1 /* 2 * GENEVE: Generic Network Virtualization Encapsulation 3 * 4 * Copyright (c) 2015 Red Hat, Inc. 5 * 6 * This program is free software; you can redistribute it and/or modify 7 * it under the terms of the GNU General Public License version 2 as 8 * published by the Free Software Foundation. 9 */ 10 11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 12 13 #include <linux/kernel.h> 14 #include <linux/module.h> 15 #include <linux/etherdevice.h> 16 #include <linux/hash.h> 17 #include <net/dst_metadata.h> 18 #include <net/gro_cells.h> 19 #include <net/rtnetlink.h> 20 #include <net/geneve.h> 21 #include <net/protocol.h> 22 23 #define GENEVE_NETDEV_VER "0.6" 24 25 #define GENEVE_UDP_PORT 6081 26 27 #define GENEVE_N_VID (1u << 24) 28 #define GENEVE_VID_MASK (GENEVE_N_VID - 1) 29 30 #define VNI_HASH_BITS 10 31 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS) 32 33 static bool log_ecn_error = true; 34 module_param(log_ecn_error, bool, 0644); 35 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN"); 36 37 #define GENEVE_VER 0 38 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr)) 39 40 /* per-network namespace private data for this module */ 41 struct geneve_net { 42 struct list_head geneve_list; 43 struct list_head sock_list; 44 }; 45 46 static unsigned int geneve_net_id; 47 48 struct geneve_dev_node { 49 struct hlist_node hlist; 50 struct geneve_dev *geneve; 51 }; 52 53 /* Pseudo network device */ 54 struct geneve_dev { 55 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */ 56 #if IS_ENABLED(CONFIG_IPV6) 57 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */ 58 #endif 59 struct net *net; /* netns for packet i/o */ 60 struct net_device *dev; /* netdev for geneve tunnel */ 61 struct ip_tunnel_info info; 62 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */ 63 #if IS_ENABLED(CONFIG_IPV6) 64 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */ 65 #endif 66 struct list_head next; /* geneve's per namespace list */ 67 struct gro_cells gro_cells; 68 bool collect_md; 69 bool use_udp6_rx_checksums; 70 }; 71 72 struct geneve_sock { 73 bool collect_md; 74 struct list_head list; 75 struct socket *sock; 76 struct rcu_head rcu; 77 int refcnt; 78 struct hlist_head vni_list[VNI_HASH_SIZE]; 79 }; 80 81 static inline __u32 geneve_net_vni_hash(u8 vni[3]) 82 { 83 __u32 vnid; 84 85 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2]; 86 return hash_32(vnid, VNI_HASH_BITS); 87 } 88 89 static __be64 vni_to_tunnel_id(const __u8 *vni) 90 { 91 #ifdef __BIG_ENDIAN 92 return (vni[0] << 16) | (vni[1] << 8) | vni[2]; 93 #else 94 return (__force __be64)(((__force u64)vni[0] << 40) | 95 ((__force u64)vni[1] << 48) | 96 ((__force u64)vni[2] << 56)); 97 #endif 98 } 99 100 /* Convert 64 bit tunnel ID to 24 bit VNI. */ 101 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni) 102 { 103 #ifdef __BIG_ENDIAN 104 vni[0] = (__force __u8)(tun_id >> 16); 105 vni[1] = (__force __u8)(tun_id >> 8); 106 vni[2] = (__force __u8)tun_id; 107 #else 108 vni[0] = (__force __u8)((__force u64)tun_id >> 40); 109 vni[1] = (__force __u8)((__force u64)tun_id >> 48); 110 vni[2] = (__force __u8)((__force u64)tun_id >> 56); 111 #endif 112 } 113 114 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni) 115 { 116 return !memcmp(vni, &tun_id[5], 3); 117 } 118 119 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs) 120 { 121 return gs->sock->sk->sk_family; 122 } 123 124 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs, 125 __be32 addr, u8 vni[]) 126 { 127 struct hlist_head *vni_list_head; 128 struct geneve_dev_node *node; 129 __u32 hash; 130 131 /* Find the device for this VNI */ 132 hash = geneve_net_vni_hash(vni); 133 vni_list_head = &gs->vni_list[hash]; 134 hlist_for_each_entry_rcu(node, vni_list_head, hlist) { 135 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) && 136 addr == node->geneve->info.key.u.ipv4.dst) 137 return node->geneve; 138 } 139 return NULL; 140 } 141 142 #if IS_ENABLED(CONFIG_IPV6) 143 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs, 144 struct in6_addr addr6, u8 vni[]) 145 { 146 struct hlist_head *vni_list_head; 147 struct geneve_dev_node *node; 148 __u32 hash; 149 150 /* Find the device for this VNI */ 151 hash = geneve_net_vni_hash(vni); 152 vni_list_head = &gs->vni_list[hash]; 153 hlist_for_each_entry_rcu(node, vni_list_head, hlist) { 154 if (eq_tun_id_and_vni((u8 *)&node->geneve->info.key.tun_id, vni) && 155 ipv6_addr_equal(&addr6, &node->geneve->info.key.u.ipv6.dst)) 156 return node->geneve; 157 } 158 return NULL; 159 } 160 #endif 161 162 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb) 163 { 164 return (struct genevehdr *)(udp_hdr(skb) + 1); 165 } 166 167 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs, 168 struct sk_buff *skb) 169 { 170 static u8 zero_vni[3]; 171 u8 *vni; 172 173 if (geneve_get_sk_family(gs) == AF_INET) { 174 struct iphdr *iph; 175 __be32 addr; 176 177 iph = ip_hdr(skb); /* outer IP header... */ 178 179 if (gs->collect_md) { 180 vni = zero_vni; 181 addr = 0; 182 } else { 183 vni = geneve_hdr(skb)->vni; 184 addr = iph->saddr; 185 } 186 187 return geneve_lookup(gs, addr, vni); 188 #if IS_ENABLED(CONFIG_IPV6) 189 } else if (geneve_get_sk_family(gs) == AF_INET6) { 190 static struct in6_addr zero_addr6; 191 struct ipv6hdr *ip6h; 192 struct in6_addr addr6; 193 194 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */ 195 196 if (gs->collect_md) { 197 vni = zero_vni; 198 addr6 = zero_addr6; 199 } else { 200 vni = geneve_hdr(skb)->vni; 201 addr6 = ip6h->saddr; 202 } 203 204 return geneve6_lookup(gs, addr6, vni); 205 #endif 206 } 207 return NULL; 208 } 209 210 /* geneve receive/decap routine */ 211 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs, 212 struct sk_buff *skb) 213 { 214 struct genevehdr *gnvh = geneve_hdr(skb); 215 struct metadata_dst *tun_dst = NULL; 216 struct pcpu_sw_netstats *stats; 217 unsigned int len; 218 int err = 0; 219 void *oiph; 220 221 if (ip_tunnel_collect_metadata() || gs->collect_md) { 222 __be16 flags; 223 224 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT | 225 (gnvh->oam ? TUNNEL_OAM : 0) | 226 (gnvh->critical ? TUNNEL_CRIT_OPT : 0); 227 228 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags, 229 vni_to_tunnel_id(gnvh->vni), 230 gnvh->opt_len * 4); 231 if (!tun_dst) { 232 geneve->dev->stats.rx_dropped++; 233 goto drop; 234 } 235 /* Update tunnel dst according to Geneve options. */ 236 ip_tunnel_info_opts_set(&tun_dst->u.tun_info, 237 gnvh->options, gnvh->opt_len * 4); 238 } else { 239 /* Drop packets w/ critical options, 240 * since we don't support any... 241 */ 242 if (gnvh->critical) { 243 geneve->dev->stats.rx_frame_errors++; 244 geneve->dev->stats.rx_errors++; 245 goto drop; 246 } 247 } 248 249 skb_reset_mac_header(skb); 250 skb->protocol = eth_type_trans(skb, geneve->dev); 251 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 252 253 if (tun_dst) 254 skb_dst_set(skb, &tun_dst->dst); 255 256 /* Ignore packet loops (and multicast echo) */ 257 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) { 258 geneve->dev->stats.rx_errors++; 259 goto drop; 260 } 261 262 oiph = skb_network_header(skb); 263 skb_reset_network_header(skb); 264 265 if (geneve_get_sk_family(gs) == AF_INET) 266 err = IP_ECN_decapsulate(oiph, skb); 267 #if IS_ENABLED(CONFIG_IPV6) 268 else 269 err = IP6_ECN_decapsulate(oiph, skb); 270 #endif 271 272 if (unlikely(err)) { 273 if (log_ecn_error) { 274 if (geneve_get_sk_family(gs) == AF_INET) 275 net_info_ratelimited("non-ECT from %pI4 " 276 "with TOS=%#x\n", 277 &((struct iphdr *)oiph)->saddr, 278 ((struct iphdr *)oiph)->tos); 279 #if IS_ENABLED(CONFIG_IPV6) 280 else 281 net_info_ratelimited("non-ECT from %pI6\n", 282 &((struct ipv6hdr *)oiph)->saddr); 283 #endif 284 } 285 if (err > 1) { 286 ++geneve->dev->stats.rx_frame_errors; 287 ++geneve->dev->stats.rx_errors; 288 goto drop; 289 } 290 } 291 292 len = skb->len; 293 err = gro_cells_receive(&geneve->gro_cells, skb); 294 if (likely(err == NET_RX_SUCCESS)) { 295 stats = this_cpu_ptr(geneve->dev->tstats); 296 u64_stats_update_begin(&stats->syncp); 297 stats->rx_packets++; 298 stats->rx_bytes += len; 299 u64_stats_update_end(&stats->syncp); 300 } 301 return; 302 drop: 303 /* Consume bad packet */ 304 kfree_skb(skb); 305 } 306 307 /* Setup stats when device is created */ 308 static int geneve_init(struct net_device *dev) 309 { 310 struct geneve_dev *geneve = netdev_priv(dev); 311 int err; 312 313 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 314 if (!dev->tstats) 315 return -ENOMEM; 316 317 err = gro_cells_init(&geneve->gro_cells, dev); 318 if (err) { 319 free_percpu(dev->tstats); 320 return err; 321 } 322 323 err = dst_cache_init(&geneve->info.dst_cache, GFP_KERNEL); 324 if (err) { 325 free_percpu(dev->tstats); 326 gro_cells_destroy(&geneve->gro_cells); 327 return err; 328 } 329 return 0; 330 } 331 332 static void geneve_uninit(struct net_device *dev) 333 { 334 struct geneve_dev *geneve = netdev_priv(dev); 335 336 dst_cache_destroy(&geneve->info.dst_cache); 337 gro_cells_destroy(&geneve->gro_cells); 338 free_percpu(dev->tstats); 339 } 340 341 /* Callback from net/ipv4/udp.c to receive packets */ 342 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb) 343 { 344 struct genevehdr *geneveh; 345 struct geneve_dev *geneve; 346 struct geneve_sock *gs; 347 int opts_len; 348 349 /* Need UDP and Geneve header to be present */ 350 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN))) 351 goto drop; 352 353 /* Return packets with reserved bits set */ 354 geneveh = geneve_hdr(skb); 355 if (unlikely(geneveh->ver != GENEVE_VER)) 356 goto drop; 357 358 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB))) 359 goto drop; 360 361 gs = rcu_dereference_sk_user_data(sk); 362 if (!gs) 363 goto drop; 364 365 geneve = geneve_lookup_skb(gs, skb); 366 if (!geneve) 367 goto drop; 368 369 opts_len = geneveh->opt_len * 4; 370 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, 371 htons(ETH_P_TEB), 372 !net_eq(geneve->net, dev_net(geneve->dev)))) { 373 geneve->dev->stats.rx_dropped++; 374 goto drop; 375 } 376 377 geneve_rx(geneve, gs, skb); 378 return 0; 379 380 drop: 381 /* Consume bad packet */ 382 kfree_skb(skb); 383 return 0; 384 } 385 386 static struct socket *geneve_create_sock(struct net *net, bool ipv6, 387 __be16 port, bool ipv6_rx_csum) 388 { 389 struct socket *sock; 390 struct udp_port_cfg udp_conf; 391 int err; 392 393 memset(&udp_conf, 0, sizeof(udp_conf)); 394 395 if (ipv6) { 396 udp_conf.family = AF_INET6; 397 udp_conf.ipv6_v6only = 1; 398 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum; 399 } else { 400 udp_conf.family = AF_INET; 401 udp_conf.local_ip.s_addr = htonl(INADDR_ANY); 402 } 403 404 udp_conf.local_udp_port = port; 405 406 /* Open UDP socket */ 407 err = udp_sock_create(net, &udp_conf, &sock); 408 if (err < 0) 409 return ERR_PTR(err); 410 411 return sock; 412 } 413 414 static int geneve_hlen(struct genevehdr *gh) 415 { 416 return sizeof(*gh) + gh->opt_len * 4; 417 } 418 419 static struct sk_buff **geneve_gro_receive(struct sock *sk, 420 struct sk_buff **head, 421 struct sk_buff *skb) 422 { 423 struct sk_buff *p, **pp = NULL; 424 struct genevehdr *gh, *gh2; 425 unsigned int hlen, gh_len, off_gnv; 426 const struct packet_offload *ptype; 427 __be16 type; 428 int flush = 1; 429 430 off_gnv = skb_gro_offset(skb); 431 hlen = off_gnv + sizeof(*gh); 432 gh = skb_gro_header_fast(skb, off_gnv); 433 if (skb_gro_header_hard(skb, hlen)) { 434 gh = skb_gro_header_slow(skb, hlen, off_gnv); 435 if (unlikely(!gh)) 436 goto out; 437 } 438 439 if (gh->ver != GENEVE_VER || gh->oam) 440 goto out; 441 gh_len = geneve_hlen(gh); 442 443 hlen = off_gnv + gh_len; 444 if (skb_gro_header_hard(skb, hlen)) { 445 gh = skb_gro_header_slow(skb, hlen, off_gnv); 446 if (unlikely(!gh)) 447 goto out; 448 } 449 450 for (p = *head; p; p = p->next) { 451 if (!NAPI_GRO_CB(p)->same_flow) 452 continue; 453 454 gh2 = (struct genevehdr *)(p->data + off_gnv); 455 if (gh->opt_len != gh2->opt_len || 456 memcmp(gh, gh2, gh_len)) { 457 NAPI_GRO_CB(p)->same_flow = 0; 458 continue; 459 } 460 } 461 462 type = gh->proto_type; 463 464 rcu_read_lock(); 465 ptype = gro_find_receive_by_type(type); 466 if (!ptype) 467 goto out_unlock; 468 469 skb_gro_pull(skb, gh_len); 470 skb_gro_postpull_rcsum(skb, gh, gh_len); 471 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb); 472 flush = 0; 473 474 out_unlock: 475 rcu_read_unlock(); 476 out: 477 NAPI_GRO_CB(skb)->flush |= flush; 478 479 return pp; 480 } 481 482 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb, 483 int nhoff) 484 { 485 struct genevehdr *gh; 486 struct packet_offload *ptype; 487 __be16 type; 488 int gh_len; 489 int err = -ENOSYS; 490 491 gh = (struct genevehdr *)(skb->data + nhoff); 492 gh_len = geneve_hlen(gh); 493 type = gh->proto_type; 494 495 rcu_read_lock(); 496 ptype = gro_find_complete_by_type(type); 497 if (ptype) 498 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len); 499 500 rcu_read_unlock(); 501 502 skb_set_inner_mac_header(skb, nhoff + gh_len); 503 504 return err; 505 } 506 507 /* Create new listen socket if needed */ 508 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port, 509 bool ipv6, bool ipv6_rx_csum) 510 { 511 struct geneve_net *gn = net_generic(net, geneve_net_id); 512 struct geneve_sock *gs; 513 struct socket *sock; 514 struct udp_tunnel_sock_cfg tunnel_cfg; 515 int h; 516 517 gs = kzalloc(sizeof(*gs), GFP_KERNEL); 518 if (!gs) 519 return ERR_PTR(-ENOMEM); 520 521 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum); 522 if (IS_ERR(sock)) { 523 kfree(gs); 524 return ERR_CAST(sock); 525 } 526 527 gs->sock = sock; 528 gs->refcnt = 1; 529 for (h = 0; h < VNI_HASH_SIZE; ++h) 530 INIT_HLIST_HEAD(&gs->vni_list[h]); 531 532 /* Initialize the geneve udp offloads structure */ 533 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 534 535 /* Mark socket as an encapsulation socket */ 536 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg)); 537 tunnel_cfg.sk_user_data = gs; 538 tunnel_cfg.encap_type = 1; 539 tunnel_cfg.gro_receive = geneve_gro_receive; 540 tunnel_cfg.gro_complete = geneve_gro_complete; 541 tunnel_cfg.encap_rcv = geneve_udp_encap_recv; 542 tunnel_cfg.encap_destroy = NULL; 543 setup_udp_tunnel_sock(net, sock, &tunnel_cfg); 544 list_add(&gs->list, &gn->sock_list); 545 return gs; 546 } 547 548 static void __geneve_sock_release(struct geneve_sock *gs) 549 { 550 if (!gs || --gs->refcnt) 551 return; 552 553 list_del(&gs->list); 554 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 555 udp_tunnel_sock_release(gs->sock); 556 kfree_rcu(gs, rcu); 557 } 558 559 static void geneve_sock_release(struct geneve_dev *geneve) 560 { 561 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4); 562 #if IS_ENABLED(CONFIG_IPV6) 563 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6); 564 565 rcu_assign_pointer(geneve->sock6, NULL); 566 #endif 567 568 rcu_assign_pointer(geneve->sock4, NULL); 569 synchronize_net(); 570 571 __geneve_sock_release(gs4); 572 #if IS_ENABLED(CONFIG_IPV6) 573 __geneve_sock_release(gs6); 574 #endif 575 } 576 577 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn, 578 sa_family_t family, 579 __be16 dst_port) 580 { 581 struct geneve_sock *gs; 582 583 list_for_each_entry(gs, &gn->sock_list, list) { 584 if (inet_sk(gs->sock->sk)->inet_sport == dst_port && 585 geneve_get_sk_family(gs) == family) { 586 return gs; 587 } 588 } 589 return NULL; 590 } 591 592 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6) 593 { 594 struct net *net = geneve->net; 595 struct geneve_net *gn = net_generic(net, geneve_net_id); 596 struct geneve_dev_node *node; 597 struct geneve_sock *gs; 598 __u8 vni[3]; 599 __u32 hash; 600 601 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->info.key.tp_dst); 602 if (gs) { 603 gs->refcnt++; 604 goto out; 605 } 606 607 gs = geneve_socket_create(net, geneve->info.key.tp_dst, ipv6, 608 geneve->use_udp6_rx_checksums); 609 if (IS_ERR(gs)) 610 return PTR_ERR(gs); 611 612 out: 613 gs->collect_md = geneve->collect_md; 614 #if IS_ENABLED(CONFIG_IPV6) 615 if (ipv6) { 616 rcu_assign_pointer(geneve->sock6, gs); 617 node = &geneve->hlist6; 618 } else 619 #endif 620 { 621 rcu_assign_pointer(geneve->sock4, gs); 622 node = &geneve->hlist4; 623 } 624 node->geneve = geneve; 625 626 tunnel_id_to_vni(geneve->info.key.tun_id, vni); 627 hash = geneve_net_vni_hash(vni); 628 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]); 629 return 0; 630 } 631 632 static int geneve_open(struct net_device *dev) 633 { 634 struct geneve_dev *geneve = netdev_priv(dev); 635 bool ipv6 = !!(geneve->info.mode & IP_TUNNEL_INFO_IPV6); 636 bool metadata = geneve->collect_md; 637 int ret = 0; 638 639 #if IS_ENABLED(CONFIG_IPV6) 640 if (ipv6 || metadata) 641 ret = geneve_sock_add(geneve, true); 642 #endif 643 if (!ret && (!ipv6 || metadata)) 644 ret = geneve_sock_add(geneve, false); 645 if (ret < 0) 646 geneve_sock_release(geneve); 647 648 return ret; 649 } 650 651 static int geneve_stop(struct net_device *dev) 652 { 653 struct geneve_dev *geneve = netdev_priv(dev); 654 655 hlist_del_init_rcu(&geneve->hlist4.hlist); 656 #if IS_ENABLED(CONFIG_IPV6) 657 hlist_del_init_rcu(&geneve->hlist6.hlist); 658 #endif 659 geneve_sock_release(geneve); 660 return 0; 661 } 662 663 static void geneve_build_header(struct genevehdr *geneveh, 664 const struct ip_tunnel_info *info) 665 { 666 geneveh->ver = GENEVE_VER; 667 geneveh->opt_len = info->options_len / 4; 668 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM); 669 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT); 670 geneveh->rsvd1 = 0; 671 tunnel_id_to_vni(info->key.tun_id, geneveh->vni); 672 geneveh->proto_type = htons(ETH_P_TEB); 673 geneveh->rsvd2 = 0; 674 675 ip_tunnel_info_opts_get(geneveh->options, info); 676 } 677 678 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb, 679 const struct ip_tunnel_info *info, 680 bool xnet, int ip_hdr_len) 681 { 682 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM); 683 struct genevehdr *gnvh; 684 int min_headroom; 685 int err; 686 687 skb_reset_mac_header(skb); 688 skb_scrub_packet(skb, xnet); 689 690 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len + 691 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len; 692 err = skb_cow_head(skb, min_headroom); 693 if (unlikely(err)) 694 goto free_dst; 695 696 err = udp_tunnel_handle_offloads(skb, udp_sum); 697 if (err) 698 goto free_dst; 699 700 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len); 701 geneve_build_header(gnvh, info); 702 skb_set_inner_protocol(skb, htons(ETH_P_TEB)); 703 return 0; 704 705 free_dst: 706 dst_release(dst); 707 return err; 708 } 709 710 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb, 711 struct net_device *dev, 712 struct geneve_sock *gs4, 713 struct flowi4 *fl4, 714 const struct ip_tunnel_info *info) 715 { 716 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 717 struct geneve_dev *geneve = netdev_priv(dev); 718 struct dst_cache *dst_cache; 719 struct rtable *rt = NULL; 720 __u8 tos; 721 722 if (!gs4) 723 return ERR_PTR(-EIO); 724 725 memset(fl4, 0, sizeof(*fl4)); 726 fl4->flowi4_mark = skb->mark; 727 fl4->flowi4_proto = IPPROTO_UDP; 728 fl4->daddr = info->key.u.ipv4.dst; 729 fl4->saddr = info->key.u.ipv4.src; 730 731 tos = info->key.tos; 732 if ((tos == 1) && !geneve->collect_md) { 733 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb); 734 use_cache = false; 735 } 736 fl4->flowi4_tos = RT_TOS(tos); 737 738 dst_cache = (struct dst_cache *)&info->dst_cache; 739 if (use_cache) { 740 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr); 741 if (rt) 742 return rt; 743 } 744 rt = ip_route_output_key(geneve->net, fl4); 745 if (IS_ERR(rt)) { 746 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr); 747 return ERR_PTR(-ENETUNREACH); 748 } 749 if (rt->dst.dev == dev) { /* is this necessary? */ 750 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr); 751 ip_rt_put(rt); 752 return ERR_PTR(-ELOOP); 753 } 754 if (use_cache) 755 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr); 756 return rt; 757 } 758 759 #if IS_ENABLED(CONFIG_IPV6) 760 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb, 761 struct net_device *dev, 762 struct geneve_sock *gs6, 763 struct flowi6 *fl6, 764 const struct ip_tunnel_info *info) 765 { 766 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 767 struct geneve_dev *geneve = netdev_priv(dev); 768 struct dst_entry *dst = NULL; 769 struct dst_cache *dst_cache; 770 __u8 prio; 771 772 if (!gs6) 773 return ERR_PTR(-EIO); 774 775 memset(fl6, 0, sizeof(*fl6)); 776 fl6->flowi6_mark = skb->mark; 777 fl6->flowi6_proto = IPPROTO_UDP; 778 fl6->daddr = info->key.u.ipv6.dst; 779 fl6->saddr = info->key.u.ipv6.src; 780 prio = info->key.tos; 781 if ((prio == 1) && !geneve->collect_md) { 782 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb); 783 use_cache = false; 784 } 785 786 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio), 787 info->key.label); 788 dst_cache = (struct dst_cache *)&info->dst_cache; 789 if (use_cache) { 790 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr); 791 if (dst) 792 return dst; 793 } 794 if (ipv6_stub->ipv6_dst_lookup(geneve->net, gs6->sock->sk, &dst, fl6)) { 795 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr); 796 return ERR_PTR(-ENETUNREACH); 797 } 798 if (dst->dev == dev) { /* is this necessary? */ 799 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr); 800 dst_release(dst); 801 return ERR_PTR(-ELOOP); 802 } 803 804 if (use_cache) 805 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr); 806 return dst; 807 } 808 #endif 809 810 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev, 811 struct geneve_dev *geneve, 812 const struct ip_tunnel_info *info) 813 { 814 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 815 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 816 const struct ip_tunnel_key *key = &info->key; 817 struct rtable *rt; 818 struct flowi4 fl4; 819 __u8 tos, ttl; 820 __be16 sport; 821 __be16 df; 822 int err; 823 824 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info); 825 if (IS_ERR(rt)) 826 return PTR_ERR(rt); 827 828 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 829 if (geneve->collect_md) { 830 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 831 ttl = key->ttl; 832 } else { 833 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb); 834 ttl = key->ttl ? : ip4_dst_hoplimit(&rt->dst); 835 } 836 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0; 837 838 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr)); 839 if (unlikely(err)) 840 return err; 841 842 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr, 843 tos, ttl, df, sport, geneve->info.key.tp_dst, 844 !net_eq(geneve->net, dev_net(geneve->dev)), 845 !(info->key.tun_flags & TUNNEL_CSUM)); 846 return 0; 847 } 848 849 #if IS_ENABLED(CONFIG_IPV6) 850 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev, 851 struct geneve_dev *geneve, 852 const struct ip_tunnel_info *info) 853 { 854 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 855 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 856 const struct ip_tunnel_key *key = &info->key; 857 struct dst_entry *dst = NULL; 858 struct flowi6 fl6; 859 __u8 prio, ttl; 860 __be16 sport; 861 int err; 862 863 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info); 864 if (IS_ERR(dst)) 865 return PTR_ERR(dst); 866 867 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 868 if (geneve->collect_md) { 869 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 870 ttl = key->ttl; 871 } else { 872 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel), 873 ip_hdr(skb), skb); 874 ttl = key->ttl ? : ip6_dst_hoplimit(dst); 875 } 876 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr)); 877 if (unlikely(err)) 878 return err; 879 880 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev, 881 &fl6.saddr, &fl6.daddr, prio, ttl, 882 info->key.label, sport, geneve->info.key.tp_dst, 883 !(info->key.tun_flags & TUNNEL_CSUM)); 884 return 0; 885 } 886 #endif 887 888 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev) 889 { 890 struct geneve_dev *geneve = netdev_priv(dev); 891 struct ip_tunnel_info *info = NULL; 892 int err; 893 894 if (geneve->collect_md) { 895 info = skb_tunnel_info(skb); 896 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) { 897 err = -EINVAL; 898 netdev_dbg(dev, "no tunnel metadata\n"); 899 goto tx_error; 900 } 901 } else { 902 info = &geneve->info; 903 } 904 905 rcu_read_lock(); 906 #if IS_ENABLED(CONFIG_IPV6) 907 if (info->mode & IP_TUNNEL_INFO_IPV6) 908 err = geneve6_xmit_skb(skb, dev, geneve, info); 909 else 910 #endif 911 err = geneve_xmit_skb(skb, dev, geneve, info); 912 rcu_read_unlock(); 913 914 if (likely(!err)) 915 return NETDEV_TX_OK; 916 tx_error: 917 dev_kfree_skb(skb); 918 919 if (err == -ELOOP) 920 dev->stats.collisions++; 921 else if (err == -ENETUNREACH) 922 dev->stats.tx_carrier_errors++; 923 924 dev->stats.tx_errors++; 925 return NETDEV_TX_OK; 926 } 927 928 static int geneve_change_mtu(struct net_device *dev, int new_mtu) 929 { 930 /* Only possible if called internally, ndo_change_mtu path's new_mtu 931 * is guaranteed to be between dev->min_mtu and dev->max_mtu. 932 */ 933 if (new_mtu > dev->max_mtu) 934 new_mtu = dev->max_mtu; 935 936 dev->mtu = new_mtu; 937 return 0; 938 } 939 940 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 941 { 942 struct ip_tunnel_info *info = skb_tunnel_info(skb); 943 struct geneve_dev *geneve = netdev_priv(dev); 944 945 if (ip_tunnel_info_af(info) == AF_INET) { 946 struct rtable *rt; 947 struct flowi4 fl4; 948 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 949 950 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info); 951 if (IS_ERR(rt)) 952 return PTR_ERR(rt); 953 954 ip_rt_put(rt); 955 info->key.u.ipv4.src = fl4.saddr; 956 #if IS_ENABLED(CONFIG_IPV6) 957 } else if (ip_tunnel_info_af(info) == AF_INET6) { 958 struct dst_entry *dst; 959 struct flowi6 fl6; 960 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 961 962 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info); 963 if (IS_ERR(dst)) 964 return PTR_ERR(dst); 965 966 dst_release(dst); 967 info->key.u.ipv6.src = fl6.saddr; 968 #endif 969 } else { 970 return -EINVAL; 971 } 972 973 info->key.tp_src = udp_flow_src_port(geneve->net, skb, 974 1, USHRT_MAX, true); 975 info->key.tp_dst = geneve->info.key.tp_dst; 976 return 0; 977 } 978 979 static const struct net_device_ops geneve_netdev_ops = { 980 .ndo_init = geneve_init, 981 .ndo_uninit = geneve_uninit, 982 .ndo_open = geneve_open, 983 .ndo_stop = geneve_stop, 984 .ndo_start_xmit = geneve_xmit, 985 .ndo_get_stats64 = ip_tunnel_get_stats64, 986 .ndo_change_mtu = geneve_change_mtu, 987 .ndo_validate_addr = eth_validate_addr, 988 .ndo_set_mac_address = eth_mac_addr, 989 .ndo_fill_metadata_dst = geneve_fill_metadata_dst, 990 }; 991 992 static void geneve_get_drvinfo(struct net_device *dev, 993 struct ethtool_drvinfo *drvinfo) 994 { 995 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version)); 996 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver)); 997 } 998 999 static const struct ethtool_ops geneve_ethtool_ops = { 1000 .get_drvinfo = geneve_get_drvinfo, 1001 .get_link = ethtool_op_get_link, 1002 }; 1003 1004 /* Info for udev, that this is a virtual tunnel endpoint */ 1005 static struct device_type geneve_type = { 1006 .name = "geneve", 1007 }; 1008 1009 /* Calls the ndo_udp_tunnel_add of the caller in order to 1010 * supply the listening GENEVE udp ports. Callers are expected 1011 * to implement the ndo_udp_tunnel_add. 1012 */ 1013 static void geneve_offload_rx_ports(struct net_device *dev, bool push) 1014 { 1015 struct net *net = dev_net(dev); 1016 struct geneve_net *gn = net_generic(net, geneve_net_id); 1017 struct geneve_sock *gs; 1018 1019 rcu_read_lock(); 1020 list_for_each_entry_rcu(gs, &gn->sock_list, list) { 1021 if (push) { 1022 udp_tunnel_push_rx_port(dev, gs->sock, 1023 UDP_TUNNEL_TYPE_GENEVE); 1024 } else { 1025 udp_tunnel_drop_rx_port(dev, gs->sock, 1026 UDP_TUNNEL_TYPE_GENEVE); 1027 } 1028 } 1029 rcu_read_unlock(); 1030 } 1031 1032 /* Initialize the device structure. */ 1033 static void geneve_setup(struct net_device *dev) 1034 { 1035 ether_setup(dev); 1036 1037 dev->netdev_ops = &geneve_netdev_ops; 1038 dev->ethtool_ops = &geneve_ethtool_ops; 1039 dev->needs_free_netdev = true; 1040 1041 SET_NETDEV_DEVTYPE(dev, &geneve_type); 1042 1043 dev->features |= NETIF_F_LLTX; 1044 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM; 1045 dev->features |= NETIF_F_RXCSUM; 1046 dev->features |= NETIF_F_GSO_SOFTWARE; 1047 1048 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM; 1049 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 1050 1051 /* MTU range: 68 - (something less than 65535) */ 1052 dev->min_mtu = ETH_MIN_MTU; 1053 /* The max_mtu calculation does not take account of GENEVE 1054 * options, to avoid excluding potentially valid 1055 * configurations. This will be further reduced by IPvX hdr size. 1056 */ 1057 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len; 1058 1059 netif_keep_dst(dev); 1060 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1061 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE; 1062 eth_hw_addr_random(dev); 1063 } 1064 1065 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = { 1066 [IFLA_GENEVE_ID] = { .type = NLA_U32 }, 1067 [IFLA_GENEVE_REMOTE] = { .len = FIELD_SIZEOF(struct iphdr, daddr) }, 1068 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) }, 1069 [IFLA_GENEVE_TTL] = { .type = NLA_U8 }, 1070 [IFLA_GENEVE_TOS] = { .type = NLA_U8 }, 1071 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 }, 1072 [IFLA_GENEVE_PORT] = { .type = NLA_U16 }, 1073 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1074 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 }, 1075 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 }, 1076 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 }, 1077 }; 1078 1079 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[], 1080 struct netlink_ext_ack *extack) 1081 { 1082 if (tb[IFLA_ADDRESS]) { 1083 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1084 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1085 "Provided link layer address is not Ethernet"); 1086 return -EINVAL; 1087 } 1088 1089 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1090 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1091 "Provided Ethernet address is not unicast"); 1092 return -EADDRNOTAVAIL; 1093 } 1094 } 1095 1096 if (!data) { 1097 NL_SET_ERR_MSG(extack, 1098 "Not enough attributes provided to perform the operation"); 1099 return -EINVAL; 1100 } 1101 1102 if (data[IFLA_GENEVE_ID]) { 1103 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1104 1105 if (vni >= GENEVE_N_VID) { 1106 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID], 1107 "Geneve ID must be lower than 16777216"); 1108 return -ERANGE; 1109 } 1110 } 1111 1112 return 0; 1113 } 1114 1115 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn, 1116 const struct ip_tunnel_info *info, 1117 bool *tun_on_same_port, 1118 bool *tun_collect_md) 1119 { 1120 struct geneve_dev *geneve, *t = NULL; 1121 1122 *tun_on_same_port = false; 1123 *tun_collect_md = false; 1124 list_for_each_entry(geneve, &gn->geneve_list, next) { 1125 if (info->key.tp_dst == geneve->info.key.tp_dst) { 1126 *tun_collect_md = geneve->collect_md; 1127 *tun_on_same_port = true; 1128 } 1129 if (info->key.tun_id == geneve->info.key.tun_id && 1130 info->key.tp_dst == geneve->info.key.tp_dst && 1131 !memcmp(&info->key.u, &geneve->info.key.u, sizeof(info->key.u))) 1132 t = geneve; 1133 } 1134 return t; 1135 } 1136 1137 static bool is_tnl_info_zero(const struct ip_tunnel_info *info) 1138 { 1139 return !(info->key.tun_id || info->key.tun_flags || info->key.tos || 1140 info->key.ttl || info->key.label || info->key.tp_src || 1141 memchr_inv(&info->key.u, 0, sizeof(info->key.u))); 1142 } 1143 1144 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a, 1145 struct ip_tunnel_info *b) 1146 { 1147 if (ip_tunnel_info_af(a) == AF_INET) 1148 return a->key.u.ipv4.dst == b->key.u.ipv4.dst; 1149 else 1150 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst); 1151 } 1152 1153 static int geneve_configure(struct net *net, struct net_device *dev, 1154 struct netlink_ext_ack *extack, 1155 const struct ip_tunnel_info *info, 1156 bool metadata, bool ipv6_rx_csum) 1157 { 1158 struct geneve_net *gn = net_generic(net, geneve_net_id); 1159 struct geneve_dev *t, *geneve = netdev_priv(dev); 1160 bool tun_collect_md, tun_on_same_port; 1161 int err, encap_len; 1162 1163 if (metadata && !is_tnl_info_zero(info)) { 1164 NL_SET_ERR_MSG(extack, 1165 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified"); 1166 return -EINVAL; 1167 } 1168 1169 geneve->net = net; 1170 geneve->dev = dev; 1171 1172 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md); 1173 if (t) 1174 return -EBUSY; 1175 1176 /* make enough headroom for basic scenario */ 1177 encap_len = GENEVE_BASE_HLEN + ETH_HLEN; 1178 if (!metadata && ip_tunnel_info_af(info) == AF_INET) { 1179 encap_len += sizeof(struct iphdr); 1180 dev->max_mtu -= sizeof(struct iphdr); 1181 } else { 1182 encap_len += sizeof(struct ipv6hdr); 1183 dev->max_mtu -= sizeof(struct ipv6hdr); 1184 } 1185 dev->needed_headroom = encap_len + ETH_HLEN; 1186 1187 if (metadata) { 1188 if (tun_on_same_port) { 1189 NL_SET_ERR_MSG(extack, 1190 "There can be only one externally controlled device on a destination port"); 1191 return -EPERM; 1192 } 1193 } else { 1194 if (tun_collect_md) { 1195 NL_SET_ERR_MSG(extack, 1196 "There already exists an externally controlled device on this destination port"); 1197 return -EPERM; 1198 } 1199 } 1200 1201 dst_cache_reset(&geneve->info.dst_cache); 1202 geneve->info = *info; 1203 geneve->collect_md = metadata; 1204 geneve->use_udp6_rx_checksums = ipv6_rx_csum; 1205 1206 err = register_netdevice(dev); 1207 if (err) 1208 return err; 1209 1210 list_add(&geneve->next, &gn->geneve_list); 1211 return 0; 1212 } 1213 1214 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port) 1215 { 1216 memset(info, 0, sizeof(*info)); 1217 info->key.tp_dst = htons(dst_port); 1218 } 1219 1220 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[], 1221 struct netlink_ext_ack *extack, 1222 struct ip_tunnel_info *info, bool *metadata, 1223 bool *use_udp6_rx_checksums, bool changelink) 1224 { 1225 int attrtype; 1226 1227 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) { 1228 NL_SET_ERR_MSG(extack, 1229 "Cannot specify both IPv4 and IPv6 Remote addresses"); 1230 return -EINVAL; 1231 } 1232 1233 if (data[IFLA_GENEVE_REMOTE]) { 1234 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) { 1235 attrtype = IFLA_GENEVE_REMOTE; 1236 goto change_notsup; 1237 } 1238 1239 info->key.u.ipv4.dst = 1240 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]); 1241 1242 if (IN_MULTICAST(ntohl(info->key.u.ipv4.dst))) { 1243 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE], 1244 "Remote IPv4 address cannot be Multicast"); 1245 return -EINVAL; 1246 } 1247 } 1248 1249 if (data[IFLA_GENEVE_REMOTE6]) { 1250 #if IS_ENABLED(CONFIG_IPV6) 1251 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) { 1252 attrtype = IFLA_GENEVE_REMOTE6; 1253 goto change_notsup; 1254 } 1255 1256 info->mode = IP_TUNNEL_INFO_IPV6; 1257 info->key.u.ipv6.dst = 1258 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]); 1259 1260 if (ipv6_addr_type(&info->key.u.ipv6.dst) & 1261 IPV6_ADDR_LINKLOCAL) { 1262 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1263 "Remote IPv6 address cannot be link-local"); 1264 return -EINVAL; 1265 } 1266 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) { 1267 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1268 "Remote IPv6 address cannot be Multicast"); 1269 return -EINVAL; 1270 } 1271 info->key.tun_flags |= TUNNEL_CSUM; 1272 *use_udp6_rx_checksums = true; 1273 #else 1274 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1275 "IPv6 support not enabled in the kernel"); 1276 return -EPFNOSUPPORT; 1277 #endif 1278 } 1279 1280 if (data[IFLA_GENEVE_ID]) { 1281 __u32 vni; 1282 __u8 tvni[3]; 1283 __be64 tunid; 1284 1285 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1286 tvni[0] = (vni & 0x00ff0000) >> 16; 1287 tvni[1] = (vni & 0x0000ff00) >> 8; 1288 tvni[2] = vni & 0x000000ff; 1289 1290 tunid = vni_to_tunnel_id(tvni); 1291 if (changelink && (tunid != info->key.tun_id)) { 1292 attrtype = IFLA_GENEVE_ID; 1293 goto change_notsup; 1294 } 1295 info->key.tun_id = tunid; 1296 } 1297 1298 if (data[IFLA_GENEVE_TTL]) 1299 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]); 1300 1301 if (data[IFLA_GENEVE_TOS]) 1302 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]); 1303 1304 if (data[IFLA_GENEVE_LABEL]) { 1305 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) & 1306 IPV6_FLOWLABEL_MASK; 1307 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) { 1308 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL], 1309 "Label attribute only applies for IPv6 Geneve devices"); 1310 return -EINVAL; 1311 } 1312 } 1313 1314 if (data[IFLA_GENEVE_PORT]) { 1315 if (changelink) { 1316 attrtype = IFLA_GENEVE_PORT; 1317 goto change_notsup; 1318 } 1319 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]); 1320 } 1321 1322 if (data[IFLA_GENEVE_COLLECT_METADATA]) { 1323 if (changelink) { 1324 attrtype = IFLA_GENEVE_COLLECT_METADATA; 1325 goto change_notsup; 1326 } 1327 *metadata = true; 1328 } 1329 1330 if (data[IFLA_GENEVE_UDP_CSUM]) { 1331 if (changelink) { 1332 attrtype = IFLA_GENEVE_UDP_CSUM; 1333 goto change_notsup; 1334 } 1335 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM])) 1336 info->key.tun_flags |= TUNNEL_CSUM; 1337 } 1338 1339 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) { 1340 if (changelink) { 1341 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX; 1342 goto change_notsup; 1343 } 1344 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX])) 1345 info->key.tun_flags &= ~TUNNEL_CSUM; 1346 } 1347 1348 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) { 1349 if (changelink) { 1350 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX; 1351 goto change_notsup; 1352 } 1353 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX])) 1354 *use_udp6_rx_checksums = false; 1355 } 1356 1357 return 0; 1358 change_notsup: 1359 NL_SET_ERR_MSG_ATTR(extack, data[attrtype], 1360 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported"); 1361 return -EOPNOTSUPP; 1362 } 1363 1364 static int geneve_newlink(struct net *net, struct net_device *dev, 1365 struct nlattr *tb[], struct nlattr *data[], 1366 struct netlink_ext_ack *extack) 1367 { 1368 bool use_udp6_rx_checksums = false; 1369 struct ip_tunnel_info info; 1370 bool metadata = false; 1371 int err; 1372 1373 init_tnl_info(&info, GENEVE_UDP_PORT); 1374 err = geneve_nl2info(tb, data, extack, &info, &metadata, 1375 &use_udp6_rx_checksums, false); 1376 if (err) 1377 return err; 1378 1379 return geneve_configure(net, dev, extack, &info, metadata, 1380 use_udp6_rx_checksums); 1381 } 1382 1383 /* Quiesces the geneve device data path for both TX and RX. 1384 * 1385 * On transmit geneve checks for non-NULL geneve_sock before it proceeds. 1386 * So, if we set that socket to NULL under RCU and wait for synchronize_net() 1387 * to complete for the existing set of in-flight packets to be transmitted, 1388 * then we would have quiesced the transmit data path. All the future packets 1389 * will get dropped until we unquiesce the data path. 1390 * 1391 * On receive geneve dereference the geneve_sock stashed in the socket. So, 1392 * if we set that to NULL under RCU and wait for synchronize_net() to 1393 * complete, then we would have quiesced the receive data path. 1394 */ 1395 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4, 1396 struct geneve_sock **gs6) 1397 { 1398 *gs4 = rtnl_dereference(geneve->sock4); 1399 rcu_assign_pointer(geneve->sock4, NULL); 1400 if (*gs4) 1401 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL); 1402 #if IS_ENABLED(CONFIG_IPV6) 1403 *gs6 = rtnl_dereference(geneve->sock6); 1404 rcu_assign_pointer(geneve->sock6, NULL); 1405 if (*gs6) 1406 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL); 1407 #else 1408 *gs6 = NULL; 1409 #endif 1410 synchronize_net(); 1411 } 1412 1413 /* Resumes the geneve device data path for both TX and RX. */ 1414 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4, 1415 struct geneve_sock __maybe_unused *gs6) 1416 { 1417 rcu_assign_pointer(geneve->sock4, gs4); 1418 if (gs4) 1419 rcu_assign_sk_user_data(gs4->sock->sk, gs4); 1420 #if IS_ENABLED(CONFIG_IPV6) 1421 rcu_assign_pointer(geneve->sock6, gs6); 1422 if (gs6) 1423 rcu_assign_sk_user_data(gs6->sock->sk, gs6); 1424 #endif 1425 synchronize_net(); 1426 } 1427 1428 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[], 1429 struct nlattr *data[], 1430 struct netlink_ext_ack *extack) 1431 { 1432 struct geneve_dev *geneve = netdev_priv(dev); 1433 struct geneve_sock *gs4, *gs6; 1434 struct ip_tunnel_info info; 1435 bool metadata; 1436 bool use_udp6_rx_checksums; 1437 int err; 1438 1439 /* If the geneve device is configured for metadata (or externally 1440 * controlled, for example, OVS), then nothing can be changed. 1441 */ 1442 if (geneve->collect_md) 1443 return -EOPNOTSUPP; 1444 1445 /* Start with the existing info. */ 1446 memcpy(&info, &geneve->info, sizeof(info)); 1447 metadata = geneve->collect_md; 1448 use_udp6_rx_checksums = geneve->use_udp6_rx_checksums; 1449 err = geneve_nl2info(tb, data, extack, &info, &metadata, 1450 &use_udp6_rx_checksums, true); 1451 if (err) 1452 return err; 1453 1454 if (!geneve_dst_addr_equal(&geneve->info, &info)) 1455 dst_cache_reset(&info.dst_cache); 1456 1457 geneve_quiesce(geneve, &gs4, &gs6); 1458 geneve->info = info; 1459 geneve->collect_md = metadata; 1460 geneve->use_udp6_rx_checksums = use_udp6_rx_checksums; 1461 geneve_unquiesce(geneve, gs4, gs6); 1462 1463 return 0; 1464 } 1465 1466 static void geneve_dellink(struct net_device *dev, struct list_head *head) 1467 { 1468 struct geneve_dev *geneve = netdev_priv(dev); 1469 1470 list_del(&geneve->next); 1471 unregister_netdevice_queue(dev, head); 1472 } 1473 1474 static size_t geneve_get_size(const struct net_device *dev) 1475 { 1476 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */ 1477 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */ 1478 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */ 1479 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */ 1480 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */ 1481 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */ 1482 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */ 1483 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */ 1484 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */ 1485 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */ 1486 0; 1487 } 1488 1489 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev) 1490 { 1491 struct geneve_dev *geneve = netdev_priv(dev); 1492 struct ip_tunnel_info *info = &geneve->info; 1493 bool metadata = geneve->collect_md; 1494 __u8 tmp_vni[3]; 1495 __u32 vni; 1496 1497 tunnel_id_to_vni(info->key.tun_id, tmp_vni); 1498 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2]; 1499 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni)) 1500 goto nla_put_failure; 1501 1502 if (!metadata && ip_tunnel_info_af(info) == AF_INET) { 1503 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE, 1504 info->key.u.ipv4.dst)) 1505 goto nla_put_failure; 1506 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM, 1507 !!(info->key.tun_flags & TUNNEL_CSUM))) 1508 goto nla_put_failure; 1509 1510 #if IS_ENABLED(CONFIG_IPV6) 1511 } else if (!metadata) { 1512 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6, 1513 &info->key.u.ipv6.dst)) 1514 goto nla_put_failure; 1515 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX, 1516 !(info->key.tun_flags & TUNNEL_CSUM))) 1517 goto nla_put_failure; 1518 #endif 1519 } 1520 1521 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) || 1522 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) || 1523 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label)) 1524 goto nla_put_failure; 1525 1526 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst)) 1527 goto nla_put_failure; 1528 1529 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA)) 1530 goto nla_put_failure; 1531 1532 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX, 1533 !geneve->use_udp6_rx_checksums)) 1534 goto nla_put_failure; 1535 1536 return 0; 1537 1538 nla_put_failure: 1539 return -EMSGSIZE; 1540 } 1541 1542 static struct rtnl_link_ops geneve_link_ops __read_mostly = { 1543 .kind = "geneve", 1544 .maxtype = IFLA_GENEVE_MAX, 1545 .policy = geneve_policy, 1546 .priv_size = sizeof(struct geneve_dev), 1547 .setup = geneve_setup, 1548 .validate = geneve_validate, 1549 .newlink = geneve_newlink, 1550 .changelink = geneve_changelink, 1551 .dellink = geneve_dellink, 1552 .get_size = geneve_get_size, 1553 .fill_info = geneve_fill_info, 1554 }; 1555 1556 struct net_device *geneve_dev_create_fb(struct net *net, const char *name, 1557 u8 name_assign_type, u16 dst_port) 1558 { 1559 struct nlattr *tb[IFLA_MAX + 1]; 1560 struct ip_tunnel_info info; 1561 struct net_device *dev; 1562 LIST_HEAD(list_kill); 1563 int err; 1564 1565 memset(tb, 0, sizeof(tb)); 1566 dev = rtnl_create_link(net, name, name_assign_type, 1567 &geneve_link_ops, tb); 1568 if (IS_ERR(dev)) 1569 return dev; 1570 1571 init_tnl_info(&info, dst_port); 1572 err = geneve_configure(net, dev, NULL, &info, true, true); 1573 if (err) { 1574 free_netdev(dev); 1575 return ERR_PTR(err); 1576 } 1577 1578 /* openvswitch users expect packet sizes to be unrestricted, 1579 * so set the largest MTU we can. 1580 */ 1581 err = geneve_change_mtu(dev, IP_MAX_MTU); 1582 if (err) 1583 goto err; 1584 1585 err = rtnl_configure_link(dev, NULL); 1586 if (err < 0) 1587 goto err; 1588 1589 return dev; 1590 err: 1591 geneve_dellink(dev, &list_kill); 1592 unregister_netdevice_many(&list_kill); 1593 return ERR_PTR(err); 1594 } 1595 EXPORT_SYMBOL_GPL(geneve_dev_create_fb); 1596 1597 static int geneve_netdevice_event(struct notifier_block *unused, 1598 unsigned long event, void *ptr) 1599 { 1600 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1601 1602 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO || 1603 event == NETDEV_UDP_TUNNEL_DROP_INFO) { 1604 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO); 1605 } else if (event == NETDEV_UNREGISTER) { 1606 geneve_offload_rx_ports(dev, false); 1607 } else if (event == NETDEV_REGISTER) { 1608 geneve_offload_rx_ports(dev, true); 1609 } 1610 1611 return NOTIFY_DONE; 1612 } 1613 1614 static struct notifier_block geneve_notifier_block __read_mostly = { 1615 .notifier_call = geneve_netdevice_event, 1616 }; 1617 1618 static __net_init int geneve_init_net(struct net *net) 1619 { 1620 struct geneve_net *gn = net_generic(net, geneve_net_id); 1621 1622 INIT_LIST_HEAD(&gn->geneve_list); 1623 INIT_LIST_HEAD(&gn->sock_list); 1624 return 0; 1625 } 1626 1627 static void __net_exit geneve_exit_net(struct net *net) 1628 { 1629 struct geneve_net *gn = net_generic(net, geneve_net_id); 1630 struct geneve_dev *geneve, *next; 1631 struct net_device *dev, *aux; 1632 LIST_HEAD(list); 1633 1634 rtnl_lock(); 1635 1636 /* gather any geneve devices that were moved into this ns */ 1637 for_each_netdev_safe(net, dev, aux) 1638 if (dev->rtnl_link_ops == &geneve_link_ops) 1639 unregister_netdevice_queue(dev, &list); 1640 1641 /* now gather any other geneve devices that were created in this ns */ 1642 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) { 1643 /* If geneve->dev is in the same netns, it was already added 1644 * to the list by the previous loop. 1645 */ 1646 if (!net_eq(dev_net(geneve->dev), net)) 1647 unregister_netdevice_queue(geneve->dev, &list); 1648 } 1649 1650 /* unregister the devices gathered above */ 1651 unregister_netdevice_many(&list); 1652 rtnl_unlock(); 1653 WARN_ON_ONCE(!list_empty(&gn->sock_list)); 1654 } 1655 1656 static struct pernet_operations geneve_net_ops = { 1657 .init = geneve_init_net, 1658 .exit = geneve_exit_net, 1659 .id = &geneve_net_id, 1660 .size = sizeof(struct geneve_net), 1661 }; 1662 1663 static int __init geneve_init_module(void) 1664 { 1665 int rc; 1666 1667 rc = register_pernet_subsys(&geneve_net_ops); 1668 if (rc) 1669 goto out1; 1670 1671 rc = register_netdevice_notifier(&geneve_notifier_block); 1672 if (rc) 1673 goto out2; 1674 1675 rc = rtnl_link_register(&geneve_link_ops); 1676 if (rc) 1677 goto out3; 1678 1679 return 0; 1680 out3: 1681 unregister_netdevice_notifier(&geneve_notifier_block); 1682 out2: 1683 unregister_pernet_subsys(&geneve_net_ops); 1684 out1: 1685 return rc; 1686 } 1687 late_initcall(geneve_init_module); 1688 1689 static void __exit geneve_cleanup_module(void) 1690 { 1691 rtnl_link_unregister(&geneve_link_ops); 1692 unregister_netdevice_notifier(&geneve_notifier_block); 1693 unregister_pernet_subsys(&geneve_net_ops); 1694 } 1695 module_exit(geneve_cleanup_module); 1696 1697 MODULE_LICENSE("GPL"); 1698 MODULE_VERSION(GENEVE_NETDEV_VER); 1699 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>"); 1700 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic"); 1701 MODULE_ALIAS_RTNL_LINK("geneve"); 1702