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 md->gbp &= VXLAN_GBP_MASK; 558 info->key.tun_flags |= TUNNEL_VXLAN_OPT; 559 } 560 561 return sizeof(struct vxlan_metadata); 562 } 563 564 static int ip_tun_parse_opts_erspan(struct nlattr *attr, 565 struct ip_tunnel_info *info, int opts_len, 566 struct netlink_ext_ack *extack) 567 { 568 struct nlattr *tb[LWTUNNEL_IP_OPT_ERSPAN_MAX + 1]; 569 int err; 570 u8 ver; 571 572 err = nla_parse_nested(tb, LWTUNNEL_IP_OPT_ERSPAN_MAX, attr, 573 erspan_opt_policy, extack); 574 if (err) 575 return err; 576 577 if (!tb[LWTUNNEL_IP_OPT_ERSPAN_VER]) 578 return -EINVAL; 579 580 ver = nla_get_u8(tb[LWTUNNEL_IP_OPT_ERSPAN_VER]); 581 if (ver == 1) { 582 if (!tb[LWTUNNEL_IP_OPT_ERSPAN_INDEX]) 583 return -EINVAL; 584 } else if (ver == 2) { 585 if (!tb[LWTUNNEL_IP_OPT_ERSPAN_DIR] || 586 !tb[LWTUNNEL_IP_OPT_ERSPAN_HWID]) 587 return -EINVAL; 588 } else { 589 return -EINVAL; 590 } 591 592 if (info) { 593 struct erspan_metadata *md = 594 ip_tunnel_info_opts(info) + opts_len; 595 596 md->version = ver; 597 if (ver == 1) { 598 attr = tb[LWTUNNEL_IP_OPT_ERSPAN_INDEX]; 599 md->u.index = nla_get_be32(attr); 600 } else { 601 attr = tb[LWTUNNEL_IP_OPT_ERSPAN_DIR]; 602 md->u.md2.dir = nla_get_u8(attr); 603 attr = tb[LWTUNNEL_IP_OPT_ERSPAN_HWID]; 604 set_hwid(&md->u.md2, nla_get_u8(attr)); 605 } 606 607 info->key.tun_flags |= TUNNEL_ERSPAN_OPT; 608 } 609 610 return sizeof(struct erspan_metadata); 611 } 612 613 static int ip_tun_parse_opts(struct nlattr *attr, struct ip_tunnel_info *info, 614 struct netlink_ext_ack *extack) 615 { 616 int err, rem, opt_len, opts_len = 0, type = 0; 617 struct nlattr *nla; 618 619 if (!attr) 620 return 0; 621 622 err = nla_validate(nla_data(attr), nla_len(attr), LWTUNNEL_IP_OPTS_MAX, 623 ip_opts_policy, extack); 624 if (err) 625 return err; 626 627 nla_for_each_attr(nla, nla_data(attr), nla_len(attr), rem) { 628 switch (nla_type(nla)) { 629 case LWTUNNEL_IP_OPTS_GENEVE: 630 if (type && type != TUNNEL_GENEVE_OPT) 631 return -EINVAL; 632 opt_len = ip_tun_parse_opts_geneve(nla, info, opts_len, 633 extack); 634 if (opt_len < 0) 635 return opt_len; 636 opts_len += opt_len; 637 if (opts_len > IP_TUNNEL_OPTS_MAX) 638 return -EINVAL; 639 type = TUNNEL_GENEVE_OPT; 640 break; 641 case LWTUNNEL_IP_OPTS_VXLAN: 642 if (type) 643 return -EINVAL; 644 opt_len = ip_tun_parse_opts_vxlan(nla, info, opts_len, 645 extack); 646 if (opt_len < 0) 647 return opt_len; 648 opts_len += opt_len; 649 type = TUNNEL_VXLAN_OPT; 650 break; 651 case LWTUNNEL_IP_OPTS_ERSPAN: 652 if (type) 653 return -EINVAL; 654 opt_len = ip_tun_parse_opts_erspan(nla, info, opts_len, 655 extack); 656 if (opt_len < 0) 657 return opt_len; 658 opts_len += opt_len; 659 type = TUNNEL_ERSPAN_OPT; 660 break; 661 default: 662 return -EINVAL; 663 } 664 } 665 666 return opts_len; 667 } 668 669 static int ip_tun_get_optlen(struct nlattr *attr, 670 struct netlink_ext_ack *extack) 671 { 672 return ip_tun_parse_opts(attr, NULL, extack); 673 } 674 675 static int ip_tun_set_opts(struct nlattr *attr, struct ip_tunnel_info *info, 676 struct netlink_ext_ack *extack) 677 { 678 return ip_tun_parse_opts(attr, info, extack); 679 } 680 681 static int ip_tun_build_state(struct net *net, struct nlattr *attr, 682 unsigned int family, const void *cfg, 683 struct lwtunnel_state **ts, 684 struct netlink_ext_ack *extack) 685 { 686 struct nlattr *tb[LWTUNNEL_IP_MAX + 1]; 687 struct lwtunnel_state *new_state; 688 struct ip_tunnel_info *tun_info; 689 int err, opt_len; 690 691 err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP_MAX, attr, 692 ip_tun_policy, extack); 693 if (err < 0) 694 return err; 695 696 opt_len = ip_tun_get_optlen(tb[LWTUNNEL_IP_OPTS], extack); 697 if (opt_len < 0) 698 return opt_len; 699 700 new_state = lwtunnel_state_alloc(sizeof(*tun_info) + opt_len); 701 if (!new_state) 702 return -ENOMEM; 703 704 new_state->type = LWTUNNEL_ENCAP_IP; 705 706 tun_info = lwt_tun_info(new_state); 707 708 err = ip_tun_set_opts(tb[LWTUNNEL_IP_OPTS], tun_info, extack); 709 if (err < 0) { 710 lwtstate_free(new_state); 711 return err; 712 } 713 714 #ifdef CONFIG_DST_CACHE 715 err = dst_cache_init(&tun_info->dst_cache, GFP_KERNEL); 716 if (err) { 717 lwtstate_free(new_state); 718 return err; 719 } 720 #endif 721 722 if (tb[LWTUNNEL_IP_ID]) 723 tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP_ID]); 724 725 if (tb[LWTUNNEL_IP_DST]) 726 tun_info->key.u.ipv4.dst = nla_get_in_addr(tb[LWTUNNEL_IP_DST]); 727 728 if (tb[LWTUNNEL_IP_SRC]) 729 tun_info->key.u.ipv4.src = nla_get_in_addr(tb[LWTUNNEL_IP_SRC]); 730 731 if (tb[LWTUNNEL_IP_TTL]) 732 tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP_TTL]); 733 734 if (tb[LWTUNNEL_IP_TOS]) 735 tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP_TOS]); 736 737 if (tb[LWTUNNEL_IP_FLAGS]) 738 tun_info->key.tun_flags |= 739 (nla_get_be16(tb[LWTUNNEL_IP_FLAGS]) & 740 ~TUNNEL_OPTIONS_PRESENT); 741 742 tun_info->mode = IP_TUNNEL_INFO_TX; 743 tun_info->options_len = opt_len; 744 745 *ts = new_state; 746 747 return 0; 748 } 749 750 static void ip_tun_destroy_state(struct lwtunnel_state *lwtstate) 751 { 752 #ifdef CONFIG_DST_CACHE 753 struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate); 754 755 dst_cache_destroy(&tun_info->dst_cache); 756 #endif 757 } 758 759 static int ip_tun_fill_encap_opts_geneve(struct sk_buff *skb, 760 struct ip_tunnel_info *tun_info) 761 { 762 struct geneve_opt *opt; 763 struct nlattr *nest; 764 int offset = 0; 765 766 nest = nla_nest_start_noflag(skb, LWTUNNEL_IP_OPTS_GENEVE); 767 if (!nest) 768 return -ENOMEM; 769 770 while (tun_info->options_len > offset) { 771 opt = ip_tunnel_info_opts(tun_info) + offset; 772 if (nla_put_be16(skb, LWTUNNEL_IP_OPT_GENEVE_CLASS, 773 opt->opt_class) || 774 nla_put_u8(skb, LWTUNNEL_IP_OPT_GENEVE_TYPE, opt->type) || 775 nla_put(skb, LWTUNNEL_IP_OPT_GENEVE_DATA, opt->length * 4, 776 opt->opt_data)) { 777 nla_nest_cancel(skb, nest); 778 return -ENOMEM; 779 } 780 offset += sizeof(*opt) + opt->length * 4; 781 } 782 783 nla_nest_end(skb, nest); 784 return 0; 785 } 786 787 static int ip_tun_fill_encap_opts_vxlan(struct sk_buff *skb, 788 struct ip_tunnel_info *tun_info) 789 { 790 struct vxlan_metadata *md; 791 struct nlattr *nest; 792 793 nest = nla_nest_start_noflag(skb, LWTUNNEL_IP_OPTS_VXLAN); 794 if (!nest) 795 return -ENOMEM; 796 797 md = ip_tunnel_info_opts(tun_info); 798 if (nla_put_u32(skb, LWTUNNEL_IP_OPT_VXLAN_GBP, md->gbp)) { 799 nla_nest_cancel(skb, nest); 800 return -ENOMEM; 801 } 802 803 nla_nest_end(skb, nest); 804 return 0; 805 } 806 807 static int ip_tun_fill_encap_opts_erspan(struct sk_buff *skb, 808 struct ip_tunnel_info *tun_info) 809 { 810 struct erspan_metadata *md; 811 struct nlattr *nest; 812 813 nest = nla_nest_start_noflag(skb, LWTUNNEL_IP_OPTS_ERSPAN); 814 if (!nest) 815 return -ENOMEM; 816 817 md = ip_tunnel_info_opts(tun_info); 818 if (nla_put_u8(skb, LWTUNNEL_IP_OPT_ERSPAN_VER, md->version)) 819 goto err; 820 821 if (md->version == 1 && 822 nla_put_be32(skb, LWTUNNEL_IP_OPT_ERSPAN_INDEX, md->u.index)) 823 goto err; 824 825 if (md->version == 2 && 826 (nla_put_u8(skb, LWTUNNEL_IP_OPT_ERSPAN_DIR, md->u.md2.dir) || 827 nla_put_u8(skb, LWTUNNEL_IP_OPT_ERSPAN_HWID, 828 get_hwid(&md->u.md2)))) 829 goto err; 830 831 nla_nest_end(skb, nest); 832 return 0; 833 err: 834 nla_nest_cancel(skb, nest); 835 return -ENOMEM; 836 } 837 838 static int ip_tun_fill_encap_opts(struct sk_buff *skb, int type, 839 struct ip_tunnel_info *tun_info) 840 { 841 struct nlattr *nest; 842 int err = 0; 843 844 if (!(tun_info->key.tun_flags & TUNNEL_OPTIONS_PRESENT)) 845 return 0; 846 847 nest = nla_nest_start_noflag(skb, type); 848 if (!nest) 849 return -ENOMEM; 850 851 if (tun_info->key.tun_flags & TUNNEL_GENEVE_OPT) 852 err = ip_tun_fill_encap_opts_geneve(skb, tun_info); 853 else if (tun_info->key.tun_flags & TUNNEL_VXLAN_OPT) 854 err = ip_tun_fill_encap_opts_vxlan(skb, tun_info); 855 else if (tun_info->key.tun_flags & TUNNEL_ERSPAN_OPT) 856 err = ip_tun_fill_encap_opts_erspan(skb, tun_info); 857 858 if (err) { 859 nla_nest_cancel(skb, nest); 860 return err; 861 } 862 863 nla_nest_end(skb, nest); 864 return 0; 865 } 866 867 static int ip_tun_fill_encap_info(struct sk_buff *skb, 868 struct lwtunnel_state *lwtstate) 869 { 870 struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate); 871 872 if (nla_put_be64(skb, LWTUNNEL_IP_ID, tun_info->key.tun_id, 873 LWTUNNEL_IP_PAD) || 874 nla_put_in_addr(skb, LWTUNNEL_IP_DST, tun_info->key.u.ipv4.dst) || 875 nla_put_in_addr(skb, LWTUNNEL_IP_SRC, tun_info->key.u.ipv4.src) || 876 nla_put_u8(skb, LWTUNNEL_IP_TOS, tun_info->key.tos) || 877 nla_put_u8(skb, LWTUNNEL_IP_TTL, tun_info->key.ttl) || 878 nla_put_be16(skb, LWTUNNEL_IP_FLAGS, tun_info->key.tun_flags) || 879 ip_tun_fill_encap_opts(skb, LWTUNNEL_IP_OPTS, tun_info)) 880 return -ENOMEM; 881 882 return 0; 883 } 884 885 static int ip_tun_opts_nlsize(struct ip_tunnel_info *info) 886 { 887 int opt_len; 888 889 if (!(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT)) 890 return 0; 891 892 opt_len = nla_total_size(0); /* LWTUNNEL_IP_OPTS */ 893 if (info->key.tun_flags & TUNNEL_GENEVE_OPT) { 894 struct geneve_opt *opt; 895 int offset = 0; 896 897 opt_len += nla_total_size(0); /* LWTUNNEL_IP_OPTS_GENEVE */ 898 while (info->options_len > offset) { 899 opt = ip_tunnel_info_opts(info) + offset; 900 opt_len += nla_total_size(2) /* OPT_GENEVE_CLASS */ 901 + nla_total_size(1) /* OPT_GENEVE_TYPE */ 902 + nla_total_size(opt->length * 4); 903 /* OPT_GENEVE_DATA */ 904 offset += sizeof(*opt) + opt->length * 4; 905 } 906 } else if (info->key.tun_flags & TUNNEL_VXLAN_OPT) { 907 opt_len += nla_total_size(0) /* LWTUNNEL_IP_OPTS_VXLAN */ 908 + nla_total_size(4); /* OPT_VXLAN_GBP */ 909 } else if (info->key.tun_flags & TUNNEL_ERSPAN_OPT) { 910 struct erspan_metadata *md = ip_tunnel_info_opts(info); 911 912 opt_len += nla_total_size(0) /* LWTUNNEL_IP_OPTS_ERSPAN */ 913 + nla_total_size(1) /* OPT_ERSPAN_VER */ 914 + (md->version == 1 ? nla_total_size(4) 915 /* OPT_ERSPAN_INDEX (v1) */ 916 : nla_total_size(1) + 917 nla_total_size(1)); 918 /* OPT_ERSPAN_DIR + HWID (v2) */ 919 } 920 921 return opt_len; 922 } 923 924 static int ip_tun_encap_nlsize(struct lwtunnel_state *lwtstate) 925 { 926 return nla_total_size_64bit(8) /* LWTUNNEL_IP_ID */ 927 + nla_total_size(4) /* LWTUNNEL_IP_DST */ 928 + nla_total_size(4) /* LWTUNNEL_IP_SRC */ 929 + nla_total_size(1) /* LWTUNNEL_IP_TOS */ 930 + nla_total_size(1) /* LWTUNNEL_IP_TTL */ 931 + nla_total_size(2) /* LWTUNNEL_IP_FLAGS */ 932 + ip_tun_opts_nlsize(lwt_tun_info(lwtstate)); 933 /* LWTUNNEL_IP_OPTS */ 934 } 935 936 static int ip_tun_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b) 937 { 938 struct ip_tunnel_info *info_a = lwt_tun_info(a); 939 struct ip_tunnel_info *info_b = lwt_tun_info(b); 940 941 return memcmp(info_a, info_b, sizeof(info_a->key)) || 942 info_a->mode != info_b->mode || 943 info_a->options_len != info_b->options_len || 944 memcmp(ip_tunnel_info_opts(info_a), 945 ip_tunnel_info_opts(info_b), info_a->options_len); 946 } 947 948 static const struct lwtunnel_encap_ops ip_tun_lwt_ops = { 949 .build_state = ip_tun_build_state, 950 .destroy_state = ip_tun_destroy_state, 951 .fill_encap = ip_tun_fill_encap_info, 952 .get_encap_size = ip_tun_encap_nlsize, 953 .cmp_encap = ip_tun_cmp_encap, 954 .owner = THIS_MODULE, 955 }; 956 957 static const struct nla_policy ip6_tun_policy[LWTUNNEL_IP6_MAX + 1] = { 958 [LWTUNNEL_IP6_UNSPEC] = { .strict_start_type = LWTUNNEL_IP6_OPTS }, 959 [LWTUNNEL_IP6_ID] = { .type = NLA_U64 }, 960 [LWTUNNEL_IP6_DST] = { .len = sizeof(struct in6_addr) }, 961 [LWTUNNEL_IP6_SRC] = { .len = sizeof(struct in6_addr) }, 962 [LWTUNNEL_IP6_HOPLIMIT] = { .type = NLA_U8 }, 963 [LWTUNNEL_IP6_TC] = { .type = NLA_U8 }, 964 [LWTUNNEL_IP6_FLAGS] = { .type = NLA_U16 }, 965 [LWTUNNEL_IP6_OPTS] = { .type = NLA_NESTED }, 966 }; 967 968 static int ip6_tun_build_state(struct net *net, struct nlattr *attr, 969 unsigned int family, const void *cfg, 970 struct lwtunnel_state **ts, 971 struct netlink_ext_ack *extack) 972 { 973 struct nlattr *tb[LWTUNNEL_IP6_MAX + 1]; 974 struct lwtunnel_state *new_state; 975 struct ip_tunnel_info *tun_info; 976 int err, opt_len; 977 978 err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP6_MAX, attr, 979 ip6_tun_policy, extack); 980 if (err < 0) 981 return err; 982 983 opt_len = ip_tun_get_optlen(tb[LWTUNNEL_IP6_OPTS], extack); 984 if (opt_len < 0) 985 return opt_len; 986 987 new_state = lwtunnel_state_alloc(sizeof(*tun_info) + opt_len); 988 if (!new_state) 989 return -ENOMEM; 990 991 new_state->type = LWTUNNEL_ENCAP_IP6; 992 993 tun_info = lwt_tun_info(new_state); 994 995 err = ip_tun_set_opts(tb[LWTUNNEL_IP6_OPTS], tun_info, extack); 996 if (err < 0) { 997 lwtstate_free(new_state); 998 return err; 999 } 1000 1001 if (tb[LWTUNNEL_IP6_ID]) 1002 tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP6_ID]); 1003 1004 if (tb[LWTUNNEL_IP6_DST]) 1005 tun_info->key.u.ipv6.dst = nla_get_in6_addr(tb[LWTUNNEL_IP6_DST]); 1006 1007 if (tb[LWTUNNEL_IP6_SRC]) 1008 tun_info->key.u.ipv6.src = nla_get_in6_addr(tb[LWTUNNEL_IP6_SRC]); 1009 1010 if (tb[LWTUNNEL_IP6_HOPLIMIT]) 1011 tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP6_HOPLIMIT]); 1012 1013 if (tb[LWTUNNEL_IP6_TC]) 1014 tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP6_TC]); 1015 1016 if (tb[LWTUNNEL_IP6_FLAGS]) 1017 tun_info->key.tun_flags |= 1018 (nla_get_be16(tb[LWTUNNEL_IP6_FLAGS]) & 1019 ~TUNNEL_OPTIONS_PRESENT); 1020 1021 tun_info->mode = IP_TUNNEL_INFO_TX | IP_TUNNEL_INFO_IPV6; 1022 tun_info->options_len = opt_len; 1023 1024 *ts = new_state; 1025 1026 return 0; 1027 } 1028 1029 static int ip6_tun_fill_encap_info(struct sk_buff *skb, 1030 struct lwtunnel_state *lwtstate) 1031 { 1032 struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate); 1033 1034 if (nla_put_be64(skb, LWTUNNEL_IP6_ID, tun_info->key.tun_id, 1035 LWTUNNEL_IP6_PAD) || 1036 nla_put_in6_addr(skb, LWTUNNEL_IP6_DST, &tun_info->key.u.ipv6.dst) || 1037 nla_put_in6_addr(skb, LWTUNNEL_IP6_SRC, &tun_info->key.u.ipv6.src) || 1038 nla_put_u8(skb, LWTUNNEL_IP6_TC, tun_info->key.tos) || 1039 nla_put_u8(skb, LWTUNNEL_IP6_HOPLIMIT, tun_info->key.ttl) || 1040 nla_put_be16(skb, LWTUNNEL_IP6_FLAGS, tun_info->key.tun_flags) || 1041 ip_tun_fill_encap_opts(skb, LWTUNNEL_IP6_OPTS, tun_info)) 1042 return -ENOMEM; 1043 1044 return 0; 1045 } 1046 1047 static int ip6_tun_encap_nlsize(struct lwtunnel_state *lwtstate) 1048 { 1049 return nla_total_size_64bit(8) /* LWTUNNEL_IP6_ID */ 1050 + nla_total_size(16) /* LWTUNNEL_IP6_DST */ 1051 + nla_total_size(16) /* LWTUNNEL_IP6_SRC */ 1052 + nla_total_size(1) /* LWTUNNEL_IP6_HOPLIMIT */ 1053 + nla_total_size(1) /* LWTUNNEL_IP6_TC */ 1054 + nla_total_size(2) /* LWTUNNEL_IP6_FLAGS */ 1055 + ip_tun_opts_nlsize(lwt_tun_info(lwtstate)); 1056 /* LWTUNNEL_IP6_OPTS */ 1057 } 1058 1059 static const struct lwtunnel_encap_ops ip6_tun_lwt_ops = { 1060 .build_state = ip6_tun_build_state, 1061 .fill_encap = ip6_tun_fill_encap_info, 1062 .get_encap_size = ip6_tun_encap_nlsize, 1063 .cmp_encap = ip_tun_cmp_encap, 1064 .owner = THIS_MODULE, 1065 }; 1066 1067 void __init ip_tunnel_core_init(void) 1068 { 1069 /* If you land here, make sure whether increasing ip_tunnel_info's 1070 * options_len is a reasonable choice with its usage in front ends 1071 * (f.e., it's part of flow keys, etc). 1072 */ 1073 BUILD_BUG_ON(IP_TUNNEL_OPTS_MAX != 255); 1074 1075 lwtunnel_encap_add_ops(&ip_tun_lwt_ops, LWTUNNEL_ENCAP_IP); 1076 lwtunnel_encap_add_ops(&ip6_tun_lwt_ops, LWTUNNEL_ENCAP_IP6); 1077 } 1078 1079 DEFINE_STATIC_KEY_FALSE(ip_tunnel_metadata_cnt); 1080 EXPORT_SYMBOL(ip_tunnel_metadata_cnt); 1081 1082 void ip_tunnel_need_metadata(void) 1083 { 1084 static_branch_inc(&ip_tunnel_metadata_cnt); 1085 } 1086 EXPORT_SYMBOL_GPL(ip_tunnel_need_metadata); 1087 1088 void ip_tunnel_unneed_metadata(void) 1089 { 1090 static_branch_dec(&ip_tunnel_metadata_cnt); 1091 } 1092 EXPORT_SYMBOL_GPL(ip_tunnel_unneed_metadata); 1093 1094 /* Returns either the correct skb->protocol value, or 0 if invalid. */ 1095 __be16 ip_tunnel_parse_protocol(const struct sk_buff *skb) 1096 { 1097 if (skb_network_header(skb) >= skb->head && 1098 (skb_network_header(skb) + sizeof(struct iphdr)) <= skb_tail_pointer(skb) && 1099 ip_hdr(skb)->version == 4) 1100 return htons(ETH_P_IP); 1101 if (skb_network_header(skb) >= skb->head && 1102 (skb_network_header(skb) + sizeof(struct ipv6hdr)) <= skb_tail_pointer(skb) && 1103 ipv6_hdr(skb)->version == 6) 1104 return htons(ETH_P_IPV6); 1105 return 0; 1106 } 1107 EXPORT_SYMBOL(ip_tunnel_parse_protocol); 1108 1109 const struct header_ops ip_tunnel_header_ops = { .parse_protocol = ip_tunnel_parse_protocol }; 1110 EXPORT_SYMBOL(ip_tunnel_header_ops); 1111