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 if (skb_dst(skb)) { 834 int mtu = dst_mtu(&rt->dst) - GENEVE_IPV4_HLEN - 835 info->options_len; 836 837 skb_dst_update_pmtu(skb, mtu); 838 } 839 840 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 841 if (geneve->collect_md) { 842 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 843 ttl = key->ttl; 844 } else { 845 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb); 846 ttl = key->ttl ? : ip4_dst_hoplimit(&rt->dst); 847 } 848 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0; 849 850 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr)); 851 if (unlikely(err)) 852 return err; 853 854 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr, 855 tos, ttl, df, sport, geneve->info.key.tp_dst, 856 !net_eq(geneve->net, dev_net(geneve->dev)), 857 !(info->key.tun_flags & TUNNEL_CSUM)); 858 return 0; 859 } 860 861 #if IS_ENABLED(CONFIG_IPV6) 862 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev, 863 struct geneve_dev *geneve, 864 const struct ip_tunnel_info *info) 865 { 866 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 867 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 868 const struct ip_tunnel_key *key = &info->key; 869 struct dst_entry *dst = NULL; 870 struct flowi6 fl6; 871 __u8 prio, ttl; 872 __be16 sport; 873 int err; 874 875 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info); 876 if (IS_ERR(dst)) 877 return PTR_ERR(dst); 878 879 if (skb_dst(skb)) { 880 int mtu = dst_mtu(dst) - GENEVE_IPV6_HLEN - info->options_len; 881 882 skb_dst_update_pmtu(skb, mtu); 883 } 884 885 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 886 if (geneve->collect_md) { 887 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 888 ttl = key->ttl; 889 } else { 890 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel), 891 ip_hdr(skb), skb); 892 ttl = key->ttl ? : ip6_dst_hoplimit(dst); 893 } 894 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr)); 895 if (unlikely(err)) 896 return err; 897 898 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev, 899 &fl6.saddr, &fl6.daddr, prio, ttl, 900 info->key.label, sport, geneve->info.key.tp_dst, 901 !(info->key.tun_flags & TUNNEL_CSUM)); 902 return 0; 903 } 904 #endif 905 906 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev) 907 { 908 struct geneve_dev *geneve = netdev_priv(dev); 909 struct ip_tunnel_info *info = NULL; 910 int err; 911 912 if (geneve->collect_md) { 913 info = skb_tunnel_info(skb); 914 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) { 915 err = -EINVAL; 916 netdev_dbg(dev, "no tunnel metadata\n"); 917 goto tx_error; 918 } 919 } else { 920 info = &geneve->info; 921 } 922 923 rcu_read_lock(); 924 #if IS_ENABLED(CONFIG_IPV6) 925 if (info->mode & IP_TUNNEL_INFO_IPV6) 926 err = geneve6_xmit_skb(skb, dev, geneve, info); 927 else 928 #endif 929 err = geneve_xmit_skb(skb, dev, geneve, info); 930 rcu_read_unlock(); 931 932 if (likely(!err)) 933 return NETDEV_TX_OK; 934 tx_error: 935 dev_kfree_skb(skb); 936 937 if (err == -ELOOP) 938 dev->stats.collisions++; 939 else if (err == -ENETUNREACH) 940 dev->stats.tx_carrier_errors++; 941 942 dev->stats.tx_errors++; 943 return NETDEV_TX_OK; 944 } 945 946 static int geneve_change_mtu(struct net_device *dev, int new_mtu) 947 { 948 if (new_mtu > dev->max_mtu) 949 new_mtu = dev->max_mtu; 950 else if (new_mtu < dev->min_mtu) 951 new_mtu = dev->min_mtu; 952 953 dev->mtu = new_mtu; 954 return 0; 955 } 956 957 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 958 { 959 struct ip_tunnel_info *info = skb_tunnel_info(skb); 960 struct geneve_dev *geneve = netdev_priv(dev); 961 962 if (ip_tunnel_info_af(info) == AF_INET) { 963 struct rtable *rt; 964 struct flowi4 fl4; 965 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 966 967 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info); 968 if (IS_ERR(rt)) 969 return PTR_ERR(rt); 970 971 ip_rt_put(rt); 972 info->key.u.ipv4.src = fl4.saddr; 973 #if IS_ENABLED(CONFIG_IPV6) 974 } else if (ip_tunnel_info_af(info) == AF_INET6) { 975 struct dst_entry *dst; 976 struct flowi6 fl6; 977 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 978 979 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info); 980 if (IS_ERR(dst)) 981 return PTR_ERR(dst); 982 983 dst_release(dst); 984 info->key.u.ipv6.src = fl6.saddr; 985 #endif 986 } else { 987 return -EINVAL; 988 } 989 990 info->key.tp_src = udp_flow_src_port(geneve->net, skb, 991 1, USHRT_MAX, true); 992 info->key.tp_dst = geneve->info.key.tp_dst; 993 return 0; 994 } 995 996 static const struct net_device_ops geneve_netdev_ops = { 997 .ndo_init = geneve_init, 998 .ndo_uninit = geneve_uninit, 999 .ndo_open = geneve_open, 1000 .ndo_stop = geneve_stop, 1001 .ndo_start_xmit = geneve_xmit, 1002 .ndo_get_stats64 = ip_tunnel_get_stats64, 1003 .ndo_change_mtu = geneve_change_mtu, 1004 .ndo_validate_addr = eth_validate_addr, 1005 .ndo_set_mac_address = eth_mac_addr, 1006 .ndo_fill_metadata_dst = geneve_fill_metadata_dst, 1007 }; 1008 1009 static void geneve_get_drvinfo(struct net_device *dev, 1010 struct ethtool_drvinfo *drvinfo) 1011 { 1012 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version)); 1013 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver)); 1014 } 1015 1016 static const struct ethtool_ops geneve_ethtool_ops = { 1017 .get_drvinfo = geneve_get_drvinfo, 1018 .get_link = ethtool_op_get_link, 1019 }; 1020 1021 /* Info for udev, that this is a virtual tunnel endpoint */ 1022 static struct device_type geneve_type = { 1023 .name = "geneve", 1024 }; 1025 1026 /* Calls the ndo_udp_tunnel_add of the caller in order to 1027 * supply the listening GENEVE udp ports. Callers are expected 1028 * to implement the ndo_udp_tunnel_add. 1029 */ 1030 static void geneve_offload_rx_ports(struct net_device *dev, bool push) 1031 { 1032 struct net *net = dev_net(dev); 1033 struct geneve_net *gn = net_generic(net, geneve_net_id); 1034 struct geneve_sock *gs; 1035 1036 rcu_read_lock(); 1037 list_for_each_entry_rcu(gs, &gn->sock_list, list) { 1038 if (push) { 1039 udp_tunnel_push_rx_port(dev, gs->sock, 1040 UDP_TUNNEL_TYPE_GENEVE); 1041 } else { 1042 udp_tunnel_drop_rx_port(dev, gs->sock, 1043 UDP_TUNNEL_TYPE_GENEVE); 1044 } 1045 } 1046 rcu_read_unlock(); 1047 } 1048 1049 /* Initialize the device structure. */ 1050 static void geneve_setup(struct net_device *dev) 1051 { 1052 ether_setup(dev); 1053 1054 dev->netdev_ops = &geneve_netdev_ops; 1055 dev->ethtool_ops = &geneve_ethtool_ops; 1056 dev->needs_free_netdev = true; 1057 1058 SET_NETDEV_DEVTYPE(dev, &geneve_type); 1059 1060 dev->features |= NETIF_F_LLTX; 1061 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM; 1062 dev->features |= NETIF_F_RXCSUM; 1063 dev->features |= NETIF_F_GSO_SOFTWARE; 1064 1065 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM; 1066 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 1067 1068 /* MTU range: 68 - (something less than 65535) */ 1069 dev->min_mtu = ETH_MIN_MTU; 1070 /* The max_mtu calculation does not take account of GENEVE 1071 * options, to avoid excluding potentially valid 1072 * configurations. This will be further reduced by IPvX hdr size. 1073 */ 1074 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len; 1075 1076 netif_keep_dst(dev); 1077 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1078 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE; 1079 eth_hw_addr_random(dev); 1080 } 1081 1082 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = { 1083 [IFLA_GENEVE_ID] = { .type = NLA_U32 }, 1084 [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) }, 1085 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) }, 1086 [IFLA_GENEVE_TTL] = { .type = NLA_U8 }, 1087 [IFLA_GENEVE_TOS] = { .type = NLA_U8 }, 1088 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 }, 1089 [IFLA_GENEVE_PORT] = { .type = NLA_U16 }, 1090 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1091 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 }, 1092 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 }, 1093 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 }, 1094 }; 1095 1096 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[], 1097 struct netlink_ext_ack *extack) 1098 { 1099 if (tb[IFLA_ADDRESS]) { 1100 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1101 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1102 "Provided link layer address is not Ethernet"); 1103 return -EINVAL; 1104 } 1105 1106 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1107 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1108 "Provided Ethernet address is not unicast"); 1109 return -EADDRNOTAVAIL; 1110 } 1111 } 1112 1113 if (!data) { 1114 NL_SET_ERR_MSG(extack, 1115 "Not enough attributes provided to perform the operation"); 1116 return -EINVAL; 1117 } 1118 1119 if (data[IFLA_GENEVE_ID]) { 1120 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1121 1122 if (vni >= GENEVE_N_VID) { 1123 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID], 1124 "Geneve ID must be lower than 16777216"); 1125 return -ERANGE; 1126 } 1127 } 1128 1129 return 0; 1130 } 1131 1132 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn, 1133 const struct ip_tunnel_info *info, 1134 bool *tun_on_same_port, 1135 bool *tun_collect_md) 1136 { 1137 struct geneve_dev *geneve, *t = NULL; 1138 1139 *tun_on_same_port = false; 1140 *tun_collect_md = false; 1141 list_for_each_entry(geneve, &gn->geneve_list, next) { 1142 if (info->key.tp_dst == geneve->info.key.tp_dst) { 1143 *tun_collect_md = geneve->collect_md; 1144 *tun_on_same_port = true; 1145 } 1146 if (info->key.tun_id == geneve->info.key.tun_id && 1147 info->key.tp_dst == geneve->info.key.tp_dst && 1148 !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u))) 1149 t = geneve; 1150 } 1151 return t; 1152 } 1153 1154 static bool is_tnl_info_zero(const struct ip_tunnel_info *info) 1155 { 1156 return !(info->key.tun_id || info->key.tun_flags || info->key.tos || 1157 info->key.ttl || info->key.label || info->key.tp_src || 1158 memchr_inv(&info->key.u, 0, sizeof(info->key.u))); 1159 } 1160 1161 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a, 1162 struct ip_tunnel_info *b) 1163 { 1164 if (ip_tunnel_info_af(a) == AF_INET) 1165 return a->key.u.ipv4.dst == b->key.u.ipv4.dst; 1166 else 1167 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst); 1168 } 1169 1170 static int geneve_configure(struct net *net, struct net_device *dev, 1171 struct netlink_ext_ack *extack, 1172 const struct ip_tunnel_info *info, 1173 bool metadata, bool ipv6_rx_csum) 1174 { 1175 struct geneve_net *gn = net_generic(net, geneve_net_id); 1176 struct geneve_dev *t, *geneve = netdev_priv(dev); 1177 bool tun_collect_md, tun_on_same_port; 1178 int err, encap_len; 1179 1180 if (metadata && !is_tnl_info_zero(info)) { 1181 NL_SET_ERR_MSG(extack, 1182 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified"); 1183 return -EINVAL; 1184 } 1185 1186 geneve->net = net; 1187 geneve->dev = dev; 1188 1189 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md); 1190 if (t) 1191 return -EBUSY; 1192 1193 /* make enough headroom for basic scenario */ 1194 encap_len = GENEVE_BASE_HLEN + ETH_HLEN; 1195 if (!metadata && ip_tunnel_info_af(info) == AF_INET) { 1196 encap_len += sizeof(struct iphdr); 1197 dev->max_mtu -= sizeof(struct iphdr); 1198 } else { 1199 encap_len += sizeof(struct ipv6hdr); 1200 dev->max_mtu -= sizeof(struct ipv6hdr); 1201 } 1202 dev->needed_headroom = encap_len + ETH_HLEN; 1203 1204 if (metadata) { 1205 if (tun_on_same_port) { 1206 NL_SET_ERR_MSG(extack, 1207 "There can be only one externally controlled device on a destination port"); 1208 return -EPERM; 1209 } 1210 } else { 1211 if (tun_collect_md) { 1212 NL_SET_ERR_MSG(extack, 1213 "There already exists an externally controlled device on this destination port"); 1214 return -EPERM; 1215 } 1216 } 1217 1218 dst_cache_reset(&geneve->info.dst_cache); 1219 geneve->info = *info; 1220 geneve->collect_md = metadata; 1221 geneve->use_udp6_rx_checksums = ipv6_rx_csum; 1222 1223 err = register_netdevice(dev); 1224 if (err) 1225 return err; 1226 1227 list_add(&geneve->next, &gn->geneve_list); 1228 return 0; 1229 } 1230 1231 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port) 1232 { 1233 memset(info, 0, sizeof(*info)); 1234 info->key.tp_dst = htons(dst_port); 1235 } 1236 1237 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[], 1238 struct netlink_ext_ack *extack, 1239 struct ip_tunnel_info *info, bool *metadata, 1240 bool *use_udp6_rx_checksums, bool changelink) 1241 { 1242 int attrtype; 1243 1244 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) { 1245 NL_SET_ERR_MSG(extack, 1246 "Cannot specify both IPv4 and IPv6 Remote addresses"); 1247 return -EINVAL; 1248 } 1249 1250 if (data[IFLA_GENEVE_REMOTE]) { 1251 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) { 1252 attrtype = IFLA_GENEVE_REMOTE; 1253 goto change_notsup; 1254 } 1255 1256 info->key.u.ipv4.dst = 1257 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]); 1258 1259 if (IN_MULTICAST(ntohl(info->key.u.ipv4.dst))) { 1260 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE], 1261 "Remote IPv4 address cannot be Multicast"); 1262 return -EINVAL; 1263 } 1264 } 1265 1266 if (data[IFLA_GENEVE_REMOTE6]) { 1267 #if IS_ENABLED(CONFIG_IPV6) 1268 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) { 1269 attrtype = IFLA_GENEVE_REMOTE6; 1270 goto change_notsup; 1271 } 1272 1273 info->mode = IP_TUNNEL_INFO_IPV6; 1274 info->key.u.ipv6.dst = 1275 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]); 1276 1277 if (ipv6_addr_type(&info->key.u.ipv6.dst) & 1278 IPV6_ADDR_LINKLOCAL) { 1279 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1280 "Remote IPv6 address cannot be link-local"); 1281 return -EINVAL; 1282 } 1283 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) { 1284 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1285 "Remote IPv6 address cannot be Multicast"); 1286 return -EINVAL; 1287 } 1288 info->key.tun_flags |= TUNNEL_CSUM; 1289 *use_udp6_rx_checksums = true; 1290 #else 1291 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1292 "IPv6 support not enabled in the kernel"); 1293 return -EPFNOSUPPORT; 1294 #endif 1295 } 1296 1297 if (data[IFLA_GENEVE_ID]) { 1298 __u32 vni; 1299 __u8 tvni[3]; 1300 __be64 tunid; 1301 1302 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1303 tvni[0] = (vni & 0x00ff0000) >> 16; 1304 tvni[1] = (vni & 0x0000ff00) >> 8; 1305 tvni[2] = vni & 0x000000ff; 1306 1307 tunid = vni_to_tunnel_id(tvni); 1308 if (changelink && (tunid != info->key.tun_id)) { 1309 attrtype = IFLA_GENEVE_ID; 1310 goto change_notsup; 1311 } 1312 info->key.tun_id = tunid; 1313 } 1314 1315 if (data[IFLA_GENEVE_TTL]) 1316 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]); 1317 1318 if (data[IFLA_GENEVE_TOS]) 1319 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]); 1320 1321 if (data[IFLA_GENEVE_LABEL]) { 1322 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) & 1323 IPV6_FLOWLABEL_MASK; 1324 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) { 1325 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL], 1326 "Label attribute only applies for IPv6 Geneve devices"); 1327 return -EINVAL; 1328 } 1329 } 1330 1331 if (data[IFLA_GENEVE_PORT]) { 1332 if (changelink) { 1333 attrtype = IFLA_GENEVE_PORT; 1334 goto change_notsup; 1335 } 1336 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]); 1337 } 1338 1339 if (data[IFLA_GENEVE_COLLECT_METADATA]) { 1340 if (changelink) { 1341 attrtype = IFLA_GENEVE_COLLECT_METADATA; 1342 goto change_notsup; 1343 } 1344 *metadata = true; 1345 } 1346 1347 if (data[IFLA_GENEVE_UDP_CSUM]) { 1348 if (changelink) { 1349 attrtype = IFLA_GENEVE_UDP_CSUM; 1350 goto change_notsup; 1351 } 1352 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM])) 1353 info->key.tun_flags |= TUNNEL_CSUM; 1354 } 1355 1356 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) { 1357 #if IS_ENABLED(CONFIG_IPV6) 1358 if (changelink) { 1359 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX; 1360 goto change_notsup; 1361 } 1362 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX])) 1363 info->key.tun_flags &= ~TUNNEL_CSUM; 1364 #else 1365 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX], 1366 "IPv6 support not enabled in the kernel"); 1367 return -EPFNOSUPPORT; 1368 #endif 1369 } 1370 1371 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) { 1372 #if IS_ENABLED(CONFIG_IPV6) 1373 if (changelink) { 1374 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX; 1375 goto change_notsup; 1376 } 1377 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX])) 1378 *use_udp6_rx_checksums = false; 1379 #else 1380 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX], 1381 "IPv6 support not enabled in the kernel"); 1382 return -EPFNOSUPPORT; 1383 #endif 1384 } 1385 1386 return 0; 1387 change_notsup: 1388 NL_SET_ERR_MSG_ATTR(extack, data[attrtype], 1389 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported"); 1390 return -EOPNOTSUPP; 1391 } 1392 1393 static void geneve_link_config(struct net_device *dev, 1394 struct ip_tunnel_info *info, struct nlattr *tb[]) 1395 { 1396 struct geneve_dev *geneve = netdev_priv(dev); 1397 int ldev_mtu = 0; 1398 1399 if (tb[IFLA_MTU]) { 1400 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU])); 1401 return; 1402 } 1403 1404 switch (ip_tunnel_info_af(info)) { 1405 case AF_INET: { 1406 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst }; 1407 struct rtable *rt = ip_route_output_key(geneve->net, &fl4); 1408 1409 if (!IS_ERR(rt) && rt->dst.dev) { 1410 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN; 1411 ip_rt_put(rt); 1412 } 1413 break; 1414 } 1415 #if IS_ENABLED(CONFIG_IPV6) 1416 case AF_INET6: { 1417 struct rt6_info *rt = rt6_lookup(geneve->net, 1418 &info->key.u.ipv6.dst, NULL, 0, 1419 NULL, 0); 1420 1421 if (rt && rt->dst.dev) 1422 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN; 1423 ip6_rt_put(rt); 1424 break; 1425 } 1426 #endif 1427 } 1428 1429 if (ldev_mtu <= 0) 1430 return; 1431 1432 geneve_change_mtu(dev, ldev_mtu - info->options_len); 1433 } 1434 1435 static int geneve_newlink(struct net *net, struct net_device *dev, 1436 struct nlattr *tb[], struct nlattr *data[], 1437 struct netlink_ext_ack *extack) 1438 { 1439 bool use_udp6_rx_checksums = false; 1440 struct ip_tunnel_info info; 1441 bool metadata = false; 1442 int err; 1443 1444 init_tnl_info(&info, GENEVE_UDP_PORT); 1445 err = geneve_nl2info(tb, data, extack, &info, &metadata, 1446 &use_udp6_rx_checksums, false); 1447 if (err) 1448 return err; 1449 1450 err = geneve_configure(net, dev, extack, &info, metadata, 1451 use_udp6_rx_checksums); 1452 if (err) 1453 return err; 1454 1455 geneve_link_config(dev, &info, tb); 1456 1457 return 0; 1458 } 1459 1460 /* Quiesces the geneve device data path for both TX and RX. 1461 * 1462 * On transmit geneve checks for non-NULL geneve_sock before it proceeds. 1463 * So, if we set that socket to NULL under RCU and wait for synchronize_net() 1464 * to complete for the existing set of in-flight packets to be transmitted, 1465 * then we would have quiesced the transmit data path. All the future packets 1466 * will get dropped until we unquiesce the data path. 1467 * 1468 * On receive geneve dereference the geneve_sock stashed in the socket. So, 1469 * if we set that to NULL under RCU and wait for synchronize_net() to 1470 * complete, then we would have quiesced the receive data path. 1471 */ 1472 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4, 1473 struct geneve_sock **gs6) 1474 { 1475 *gs4 = rtnl_dereference(geneve->sock4); 1476 rcu_assign_pointer(geneve->sock4, NULL); 1477 if (*gs4) 1478 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL); 1479 #if IS_ENABLED(CONFIG_IPV6) 1480 *gs6 = rtnl_dereference(geneve->sock6); 1481 rcu_assign_pointer(geneve->sock6, NULL); 1482 if (*gs6) 1483 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL); 1484 #else 1485 *gs6 = NULL; 1486 #endif 1487 synchronize_net(); 1488 } 1489 1490 /* Resumes the geneve device data path for both TX and RX. */ 1491 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4, 1492 struct geneve_sock __maybe_unused *gs6) 1493 { 1494 rcu_assign_pointer(geneve->sock4, gs4); 1495 if (gs4) 1496 rcu_assign_sk_user_data(gs4->sock->sk, gs4); 1497 #if IS_ENABLED(CONFIG_IPV6) 1498 rcu_assign_pointer(geneve->sock6, gs6); 1499 if (gs6) 1500 rcu_assign_sk_user_data(gs6->sock->sk, gs6); 1501 #endif 1502 synchronize_net(); 1503 } 1504 1505 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[], 1506 struct nlattr *data[], 1507 struct netlink_ext_ack *extack) 1508 { 1509 struct geneve_dev *geneve = netdev_priv(dev); 1510 struct geneve_sock *gs4, *gs6; 1511 struct ip_tunnel_info info; 1512 bool metadata; 1513 bool use_udp6_rx_checksums; 1514 int err; 1515 1516 /* If the geneve device is configured for metadata (or externally 1517 * controlled, for example, OVS), then nothing can be changed. 1518 */ 1519 if (geneve->collect_md) 1520 return -EOPNOTSUPP; 1521 1522 /* Start with the existing info. */ 1523 memcpy(&info, &geneve->info, sizeof(info)); 1524 metadata = geneve->collect_md; 1525 use_udp6_rx_checksums = geneve->use_udp6_rx_checksums; 1526 err = geneve_nl2info(tb, data, extack, &info, &metadata, 1527 &use_udp6_rx_checksums, true); 1528 if (err) 1529 return err; 1530 1531 if (!geneve_dst_addr_equal(&geneve->info, &info)) { 1532 dst_cache_reset(&info.dst_cache); 1533 geneve_link_config(dev, &info, tb); 1534 } 1535 1536 geneve_quiesce(geneve, &gs4, &gs6); 1537 geneve->info = info; 1538 geneve->collect_md = metadata; 1539 geneve->use_udp6_rx_checksums = use_udp6_rx_checksums; 1540 geneve_unquiesce(geneve, gs4, gs6); 1541 1542 return 0; 1543 } 1544 1545 static void geneve_dellink(struct net_device *dev, struct list_head *head) 1546 { 1547 struct geneve_dev *geneve = netdev_priv(dev); 1548 1549 list_del(&geneve->next); 1550 unregister_netdevice_queue(dev, head); 1551 } 1552 1553 static size_t geneve_get_size(const struct net_device *dev) 1554 { 1555 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */ 1556 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */ 1557 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */ 1558 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */ 1559 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */ 1560 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */ 1561 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */ 1562 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */ 1563 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */ 1564 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */ 1565 0; 1566 } 1567 1568 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev) 1569 { 1570 struct geneve_dev *geneve = netdev_priv(dev); 1571 struct ip_tunnel_info *info = &geneve->info; 1572 bool metadata = geneve->collect_md; 1573 __u8 tmp_vni[3]; 1574 __u32 vni; 1575 1576 tunnel_id_to_vni(info->key.tun_id, tmp_vni); 1577 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2]; 1578 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni)) 1579 goto nla_put_failure; 1580 1581 if (!metadata && ip_tunnel_info_af(info) == AF_INET) { 1582 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE, 1583 info->key.u.ipv4.dst)) 1584 goto nla_put_failure; 1585 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM, 1586 !!(info->key.tun_flags & TUNNEL_CSUM))) 1587 goto nla_put_failure; 1588 1589 #if IS_ENABLED(CONFIG_IPV6) 1590 } else if (!metadata) { 1591 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6, 1592 &info->key.u.ipv6.dst)) 1593 goto nla_put_failure; 1594 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX, 1595 !(info->key.tun_flags & TUNNEL_CSUM))) 1596 goto nla_put_failure; 1597 #endif 1598 } 1599 1600 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) || 1601 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) || 1602 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label)) 1603 goto nla_put_failure; 1604 1605 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst)) 1606 goto nla_put_failure; 1607 1608 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA)) 1609 goto nla_put_failure; 1610 1611 #if IS_ENABLED(CONFIG_IPV6) 1612 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX, 1613 !geneve->use_udp6_rx_checksums)) 1614 goto nla_put_failure; 1615 #endif 1616 1617 return 0; 1618 1619 nla_put_failure: 1620 return -EMSGSIZE; 1621 } 1622 1623 static struct rtnl_link_ops geneve_link_ops __read_mostly = { 1624 .kind = "geneve", 1625 .maxtype = IFLA_GENEVE_MAX, 1626 .policy = geneve_policy, 1627 .priv_size = sizeof(struct geneve_dev), 1628 .setup = geneve_setup, 1629 .validate = geneve_validate, 1630 .newlink = geneve_newlink, 1631 .changelink = geneve_changelink, 1632 .dellink = geneve_dellink, 1633 .get_size = geneve_get_size, 1634 .fill_info = geneve_fill_info, 1635 }; 1636 1637 struct net_device *geneve_dev_create_fb(struct net *net, const char *name, 1638 u8 name_assign_type, u16 dst_port) 1639 { 1640 struct nlattr *tb[IFLA_MAX + 1]; 1641 struct ip_tunnel_info info; 1642 struct net_device *dev; 1643 LIST_HEAD(list_kill); 1644 int err; 1645 1646 memset(tb, 0, sizeof(tb)); 1647 dev = rtnl_create_link(net, name, name_assign_type, 1648 &geneve_link_ops, tb); 1649 if (IS_ERR(dev)) 1650 return dev; 1651 1652 init_tnl_info(&info, dst_port); 1653 err = geneve_configure(net, dev, NULL, &info, true, true); 1654 if (err) { 1655 free_netdev(dev); 1656 return ERR_PTR(err); 1657 } 1658 1659 /* openvswitch users expect packet sizes to be unrestricted, 1660 * so set the largest MTU we can. 1661 */ 1662 err = geneve_change_mtu(dev, IP_MAX_MTU); 1663 if (err) 1664 goto err; 1665 1666 err = rtnl_configure_link(dev, NULL); 1667 if (err < 0) 1668 goto err; 1669 1670 return dev; 1671 err: 1672 geneve_dellink(dev, &list_kill); 1673 unregister_netdevice_many(&list_kill); 1674 return ERR_PTR(err); 1675 } 1676 EXPORT_SYMBOL_GPL(geneve_dev_create_fb); 1677 1678 static int geneve_netdevice_event(struct notifier_block *unused, 1679 unsigned long event, void *ptr) 1680 { 1681 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1682 1683 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO || 1684 event == NETDEV_UDP_TUNNEL_DROP_INFO) { 1685 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO); 1686 } else if (event == NETDEV_UNREGISTER) { 1687 geneve_offload_rx_ports(dev, false); 1688 } else if (event == NETDEV_REGISTER) { 1689 geneve_offload_rx_ports(dev, true); 1690 } 1691 1692 return NOTIFY_DONE; 1693 } 1694 1695 static struct notifier_block geneve_notifier_block __read_mostly = { 1696 .notifier_call = geneve_netdevice_event, 1697 }; 1698 1699 static __net_init int geneve_init_net(struct net *net) 1700 { 1701 struct geneve_net *gn = net_generic(net, geneve_net_id); 1702 1703 INIT_LIST_HEAD(&gn->geneve_list); 1704 INIT_LIST_HEAD(&gn->sock_list); 1705 return 0; 1706 } 1707 1708 static void geneve_destroy_tunnels(struct net *net, struct list_head *head) 1709 { 1710 struct geneve_net *gn = net_generic(net, geneve_net_id); 1711 struct geneve_dev *geneve, *next; 1712 struct net_device *dev, *aux; 1713 1714 /* gather any geneve devices that were moved into this ns */ 1715 for_each_netdev_safe(net, dev, aux) 1716 if (dev->rtnl_link_ops == &geneve_link_ops) 1717 unregister_netdevice_queue(dev, head); 1718 1719 /* now gather any other geneve devices that were created in this ns */ 1720 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) { 1721 /* If geneve->dev is in the same netns, it was already added 1722 * to the list by the previous loop. 1723 */ 1724 if (!net_eq(dev_net(geneve->dev), net)) 1725 unregister_netdevice_queue(geneve->dev, head); 1726 } 1727 1728 WARN_ON_ONCE(!list_empty(&gn->sock_list)); 1729 } 1730 1731 static void __net_exit geneve_exit_batch_net(struct list_head *net_list) 1732 { 1733 struct net *net; 1734 LIST_HEAD(list); 1735 1736 rtnl_lock(); 1737 list_for_each_entry(net, net_list, exit_list) 1738 geneve_destroy_tunnels(net, &list); 1739 1740 /* unregister the devices gathered above */ 1741 unregister_netdevice_many(&list); 1742 rtnl_unlock(); 1743 } 1744 1745 static struct pernet_operations geneve_net_ops = { 1746 .init = geneve_init_net, 1747 .exit_batch = geneve_exit_batch_net, 1748 .id = &geneve_net_id, 1749 .size = sizeof(struct geneve_net), 1750 }; 1751 1752 static int __init geneve_init_module(void) 1753 { 1754 int rc; 1755 1756 rc = register_pernet_subsys(&geneve_net_ops); 1757 if (rc) 1758 goto out1; 1759 1760 rc = register_netdevice_notifier(&geneve_notifier_block); 1761 if (rc) 1762 goto out2; 1763 1764 rc = rtnl_link_register(&geneve_link_ops); 1765 if (rc) 1766 goto out3; 1767 1768 return 0; 1769 out3: 1770 unregister_netdevice_notifier(&geneve_notifier_block); 1771 out2: 1772 unregister_pernet_subsys(&geneve_net_ops); 1773 out1: 1774 return rc; 1775 } 1776 late_initcall(geneve_init_module); 1777 1778 static void __exit geneve_cleanup_module(void) 1779 { 1780 rtnl_link_unregister(&geneve_link_ops); 1781 unregister_netdevice_notifier(&geneve_notifier_block); 1782 unregister_pernet_subsys(&geneve_net_ops); 1783 } 1784 module_exit(geneve_cleanup_module); 1785 1786 MODULE_LICENSE("GPL"); 1787 MODULE_VERSION(GENEVE_NETDEV_VER); 1788 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>"); 1789 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic"); 1790 MODULE_ALIAS_RTNL_LINK("geneve"); 1791