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 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN) 40 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN) 41 42 /* per-network namespace private data for this module */ 43 struct geneve_net { 44 struct list_head geneve_list; 45 struct list_head sock_list; 46 }; 47 48 static unsigned int geneve_net_id; 49 50 struct geneve_dev_node { 51 struct hlist_node hlist; 52 struct geneve_dev *geneve; 53 }; 54 55 /* Pseudo network device */ 56 struct geneve_dev { 57 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */ 58 #if IS_ENABLED(CONFIG_IPV6) 59 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */ 60 #endif 61 struct net *net; /* netns for packet i/o */ 62 struct net_device *dev; /* netdev for geneve tunnel */ 63 struct ip_tunnel_info info; 64 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */ 65 #if IS_ENABLED(CONFIG_IPV6) 66 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */ 67 #endif 68 struct list_head next; /* geneve's per namespace list */ 69 struct gro_cells gro_cells; 70 bool collect_md; 71 bool use_udp6_rx_checksums; 72 bool ttl_inherit; 73 }; 74 75 struct geneve_sock { 76 bool collect_md; 77 struct list_head list; 78 struct socket *sock; 79 struct rcu_head rcu; 80 int refcnt; 81 struct hlist_head vni_list[VNI_HASH_SIZE]; 82 }; 83 84 static inline __u32 geneve_net_vni_hash(u8 vni[3]) 85 { 86 __u32 vnid; 87 88 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2]; 89 return hash_32(vnid, VNI_HASH_BITS); 90 } 91 92 static __be64 vni_to_tunnel_id(const __u8 *vni) 93 { 94 #ifdef __BIG_ENDIAN 95 return (vni[0] << 16) | (vni[1] << 8) | vni[2]; 96 #else 97 return (__force __be64)(((__force u64)vni[0] << 40) | 98 ((__force u64)vni[1] << 48) | 99 ((__force u64)vni[2] << 56)); 100 #endif 101 } 102 103 /* Convert 64 bit tunnel ID to 24 bit VNI. */ 104 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni) 105 { 106 #ifdef __BIG_ENDIAN 107 vni[0] = (__force __u8)(tun_id >> 16); 108 vni[1] = (__force __u8)(tun_id >> 8); 109 vni[2] = (__force __u8)tun_id; 110 #else 111 vni[0] = (__force __u8)((__force u64)tun_id >> 40); 112 vni[1] = (__force __u8)((__force u64)tun_id >> 48); 113 vni[2] = (__force __u8)((__force u64)tun_id >> 56); 114 #endif 115 } 116 117 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni) 118 { 119 return !memcmp(vni, &tun_id[5], 3); 120 } 121 122 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs) 123 { 124 return gs->sock->sk->sk_family; 125 } 126 127 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs, 128 __be32 addr, u8 vni[]) 129 { 130 struct hlist_head *vni_list_head; 131 struct geneve_dev_node *node; 132 __u32 hash; 133 134 /* Find the device for this VNI */ 135 hash = geneve_net_vni_hash(vni); 136 vni_list_head = &gs->vni_list[hash]; 137 hlist_for_each_entry_rcu(node, vni_list_head, hlist) { 138 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) && 139 addr == node->geneve->info.key.u.ipv4.dst) 140 return node->geneve; 141 } 142 return NULL; 143 } 144 145 #if IS_ENABLED(CONFIG_IPV6) 146 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs, 147 struct in6_addr addr6, u8 vni[]) 148 { 149 struct hlist_head *vni_list_head; 150 struct geneve_dev_node *node; 151 __u32 hash; 152 153 /* Find the device for this VNI */ 154 hash = geneve_net_vni_hash(vni); 155 vni_list_head = &gs->vni_list[hash]; 156 hlist_for_each_entry_rcu(node, vni_list_head, hlist) { 157 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) && 158 ipv6_addr_equal(&addr6, &node->geneve->info.key.u.ipv6.dst)) 159 return node->geneve; 160 } 161 return NULL; 162 } 163 #endif 164 165 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb) 166 { 167 return (struct genevehdr *)(udp_hdr(skb) + 1); 168 } 169 170 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs, 171 struct sk_buff *skb) 172 { 173 static u8 zero_vni[3]; 174 u8 *vni; 175 176 if (geneve_get_sk_family(gs) == AF_INET) { 177 struct iphdr *iph; 178 __be32 addr; 179 180 iph = ip_hdr(skb); /* outer IP header... */ 181 182 if (gs->collect_md) { 183 vni = zero_vni; 184 addr = 0; 185 } else { 186 vni = geneve_hdr(skb)->vni; 187 addr = iph->saddr; 188 } 189 190 return geneve_lookup(gs, addr, vni); 191 #if IS_ENABLED(CONFIG_IPV6) 192 } else if (geneve_get_sk_family(gs) == AF_INET6) { 193 static struct in6_addr zero_addr6; 194 struct ipv6hdr *ip6h; 195 struct in6_addr addr6; 196 197 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */ 198 199 if (gs->collect_md) { 200 vni = zero_vni; 201 addr6 = zero_addr6; 202 } else { 203 vni = geneve_hdr(skb)->vni; 204 addr6 = ip6h->saddr; 205 } 206 207 return geneve6_lookup(gs, addr6, vni); 208 #endif 209 } 210 return NULL; 211 } 212 213 /* geneve receive/decap routine */ 214 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs, 215 struct sk_buff *skb) 216 { 217 struct genevehdr *gnvh = geneve_hdr(skb); 218 struct metadata_dst *tun_dst = NULL; 219 struct pcpu_sw_netstats *stats; 220 unsigned int len; 221 int err = 0; 222 void *oiph; 223 224 if (ip_tunnel_collect_metadata() || gs->collect_md) { 225 __be16 flags; 226 227 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT | 228 (gnvh->oam ? TUNNEL_OAM : 0) | 229 (gnvh->critical ? TUNNEL_CRIT_OPT : 0); 230 231 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags, 232 vni_to_tunnel_id(gnvh->vni), 233 gnvh->opt_len * 4); 234 if (!tun_dst) { 235 geneve->dev->stats.rx_dropped++; 236 goto drop; 237 } 238 /* Update tunnel dst according to Geneve options. */ 239 ip_tunnel_info_opts_set(&tun_dst->u.tun_info, 240 gnvh->options, gnvh->opt_len * 4, 241 TUNNEL_GENEVE_OPT); 242 } else { 243 /* Drop packets w/ critical options, 244 * since we don't support any... 245 */ 246 if (gnvh->critical) { 247 geneve->dev->stats.rx_frame_errors++; 248 geneve->dev->stats.rx_errors++; 249 goto drop; 250 } 251 } 252 253 skb_reset_mac_header(skb); 254 skb->protocol = eth_type_trans(skb, geneve->dev); 255 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 256 257 if (tun_dst) 258 skb_dst_set(skb, &tun_dst->dst); 259 260 /* Ignore packet loops (and multicast echo) */ 261 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) { 262 geneve->dev->stats.rx_errors++; 263 goto drop; 264 } 265 266 oiph = skb_network_header(skb); 267 skb_reset_network_header(skb); 268 269 if (geneve_get_sk_family(gs) == AF_INET) 270 err = IP_ECN_decapsulate(oiph, skb); 271 #if IS_ENABLED(CONFIG_IPV6) 272 else 273 err = IP6_ECN_decapsulate(oiph, skb); 274 #endif 275 276 if (unlikely(err)) { 277 if (log_ecn_error) { 278 if (geneve_get_sk_family(gs) == AF_INET) 279 net_info_ratelimited("non-ECT from %pI4 " 280 "with TOS=%#x\n", 281 &((struct iphdr *)oiph)->saddr, 282 ((struct iphdr *)oiph)->tos); 283 #if IS_ENABLED(CONFIG_IPV6) 284 else 285 net_info_ratelimited("non-ECT from %pI6\n", 286 &((struct ipv6hdr *)oiph)->saddr); 287 #endif 288 } 289 if (err > 1) { 290 ++geneve->dev->stats.rx_frame_errors; 291 ++geneve->dev->stats.rx_errors; 292 goto drop; 293 } 294 } 295 296 len = skb->len; 297 err = gro_cells_receive(&geneve->gro_cells, skb); 298 if (likely(err == NET_RX_SUCCESS)) { 299 stats = this_cpu_ptr(geneve->dev->tstats); 300 u64_stats_update_begin(&stats->syncp); 301 stats->rx_packets++; 302 stats->rx_bytes += len; 303 u64_stats_update_end(&stats->syncp); 304 } 305 return; 306 drop: 307 /* Consume bad packet */ 308 kfree_skb(skb); 309 } 310 311 /* Setup stats when device is created */ 312 static int geneve_init(struct net_device *dev) 313 { 314 struct geneve_dev *geneve = netdev_priv(dev); 315 int err; 316 317 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 318 if (!dev->tstats) 319 return -ENOMEM; 320 321 err = gro_cells_init(&geneve->gro_cells, dev); 322 if (err) { 323 free_percpu(dev->tstats); 324 return err; 325 } 326 327 err = dst_cache_init(&geneve->info.dst_cache, GFP_KERNEL); 328 if (err) { 329 free_percpu(dev->tstats); 330 gro_cells_destroy(&geneve->gro_cells); 331 return err; 332 } 333 return 0; 334 } 335 336 static void geneve_uninit(struct net_device *dev) 337 { 338 struct geneve_dev *geneve = netdev_priv(dev); 339 340 dst_cache_destroy(&geneve->info.dst_cache); 341 gro_cells_destroy(&geneve->gro_cells); 342 free_percpu(dev->tstats); 343 } 344 345 /* Callback from net/ipv4/udp.c to receive packets */ 346 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb) 347 { 348 struct genevehdr *geneveh; 349 struct geneve_dev *geneve; 350 struct geneve_sock *gs; 351 int opts_len; 352 353 /* Need UDP and Geneve header to be present */ 354 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN))) 355 goto drop; 356 357 /* Return packets with reserved bits set */ 358 geneveh = geneve_hdr(skb); 359 if (unlikely(geneveh->ver != GENEVE_VER)) 360 goto drop; 361 362 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB))) 363 goto drop; 364 365 gs = rcu_dereference_sk_user_data(sk); 366 if (!gs) 367 goto drop; 368 369 geneve = geneve_lookup_skb(gs, skb); 370 if (!geneve) 371 goto drop; 372 373 opts_len = geneveh->opt_len * 4; 374 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, 375 htons(ETH_P_TEB), 376 !net_eq(geneve->net, dev_net(geneve->dev)))) { 377 geneve->dev->stats.rx_dropped++; 378 goto drop; 379 } 380 381 geneve_rx(geneve, gs, skb); 382 return 0; 383 384 drop: 385 /* Consume bad packet */ 386 kfree_skb(skb); 387 return 0; 388 } 389 390 static struct socket *geneve_create_sock(struct net *net, bool ipv6, 391 __be16 port, bool ipv6_rx_csum) 392 { 393 struct socket *sock; 394 struct udp_port_cfg udp_conf; 395 int err; 396 397 memset(&udp_conf, 0, sizeof(udp_conf)); 398 399 if (ipv6) { 400 udp_conf.family = AF_INET6; 401 udp_conf.ipv6_v6only = 1; 402 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum; 403 } else { 404 udp_conf.family = AF_INET; 405 udp_conf.local_ip.s_addr = htonl(INADDR_ANY); 406 } 407 408 udp_conf.local_udp_port = port; 409 410 /* Open UDP socket */ 411 err = udp_sock_create(net, &udp_conf, &sock); 412 if (err < 0) 413 return ERR_PTR(err); 414 415 return sock; 416 } 417 418 static int geneve_hlen(struct genevehdr *gh) 419 { 420 return sizeof(*gh) + gh->opt_len * 4; 421 } 422 423 static struct sk_buff *geneve_gro_receive(struct sock *sk, 424 struct list_head *head, 425 struct sk_buff *skb) 426 { 427 struct sk_buff *pp = NULL; 428 struct sk_buff *p; 429 struct genevehdr *gh, *gh2; 430 unsigned int hlen, gh_len, off_gnv; 431 const struct packet_offload *ptype; 432 __be16 type; 433 int flush = 1; 434 435 off_gnv = skb_gro_offset(skb); 436 hlen = off_gnv + sizeof(*gh); 437 gh = skb_gro_header_fast(skb, off_gnv); 438 if (skb_gro_header_hard(skb, hlen)) { 439 gh = skb_gro_header_slow(skb, hlen, off_gnv); 440 if (unlikely(!gh)) 441 goto out; 442 } 443 444 if (gh->ver != GENEVE_VER || gh->oam) 445 goto out; 446 gh_len = geneve_hlen(gh); 447 448 hlen = off_gnv + gh_len; 449 if (skb_gro_header_hard(skb, hlen)) { 450 gh = skb_gro_header_slow(skb, hlen, off_gnv); 451 if (unlikely(!gh)) 452 goto out; 453 } 454 455 list_for_each_entry(p, head, list) { 456 if (!NAPI_GRO_CB(p)->same_flow) 457 continue; 458 459 gh2 = (struct genevehdr *)(p->data + off_gnv); 460 if (gh->opt_len != gh2->opt_len || 461 memcmp(gh, gh2, gh_len)) { 462 NAPI_GRO_CB(p)->same_flow = 0; 463 continue; 464 } 465 } 466 467 type = gh->proto_type; 468 469 rcu_read_lock(); 470 ptype = gro_find_receive_by_type(type); 471 if (!ptype) 472 goto out_unlock; 473 474 skb_gro_pull(skb, gh_len); 475 skb_gro_postpull_rcsum(skb, gh, gh_len); 476 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb); 477 flush = 0; 478 479 out_unlock: 480 rcu_read_unlock(); 481 out: 482 skb_gro_flush_final(skb, pp, flush); 483 484 return pp; 485 } 486 487 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb, 488 int nhoff) 489 { 490 struct genevehdr *gh; 491 struct packet_offload *ptype; 492 __be16 type; 493 int gh_len; 494 int err = -ENOSYS; 495 496 gh = (struct genevehdr *)(skb->data + nhoff); 497 gh_len = geneve_hlen(gh); 498 type = gh->proto_type; 499 500 rcu_read_lock(); 501 ptype = gro_find_complete_by_type(type); 502 if (ptype) 503 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len); 504 505 rcu_read_unlock(); 506 507 skb_set_inner_mac_header(skb, nhoff + gh_len); 508 509 return err; 510 } 511 512 /* Create new listen socket if needed */ 513 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port, 514 bool ipv6, bool ipv6_rx_csum) 515 { 516 struct geneve_net *gn = net_generic(net, geneve_net_id); 517 struct geneve_sock *gs; 518 struct socket *sock; 519 struct udp_tunnel_sock_cfg tunnel_cfg; 520 int h; 521 522 gs = kzalloc(sizeof(*gs), GFP_KERNEL); 523 if (!gs) 524 return ERR_PTR(-ENOMEM); 525 526 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum); 527 if (IS_ERR(sock)) { 528 kfree(gs); 529 return ERR_CAST(sock); 530 } 531 532 gs->sock = sock; 533 gs->refcnt = 1; 534 for (h = 0; h < VNI_HASH_SIZE; ++h) 535 INIT_HLIST_HEAD(&gs->vni_list[h]); 536 537 /* Initialize the geneve udp offloads structure */ 538 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 539 540 /* Mark socket as an encapsulation socket */ 541 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg)); 542 tunnel_cfg.sk_user_data = gs; 543 tunnel_cfg.encap_type = 1; 544 tunnel_cfg.gro_receive = geneve_gro_receive; 545 tunnel_cfg.gro_complete = geneve_gro_complete; 546 tunnel_cfg.encap_rcv = geneve_udp_encap_recv; 547 tunnel_cfg.encap_destroy = NULL; 548 setup_udp_tunnel_sock(net, sock, &tunnel_cfg); 549 list_add(&gs->list, &gn->sock_list); 550 return gs; 551 } 552 553 static void __geneve_sock_release(struct geneve_sock *gs) 554 { 555 if (!gs || --gs->refcnt) 556 return; 557 558 list_del(&gs->list); 559 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 560 udp_tunnel_sock_release(gs->sock); 561 kfree_rcu(gs, rcu); 562 } 563 564 static void geneve_sock_release(struct geneve_dev *geneve) 565 { 566 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4); 567 #if IS_ENABLED(CONFIG_IPV6) 568 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6); 569 570 rcu_assign_pointer(geneve->sock6, NULL); 571 #endif 572 573 rcu_assign_pointer(geneve->sock4, NULL); 574 synchronize_net(); 575 576 __geneve_sock_release(gs4); 577 #if IS_ENABLED(CONFIG_IPV6) 578 __geneve_sock_release(gs6); 579 #endif 580 } 581 582 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn, 583 sa_family_t family, 584 __be16 dst_port) 585 { 586 struct geneve_sock *gs; 587 588 list_for_each_entry(gs, &gn->sock_list, list) { 589 if (inet_sk(gs->sock->sk)->inet_sport == dst_port && 590 geneve_get_sk_family(gs) == family) { 591 return gs; 592 } 593 } 594 return NULL; 595 } 596 597 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6) 598 { 599 struct net *net = geneve->net; 600 struct geneve_net *gn = net_generic(net, geneve_net_id); 601 struct geneve_dev_node *node; 602 struct geneve_sock *gs; 603 __u8 vni[3]; 604 __u32 hash; 605 606 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->info.key.tp_dst); 607 if (gs) { 608 gs->refcnt++; 609 goto out; 610 } 611 612 gs = geneve_socket_create(net, geneve->info.key.tp_dst, ipv6, 613 geneve->use_udp6_rx_checksums); 614 if (IS_ERR(gs)) 615 return PTR_ERR(gs); 616 617 out: 618 gs->collect_md = geneve->collect_md; 619 #if IS_ENABLED(CONFIG_IPV6) 620 if (ipv6) { 621 rcu_assign_pointer(geneve->sock6, gs); 622 node = &geneve->hlist6; 623 } else 624 #endif 625 { 626 rcu_assign_pointer(geneve->sock4, gs); 627 node = &geneve->hlist4; 628 } 629 node->geneve = geneve; 630 631 tunnel_id_to_vni(geneve->info.key.tun_id, vni); 632 hash = geneve_net_vni_hash(vni); 633 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]); 634 return 0; 635 } 636 637 static int geneve_open(struct net_device *dev) 638 { 639 struct geneve_dev *geneve = netdev_priv(dev); 640 bool ipv6 = !!(geneve->info.mode & IP_TUNNEL_INFO_IPV6); 641 bool metadata = geneve->collect_md; 642 int ret = 0; 643 644 #if IS_ENABLED(CONFIG_IPV6) 645 if (ipv6 || metadata) 646 ret = geneve_sock_add(geneve, true); 647 #endif 648 if (!ret && (!ipv6 || metadata)) 649 ret = geneve_sock_add(geneve, false); 650 if (ret < 0) 651 geneve_sock_release(geneve); 652 653 return ret; 654 } 655 656 static int geneve_stop(struct net_device *dev) 657 { 658 struct geneve_dev *geneve = netdev_priv(dev); 659 660 hlist_del_init_rcu(&geneve->hlist4.hlist); 661 #if IS_ENABLED(CONFIG_IPV6) 662 hlist_del_init_rcu(&geneve->hlist6.hlist); 663 #endif 664 geneve_sock_release(geneve); 665 return 0; 666 } 667 668 static void geneve_build_header(struct genevehdr *geneveh, 669 const struct ip_tunnel_info *info) 670 { 671 geneveh->ver = GENEVE_VER; 672 geneveh->opt_len = info->options_len / 4; 673 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM); 674 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT); 675 geneveh->rsvd1 = 0; 676 tunnel_id_to_vni(info->key.tun_id, geneveh->vni); 677 geneveh->proto_type = htons(ETH_P_TEB); 678 geneveh->rsvd2 = 0; 679 680 if (info->key.tun_flags & TUNNEL_GENEVE_OPT) 681 ip_tunnel_info_opts_get(geneveh->options, info); 682 } 683 684 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb, 685 const struct ip_tunnel_info *info, 686 bool xnet, int ip_hdr_len) 687 { 688 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM); 689 struct genevehdr *gnvh; 690 int min_headroom; 691 int err; 692 693 skb_reset_mac_header(skb); 694 skb_scrub_packet(skb, xnet); 695 696 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len + 697 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len; 698 err = skb_cow_head(skb, min_headroom); 699 if (unlikely(err)) 700 goto free_dst; 701 702 err = udp_tunnel_handle_offloads(skb, udp_sum); 703 if (err) 704 goto free_dst; 705 706 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len); 707 geneve_build_header(gnvh, info); 708 skb_set_inner_protocol(skb, htons(ETH_P_TEB)); 709 return 0; 710 711 free_dst: 712 dst_release(dst); 713 return err; 714 } 715 716 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb, 717 struct net_device *dev, 718 struct geneve_sock *gs4, 719 struct flowi4 *fl4, 720 const struct ip_tunnel_info *info) 721 { 722 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 723 struct geneve_dev *geneve = netdev_priv(dev); 724 struct dst_cache *dst_cache; 725 struct rtable *rt = NULL; 726 __u8 tos; 727 728 if (!gs4) 729 return ERR_PTR(-EIO); 730 731 memset(fl4, 0, sizeof(*fl4)); 732 fl4->flowi4_mark = skb->mark; 733 fl4->flowi4_proto = IPPROTO_UDP; 734 fl4->daddr = info->key.u.ipv4.dst; 735 fl4->saddr = info->key.u.ipv4.src; 736 737 tos = info->key.tos; 738 if ((tos == 1) && !geneve->collect_md) { 739 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb); 740 use_cache = false; 741 } 742 fl4->flowi4_tos = RT_TOS(tos); 743 744 dst_cache = (struct dst_cache *)&info->dst_cache; 745 if (use_cache) { 746 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr); 747 if (rt) 748 return rt; 749 } 750 rt = ip_route_output_key(geneve->net, fl4); 751 if (IS_ERR(rt)) { 752 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr); 753 return ERR_PTR(-ENETUNREACH); 754 } 755 if (rt->dst.dev == dev) { /* is this necessary? */ 756 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr); 757 ip_rt_put(rt); 758 return ERR_PTR(-ELOOP); 759 } 760 if (use_cache) 761 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr); 762 return rt; 763 } 764 765 #if IS_ENABLED(CONFIG_IPV6) 766 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb, 767 struct net_device *dev, 768 struct geneve_sock *gs6, 769 struct flowi6 *fl6, 770 const struct ip_tunnel_info *info) 771 { 772 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 773 struct geneve_dev *geneve = netdev_priv(dev); 774 struct dst_entry *dst = NULL; 775 struct dst_cache *dst_cache; 776 __u8 prio; 777 778 if (!gs6) 779 return ERR_PTR(-EIO); 780 781 memset(fl6, 0, sizeof(*fl6)); 782 fl6->flowi6_mark = skb->mark; 783 fl6->flowi6_proto = IPPROTO_UDP; 784 fl6->daddr = info->key.u.ipv6.dst; 785 fl6->saddr = info->key.u.ipv6.src; 786 prio = info->key.tos; 787 if ((prio == 1) && !geneve->collect_md) { 788 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb); 789 use_cache = false; 790 } 791 792 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio), 793 info->key.label); 794 dst_cache = (struct dst_cache *)&info->dst_cache; 795 if (use_cache) { 796 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr); 797 if (dst) 798 return dst; 799 } 800 if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) { 801 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr); 802 return ERR_PTR(-ENETUNREACH); 803 } 804 if (dst->dev == dev) { /* is this necessary? */ 805 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr); 806 dst_release(dst); 807 return ERR_PTR(-ELOOP); 808 } 809 810 if (use_cache) 811 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr); 812 return dst; 813 } 814 #endif 815 816 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev, 817 struct geneve_dev *geneve, 818 const struct ip_tunnel_info *info) 819 { 820 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 821 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 822 const struct ip_tunnel_key *key = &info->key; 823 struct rtable *rt; 824 struct flowi4 fl4; 825 __u8 tos, ttl; 826 __be16 sport; 827 __be16 df; 828 int err; 829 830 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info); 831 if (IS_ERR(rt)) 832 return PTR_ERR(rt); 833 834 if (skb_dst(skb)) { 835 int mtu = dst_mtu(&rt->dst) - GENEVE_IPV4_HLEN - 836 info->options_len; 837 838 skb_dst_update_pmtu(skb, mtu); 839 } 840 841 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 842 if (geneve->collect_md) { 843 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 844 ttl = key->ttl; 845 } else { 846 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb); 847 if (geneve->ttl_inherit) 848 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb); 849 else 850 ttl = key->ttl; 851 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst); 852 } 853 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0; 854 855 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr)); 856 if (unlikely(err)) 857 return err; 858 859 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr, 860 tos, ttl, df, sport, geneve->info.key.tp_dst, 861 !net_eq(geneve->net, dev_net(geneve->dev)), 862 !(info->key.tun_flags & TUNNEL_CSUM)); 863 return 0; 864 } 865 866 #if IS_ENABLED(CONFIG_IPV6) 867 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev, 868 struct geneve_dev *geneve, 869 const struct ip_tunnel_info *info) 870 { 871 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 872 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 873 const struct ip_tunnel_key *key = &info->key; 874 struct dst_entry *dst = NULL; 875 struct flowi6 fl6; 876 __u8 prio, ttl; 877 __be16 sport; 878 int err; 879 880 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info); 881 if (IS_ERR(dst)) 882 return PTR_ERR(dst); 883 884 if (skb_dst(skb)) { 885 int mtu = dst_mtu(dst) - GENEVE_IPV6_HLEN - info->options_len; 886 887 skb_dst_update_pmtu(skb, mtu); 888 } 889 890 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 891 if (geneve->collect_md) { 892 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 893 ttl = key->ttl; 894 } else { 895 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel), 896 ip_hdr(skb), skb); 897 if (geneve->ttl_inherit) 898 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb); 899 else 900 ttl = key->ttl; 901 ttl = ttl ? : ip6_dst_hoplimit(dst); 902 } 903 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr)); 904 if (unlikely(err)) 905 return err; 906 907 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev, 908 &fl6.saddr, &fl6.daddr, prio, ttl, 909 info->key.label, sport, geneve->info.key.tp_dst, 910 !(info->key.tun_flags & TUNNEL_CSUM)); 911 return 0; 912 } 913 #endif 914 915 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev) 916 { 917 struct geneve_dev *geneve = netdev_priv(dev); 918 struct ip_tunnel_info *info = NULL; 919 int err; 920 921 if (geneve->collect_md) { 922 info = skb_tunnel_info(skb); 923 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) { 924 err = -EINVAL; 925 netdev_dbg(dev, "no tunnel metadata\n"); 926 goto tx_error; 927 } 928 } else { 929 info = &geneve->info; 930 } 931 932 rcu_read_lock(); 933 #if IS_ENABLED(CONFIG_IPV6) 934 if (info->mode & IP_TUNNEL_INFO_IPV6) 935 err = geneve6_xmit_skb(skb, dev, geneve, info); 936 else 937 #endif 938 err = geneve_xmit_skb(skb, dev, geneve, info); 939 rcu_read_unlock(); 940 941 if (likely(!err)) 942 return NETDEV_TX_OK; 943 tx_error: 944 dev_kfree_skb(skb); 945 946 if (err == -ELOOP) 947 dev->stats.collisions++; 948 else if (err == -ENETUNREACH) 949 dev->stats.tx_carrier_errors++; 950 951 dev->stats.tx_errors++; 952 return NETDEV_TX_OK; 953 } 954 955 static int geneve_change_mtu(struct net_device *dev, int new_mtu) 956 { 957 if (new_mtu > dev->max_mtu) 958 new_mtu = dev->max_mtu; 959 else if (new_mtu < dev->min_mtu) 960 new_mtu = dev->min_mtu; 961 962 dev->mtu = new_mtu; 963 return 0; 964 } 965 966 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 967 { 968 struct ip_tunnel_info *info = skb_tunnel_info(skb); 969 struct geneve_dev *geneve = netdev_priv(dev); 970 971 if (ip_tunnel_info_af(info) == AF_INET) { 972 struct rtable *rt; 973 struct flowi4 fl4; 974 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 975 976 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info); 977 if (IS_ERR(rt)) 978 return PTR_ERR(rt); 979 980 ip_rt_put(rt); 981 info->key.u.ipv4.src = fl4.saddr; 982 #if IS_ENABLED(CONFIG_IPV6) 983 } else if (ip_tunnel_info_af(info) == AF_INET6) { 984 struct dst_entry *dst; 985 struct flowi6 fl6; 986 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 987 988 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info); 989 if (IS_ERR(dst)) 990 return PTR_ERR(dst); 991 992 dst_release(dst); 993 info->key.u.ipv6.src = fl6.saddr; 994 #endif 995 } else { 996 return -EINVAL; 997 } 998 999 info->key.tp_src = udp_flow_src_port(geneve->net, skb, 1000 1, USHRT_MAX, true); 1001 info->key.tp_dst = geneve->info.key.tp_dst; 1002 return 0; 1003 } 1004 1005 static const struct net_device_ops geneve_netdev_ops = { 1006 .ndo_init = geneve_init, 1007 .ndo_uninit = geneve_uninit, 1008 .ndo_open = geneve_open, 1009 .ndo_stop = geneve_stop, 1010 .ndo_start_xmit = geneve_xmit, 1011 .ndo_get_stats64 = ip_tunnel_get_stats64, 1012 .ndo_change_mtu = geneve_change_mtu, 1013 .ndo_validate_addr = eth_validate_addr, 1014 .ndo_set_mac_address = eth_mac_addr, 1015 .ndo_fill_metadata_dst = geneve_fill_metadata_dst, 1016 }; 1017 1018 static void geneve_get_drvinfo(struct net_device *dev, 1019 struct ethtool_drvinfo *drvinfo) 1020 { 1021 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version)); 1022 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver)); 1023 } 1024 1025 static const struct ethtool_ops geneve_ethtool_ops = { 1026 .get_drvinfo = geneve_get_drvinfo, 1027 .get_link = ethtool_op_get_link, 1028 }; 1029 1030 /* Info for udev, that this is a virtual tunnel endpoint */ 1031 static struct device_type geneve_type = { 1032 .name = "geneve", 1033 }; 1034 1035 /* Calls the ndo_udp_tunnel_add of the caller in order to 1036 * supply the listening GENEVE udp ports. Callers are expected 1037 * to implement the ndo_udp_tunnel_add. 1038 */ 1039 static void geneve_offload_rx_ports(struct net_device *dev, bool push) 1040 { 1041 struct net *net = dev_net(dev); 1042 struct geneve_net *gn = net_generic(net, geneve_net_id); 1043 struct geneve_sock *gs; 1044 1045 rcu_read_lock(); 1046 list_for_each_entry_rcu(gs, &gn->sock_list, list) { 1047 if (push) { 1048 udp_tunnel_push_rx_port(dev, gs->sock, 1049 UDP_TUNNEL_TYPE_GENEVE); 1050 } else { 1051 udp_tunnel_drop_rx_port(dev, gs->sock, 1052 UDP_TUNNEL_TYPE_GENEVE); 1053 } 1054 } 1055 rcu_read_unlock(); 1056 } 1057 1058 /* Initialize the device structure. */ 1059 static void geneve_setup(struct net_device *dev) 1060 { 1061 ether_setup(dev); 1062 1063 dev->netdev_ops = &geneve_netdev_ops; 1064 dev->ethtool_ops = &geneve_ethtool_ops; 1065 dev->needs_free_netdev = true; 1066 1067 SET_NETDEV_DEVTYPE(dev, &geneve_type); 1068 1069 dev->features |= NETIF_F_LLTX; 1070 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM; 1071 dev->features |= NETIF_F_RXCSUM; 1072 dev->features |= NETIF_F_GSO_SOFTWARE; 1073 1074 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM; 1075 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 1076 1077 /* MTU range: 68 - (something less than 65535) */ 1078 dev->min_mtu = ETH_MIN_MTU; 1079 /* The max_mtu calculation does not take account of GENEVE 1080 * options, to avoid excluding potentially valid 1081 * configurations. This will be further reduced by IPvX hdr size. 1082 */ 1083 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len; 1084 1085 netif_keep_dst(dev); 1086 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1087 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE; 1088 eth_hw_addr_random(dev); 1089 } 1090 1091 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = { 1092 [IFLA_GENEVE_ID] = { .type = NLA_U32 }, 1093 [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) }, 1094 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) }, 1095 [IFLA_GENEVE_TTL] = { .type = NLA_U8 }, 1096 [IFLA_GENEVE_TOS] = { .type = NLA_U8 }, 1097 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 }, 1098 [IFLA_GENEVE_PORT] = { .type = NLA_U16 }, 1099 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1100 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 }, 1101 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 }, 1102 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 }, 1103 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 }, 1104 }; 1105 1106 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[], 1107 struct netlink_ext_ack *extack) 1108 { 1109 if (tb[IFLA_ADDRESS]) { 1110 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1111 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1112 "Provided link layer address is not Ethernet"); 1113 return -EINVAL; 1114 } 1115 1116 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1117 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1118 "Provided Ethernet address is not unicast"); 1119 return -EADDRNOTAVAIL; 1120 } 1121 } 1122 1123 if (!data) { 1124 NL_SET_ERR_MSG(extack, 1125 "Not enough attributes provided to perform the operation"); 1126 return -EINVAL; 1127 } 1128 1129 if (data[IFLA_GENEVE_ID]) { 1130 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1131 1132 if (vni >= GENEVE_N_VID) { 1133 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID], 1134 "Geneve ID must be lower than 16777216"); 1135 return -ERANGE; 1136 } 1137 } 1138 1139 return 0; 1140 } 1141 1142 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn, 1143 const struct ip_tunnel_info *info, 1144 bool *tun_on_same_port, 1145 bool *tun_collect_md) 1146 { 1147 struct geneve_dev *geneve, *t = NULL; 1148 1149 *tun_on_same_port = false; 1150 *tun_collect_md = false; 1151 list_for_each_entry(geneve, &gn->geneve_list, next) { 1152 if (info->key.tp_dst == geneve->info.key.tp_dst) { 1153 *tun_collect_md = geneve->collect_md; 1154 *tun_on_same_port = true; 1155 } 1156 if (info->key.tun_id == geneve->info.key.tun_id && 1157 info->key.tp_dst == geneve->info.key.tp_dst && 1158 !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u))) 1159 t = geneve; 1160 } 1161 return t; 1162 } 1163 1164 static bool is_tnl_info_zero(const struct ip_tunnel_info *info) 1165 { 1166 return !(info->key.tun_id || info->key.tun_flags || info->key.tos || 1167 info->key.ttl || info->key.label || info->key.tp_src || 1168 memchr_inv(&info->key.u, 0, sizeof(info->key.u))); 1169 } 1170 1171 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a, 1172 struct ip_tunnel_info *b) 1173 { 1174 if (ip_tunnel_info_af(a) == AF_INET) 1175 return a->key.u.ipv4.dst == b->key.u.ipv4.dst; 1176 else 1177 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst); 1178 } 1179 1180 static int geneve_configure(struct net *net, struct net_device *dev, 1181 struct netlink_ext_ack *extack, 1182 const struct ip_tunnel_info *info, 1183 bool metadata, bool ipv6_rx_csum, 1184 bool ttl_inherit) 1185 { 1186 struct geneve_net *gn = net_generic(net, geneve_net_id); 1187 struct geneve_dev *t, *geneve = netdev_priv(dev); 1188 bool tun_collect_md, tun_on_same_port; 1189 int err, encap_len; 1190 1191 if (metadata && !is_tnl_info_zero(info)) { 1192 NL_SET_ERR_MSG(extack, 1193 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified"); 1194 return -EINVAL; 1195 } 1196 1197 geneve->net = net; 1198 geneve->dev = dev; 1199 1200 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md); 1201 if (t) 1202 return -EBUSY; 1203 1204 /* make enough headroom for basic scenario */ 1205 encap_len = GENEVE_BASE_HLEN + ETH_HLEN; 1206 if (!metadata && ip_tunnel_info_af(info) == AF_INET) { 1207 encap_len += sizeof(struct iphdr); 1208 dev->max_mtu -= sizeof(struct iphdr); 1209 } else { 1210 encap_len += sizeof(struct ipv6hdr); 1211 dev->max_mtu -= sizeof(struct ipv6hdr); 1212 } 1213 dev->needed_headroom = encap_len + ETH_HLEN; 1214 1215 if (metadata) { 1216 if (tun_on_same_port) { 1217 NL_SET_ERR_MSG(extack, 1218 "There can be only one externally controlled device on a destination port"); 1219 return -EPERM; 1220 } 1221 } else { 1222 if (tun_collect_md) { 1223 NL_SET_ERR_MSG(extack, 1224 "There already exists an externally controlled device on this destination port"); 1225 return -EPERM; 1226 } 1227 } 1228 1229 dst_cache_reset(&geneve->info.dst_cache); 1230 geneve->info = *info; 1231 geneve->collect_md = metadata; 1232 geneve->use_udp6_rx_checksums = ipv6_rx_csum; 1233 geneve->ttl_inherit = ttl_inherit; 1234 1235 err = register_netdevice(dev); 1236 if (err) 1237 return err; 1238 1239 list_add(&geneve->next, &gn->geneve_list); 1240 return 0; 1241 } 1242 1243 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port) 1244 { 1245 memset(info, 0, sizeof(*info)); 1246 info->key.tp_dst = htons(dst_port); 1247 } 1248 1249 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[], 1250 struct netlink_ext_ack *extack, 1251 struct ip_tunnel_info *info, bool *metadata, 1252 bool *use_udp6_rx_checksums, bool *ttl_inherit, 1253 bool changelink) 1254 { 1255 int attrtype; 1256 1257 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) { 1258 NL_SET_ERR_MSG(extack, 1259 "Cannot specify both IPv4 and IPv6 Remote addresses"); 1260 return -EINVAL; 1261 } 1262 1263 if (data[IFLA_GENEVE_REMOTE]) { 1264 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) { 1265 attrtype = IFLA_GENEVE_REMOTE; 1266 goto change_notsup; 1267 } 1268 1269 info->key.u.ipv4.dst = 1270 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]); 1271 1272 if (IN_MULTICAST(ntohl(info->key.u.ipv4.dst))) { 1273 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE], 1274 "Remote IPv4 address cannot be Multicast"); 1275 return -EINVAL; 1276 } 1277 } 1278 1279 if (data[IFLA_GENEVE_REMOTE6]) { 1280 #if IS_ENABLED(CONFIG_IPV6) 1281 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) { 1282 attrtype = IFLA_GENEVE_REMOTE6; 1283 goto change_notsup; 1284 } 1285 1286 info->mode = IP_TUNNEL_INFO_IPV6; 1287 info->key.u.ipv6.dst = 1288 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]); 1289 1290 if (ipv6_addr_type(&info->key.u.ipv6.dst) & 1291 IPV6_ADDR_LINKLOCAL) { 1292 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1293 "Remote IPv6 address cannot be link-local"); 1294 return -EINVAL; 1295 } 1296 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) { 1297 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1298 "Remote IPv6 address cannot be Multicast"); 1299 return -EINVAL; 1300 } 1301 info->key.tun_flags |= TUNNEL_CSUM; 1302 *use_udp6_rx_checksums = true; 1303 #else 1304 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1305 "IPv6 support not enabled in the kernel"); 1306 return -EPFNOSUPPORT; 1307 #endif 1308 } 1309 1310 if (data[IFLA_GENEVE_ID]) { 1311 __u32 vni; 1312 __u8 tvni[3]; 1313 __be64 tunid; 1314 1315 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1316 tvni[0] = (vni & 0x00ff0000) >> 16; 1317 tvni[1] = (vni & 0x0000ff00) >> 8; 1318 tvni[2] = vni & 0x000000ff; 1319 1320 tunid = vni_to_tunnel_id(tvni); 1321 if (changelink && (tunid != info->key.tun_id)) { 1322 attrtype = IFLA_GENEVE_ID; 1323 goto change_notsup; 1324 } 1325 info->key.tun_id = tunid; 1326 } 1327 1328 if (data[IFLA_GENEVE_TTL_INHERIT]) { 1329 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT])) 1330 *ttl_inherit = true; 1331 else 1332 *ttl_inherit = false; 1333 } else if (data[IFLA_GENEVE_TTL]) { 1334 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]); 1335 *ttl_inherit = false; 1336 } 1337 1338 if (data[IFLA_GENEVE_TOS]) 1339 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]); 1340 1341 if (data[IFLA_GENEVE_LABEL]) { 1342 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) & 1343 IPV6_FLOWLABEL_MASK; 1344 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) { 1345 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL], 1346 "Label attribute only applies for IPv6 Geneve devices"); 1347 return -EINVAL; 1348 } 1349 } 1350 1351 if (data[IFLA_GENEVE_PORT]) { 1352 if (changelink) { 1353 attrtype = IFLA_GENEVE_PORT; 1354 goto change_notsup; 1355 } 1356 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]); 1357 } 1358 1359 if (data[IFLA_GENEVE_COLLECT_METADATA]) { 1360 if (changelink) { 1361 attrtype = IFLA_GENEVE_COLLECT_METADATA; 1362 goto change_notsup; 1363 } 1364 *metadata = true; 1365 } 1366 1367 if (data[IFLA_GENEVE_UDP_CSUM]) { 1368 if (changelink) { 1369 attrtype = IFLA_GENEVE_UDP_CSUM; 1370 goto change_notsup; 1371 } 1372 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM])) 1373 info->key.tun_flags |= TUNNEL_CSUM; 1374 } 1375 1376 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) { 1377 #if IS_ENABLED(CONFIG_IPV6) 1378 if (changelink) { 1379 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX; 1380 goto change_notsup; 1381 } 1382 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX])) 1383 info->key.tun_flags &= ~TUNNEL_CSUM; 1384 #else 1385 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX], 1386 "IPv6 support not enabled in the kernel"); 1387 return -EPFNOSUPPORT; 1388 #endif 1389 } 1390 1391 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) { 1392 #if IS_ENABLED(CONFIG_IPV6) 1393 if (changelink) { 1394 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX; 1395 goto change_notsup; 1396 } 1397 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX])) 1398 *use_udp6_rx_checksums = false; 1399 #else 1400 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX], 1401 "IPv6 support not enabled in the kernel"); 1402 return -EPFNOSUPPORT; 1403 #endif 1404 } 1405 1406 return 0; 1407 change_notsup: 1408 NL_SET_ERR_MSG_ATTR(extack, data[attrtype], 1409 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported"); 1410 return -EOPNOTSUPP; 1411 } 1412 1413 static void geneve_link_config(struct net_device *dev, 1414 struct ip_tunnel_info *info, struct nlattr *tb[]) 1415 { 1416 struct geneve_dev *geneve = netdev_priv(dev); 1417 int ldev_mtu = 0; 1418 1419 if (tb[IFLA_MTU]) { 1420 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU])); 1421 return; 1422 } 1423 1424 switch (ip_tunnel_info_af(info)) { 1425 case AF_INET: { 1426 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst }; 1427 struct rtable *rt = ip_route_output_key(geneve->net, &fl4); 1428 1429 if (!IS_ERR(rt) && rt->dst.dev) { 1430 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN; 1431 ip_rt_put(rt); 1432 } 1433 break; 1434 } 1435 #if IS_ENABLED(CONFIG_IPV6) 1436 case AF_INET6: { 1437 struct rt6_info *rt = rt6_lookup(geneve->net, 1438 &info->key.u.ipv6.dst, NULL, 0, 1439 NULL, 0); 1440 1441 if (rt && rt->dst.dev) 1442 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN; 1443 ip6_rt_put(rt); 1444 break; 1445 } 1446 #endif 1447 } 1448 1449 if (ldev_mtu <= 0) 1450 return; 1451 1452 geneve_change_mtu(dev, ldev_mtu - info->options_len); 1453 } 1454 1455 static int geneve_newlink(struct net *net, struct net_device *dev, 1456 struct nlattr *tb[], struct nlattr *data[], 1457 struct netlink_ext_ack *extack) 1458 { 1459 bool use_udp6_rx_checksums = false; 1460 struct ip_tunnel_info info; 1461 bool ttl_inherit = false; 1462 bool metadata = false; 1463 int err; 1464 1465 init_tnl_info(&info, GENEVE_UDP_PORT); 1466 err = geneve_nl2info(tb, data, extack, &info, &metadata, 1467 &use_udp6_rx_checksums, &ttl_inherit, false); 1468 if (err) 1469 return err; 1470 1471 err = geneve_configure(net, dev, extack, &info, metadata, 1472 use_udp6_rx_checksums, ttl_inherit); 1473 if (err) 1474 return err; 1475 1476 geneve_link_config(dev, &info, tb); 1477 1478 return 0; 1479 } 1480 1481 /* Quiesces the geneve device data path for both TX and RX. 1482 * 1483 * On transmit geneve checks for non-NULL geneve_sock before it proceeds. 1484 * So, if we set that socket to NULL under RCU and wait for synchronize_net() 1485 * to complete for the existing set of in-flight packets to be transmitted, 1486 * then we would have quiesced the transmit data path. All the future packets 1487 * will get dropped until we unquiesce the data path. 1488 * 1489 * On receive geneve dereference the geneve_sock stashed in the socket. So, 1490 * if we set that to NULL under RCU and wait for synchronize_net() to 1491 * complete, then we would have quiesced the receive data path. 1492 */ 1493 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4, 1494 struct geneve_sock **gs6) 1495 { 1496 *gs4 = rtnl_dereference(geneve->sock4); 1497 rcu_assign_pointer(geneve->sock4, NULL); 1498 if (*gs4) 1499 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL); 1500 #if IS_ENABLED(CONFIG_IPV6) 1501 *gs6 = rtnl_dereference(geneve->sock6); 1502 rcu_assign_pointer(geneve->sock6, NULL); 1503 if (*gs6) 1504 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL); 1505 #else 1506 *gs6 = NULL; 1507 #endif 1508 synchronize_net(); 1509 } 1510 1511 /* Resumes the geneve device data path for both TX and RX. */ 1512 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4, 1513 struct geneve_sock __maybe_unused *gs6) 1514 { 1515 rcu_assign_pointer(geneve->sock4, gs4); 1516 if (gs4) 1517 rcu_assign_sk_user_data(gs4->sock->sk, gs4); 1518 #if IS_ENABLED(CONFIG_IPV6) 1519 rcu_assign_pointer(geneve->sock6, gs6); 1520 if (gs6) 1521 rcu_assign_sk_user_data(gs6->sock->sk, gs6); 1522 #endif 1523 synchronize_net(); 1524 } 1525 1526 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[], 1527 struct nlattr *data[], 1528 struct netlink_ext_ack *extack) 1529 { 1530 struct geneve_dev *geneve = netdev_priv(dev); 1531 struct geneve_sock *gs4, *gs6; 1532 struct ip_tunnel_info info; 1533 bool metadata; 1534 bool use_udp6_rx_checksums; 1535 bool ttl_inherit; 1536 int err; 1537 1538 /* If the geneve device is configured for metadata (or externally 1539 * controlled, for example, OVS), then nothing can be changed. 1540 */ 1541 if (geneve->collect_md) 1542 return -EOPNOTSUPP; 1543 1544 /* Start with the existing info. */ 1545 memcpy(&info, &geneve->info, sizeof(info)); 1546 metadata = geneve->collect_md; 1547 use_udp6_rx_checksums = geneve->use_udp6_rx_checksums; 1548 ttl_inherit = geneve->ttl_inherit; 1549 err = geneve_nl2info(tb, data, extack, &info, &metadata, 1550 &use_udp6_rx_checksums, &ttl_inherit, true); 1551 if (err) 1552 return err; 1553 1554 if (!geneve_dst_addr_equal(&geneve->info, &info)) { 1555 dst_cache_reset(&info.dst_cache); 1556 geneve_link_config(dev, &info, tb); 1557 } 1558 1559 geneve_quiesce(geneve, &gs4, &gs6); 1560 geneve->info = info; 1561 geneve->collect_md = metadata; 1562 geneve->use_udp6_rx_checksums = use_udp6_rx_checksums; 1563 geneve->ttl_inherit = ttl_inherit; 1564 geneve_unquiesce(geneve, gs4, gs6); 1565 1566 return 0; 1567 } 1568 1569 static void geneve_dellink(struct net_device *dev, struct list_head *head) 1570 { 1571 struct geneve_dev *geneve = netdev_priv(dev); 1572 1573 list_del(&geneve->next); 1574 unregister_netdevice_queue(dev, head); 1575 } 1576 1577 static size_t geneve_get_size(const struct net_device *dev) 1578 { 1579 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */ 1580 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */ 1581 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */ 1582 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */ 1583 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */ 1584 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */ 1585 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */ 1586 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */ 1587 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */ 1588 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */ 1589 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */ 1590 0; 1591 } 1592 1593 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev) 1594 { 1595 struct geneve_dev *geneve = netdev_priv(dev); 1596 struct ip_tunnel_info *info = &geneve->info; 1597 bool ttl_inherit = geneve->ttl_inherit; 1598 bool metadata = geneve->collect_md; 1599 __u8 tmp_vni[3]; 1600 __u32 vni; 1601 1602 tunnel_id_to_vni(info->key.tun_id, tmp_vni); 1603 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2]; 1604 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni)) 1605 goto nla_put_failure; 1606 1607 if (!metadata && ip_tunnel_info_af(info) == AF_INET) { 1608 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE, 1609 info->key.u.ipv4.dst)) 1610 goto nla_put_failure; 1611 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM, 1612 !!(info->key.tun_flags & TUNNEL_CSUM))) 1613 goto nla_put_failure; 1614 1615 #if IS_ENABLED(CONFIG_IPV6) 1616 } else if (!metadata) { 1617 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6, 1618 &info->key.u.ipv6.dst)) 1619 goto nla_put_failure; 1620 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX, 1621 !(info->key.tun_flags & TUNNEL_CSUM))) 1622 goto nla_put_failure; 1623 #endif 1624 } 1625 1626 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) || 1627 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) || 1628 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label)) 1629 goto nla_put_failure; 1630 1631 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst)) 1632 goto nla_put_failure; 1633 1634 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA)) 1635 goto nla_put_failure; 1636 1637 #if IS_ENABLED(CONFIG_IPV6) 1638 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX, 1639 !geneve->use_udp6_rx_checksums)) 1640 goto nla_put_failure; 1641 #endif 1642 1643 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit)) 1644 goto nla_put_failure; 1645 1646 return 0; 1647 1648 nla_put_failure: 1649 return -EMSGSIZE; 1650 } 1651 1652 static struct rtnl_link_ops geneve_link_ops __read_mostly = { 1653 .kind = "geneve", 1654 .maxtype = IFLA_GENEVE_MAX, 1655 .policy = geneve_policy, 1656 .priv_size = sizeof(struct geneve_dev), 1657 .setup = geneve_setup, 1658 .validate = geneve_validate, 1659 .newlink = geneve_newlink, 1660 .changelink = geneve_changelink, 1661 .dellink = geneve_dellink, 1662 .get_size = geneve_get_size, 1663 .fill_info = geneve_fill_info, 1664 }; 1665 1666 struct net_device *geneve_dev_create_fb(struct net *net, const char *name, 1667 u8 name_assign_type, u16 dst_port) 1668 { 1669 struct nlattr *tb[IFLA_MAX + 1]; 1670 struct ip_tunnel_info info; 1671 struct net_device *dev; 1672 LIST_HEAD(list_kill); 1673 int err; 1674 1675 memset(tb, 0, sizeof(tb)); 1676 dev = rtnl_create_link(net, name, name_assign_type, 1677 &geneve_link_ops, tb); 1678 if (IS_ERR(dev)) 1679 return dev; 1680 1681 init_tnl_info(&info, dst_port); 1682 err = geneve_configure(net, dev, NULL, &info, true, true, false); 1683 if (err) { 1684 free_netdev(dev); 1685 return ERR_PTR(err); 1686 } 1687 1688 /* openvswitch users expect packet sizes to be unrestricted, 1689 * so set the largest MTU we can. 1690 */ 1691 err = geneve_change_mtu(dev, IP_MAX_MTU); 1692 if (err) 1693 goto err; 1694 1695 err = rtnl_configure_link(dev, NULL); 1696 if (err < 0) 1697 goto err; 1698 1699 return dev; 1700 err: 1701 geneve_dellink(dev, &list_kill); 1702 unregister_netdevice_many(&list_kill); 1703 return ERR_PTR(err); 1704 } 1705 EXPORT_SYMBOL_GPL(geneve_dev_create_fb); 1706 1707 static int geneve_netdevice_event(struct notifier_block *unused, 1708 unsigned long event, void *ptr) 1709 { 1710 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1711 1712 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO || 1713 event == NETDEV_UDP_TUNNEL_DROP_INFO) { 1714 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO); 1715 } else if (event == NETDEV_UNREGISTER) { 1716 geneve_offload_rx_ports(dev, false); 1717 } else if (event == NETDEV_REGISTER) { 1718 geneve_offload_rx_ports(dev, true); 1719 } 1720 1721 return NOTIFY_DONE; 1722 } 1723 1724 static struct notifier_block geneve_notifier_block __read_mostly = { 1725 .notifier_call = geneve_netdevice_event, 1726 }; 1727 1728 static __net_init int geneve_init_net(struct net *net) 1729 { 1730 struct geneve_net *gn = net_generic(net, geneve_net_id); 1731 1732 INIT_LIST_HEAD(&gn->geneve_list); 1733 INIT_LIST_HEAD(&gn->sock_list); 1734 return 0; 1735 } 1736 1737 static void geneve_destroy_tunnels(struct net *net, struct list_head *head) 1738 { 1739 struct geneve_net *gn = net_generic(net, geneve_net_id); 1740 struct geneve_dev *geneve, *next; 1741 struct net_device *dev, *aux; 1742 1743 /* gather any geneve devices that were moved into this ns */ 1744 for_each_netdev_safe(net, dev, aux) 1745 if (dev->rtnl_link_ops == &geneve_link_ops) 1746 unregister_netdevice_queue(dev, head); 1747 1748 /* now gather any other geneve devices that were created in this ns */ 1749 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) { 1750 /* If geneve->dev is in the same netns, it was already added 1751 * to the list by the previous loop. 1752 */ 1753 if (!net_eq(dev_net(geneve->dev), net)) 1754 unregister_netdevice_queue(geneve->dev, head); 1755 } 1756 1757 WARN_ON_ONCE(!list_empty(&gn->sock_list)); 1758 } 1759 1760 static void __net_exit geneve_exit_batch_net(struct list_head *net_list) 1761 { 1762 struct net *net; 1763 LIST_HEAD(list); 1764 1765 rtnl_lock(); 1766 list_for_each_entry(net, net_list, exit_list) 1767 geneve_destroy_tunnels(net, &list); 1768 1769 /* unregister the devices gathered above */ 1770 unregister_netdevice_many(&list); 1771 rtnl_unlock(); 1772 } 1773 1774 static struct pernet_operations geneve_net_ops = { 1775 .init = geneve_init_net, 1776 .exit_batch = geneve_exit_batch_net, 1777 .id = &geneve_net_id, 1778 .size = sizeof(struct geneve_net), 1779 }; 1780 1781 static int __init geneve_init_module(void) 1782 { 1783 int rc; 1784 1785 rc = register_pernet_subsys(&geneve_net_ops); 1786 if (rc) 1787 goto out1; 1788 1789 rc = register_netdevice_notifier(&geneve_notifier_block); 1790 if (rc) 1791 goto out2; 1792 1793 rc = rtnl_link_register(&geneve_link_ops); 1794 if (rc) 1795 goto out3; 1796 1797 return 0; 1798 out3: 1799 unregister_netdevice_notifier(&geneve_notifier_block); 1800 out2: 1801 unregister_pernet_subsys(&geneve_net_ops); 1802 out1: 1803 return rc; 1804 } 1805 late_initcall(geneve_init_module); 1806 1807 static void __exit geneve_cleanup_module(void) 1808 { 1809 rtnl_link_unregister(&geneve_link_ops); 1810 unregister_netdevice_notifier(&geneve_notifier_block); 1811 unregister_pernet_subsys(&geneve_net_ops); 1812 } 1813 module_exit(geneve_cleanup_module); 1814 1815 MODULE_LICENSE("GPL"); 1816 MODULE_VERSION(GENEVE_NETDEV_VER); 1817 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>"); 1818 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic"); 1819 MODULE_ALIAS_RTNL_LINK("geneve"); 1820