094 __be16 ip_tunnel_parse_protocol(const struct sk_buff *skb) 1095 { 1096 if (skb_network_header(skb) >= skb->head && 1097 (skb_network_header(skb) + sizeof(struct iphdr)) <= skb_tail_pointer(skb) && 1098 ip_hdr(skb)->version == 4) 1099 return htons(ETH_P_IP); 1100 if (skb_network_header(skb) >= skb->head && 1101 (skb_network_header(skb) + sizeof(struct ipv6hdr)) <= skb_tail_pointer(skb) && 1102 ipv6_hdr(skb)->version == 6) 1103 return htons(ETH_P_IPV6); 1104 return 0; 1105 } 1106 EXPORT_SYMBOL(ip_tunnel_parse_protocol); 1107 1108 const struct header_ops ip_tunnel_header_ops = { .parse_protocol = ip_tunnel_parse_protocol }; 1109 EXPORT_SYMBOL(ip_tunnel_header_ops); 1110