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