1 /* 2 * GENEVE: Generic Network Virtualization Encapsulation 3 * 4 * Copyright (c) 2015 Red Hat, Inc. 5 * 6 * This program is free software; you can redistribute it and/or modify 7 * it under the terms of the GNU General Public License version 2 as 8 * published by the Free Software Foundation. 9 */ 10 11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 12 13 #include <linux/kernel.h> 14 #include <linux/module.h> 15 #include <linux/etherdevice.h> 16 #include <linux/hash.h> 17 #include <net/dst_metadata.h> 18 #include <net/gro_cells.h> 19 #include <net/rtnetlink.h> 20 #include <net/geneve.h> 21 #include <net/protocol.h> 22 23 #define GENEVE_NETDEV_VER "0.6" 24 25 #define GENEVE_UDP_PORT 6081 26 27 #define GENEVE_N_VID (1u << 24) 28 #define GENEVE_VID_MASK (GENEVE_N_VID - 1) 29 30 #define VNI_HASH_BITS 10 31 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS) 32 33 static bool log_ecn_error = true; 34 module_param(log_ecn_error, bool, 0644); 35 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN"); 36 37 #define GENEVE_VER 0 38 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr)) 39 40 /* per-network namespace private data for this module */ 41 struct geneve_net { 42 struct list_head geneve_list; 43 struct list_head sock_list; 44 }; 45 46 static int geneve_net_id; 47 48 union geneve_addr { 49 struct sockaddr_in sin; 50 struct sockaddr_in6 sin6; 51 struct sockaddr sa; 52 }; 53 54 static union geneve_addr geneve_remote_unspec = { .sa.sa_family = AF_UNSPEC, }; 55 56 /* Pseudo network device */ 57 struct geneve_dev { 58 struct hlist_node hlist; /* vni hash table */ 59 struct net *net; /* netns for packet i/o */ 60 struct net_device *dev; /* netdev for geneve tunnel */ 61 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */ 62 #if IS_ENABLED(CONFIG_IPV6) 63 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */ 64 #endif 65 u8 vni[3]; /* virtual network ID for tunnel */ 66 u8 ttl; /* TTL override */ 67 u8 tos; /* TOS override */ 68 union geneve_addr remote; /* IP address for link partner */ 69 struct list_head next; /* geneve's per namespace list */ 70 __be32 label; /* IPv6 flowlabel override */ 71 __be16 dst_port; 72 bool collect_md; 73 struct gro_cells gro_cells; 74 u32 flags; 75 struct dst_cache dst_cache; 76 }; 77 78 /* Geneve device flags */ 79 #define GENEVE_F_UDP_ZERO_CSUM_TX BIT(0) 80 #define GENEVE_F_UDP_ZERO_CSUM6_TX BIT(1) 81 #define GENEVE_F_UDP_ZERO_CSUM6_RX BIT(2) 82 83 struct geneve_sock { 84 bool collect_md; 85 struct list_head list; 86 struct socket *sock; 87 struct rcu_head rcu; 88 int refcnt; 89 struct hlist_head vni_list[VNI_HASH_SIZE]; 90 u32 flags; 91 }; 92 93 static inline __u32 geneve_net_vni_hash(u8 vni[3]) 94 { 95 __u32 vnid; 96 97 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2]; 98 return hash_32(vnid, VNI_HASH_BITS); 99 } 100 101 static __be64 vni_to_tunnel_id(const __u8 *vni) 102 { 103 #ifdef __BIG_ENDIAN 104 return (vni[0] << 16) | (vni[1] << 8) | vni[2]; 105 #else 106 return (__force __be64)(((__force u64)vni[0] << 40) | 107 ((__force u64)vni[1] << 48) | 108 ((__force u64)vni[2] << 56)); 109 #endif 110 } 111 112 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs) 113 { 114 return gs->sock->sk->sk_family; 115 } 116 117 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs, 118 __be32 addr, u8 vni[]) 119 { 120 struct hlist_head *vni_list_head; 121 struct geneve_dev *geneve; 122 __u32 hash; 123 124 /* Find the device for this VNI */ 125 hash = geneve_net_vni_hash(vni); 126 vni_list_head = &gs->vni_list[hash]; 127 hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) { 128 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) && 129 addr == geneve->remote.sin.sin_addr.s_addr) 130 return geneve; 131 } 132 return NULL; 133 } 134 135 #if IS_ENABLED(CONFIG_IPV6) 136 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs, 137 struct in6_addr addr6, u8 vni[]) 138 { 139 struct hlist_head *vni_list_head; 140 struct geneve_dev *geneve; 141 __u32 hash; 142 143 /* Find the device for this VNI */ 144 hash = geneve_net_vni_hash(vni); 145 vni_list_head = &gs->vni_list[hash]; 146 hlist_for_each_entry_rcu(geneve, vni_list_head, hlist) { 147 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) && 148 ipv6_addr_equal(&addr6, &geneve->remote.sin6.sin6_addr)) 149 return geneve; 150 } 151 return NULL; 152 } 153 #endif 154 155 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb) 156 { 157 return (struct genevehdr *)(udp_hdr(skb) + 1); 158 } 159 160 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs, 161 struct sk_buff *skb) 162 { 163 u8 *vni; 164 __be32 addr; 165 static u8 zero_vni[3]; 166 #if IS_ENABLED(CONFIG_IPV6) 167 static struct in6_addr zero_addr6; 168 #endif 169 170 if (geneve_get_sk_family(gs) == AF_INET) { 171 struct iphdr *iph; 172 173 iph = ip_hdr(skb); /* outer IP header... */ 174 175 if (gs->collect_md) { 176 vni = zero_vni; 177 addr = 0; 178 } else { 179 vni = geneve_hdr(skb)->vni; 180 addr = iph->saddr; 181 } 182 183 return geneve_lookup(gs, addr, vni); 184 #if IS_ENABLED(CONFIG_IPV6) 185 } else if (geneve_get_sk_family(gs) == AF_INET6) { 186 struct ipv6hdr *ip6h; 187 struct in6_addr addr6; 188 189 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */ 190 191 if (gs->collect_md) { 192 vni = zero_vni; 193 addr6 = zero_addr6; 194 } else { 195 vni = geneve_hdr(skb)->vni; 196 addr6 = ip6h->saddr; 197 } 198 199 return geneve6_lookup(gs, addr6, vni); 200 #endif 201 } 202 return NULL; 203 } 204 205 /* geneve receive/decap routine */ 206 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs, 207 struct sk_buff *skb) 208 { 209 struct genevehdr *gnvh = geneve_hdr(skb); 210 struct metadata_dst *tun_dst = NULL; 211 struct pcpu_sw_netstats *stats; 212 int err = 0; 213 void *oiph; 214 215 if (ip_tunnel_collect_metadata() || gs->collect_md) { 216 __be16 flags; 217 218 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT | 219 (gnvh->oam ? TUNNEL_OAM : 0) | 220 (gnvh->critical ? TUNNEL_CRIT_OPT : 0); 221 222 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags, 223 vni_to_tunnel_id(gnvh->vni), 224 gnvh->opt_len * 4); 225 if (!tun_dst) 226 goto drop; 227 /* Update tunnel dst according to Geneve options. */ 228 ip_tunnel_info_opts_set(&tun_dst->u.tun_info, 229 gnvh->options, gnvh->opt_len * 4); 230 } else { 231 /* Drop packets w/ critical options, 232 * since we don't support any... 233 */ 234 if (gnvh->critical) 235 goto drop; 236 } 237 238 skb_reset_mac_header(skb); 239 skb->protocol = eth_type_trans(skb, geneve->dev); 240 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 241 242 if (tun_dst) 243 skb_dst_set(skb, &tun_dst->dst); 244 245 /* Ignore packet loops (and multicast echo) */ 246 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) 247 goto drop; 248 249 oiph = skb_network_header(skb); 250 skb_reset_network_header(skb); 251 252 if (geneve_get_sk_family(gs) == AF_INET) 253 err = IP_ECN_decapsulate(oiph, skb); 254 #if IS_ENABLED(CONFIG_IPV6) 255 else 256 err = IP6_ECN_decapsulate(oiph, skb); 257 #endif 258 259 if (unlikely(err)) { 260 if (log_ecn_error) { 261 if (geneve_get_sk_family(gs) == AF_INET) 262 net_info_ratelimited("non-ECT from %pI4 " 263 "with TOS=%#x\n", 264 &((struct iphdr *)oiph)->saddr, 265 ((struct iphdr *)oiph)->tos); 266 #if IS_ENABLED(CONFIG_IPV6) 267 else 268 net_info_ratelimited("non-ECT from %pI6\n", 269 &((struct ipv6hdr *)oiph)->saddr); 270 #endif 271 } 272 if (err > 1) { 273 ++geneve->dev->stats.rx_frame_errors; 274 ++geneve->dev->stats.rx_errors; 275 goto drop; 276 } 277 } 278 279 stats = this_cpu_ptr(geneve->dev->tstats); 280 u64_stats_update_begin(&stats->syncp); 281 stats->rx_packets++; 282 stats->rx_bytes += skb->len; 283 u64_stats_update_end(&stats->syncp); 284 285 gro_cells_receive(&geneve->gro_cells, skb); 286 return; 287 drop: 288 /* Consume bad packet */ 289 kfree_skb(skb); 290 } 291 292 /* Setup stats when device is created */ 293 static int geneve_init(struct net_device *dev) 294 { 295 struct geneve_dev *geneve = netdev_priv(dev); 296 int err; 297 298 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 299 if (!dev->tstats) 300 return -ENOMEM; 301 302 err = gro_cells_init(&geneve->gro_cells, dev); 303 if (err) { 304 free_percpu(dev->tstats); 305 return err; 306 } 307 308 err = dst_cache_init(&geneve->dst_cache, GFP_KERNEL); 309 if (err) { 310 free_percpu(dev->tstats); 311 gro_cells_destroy(&geneve->gro_cells); 312 return err; 313 } 314 315 return 0; 316 } 317 318 static void geneve_uninit(struct net_device *dev) 319 { 320 struct geneve_dev *geneve = netdev_priv(dev); 321 322 dst_cache_destroy(&geneve->dst_cache); 323 gro_cells_destroy(&geneve->gro_cells); 324 free_percpu(dev->tstats); 325 } 326 327 /* Callback from net/ipv4/udp.c to receive packets */ 328 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb) 329 { 330 struct genevehdr *geneveh; 331 struct geneve_dev *geneve; 332 struct geneve_sock *gs; 333 int opts_len; 334 335 /* Need Geneve and inner Ethernet header to be present */ 336 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN))) 337 goto drop; 338 339 /* Return packets with reserved bits set */ 340 geneveh = geneve_hdr(skb); 341 if (unlikely(geneveh->ver != GENEVE_VER)) 342 goto drop; 343 344 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB))) 345 goto drop; 346 347 gs = rcu_dereference_sk_user_data(sk); 348 if (!gs) 349 goto drop; 350 351 geneve = geneve_lookup_skb(gs, skb); 352 if (!geneve) 353 goto drop; 354 355 opts_len = geneveh->opt_len * 4; 356 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, 357 htons(ETH_P_TEB), 358 !net_eq(geneve->net, dev_net(geneve->dev)))) 359 goto drop; 360 361 geneve_rx(geneve, gs, skb); 362 return 0; 363 364 drop: 365 /* Consume bad packet */ 366 kfree_skb(skb); 367 return 0; 368 } 369 370 static struct socket *geneve_create_sock(struct net *net, bool ipv6, 371 __be16 port, u32 flags) 372 { 373 struct socket *sock; 374 struct udp_port_cfg udp_conf; 375 int err; 376 377 memset(&udp_conf, 0, sizeof(udp_conf)); 378 379 if (ipv6) { 380 udp_conf.family = AF_INET6; 381 udp_conf.ipv6_v6only = 1; 382 udp_conf.use_udp6_rx_checksums = 383 !(flags & GENEVE_F_UDP_ZERO_CSUM6_RX); 384 } else { 385 udp_conf.family = AF_INET; 386 udp_conf.local_ip.s_addr = htonl(INADDR_ANY); 387 } 388 389 udp_conf.local_udp_port = port; 390 391 /* Open UDP socket */ 392 err = udp_sock_create(net, &udp_conf, &sock); 393 if (err < 0) 394 return ERR_PTR(err); 395 396 return sock; 397 } 398 399 static int geneve_hlen(struct genevehdr *gh) 400 { 401 return sizeof(*gh) + gh->opt_len * 4; 402 } 403 404 static struct sk_buff **geneve_gro_receive(struct sock *sk, 405 struct sk_buff **head, 406 struct sk_buff *skb) 407 { 408 struct sk_buff *p, **pp = NULL; 409 struct genevehdr *gh, *gh2; 410 unsigned int hlen, gh_len, off_gnv; 411 const struct packet_offload *ptype; 412 __be16 type; 413 int flush = 1; 414 415 off_gnv = skb_gro_offset(skb); 416 hlen = off_gnv + sizeof(*gh); 417 gh = skb_gro_header_fast(skb, off_gnv); 418 if (skb_gro_header_hard(skb, hlen)) { 419 gh = skb_gro_header_slow(skb, hlen, off_gnv); 420 if (unlikely(!gh)) 421 goto out; 422 } 423 424 if (gh->ver != GENEVE_VER || gh->oam) 425 goto out; 426 gh_len = geneve_hlen(gh); 427 428 hlen = off_gnv + gh_len; 429 if (skb_gro_header_hard(skb, hlen)) { 430 gh = skb_gro_header_slow(skb, hlen, off_gnv); 431 if (unlikely(!gh)) 432 goto out; 433 } 434 435 for (p = *head; p; p = p->next) { 436 if (!NAPI_GRO_CB(p)->same_flow) 437 continue; 438 439 gh2 = (struct genevehdr *)(p->data + off_gnv); 440 if (gh->opt_len != gh2->opt_len || 441 memcmp(gh, gh2, gh_len)) { 442 NAPI_GRO_CB(p)->same_flow = 0; 443 continue; 444 } 445 } 446 447 type = gh->proto_type; 448 449 rcu_read_lock(); 450 ptype = gro_find_receive_by_type(type); 451 if (!ptype) 452 goto out_unlock; 453 454 skb_gro_pull(skb, gh_len); 455 skb_gro_postpull_rcsum(skb, gh, gh_len); 456 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb); 457 flush = 0; 458 459 out_unlock: 460 rcu_read_unlock(); 461 out: 462 NAPI_GRO_CB(skb)->flush |= flush; 463 464 return pp; 465 } 466 467 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb, 468 int nhoff) 469 { 470 struct genevehdr *gh; 471 struct packet_offload *ptype; 472 __be16 type; 473 int gh_len; 474 int err = -ENOSYS; 475 476 gh = (struct genevehdr *)(skb->data + nhoff); 477 gh_len = geneve_hlen(gh); 478 type = gh->proto_type; 479 480 rcu_read_lock(); 481 ptype = gro_find_complete_by_type(type); 482 if (ptype) 483 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len); 484 485 rcu_read_unlock(); 486 487 skb_set_inner_mac_header(skb, nhoff + gh_len); 488 489 return err; 490 } 491 492 /* Create new listen socket if needed */ 493 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port, 494 bool ipv6, u32 flags) 495 { 496 struct geneve_net *gn = net_generic(net, geneve_net_id); 497 struct geneve_sock *gs; 498 struct socket *sock; 499 struct udp_tunnel_sock_cfg tunnel_cfg; 500 int h; 501 502 gs = kzalloc(sizeof(*gs), GFP_KERNEL); 503 if (!gs) 504 return ERR_PTR(-ENOMEM); 505 506 sock = geneve_create_sock(net, ipv6, port, flags); 507 if (IS_ERR(sock)) { 508 kfree(gs); 509 return ERR_CAST(sock); 510 } 511 512 gs->sock = sock; 513 gs->refcnt = 1; 514 for (h = 0; h < VNI_HASH_SIZE; ++h) 515 INIT_HLIST_HEAD(&gs->vni_list[h]); 516 517 /* Initialize the geneve udp offloads structure */ 518 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 519 520 /* Mark socket as an encapsulation socket */ 521 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg)); 522 tunnel_cfg.sk_user_data = gs; 523 tunnel_cfg.encap_type = 1; 524 tunnel_cfg.gro_receive = geneve_gro_receive; 525 tunnel_cfg.gro_complete = geneve_gro_complete; 526 tunnel_cfg.encap_rcv = geneve_udp_encap_recv; 527 tunnel_cfg.encap_destroy = NULL; 528 setup_udp_tunnel_sock(net, sock, &tunnel_cfg); 529 list_add(&gs->list, &gn->sock_list); 530 return gs; 531 } 532 533 static void __geneve_sock_release(struct geneve_sock *gs) 534 { 535 if (!gs || --gs->refcnt) 536 return; 537 538 list_del(&gs->list); 539 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 540 udp_tunnel_sock_release(gs->sock); 541 kfree_rcu(gs, rcu); 542 } 543 544 static void geneve_sock_release(struct geneve_dev *geneve) 545 { 546 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4); 547 #if IS_ENABLED(CONFIG_IPV6) 548 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6); 549 550 rcu_assign_pointer(geneve->sock6, NULL); 551 #endif 552 553 rcu_assign_pointer(geneve->sock4, NULL); 554 synchronize_net(); 555 556 __geneve_sock_release(gs4); 557 #if IS_ENABLED(CONFIG_IPV6) 558 __geneve_sock_release(gs6); 559 #endif 560 } 561 562 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn, 563 sa_family_t family, 564 __be16 dst_port) 565 { 566 struct geneve_sock *gs; 567 568 list_for_each_entry(gs, &gn->sock_list, list) { 569 if (inet_sk(gs->sock->sk)->inet_sport == dst_port && 570 geneve_get_sk_family(gs) == family) { 571 return gs; 572 } 573 } 574 return NULL; 575 } 576 577 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6) 578 { 579 struct net *net = geneve->net; 580 struct geneve_net *gn = net_generic(net, geneve_net_id); 581 struct geneve_sock *gs; 582 __u32 hash; 583 584 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->dst_port); 585 if (gs) { 586 gs->refcnt++; 587 goto out; 588 } 589 590 gs = geneve_socket_create(net, geneve->dst_port, ipv6, geneve->flags); 591 if (IS_ERR(gs)) 592 return PTR_ERR(gs); 593 594 out: 595 gs->collect_md = geneve->collect_md; 596 gs->flags = geneve->flags; 597 #if IS_ENABLED(CONFIG_IPV6) 598 if (ipv6) 599 rcu_assign_pointer(geneve->sock6, gs); 600 else 601 #endif 602 rcu_assign_pointer(geneve->sock4, gs); 603 604 hash = geneve_net_vni_hash(geneve->vni); 605 hlist_add_head_rcu(&geneve->hlist, &gs->vni_list[hash]); 606 return 0; 607 } 608 609 static int geneve_open(struct net_device *dev) 610 { 611 struct geneve_dev *geneve = netdev_priv(dev); 612 bool ipv6 = geneve->remote.sa.sa_family == AF_INET6; 613 bool metadata = geneve->collect_md; 614 int ret = 0; 615 616 #if IS_ENABLED(CONFIG_IPV6) 617 if (ipv6 || metadata) 618 ret = geneve_sock_add(geneve, true); 619 #endif 620 if (!ret && (!ipv6 || metadata)) 621 ret = geneve_sock_add(geneve, false); 622 if (ret < 0) 623 geneve_sock_release(geneve); 624 625 return ret; 626 } 627 628 static int geneve_stop(struct net_device *dev) 629 { 630 struct geneve_dev *geneve = netdev_priv(dev); 631 632 if (!hlist_unhashed(&geneve->hlist)) 633 hlist_del_rcu(&geneve->hlist); 634 geneve_sock_release(geneve); 635 return 0; 636 } 637 638 static void geneve_build_header(struct genevehdr *geneveh, 639 __be16 tun_flags, u8 vni[3], 640 u8 options_len, u8 *options) 641 { 642 geneveh->ver = GENEVE_VER; 643 geneveh->opt_len = options_len / 4; 644 geneveh->oam = !!(tun_flags & TUNNEL_OAM); 645 geneveh->critical = !!(tun_flags & TUNNEL_CRIT_OPT); 646 geneveh->rsvd1 = 0; 647 memcpy(geneveh->vni, vni, 3); 648 geneveh->proto_type = htons(ETH_P_TEB); 649 geneveh->rsvd2 = 0; 650 651 memcpy(geneveh->options, options, options_len); 652 } 653 654 static int geneve_build_skb(struct rtable *rt, struct sk_buff *skb, 655 __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt, 656 u32 flags, bool xnet) 657 { 658 struct genevehdr *gnvh; 659 int min_headroom; 660 int err; 661 bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM_TX); 662 663 skb_scrub_packet(skb, xnet); 664 665 min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len 666 + GENEVE_BASE_HLEN + opt_len + sizeof(struct iphdr); 667 err = skb_cow_head(skb, min_headroom); 668 if (unlikely(err)) 669 goto free_rt; 670 671 err = udp_tunnel_handle_offloads(skb, udp_sum); 672 if (err) 673 goto free_rt; 674 675 gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len); 676 geneve_build_header(gnvh, tun_flags, vni, opt_len, opt); 677 678 skb_set_inner_protocol(skb, htons(ETH_P_TEB)); 679 return 0; 680 681 free_rt: 682 ip_rt_put(rt); 683 return err; 684 } 685 686 #if IS_ENABLED(CONFIG_IPV6) 687 static int geneve6_build_skb(struct dst_entry *dst, struct sk_buff *skb, 688 __be16 tun_flags, u8 vni[3], u8 opt_len, u8 *opt, 689 u32 flags, bool xnet) 690 { 691 struct genevehdr *gnvh; 692 int min_headroom; 693 int err; 694 bool udp_sum = !(flags & GENEVE_F_UDP_ZERO_CSUM6_TX); 695 696 skb_scrub_packet(skb, xnet); 697 698 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len 699 + GENEVE_BASE_HLEN + opt_len + sizeof(struct ipv6hdr); 700 err = skb_cow_head(skb, min_headroom); 701 if (unlikely(err)) 702 goto free_dst; 703 704 err = udp_tunnel_handle_offloads(skb, udp_sum); 705 if (err) 706 goto free_dst; 707 708 gnvh = (struct genevehdr *)__skb_push(skb, sizeof(*gnvh) + opt_len); 709 geneve_build_header(gnvh, tun_flags, vni, opt_len, opt); 710 711 skb_set_inner_protocol(skb, htons(ETH_P_TEB)); 712 return 0; 713 714 free_dst: 715 dst_release(dst); 716 return err; 717 } 718 #endif 719 720 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb, 721 struct net_device *dev, 722 struct flowi4 *fl4, 723 struct ip_tunnel_info *info) 724 { 725 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 726 struct geneve_dev *geneve = netdev_priv(dev); 727 struct dst_cache *dst_cache; 728 struct rtable *rt = NULL; 729 __u8 tos; 730 731 if (!rcu_dereference(geneve->sock4)) 732 return ERR_PTR(-EIO); 733 734 memset(fl4, 0, sizeof(*fl4)); 735 fl4->flowi4_mark = skb->mark; 736 fl4->flowi4_proto = IPPROTO_UDP; 737 738 if (info) { 739 fl4->daddr = info->key.u.ipv4.dst; 740 fl4->saddr = info->key.u.ipv4.src; 741 fl4->flowi4_tos = RT_TOS(info->key.tos); 742 dst_cache = &info->dst_cache; 743 } else { 744 tos = geneve->tos; 745 if (tos == 1) { 746 const struct iphdr *iip = ip_hdr(skb); 747 748 tos = ip_tunnel_get_dsfield(iip, skb); 749 use_cache = false; 750 } 751 752 fl4->flowi4_tos = RT_TOS(tos); 753 fl4->daddr = geneve->remote.sin.sin_addr.s_addr; 754 dst_cache = &geneve->dst_cache; 755 } 756 757 if (use_cache) { 758 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr); 759 if (rt) 760 return rt; 761 } 762 763 rt = ip_route_output_key(geneve->net, fl4); 764 if (IS_ERR(rt)) { 765 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr); 766 return ERR_PTR(-ENETUNREACH); 767 } 768 if (rt->dst.dev == dev) { /* is this necessary? */ 769 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr); 770 ip_rt_put(rt); 771 return ERR_PTR(-ELOOP); 772 } 773 if (use_cache) 774 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr); 775 return rt; 776 } 777 778 #if IS_ENABLED(CONFIG_IPV6) 779 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb, 780 struct net_device *dev, 781 struct flowi6 *fl6, 782 struct ip_tunnel_info *info) 783 { 784 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 785 struct geneve_dev *geneve = netdev_priv(dev); 786 struct dst_entry *dst = NULL; 787 struct dst_cache *dst_cache; 788 struct geneve_sock *gs6; 789 __u8 prio; 790 791 gs6 = rcu_dereference(geneve->sock6); 792 if (!gs6) 793 return ERR_PTR(-EIO); 794 795 memset(fl6, 0, sizeof(*fl6)); 796 fl6->flowi6_mark = skb->mark; 797 fl6->flowi6_proto = IPPROTO_UDP; 798 799 if (info) { 800 fl6->daddr = info->key.u.ipv6.dst; 801 fl6->saddr = info->key.u.ipv6.src; 802 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(info->key.tos), 803 info->key.label); 804 dst_cache = &info->dst_cache; 805 } else { 806 prio = geneve->tos; 807 if (prio == 1) { 808 const struct iphdr *iip = ip_hdr(skb); 809 810 prio = ip_tunnel_get_dsfield(iip, skb); 811 use_cache = false; 812 } 813 814 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio), 815 geneve->label); 816 fl6->daddr = geneve->remote.sin6.sin6_addr; 817 dst_cache = &geneve->dst_cache; 818 } 819 820 if (use_cache) { 821 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr); 822 if (dst) 823 return dst; 824 } 825 826 if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) { 827 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr); 828 return ERR_PTR(-ENETUNREACH); 829 } 830 if (dst->dev == dev) { /* is this necessary? */ 831 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr); 832 dst_release(dst); 833 return ERR_PTR(-ELOOP); 834 } 835 836 if (use_cache) 837 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr); 838 return dst; 839 } 840 #endif 841 842 /* Convert 64 bit tunnel ID to 24 bit VNI. */ 843 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni) 844 { 845 #ifdef __BIG_ENDIAN 846 vni[0] = (__force __u8)(tun_id >> 16); 847 vni[1] = (__force __u8)(tun_id >> 8); 848 vni[2] = (__force __u8)tun_id; 849 #else 850 vni[0] = (__force __u8)((__force u64)tun_id >> 40); 851 vni[1] = (__force __u8)((__force u64)tun_id >> 48); 852 vni[2] = (__force __u8)((__force u64)tun_id >> 56); 853 #endif 854 } 855 856 static netdev_tx_t geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev, 857 struct ip_tunnel_info *info) 858 { 859 struct geneve_dev *geneve = netdev_priv(dev); 860 struct geneve_sock *gs4; 861 struct rtable *rt = NULL; 862 int err = -EINVAL; 863 struct flowi4 fl4; 864 __u8 tos, ttl; 865 __be16 sport; 866 __be16 df; 867 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 868 u32 flags = geneve->flags; 869 870 gs4 = rcu_dereference(geneve->sock4); 871 if (!gs4) 872 goto tx_error; 873 874 if (geneve->collect_md) { 875 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) { 876 netdev_dbg(dev, "no tunnel metadata\n"); 877 goto tx_error; 878 } 879 if (info && ip_tunnel_info_af(info) != AF_INET) 880 goto tx_error; 881 } 882 883 rt = geneve_get_v4_rt(skb, dev, &fl4, info); 884 if (IS_ERR(rt)) { 885 err = PTR_ERR(rt); 886 goto tx_error; 887 } 888 889 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 890 skb_reset_mac_header(skb); 891 892 if (info) { 893 const struct ip_tunnel_key *key = &info->key; 894 u8 *opts = NULL; 895 u8 vni[3]; 896 897 tunnel_id_to_vni(key->tun_id, vni); 898 if (info->options_len) 899 opts = ip_tunnel_info_opts(info); 900 901 if (key->tun_flags & TUNNEL_CSUM) 902 flags &= ~GENEVE_F_UDP_ZERO_CSUM_TX; 903 else 904 flags |= GENEVE_F_UDP_ZERO_CSUM_TX; 905 906 err = geneve_build_skb(rt, skb, key->tun_flags, vni, 907 info->options_len, opts, flags, xnet); 908 if (unlikely(err)) 909 goto tx_error; 910 911 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 912 ttl = key->ttl; 913 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0; 914 } else { 915 err = geneve_build_skb(rt, skb, 0, geneve->vni, 916 0, NULL, flags, xnet); 917 if (unlikely(err)) 918 goto tx_error; 919 920 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb); 921 ttl = geneve->ttl; 922 if (!ttl && IN_MULTICAST(ntohl(fl4.daddr))) 923 ttl = 1; 924 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst); 925 df = 0; 926 } 927 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr, 928 tos, ttl, df, sport, geneve->dst_port, 929 !net_eq(geneve->net, dev_net(geneve->dev)), 930 !!(flags & GENEVE_F_UDP_ZERO_CSUM_TX)); 931 932 return NETDEV_TX_OK; 933 934 tx_error: 935 dev_kfree_skb(skb); 936 937 if (err == -ELOOP) 938 dev->stats.collisions++; 939 else if (err == -ENETUNREACH) 940 dev->stats.tx_carrier_errors++; 941 942 dev->stats.tx_errors++; 943 return NETDEV_TX_OK; 944 } 945 946 #if IS_ENABLED(CONFIG_IPV6) 947 static netdev_tx_t geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev, 948 struct ip_tunnel_info *info) 949 { 950 struct geneve_dev *geneve = netdev_priv(dev); 951 struct dst_entry *dst = NULL; 952 struct geneve_sock *gs6; 953 int err = -EINVAL; 954 struct flowi6 fl6; 955 __u8 prio, ttl; 956 __be16 sport; 957 __be32 label; 958 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 959 u32 flags = geneve->flags; 960 961 gs6 = rcu_dereference(geneve->sock6); 962 if (!gs6) 963 goto tx_error; 964 965 if (geneve->collect_md) { 966 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) { 967 netdev_dbg(dev, "no tunnel metadata\n"); 968 goto tx_error; 969 } 970 } 971 972 dst = geneve_get_v6_dst(skb, dev, &fl6, info); 973 if (IS_ERR(dst)) { 974 err = PTR_ERR(dst); 975 goto tx_error; 976 } 977 978 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 979 skb_reset_mac_header(skb); 980 981 if (info) { 982 const struct ip_tunnel_key *key = &info->key; 983 u8 *opts = NULL; 984 u8 vni[3]; 985 986 tunnel_id_to_vni(key->tun_id, vni); 987 if (info->options_len) 988 opts = ip_tunnel_info_opts(info); 989 990 if (key->tun_flags & TUNNEL_CSUM) 991 flags &= ~GENEVE_F_UDP_ZERO_CSUM6_TX; 992 else 993 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX; 994 995 err = geneve6_build_skb(dst, skb, key->tun_flags, vni, 996 info->options_len, opts, 997 flags, xnet); 998 if (unlikely(err)) 999 goto tx_error; 1000 1001 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 1002 ttl = key->ttl; 1003 label = info->key.label; 1004 } else { 1005 err = geneve6_build_skb(dst, skb, 0, geneve->vni, 1006 0, NULL, flags, xnet); 1007 if (unlikely(err)) 1008 goto tx_error; 1009 1010 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel), 1011 ip_hdr(skb), skb); 1012 ttl = geneve->ttl; 1013 if (!ttl && ipv6_addr_is_multicast(&fl6.daddr)) 1014 ttl = 1; 1015 ttl = ttl ? : ip6_dst_hoplimit(dst); 1016 label = geneve->label; 1017 } 1018 1019 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev, 1020 &fl6.saddr, &fl6.daddr, prio, ttl, label, 1021 sport, geneve->dst_port, 1022 !!(flags & GENEVE_F_UDP_ZERO_CSUM6_TX)); 1023 return NETDEV_TX_OK; 1024 1025 tx_error: 1026 dev_kfree_skb(skb); 1027 1028 if (err == -ELOOP) 1029 dev->stats.collisions++; 1030 else if (err == -ENETUNREACH) 1031 dev->stats.tx_carrier_errors++; 1032 1033 dev->stats.tx_errors++; 1034 return NETDEV_TX_OK; 1035 } 1036 #endif 1037 1038 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev) 1039 { 1040 struct geneve_dev *geneve = netdev_priv(dev); 1041 struct ip_tunnel_info *info = NULL; 1042 1043 if (geneve->collect_md) 1044 info = skb_tunnel_info(skb); 1045 1046 #if IS_ENABLED(CONFIG_IPV6) 1047 if ((info && ip_tunnel_info_af(info) == AF_INET6) || 1048 (!info && geneve->remote.sa.sa_family == AF_INET6)) 1049 return geneve6_xmit_skb(skb, dev, info); 1050 #endif 1051 return geneve_xmit_skb(skb, dev, info); 1052 } 1053 1054 static int __geneve_change_mtu(struct net_device *dev, int new_mtu, bool strict) 1055 { 1056 struct geneve_dev *geneve = netdev_priv(dev); 1057 /* The max_mtu calculation does not take account of GENEVE 1058 * options, to avoid excluding potentially valid 1059 * configurations. 1060 */ 1061 int max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len; 1062 1063 if (geneve->remote.sa.sa_family == AF_INET6) 1064 max_mtu -= sizeof(struct ipv6hdr); 1065 else 1066 max_mtu -= sizeof(struct iphdr); 1067 1068 if (new_mtu < 68) 1069 return -EINVAL; 1070 1071 if (new_mtu > max_mtu) { 1072 if (strict) 1073 return -EINVAL; 1074 1075 new_mtu = max_mtu; 1076 } 1077 1078 dev->mtu = new_mtu; 1079 return 0; 1080 } 1081 1082 static int geneve_change_mtu(struct net_device *dev, int new_mtu) 1083 { 1084 return __geneve_change_mtu(dev, new_mtu, true); 1085 } 1086 1087 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 1088 { 1089 struct ip_tunnel_info *info = skb_tunnel_info(skb); 1090 struct geneve_dev *geneve = netdev_priv(dev); 1091 struct rtable *rt; 1092 struct flowi4 fl4; 1093 #if IS_ENABLED(CONFIG_IPV6) 1094 struct dst_entry *dst; 1095 struct flowi6 fl6; 1096 #endif 1097 1098 if (ip_tunnel_info_af(info) == AF_INET) { 1099 rt = geneve_get_v4_rt(skb, dev, &fl4, info); 1100 if (IS_ERR(rt)) 1101 return PTR_ERR(rt); 1102 1103 ip_rt_put(rt); 1104 info->key.u.ipv4.src = fl4.saddr; 1105 #if IS_ENABLED(CONFIG_IPV6) 1106 } else if (ip_tunnel_info_af(info) == AF_INET6) { 1107 dst = geneve_get_v6_dst(skb, dev, &fl6, info); 1108 if (IS_ERR(dst)) 1109 return PTR_ERR(dst); 1110 1111 dst_release(dst); 1112 info->key.u.ipv6.src = fl6.saddr; 1113 #endif 1114 } else { 1115 return -EINVAL; 1116 } 1117 1118 info->key.tp_src = udp_flow_src_port(geneve->net, skb, 1119 1, USHRT_MAX, true); 1120 info->key.tp_dst = geneve->dst_port; 1121 return 0; 1122 } 1123 1124 static const struct net_device_ops geneve_netdev_ops = { 1125 .ndo_init = geneve_init, 1126 .ndo_uninit = geneve_uninit, 1127 .ndo_open = geneve_open, 1128 .ndo_stop = geneve_stop, 1129 .ndo_start_xmit = geneve_xmit, 1130 .ndo_get_stats64 = ip_tunnel_get_stats64, 1131 .ndo_change_mtu = geneve_change_mtu, 1132 .ndo_validate_addr = eth_validate_addr, 1133 .ndo_set_mac_address = eth_mac_addr, 1134 .ndo_fill_metadata_dst = geneve_fill_metadata_dst, 1135 }; 1136 1137 static void geneve_get_drvinfo(struct net_device *dev, 1138 struct ethtool_drvinfo *drvinfo) 1139 { 1140 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version)); 1141 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver)); 1142 } 1143 1144 static const struct ethtool_ops geneve_ethtool_ops = { 1145 .get_drvinfo = geneve_get_drvinfo, 1146 .get_link = ethtool_op_get_link, 1147 }; 1148 1149 /* Info for udev, that this is a virtual tunnel endpoint */ 1150 static struct device_type geneve_type = { 1151 .name = "geneve", 1152 }; 1153 1154 /* Calls the ndo_udp_tunnel_add of the caller in order to 1155 * supply the listening GENEVE udp ports. Callers are expected 1156 * to implement the ndo_udp_tunnel_add. 1157 */ 1158 static void geneve_push_rx_ports(struct net_device *dev) 1159 { 1160 struct net *net = dev_net(dev); 1161 struct geneve_net *gn = net_generic(net, geneve_net_id); 1162 struct geneve_sock *gs; 1163 1164 rcu_read_lock(); 1165 list_for_each_entry_rcu(gs, &gn->sock_list, list) 1166 udp_tunnel_push_rx_port(dev, gs->sock, 1167 UDP_TUNNEL_TYPE_GENEVE); 1168 rcu_read_unlock(); 1169 } 1170 1171 /* Initialize the device structure. */ 1172 static void geneve_setup(struct net_device *dev) 1173 { 1174 ether_setup(dev); 1175 1176 dev->netdev_ops = &geneve_netdev_ops; 1177 dev->ethtool_ops = &geneve_ethtool_ops; 1178 dev->destructor = free_netdev; 1179 1180 SET_NETDEV_DEVTYPE(dev, &geneve_type); 1181 1182 dev->features |= NETIF_F_LLTX; 1183 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM; 1184 dev->features |= NETIF_F_RXCSUM; 1185 dev->features |= NETIF_F_GSO_SOFTWARE; 1186 1187 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM; 1188 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 1189 1190 netif_keep_dst(dev); 1191 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1192 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE; 1193 eth_hw_addr_random(dev); 1194 } 1195 1196 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = { 1197 [IFLA_GENEVE_ID] = { .type = NLA_U32 }, 1198 [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) }, 1199 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) }, 1200 [IFLA_GENEVE_TTL] = { .type = NLA_U8 }, 1201 [IFLA_GENEVE_TOS] = { .type = NLA_U8 }, 1202 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 }, 1203 [IFLA_GENEVE_PORT] = { .type = NLA_U16 }, 1204 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1205 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 }, 1206 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 }, 1207 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 }, 1208 }; 1209 1210 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[]) 1211 { 1212 if (tb[IFLA_ADDRESS]) { 1213 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1214 return -EINVAL; 1215 1216 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1217 return -EADDRNOTAVAIL; 1218 } 1219 1220 if (!data) 1221 return -EINVAL; 1222 1223 if (data[IFLA_GENEVE_ID]) { 1224 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1225 1226 if (vni >= GENEVE_VID_MASK) 1227 return -ERANGE; 1228 } 1229 1230 return 0; 1231 } 1232 1233 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn, 1234 __be16 dst_port, 1235 union geneve_addr *remote, 1236 u8 vni[], 1237 bool *tun_on_same_port, 1238 bool *tun_collect_md) 1239 { 1240 struct geneve_dev *geneve, *t; 1241 1242 *tun_on_same_port = false; 1243 *tun_collect_md = false; 1244 t = NULL; 1245 list_for_each_entry(geneve, &gn->geneve_list, next) { 1246 if (geneve->dst_port == dst_port) { 1247 *tun_collect_md = geneve->collect_md; 1248 *tun_on_same_port = true; 1249 } 1250 if (!memcmp(vni, geneve->vni, sizeof(geneve->vni)) && 1251 !memcmp(remote, &geneve->remote, sizeof(geneve->remote)) && 1252 dst_port == geneve->dst_port) 1253 t = geneve; 1254 } 1255 return t; 1256 } 1257 1258 static int geneve_configure(struct net *net, struct net_device *dev, 1259 union geneve_addr *remote, 1260 __u32 vni, __u8 ttl, __u8 tos, __be32 label, 1261 __be16 dst_port, bool metadata, u32 flags) 1262 { 1263 struct geneve_net *gn = net_generic(net, geneve_net_id); 1264 struct geneve_dev *t, *geneve = netdev_priv(dev); 1265 bool tun_collect_md, tun_on_same_port; 1266 int err, encap_len; 1267 1268 if (!remote) 1269 return -EINVAL; 1270 if (metadata && 1271 (remote->sa.sa_family != AF_UNSPEC || vni || tos || ttl || label)) 1272 return -EINVAL; 1273 1274 geneve->net = net; 1275 geneve->dev = dev; 1276 1277 geneve->vni[0] = (vni & 0x00ff0000) >> 16; 1278 geneve->vni[1] = (vni & 0x0000ff00) >> 8; 1279 geneve->vni[2] = vni & 0x000000ff; 1280 1281 if ((remote->sa.sa_family == AF_INET && 1282 IN_MULTICAST(ntohl(remote->sin.sin_addr.s_addr))) || 1283 (remote->sa.sa_family == AF_INET6 && 1284 ipv6_addr_is_multicast(&remote->sin6.sin6_addr))) 1285 return -EINVAL; 1286 if (label && remote->sa.sa_family != AF_INET6) 1287 return -EINVAL; 1288 1289 geneve->remote = *remote; 1290 1291 geneve->ttl = ttl; 1292 geneve->tos = tos; 1293 geneve->label = label; 1294 geneve->dst_port = dst_port; 1295 geneve->collect_md = metadata; 1296 geneve->flags = flags; 1297 1298 t = geneve_find_dev(gn, dst_port, remote, geneve->vni, 1299 &tun_on_same_port, &tun_collect_md); 1300 if (t) 1301 return -EBUSY; 1302 1303 /* make enough headroom for basic scenario */ 1304 encap_len = GENEVE_BASE_HLEN + ETH_HLEN; 1305 if (remote->sa.sa_family == AF_INET) 1306 encap_len += sizeof(struct iphdr); 1307 else 1308 encap_len += sizeof(struct ipv6hdr); 1309 dev->needed_headroom = encap_len + ETH_HLEN; 1310 1311 if (metadata) { 1312 if (tun_on_same_port) 1313 return -EPERM; 1314 } else { 1315 if (tun_collect_md) 1316 return -EPERM; 1317 } 1318 1319 dst_cache_reset(&geneve->dst_cache); 1320 1321 err = register_netdevice(dev); 1322 if (err) 1323 return err; 1324 1325 list_add(&geneve->next, &gn->geneve_list); 1326 return 0; 1327 } 1328 1329 static int geneve_newlink(struct net *net, struct net_device *dev, 1330 struct nlattr *tb[], struct nlattr *data[]) 1331 { 1332 __be16 dst_port = htons(GENEVE_UDP_PORT); 1333 __u8 ttl = 0, tos = 0; 1334 bool metadata = false; 1335 union geneve_addr remote = geneve_remote_unspec; 1336 __be32 label = 0; 1337 __u32 vni = 0; 1338 u32 flags = 0; 1339 1340 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) 1341 return -EINVAL; 1342 1343 if (data[IFLA_GENEVE_REMOTE]) { 1344 remote.sa.sa_family = AF_INET; 1345 remote.sin.sin_addr.s_addr = 1346 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]); 1347 } 1348 1349 if (data[IFLA_GENEVE_REMOTE6]) { 1350 if (!IS_ENABLED(CONFIG_IPV6)) 1351 return -EPFNOSUPPORT; 1352 1353 remote.sa.sa_family = AF_INET6; 1354 remote.sin6.sin6_addr = 1355 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]); 1356 1357 if (ipv6_addr_type(&remote.sin6.sin6_addr) & 1358 IPV6_ADDR_LINKLOCAL) { 1359 netdev_dbg(dev, "link-local remote is unsupported\n"); 1360 return -EINVAL; 1361 } 1362 } 1363 1364 if (data[IFLA_GENEVE_ID]) 1365 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1366 1367 if (data[IFLA_GENEVE_TTL]) 1368 ttl = nla_get_u8(data[IFLA_GENEVE_TTL]); 1369 1370 if (data[IFLA_GENEVE_TOS]) 1371 tos = nla_get_u8(data[IFLA_GENEVE_TOS]); 1372 1373 if (data[IFLA_GENEVE_LABEL]) 1374 label = nla_get_be32(data[IFLA_GENEVE_LABEL]) & 1375 IPV6_FLOWLABEL_MASK; 1376 1377 if (data[IFLA_GENEVE_PORT]) 1378 dst_port = nla_get_be16(data[IFLA_GENEVE_PORT]); 1379 1380 if (data[IFLA_GENEVE_COLLECT_METADATA]) 1381 metadata = true; 1382 1383 if (data[IFLA_GENEVE_UDP_CSUM] && 1384 !nla_get_u8(data[IFLA_GENEVE_UDP_CSUM])) 1385 flags |= GENEVE_F_UDP_ZERO_CSUM_TX; 1386 1387 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX] && 1388 nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX])) 1389 flags |= GENEVE_F_UDP_ZERO_CSUM6_TX; 1390 1391 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX] && 1392 nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX])) 1393 flags |= GENEVE_F_UDP_ZERO_CSUM6_RX; 1394 1395 return geneve_configure(net, dev, &remote, vni, ttl, tos, label, 1396 dst_port, metadata, flags); 1397 } 1398 1399 static void geneve_dellink(struct net_device *dev, struct list_head *head) 1400 { 1401 struct geneve_dev *geneve = netdev_priv(dev); 1402 1403 list_del(&geneve->next); 1404 unregister_netdevice_queue(dev, head); 1405 } 1406 1407 static size_t geneve_get_size(const struct net_device *dev) 1408 { 1409 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */ 1410 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */ 1411 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */ 1412 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */ 1413 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */ 1414 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */ 1415 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */ 1416 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */ 1417 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */ 1418 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */ 1419 0; 1420 } 1421 1422 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev) 1423 { 1424 struct geneve_dev *geneve = netdev_priv(dev); 1425 __u32 vni; 1426 1427 vni = (geneve->vni[0] << 16) | (geneve->vni[1] << 8) | geneve->vni[2]; 1428 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni)) 1429 goto nla_put_failure; 1430 1431 if (geneve->remote.sa.sa_family == AF_INET) { 1432 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE, 1433 geneve->remote.sin.sin_addr.s_addr)) 1434 goto nla_put_failure; 1435 #if IS_ENABLED(CONFIG_IPV6) 1436 } else { 1437 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6, 1438 &geneve->remote.sin6.sin6_addr)) 1439 goto nla_put_failure; 1440 #endif 1441 } 1442 1443 if (nla_put_u8(skb, IFLA_GENEVE_TTL, geneve->ttl) || 1444 nla_put_u8(skb, IFLA_GENEVE_TOS, geneve->tos) || 1445 nla_put_be32(skb, IFLA_GENEVE_LABEL, geneve->label)) 1446 goto nla_put_failure; 1447 1448 if (nla_put_be16(skb, IFLA_GENEVE_PORT, geneve->dst_port)) 1449 goto nla_put_failure; 1450 1451 if (geneve->collect_md) { 1452 if (nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA)) 1453 goto nla_put_failure; 1454 } 1455 1456 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM, 1457 !(geneve->flags & GENEVE_F_UDP_ZERO_CSUM_TX)) || 1458 nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX, 1459 !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_TX)) || 1460 nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX, 1461 !!(geneve->flags & GENEVE_F_UDP_ZERO_CSUM6_RX))) 1462 goto nla_put_failure; 1463 1464 return 0; 1465 1466 nla_put_failure: 1467 return -EMSGSIZE; 1468 } 1469 1470 static struct rtnl_link_ops geneve_link_ops __read_mostly = { 1471 .kind = "geneve", 1472 .maxtype = IFLA_GENEVE_MAX, 1473 .policy = geneve_policy, 1474 .priv_size = sizeof(struct geneve_dev), 1475 .setup = geneve_setup, 1476 .validate = geneve_validate, 1477 .newlink = geneve_newlink, 1478 .dellink = geneve_dellink, 1479 .get_size = geneve_get_size, 1480 .fill_info = geneve_fill_info, 1481 }; 1482 1483 struct net_device *geneve_dev_create_fb(struct net *net, const char *name, 1484 u8 name_assign_type, u16 dst_port) 1485 { 1486 struct nlattr *tb[IFLA_MAX + 1]; 1487 struct net_device *dev; 1488 LIST_HEAD(list_kill); 1489 int err; 1490 1491 memset(tb, 0, sizeof(tb)); 1492 dev = rtnl_create_link(net, name, name_assign_type, 1493 &geneve_link_ops, tb); 1494 if (IS_ERR(dev)) 1495 return dev; 1496 1497 err = geneve_configure(net, dev, &geneve_remote_unspec, 1498 0, 0, 0, 0, htons(dst_port), true, 1499 GENEVE_F_UDP_ZERO_CSUM6_RX); 1500 if (err) { 1501 free_netdev(dev); 1502 return ERR_PTR(err); 1503 } 1504 1505 /* openvswitch users expect packet sizes to be unrestricted, 1506 * so set the largest MTU we can. 1507 */ 1508 err = __geneve_change_mtu(dev, IP_MAX_MTU, false); 1509 if (err) 1510 goto err; 1511 1512 err = rtnl_configure_link(dev, NULL); 1513 if (err < 0) 1514 goto err; 1515 1516 return dev; 1517 1518 err: 1519 geneve_dellink(dev, &list_kill); 1520 unregister_netdevice_many(&list_kill); 1521 return ERR_PTR(err); 1522 } 1523 EXPORT_SYMBOL_GPL(geneve_dev_create_fb); 1524 1525 static int geneve_netdevice_event(struct notifier_block *unused, 1526 unsigned long event, void *ptr) 1527 { 1528 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1529 1530 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO) 1531 geneve_push_rx_ports(dev); 1532 1533 return NOTIFY_DONE; 1534 } 1535 1536 static struct notifier_block geneve_notifier_block __read_mostly = { 1537 .notifier_call = geneve_netdevice_event, 1538 }; 1539 1540 static __net_init int geneve_init_net(struct net *net) 1541 { 1542 struct geneve_net *gn = net_generic(net, geneve_net_id); 1543 1544 INIT_LIST_HEAD(&gn->geneve_list); 1545 INIT_LIST_HEAD(&gn->sock_list); 1546 return 0; 1547 } 1548 1549 static void __net_exit geneve_exit_net(struct net *net) 1550 { 1551 struct geneve_net *gn = net_generic(net, geneve_net_id); 1552 struct geneve_dev *geneve, *next; 1553 struct net_device *dev, *aux; 1554 LIST_HEAD(list); 1555 1556 rtnl_lock(); 1557 1558 /* gather any geneve devices that were moved into this ns */ 1559 for_each_netdev_safe(net, dev, aux) 1560 if (dev->rtnl_link_ops == &geneve_link_ops) 1561 unregister_netdevice_queue(dev, &list); 1562 1563 /* now gather any other geneve devices that were created in this ns */ 1564 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) { 1565 /* If geneve->dev is in the same netns, it was already added 1566 * to the list by the previous loop. 1567 */ 1568 if (!net_eq(dev_net(geneve->dev), net)) 1569 unregister_netdevice_queue(geneve->dev, &list); 1570 } 1571 1572 /* unregister the devices gathered above */ 1573 unregister_netdevice_many(&list); 1574 rtnl_unlock(); 1575 } 1576 1577 static struct pernet_operations geneve_net_ops = { 1578 .init = geneve_init_net, 1579 .exit = geneve_exit_net, 1580 .id = &geneve_net_id, 1581 .size = sizeof(struct geneve_net), 1582 }; 1583 1584 static int __init geneve_init_module(void) 1585 { 1586 int rc; 1587 1588 rc = register_pernet_subsys(&geneve_net_ops); 1589 if (rc) 1590 goto out1; 1591 1592 rc = register_netdevice_notifier(&geneve_notifier_block); 1593 if (rc) 1594 goto out2; 1595 1596 rc = rtnl_link_register(&geneve_link_ops); 1597 if (rc) 1598 goto out3; 1599 1600 return 0; 1601 1602 out3: 1603 unregister_netdevice_notifier(&geneve_notifier_block); 1604 out2: 1605 unregister_pernet_subsys(&geneve_net_ops); 1606 out1: 1607 return rc; 1608 } 1609 late_initcall(geneve_init_module); 1610 1611 static void __exit geneve_cleanup_module(void) 1612 { 1613 rtnl_link_unregister(&geneve_link_ops); 1614 unregister_netdevice_notifier(&geneve_notifier_block); 1615 unregister_pernet_subsys(&geneve_net_ops); 1616 } 1617 module_exit(geneve_cleanup_module); 1618 1619 MODULE_LICENSE("GPL"); 1620 MODULE_VERSION(GENEVE_NETDEV_VER); 1621 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>"); 1622 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic"); 1623 MODULE_ALIAS_RTNL_LINK("geneve"); 1624