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