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