1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * GENEVE: Generic Network Virtualization Encapsulation 4 * 5 * Copyright (c) 2015 Red Hat, Inc. 6 */ 7 8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 9 10 #include <linux/ethtool.h> 11 #include <linux/kernel.h> 12 #include <linux/module.h> 13 #include <linux/etherdevice.h> 14 #include <linux/hash.h> 15 #include <net/ipv6_stubs.h> 16 #include <net/dst_metadata.h> 17 #include <net/gro_cells.h> 18 #include <net/rtnetlink.h> 19 #include <net/geneve.h> 20 #include <net/gro.h> 21 #include <net/protocol.h> 22 23 #define GENEVE_NETDEV_VER "0.6" 24 25 #define GENEVE_N_VID (1u << 24) 26 #define GENEVE_VID_MASK (GENEVE_N_VID - 1) 27 28 #define VNI_HASH_BITS 10 29 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS) 30 31 static bool log_ecn_error = true; 32 module_param(log_ecn_error, bool, 0644); 33 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN"); 34 35 #define GENEVE_VER 0 36 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr)) 37 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN) 38 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN) 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 struct geneve_config { 54 struct ip_tunnel_info info; 55 bool collect_md; 56 bool use_udp6_rx_checksums; 57 bool ttl_inherit; 58 enum ifla_geneve_df df; 59 bool inner_proto_inherit; 60 }; 61 62 /* Pseudo network device */ 63 struct geneve_dev { 64 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */ 65 #if IS_ENABLED(CONFIG_IPV6) 66 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */ 67 #endif 68 struct net *net; /* netns for packet i/o */ 69 struct net_device *dev; /* netdev for geneve tunnel */ 70 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */ 71 #if IS_ENABLED(CONFIG_IPV6) 72 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */ 73 #endif 74 struct list_head next; /* geneve's per namespace list */ 75 struct gro_cells gro_cells; 76 struct geneve_config cfg; 77 }; 78 79 struct geneve_sock { 80 bool collect_md; 81 struct list_head list; 82 struct socket *sock; 83 struct rcu_head rcu; 84 int refcnt; 85 struct hlist_head vni_list[VNI_HASH_SIZE]; 86 }; 87 88 static inline __u32 geneve_net_vni_hash(u8 vni[3]) 89 { 90 __u32 vnid; 91 92 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2]; 93 return hash_32(vnid, VNI_HASH_BITS); 94 } 95 96 static __be64 vni_to_tunnel_id(const __u8 *vni) 97 { 98 #ifdef __BIG_ENDIAN 99 return (vni[0] << 16) | (vni[1] << 8) | vni[2]; 100 #else 101 return (__force __be64)(((__force u64)vni[0] << 40) | 102 ((__force u64)vni[1] << 48) | 103 ((__force u64)vni[2] << 56)); 104 #endif 105 } 106 107 /* Convert 64 bit tunnel ID to 24 bit VNI. */ 108 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni) 109 { 110 #ifdef __BIG_ENDIAN 111 vni[0] = (__force __u8)(tun_id >> 16); 112 vni[1] = (__force __u8)(tun_id >> 8); 113 vni[2] = (__force __u8)tun_id; 114 #else 115 vni[0] = (__force __u8)((__force u64)tun_id >> 40); 116 vni[1] = (__force __u8)((__force u64)tun_id >> 48); 117 vni[2] = (__force __u8)((__force u64)tun_id >> 56); 118 #endif 119 } 120 121 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni) 122 { 123 return !memcmp(vni, &tun_id[5], 3); 124 } 125 126 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs) 127 { 128 return gs->sock->sk->sk_family; 129 } 130 131 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs, 132 __be32 addr, u8 vni[]) 133 { 134 struct hlist_head *vni_list_head; 135 struct geneve_dev_node *node; 136 __u32 hash; 137 138 /* Find the device for this VNI */ 139 hash = geneve_net_vni_hash(vni); 140 vni_list_head = &gs->vni_list[hash]; 141 hlist_for_each_entry_rcu(node, vni_list_head, hlist) { 142 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) && 143 addr == node->geneve->cfg.info.key.u.ipv4.dst) 144 return node->geneve; 145 } 146 return NULL; 147 } 148 149 #if IS_ENABLED(CONFIG_IPV6) 150 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs, 151 struct in6_addr addr6, u8 vni[]) 152 { 153 struct hlist_head *vni_list_head; 154 struct geneve_dev_node *node; 155 __u32 hash; 156 157 /* Find the device for this VNI */ 158 hash = geneve_net_vni_hash(vni); 159 vni_list_head = &gs->vni_list[hash]; 160 hlist_for_each_entry_rcu(node, vni_list_head, hlist) { 161 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) && 162 ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst)) 163 return node->geneve; 164 } 165 return NULL; 166 } 167 #endif 168 169 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb) 170 { 171 return (struct genevehdr *)(udp_hdr(skb) + 1); 172 } 173 174 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs, 175 struct sk_buff *skb) 176 { 177 static u8 zero_vni[3]; 178 u8 *vni; 179 180 if (geneve_get_sk_family(gs) == AF_INET) { 181 struct iphdr *iph; 182 __be32 addr; 183 184 iph = ip_hdr(skb); /* outer IP header... */ 185 186 if (gs->collect_md) { 187 vni = zero_vni; 188 addr = 0; 189 } else { 190 vni = geneve_hdr(skb)->vni; 191 addr = iph->saddr; 192 } 193 194 return geneve_lookup(gs, addr, vni); 195 #if IS_ENABLED(CONFIG_IPV6) 196 } else if (geneve_get_sk_family(gs) == AF_INET6) { 197 static struct in6_addr zero_addr6; 198 struct ipv6hdr *ip6h; 199 struct in6_addr addr6; 200 201 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */ 202 203 if (gs->collect_md) { 204 vni = zero_vni; 205 addr6 = zero_addr6; 206 } else { 207 vni = geneve_hdr(skb)->vni; 208 addr6 = ip6h->saddr; 209 } 210 211 return geneve6_lookup(gs, addr6, vni); 212 #endif 213 } 214 return NULL; 215 } 216 217 /* geneve receive/decap routine */ 218 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs, 219 struct sk_buff *skb) 220 { 221 struct genevehdr *gnvh = geneve_hdr(skb); 222 struct metadata_dst *tun_dst = NULL; 223 unsigned int len; 224 int err = 0; 225 void *oiph; 226 227 if (ip_tunnel_collect_metadata() || gs->collect_md) { 228 __be16 flags; 229 230 flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) | 231 (gnvh->critical ? TUNNEL_CRIT_OPT : 0); 232 233 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags, 234 vni_to_tunnel_id(gnvh->vni), 235 gnvh->opt_len * 4); 236 if (!tun_dst) { 237 geneve->dev->stats.rx_dropped++; 238 goto drop; 239 } 240 /* Update tunnel dst according to Geneve options. */ 241 ip_tunnel_info_opts_set(&tun_dst->u.tun_info, 242 gnvh->options, gnvh->opt_len * 4, 243 TUNNEL_GENEVE_OPT); 244 } else { 245 /* Drop packets w/ critical options, 246 * since we don't support any... 247 */ 248 if (gnvh->critical) { 249 geneve->dev->stats.rx_frame_errors++; 250 geneve->dev->stats.rx_errors++; 251 goto drop; 252 } 253 } 254 255 if (tun_dst) 256 skb_dst_set(skb, &tun_dst->dst); 257 258 if (gnvh->proto_type == htons(ETH_P_TEB)) { 259 skb_reset_mac_header(skb); 260 skb->protocol = eth_type_trans(skb, geneve->dev); 261 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 262 263 /* Ignore packet loops (and multicast echo) */ 264 if (ether_addr_equal(eth_hdr(skb)->h_source, 265 geneve->dev->dev_addr)) { 266 geneve->dev->stats.rx_errors++; 267 goto drop; 268 } 269 } else { 270 skb_reset_mac_header(skb); 271 skb->dev = geneve->dev; 272 skb->pkt_type = PACKET_HOST; 273 } 274 275 oiph = skb_network_header(skb); 276 skb_reset_network_header(skb); 277 278 if (geneve_get_sk_family(gs) == AF_INET) 279 err = IP_ECN_decapsulate(oiph, skb); 280 #if IS_ENABLED(CONFIG_IPV6) 281 else 282 err = IP6_ECN_decapsulate(oiph, skb); 283 #endif 284 285 if (unlikely(err)) { 286 if (log_ecn_error) { 287 if (geneve_get_sk_family(gs) == AF_INET) 288 net_info_ratelimited("non-ECT from %pI4 " 289 "with TOS=%#x\n", 290 &((struct iphdr *)oiph)->saddr, 291 ((struct iphdr *)oiph)->tos); 292 #if IS_ENABLED(CONFIG_IPV6) 293 else 294 net_info_ratelimited("non-ECT from %pI6\n", 295 &((struct ipv6hdr *)oiph)->saddr); 296 #endif 297 } 298 if (err > 1) { 299 ++geneve->dev->stats.rx_frame_errors; 300 ++geneve->dev->stats.rx_errors; 301 goto drop; 302 } 303 } 304 305 len = skb->len; 306 err = gro_cells_receive(&geneve->gro_cells, skb); 307 if (likely(err == NET_RX_SUCCESS)) 308 dev_sw_netstats_rx_add(geneve->dev, len); 309 310 return; 311 drop: 312 /* Consume bad packet */ 313 kfree_skb(skb); 314 } 315 316 /* Setup stats when device is created */ 317 static int geneve_init(struct net_device *dev) 318 { 319 struct geneve_dev *geneve = netdev_priv(dev); 320 int err; 321 322 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 323 if (!dev->tstats) 324 return -ENOMEM; 325 326 err = gro_cells_init(&geneve->gro_cells, dev); 327 if (err) { 328 free_percpu(dev->tstats); 329 return err; 330 } 331 332 err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL); 333 if (err) { 334 free_percpu(dev->tstats); 335 gro_cells_destroy(&geneve->gro_cells); 336 return err; 337 } 338 return 0; 339 } 340 341 static void geneve_uninit(struct net_device *dev) 342 { 343 struct geneve_dev *geneve = netdev_priv(dev); 344 345 dst_cache_destroy(&geneve->cfg.info.dst_cache); 346 gro_cells_destroy(&geneve->gro_cells); 347 free_percpu(dev->tstats); 348 } 349 350 /* Callback from net/ipv4/udp.c to receive packets */ 351 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb) 352 { 353 struct genevehdr *geneveh; 354 struct geneve_dev *geneve; 355 struct geneve_sock *gs; 356 __be16 inner_proto; 357 int opts_len; 358 359 /* Need UDP and Geneve header to be present */ 360 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN))) 361 goto drop; 362 363 /* Return packets with reserved bits set */ 364 geneveh = geneve_hdr(skb); 365 if (unlikely(geneveh->ver != GENEVE_VER)) 366 goto drop; 367 368 inner_proto = geneveh->proto_type; 369 370 if (unlikely((inner_proto != htons(ETH_P_TEB) && 371 inner_proto != htons(ETH_P_IP) && 372 inner_proto != htons(ETH_P_IPV6)))) 373 goto drop; 374 375 gs = rcu_dereference_sk_user_data(sk); 376 if (!gs) 377 goto drop; 378 379 geneve = geneve_lookup_skb(gs, skb); 380 if (!geneve) 381 goto drop; 382 383 if (unlikely((!geneve->cfg.inner_proto_inherit && 384 inner_proto != htons(ETH_P_TEB)))) { 385 geneve->dev->stats.rx_dropped++; 386 goto drop; 387 } 388 389 opts_len = geneveh->opt_len * 4; 390 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto, 391 !net_eq(geneve->net, dev_net(geneve->dev)))) { 392 geneve->dev->stats.rx_dropped++; 393 goto drop; 394 } 395 396 geneve_rx(geneve, gs, skb); 397 return 0; 398 399 drop: 400 /* Consume bad packet */ 401 kfree_skb(skb); 402 return 0; 403 } 404 405 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */ 406 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb) 407 { 408 struct genevehdr *geneveh; 409 struct geneve_sock *gs; 410 u8 zero_vni[3] = { 0 }; 411 u8 *vni = zero_vni; 412 413 if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN)) 414 return -EINVAL; 415 416 geneveh = geneve_hdr(skb); 417 if (geneveh->ver != GENEVE_VER) 418 return -EINVAL; 419 420 if (geneveh->proto_type != htons(ETH_P_TEB)) 421 return -EINVAL; 422 423 gs = rcu_dereference_sk_user_data(sk); 424 if (!gs) 425 return -ENOENT; 426 427 if (geneve_get_sk_family(gs) == AF_INET) { 428 struct iphdr *iph = ip_hdr(skb); 429 __be32 addr4 = 0; 430 431 if (!gs->collect_md) { 432 vni = geneve_hdr(skb)->vni; 433 addr4 = iph->daddr; 434 } 435 436 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT; 437 } 438 439 #if IS_ENABLED(CONFIG_IPV6) 440 if (geneve_get_sk_family(gs) == AF_INET6) { 441 struct ipv6hdr *ip6h = ipv6_hdr(skb); 442 struct in6_addr addr6; 443 444 memset(&addr6, 0, sizeof(struct in6_addr)); 445 446 if (!gs->collect_md) { 447 vni = geneve_hdr(skb)->vni; 448 addr6 = ip6h->daddr; 449 } 450 451 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT; 452 } 453 #endif 454 455 return -EPFNOSUPPORT; 456 } 457 458 static struct socket *geneve_create_sock(struct net *net, bool ipv6, 459 __be16 port, bool ipv6_rx_csum) 460 { 461 struct socket *sock; 462 struct udp_port_cfg udp_conf; 463 int err; 464 465 memset(&udp_conf, 0, sizeof(udp_conf)); 466 467 if (ipv6) { 468 udp_conf.family = AF_INET6; 469 udp_conf.ipv6_v6only = 1; 470 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum; 471 } else { 472 udp_conf.family = AF_INET; 473 udp_conf.local_ip.s_addr = htonl(INADDR_ANY); 474 } 475 476 udp_conf.local_udp_port = port; 477 478 /* Open UDP socket */ 479 err = udp_sock_create(net, &udp_conf, &sock); 480 if (err < 0) 481 return ERR_PTR(err); 482 483 udp_allow_gso(sock->sk); 484 return sock; 485 } 486 487 static int geneve_hlen(struct genevehdr *gh) 488 { 489 return sizeof(*gh) + gh->opt_len * 4; 490 } 491 492 static struct sk_buff *geneve_gro_receive(struct sock *sk, 493 struct list_head *head, 494 struct sk_buff *skb) 495 { 496 struct sk_buff *pp = NULL; 497 struct sk_buff *p; 498 struct genevehdr *gh, *gh2; 499 unsigned int hlen, gh_len, off_gnv; 500 const struct packet_offload *ptype; 501 __be16 type; 502 int flush = 1; 503 504 off_gnv = skb_gro_offset(skb); 505 hlen = off_gnv + sizeof(*gh); 506 gh = skb_gro_header_fast(skb, off_gnv); 507 if (skb_gro_header_hard(skb, hlen)) { 508 gh = skb_gro_header_slow(skb, hlen, off_gnv); 509 if (unlikely(!gh)) 510 goto out; 511 } 512 513 if (gh->ver != GENEVE_VER || gh->oam) 514 goto out; 515 gh_len = geneve_hlen(gh); 516 517 hlen = off_gnv + gh_len; 518 if (skb_gro_header_hard(skb, hlen)) { 519 gh = skb_gro_header_slow(skb, hlen, off_gnv); 520 if (unlikely(!gh)) 521 goto out; 522 } 523 524 list_for_each_entry(p, head, list) { 525 if (!NAPI_GRO_CB(p)->same_flow) 526 continue; 527 528 gh2 = (struct genevehdr *)(p->data + off_gnv); 529 if (gh->opt_len != gh2->opt_len || 530 memcmp(gh, gh2, gh_len)) { 531 NAPI_GRO_CB(p)->same_flow = 0; 532 continue; 533 } 534 } 535 536 skb_gro_pull(skb, gh_len); 537 skb_gro_postpull_rcsum(skb, gh, gh_len); 538 type = gh->proto_type; 539 if (likely(type == htons(ETH_P_TEB))) 540 return call_gro_receive(eth_gro_receive, head, skb); 541 542 ptype = gro_find_receive_by_type(type); 543 if (!ptype) 544 goto out; 545 546 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb); 547 flush = 0; 548 549 out: 550 skb_gro_flush_final(skb, pp, flush); 551 552 return pp; 553 } 554 555 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb, 556 int nhoff) 557 { 558 struct genevehdr *gh; 559 struct packet_offload *ptype; 560 __be16 type; 561 int gh_len; 562 int err = -ENOSYS; 563 564 gh = (struct genevehdr *)(skb->data + nhoff); 565 gh_len = geneve_hlen(gh); 566 type = gh->proto_type; 567 568 /* since skb->encapsulation is set, eth_gro_complete() sets the inner mac header */ 569 if (likely(type == htons(ETH_P_TEB))) 570 return eth_gro_complete(skb, nhoff + gh_len); 571 572 ptype = gro_find_complete_by_type(type); 573 if (ptype) 574 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len); 575 576 skb_set_inner_mac_header(skb, nhoff + gh_len); 577 578 return err; 579 } 580 581 /* Create new listen socket if needed */ 582 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port, 583 bool ipv6, bool ipv6_rx_csum) 584 { 585 struct geneve_net *gn = net_generic(net, geneve_net_id); 586 struct geneve_sock *gs; 587 struct socket *sock; 588 struct udp_tunnel_sock_cfg tunnel_cfg; 589 int h; 590 591 gs = kzalloc(sizeof(*gs), GFP_KERNEL); 592 if (!gs) 593 return ERR_PTR(-ENOMEM); 594 595 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum); 596 if (IS_ERR(sock)) { 597 kfree(gs); 598 return ERR_CAST(sock); 599 } 600 601 gs->sock = sock; 602 gs->refcnt = 1; 603 for (h = 0; h < VNI_HASH_SIZE; ++h) 604 INIT_HLIST_HEAD(&gs->vni_list[h]); 605 606 /* Initialize the geneve udp offloads structure */ 607 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 608 609 /* Mark socket as an encapsulation socket */ 610 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg)); 611 tunnel_cfg.sk_user_data = gs; 612 tunnel_cfg.encap_type = 1; 613 tunnel_cfg.gro_receive = geneve_gro_receive; 614 tunnel_cfg.gro_complete = geneve_gro_complete; 615 tunnel_cfg.encap_rcv = geneve_udp_encap_recv; 616 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup; 617 tunnel_cfg.encap_destroy = NULL; 618 setup_udp_tunnel_sock(net, sock, &tunnel_cfg); 619 list_add(&gs->list, &gn->sock_list); 620 return gs; 621 } 622 623 static void __geneve_sock_release(struct geneve_sock *gs) 624 { 625 if (!gs || --gs->refcnt) 626 return; 627 628 list_del(&gs->list); 629 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 630 udp_tunnel_sock_release(gs->sock); 631 kfree_rcu(gs, rcu); 632 } 633 634 static void geneve_sock_release(struct geneve_dev *geneve) 635 { 636 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4); 637 #if IS_ENABLED(CONFIG_IPV6) 638 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6); 639 640 rcu_assign_pointer(geneve->sock6, NULL); 641 #endif 642 643 rcu_assign_pointer(geneve->sock4, NULL); 644 synchronize_net(); 645 646 __geneve_sock_release(gs4); 647 #if IS_ENABLED(CONFIG_IPV6) 648 __geneve_sock_release(gs6); 649 #endif 650 } 651 652 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn, 653 sa_family_t family, 654 __be16 dst_port) 655 { 656 struct geneve_sock *gs; 657 658 list_for_each_entry(gs, &gn->sock_list, list) { 659 if (inet_sk(gs->sock->sk)->inet_sport == dst_port && 660 geneve_get_sk_family(gs) == family) { 661 return gs; 662 } 663 } 664 return NULL; 665 } 666 667 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6) 668 { 669 struct net *net = geneve->net; 670 struct geneve_net *gn = net_generic(net, geneve_net_id); 671 struct geneve_dev_node *node; 672 struct geneve_sock *gs; 673 __u8 vni[3]; 674 __u32 hash; 675 676 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst); 677 if (gs) { 678 gs->refcnt++; 679 goto out; 680 } 681 682 gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6, 683 geneve->cfg.use_udp6_rx_checksums); 684 if (IS_ERR(gs)) 685 return PTR_ERR(gs); 686 687 out: 688 gs->collect_md = geneve->cfg.collect_md; 689 #if IS_ENABLED(CONFIG_IPV6) 690 if (ipv6) { 691 rcu_assign_pointer(geneve->sock6, gs); 692 node = &geneve->hlist6; 693 } else 694 #endif 695 { 696 rcu_assign_pointer(geneve->sock4, gs); 697 node = &geneve->hlist4; 698 } 699 node->geneve = geneve; 700 701 tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni); 702 hash = geneve_net_vni_hash(vni); 703 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]); 704 return 0; 705 } 706 707 static int geneve_open(struct net_device *dev) 708 { 709 struct geneve_dev *geneve = netdev_priv(dev); 710 bool metadata = geneve->cfg.collect_md; 711 bool ipv4, ipv6; 712 int ret = 0; 713 714 ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata; 715 ipv4 = !ipv6 || metadata; 716 #if IS_ENABLED(CONFIG_IPV6) 717 if (ipv6) { 718 ret = geneve_sock_add(geneve, true); 719 if (ret < 0 && ret != -EAFNOSUPPORT) 720 ipv4 = false; 721 } 722 #endif 723 if (ipv4) 724 ret = geneve_sock_add(geneve, false); 725 if (ret < 0) 726 geneve_sock_release(geneve); 727 728 return ret; 729 } 730 731 static int geneve_stop(struct net_device *dev) 732 { 733 struct geneve_dev *geneve = netdev_priv(dev); 734 735 hlist_del_init_rcu(&geneve->hlist4.hlist); 736 #if IS_ENABLED(CONFIG_IPV6) 737 hlist_del_init_rcu(&geneve->hlist6.hlist); 738 #endif 739 geneve_sock_release(geneve); 740 return 0; 741 } 742 743 static void geneve_build_header(struct genevehdr *geneveh, 744 const struct ip_tunnel_info *info, 745 __be16 inner_proto) 746 { 747 geneveh->ver = GENEVE_VER; 748 geneveh->opt_len = info->options_len / 4; 749 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM); 750 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT); 751 geneveh->rsvd1 = 0; 752 tunnel_id_to_vni(info->key.tun_id, geneveh->vni); 753 geneveh->proto_type = inner_proto; 754 geneveh->rsvd2 = 0; 755 756 if (info->key.tun_flags & TUNNEL_GENEVE_OPT) 757 ip_tunnel_info_opts_get(geneveh->options, info); 758 } 759 760 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb, 761 const struct ip_tunnel_info *info, 762 bool xnet, int ip_hdr_len, 763 bool inner_proto_inherit) 764 { 765 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM); 766 struct genevehdr *gnvh; 767 __be16 inner_proto; 768 int min_headroom; 769 int err; 770 771 skb_reset_mac_header(skb); 772 skb_scrub_packet(skb, xnet); 773 774 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len + 775 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len; 776 err = skb_cow_head(skb, min_headroom); 777 if (unlikely(err)) 778 goto free_dst; 779 780 err = udp_tunnel_handle_offloads(skb, udp_sum); 781 if (err) 782 goto free_dst; 783 784 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len); 785 inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB); 786 geneve_build_header(gnvh, info, inner_proto); 787 skb_set_inner_protocol(skb, inner_proto); 788 return 0; 789 790 free_dst: 791 dst_release(dst); 792 return err; 793 } 794 795 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb, 796 struct net_device *dev, 797 struct geneve_sock *gs4, 798 struct flowi4 *fl4, 799 const struct ip_tunnel_info *info, 800 __be16 dport, __be16 sport) 801 { 802 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 803 struct geneve_dev *geneve = netdev_priv(dev); 804 struct dst_cache *dst_cache; 805 struct rtable *rt = NULL; 806 __u8 tos; 807 808 if (!gs4) 809 return ERR_PTR(-EIO); 810 811 memset(fl4, 0, sizeof(*fl4)); 812 fl4->flowi4_mark = skb->mark; 813 fl4->flowi4_proto = IPPROTO_UDP; 814 fl4->daddr = info->key.u.ipv4.dst; 815 fl4->saddr = info->key.u.ipv4.src; 816 fl4->fl4_dport = dport; 817 fl4->fl4_sport = sport; 818 819 tos = info->key.tos; 820 if ((tos == 1) && !geneve->cfg.collect_md) { 821 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb); 822 use_cache = false; 823 } 824 fl4->flowi4_tos = RT_TOS(tos); 825 826 dst_cache = (struct dst_cache *)&info->dst_cache; 827 if (use_cache) { 828 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr); 829 if (rt) 830 return rt; 831 } 832 rt = ip_route_output_key(geneve->net, fl4); 833 if (IS_ERR(rt)) { 834 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr); 835 return ERR_PTR(-ENETUNREACH); 836 } 837 if (rt->dst.dev == dev) { /* is this necessary? */ 838 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr); 839 ip_rt_put(rt); 840 return ERR_PTR(-ELOOP); 841 } 842 if (use_cache) 843 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr); 844 return rt; 845 } 846 847 #if IS_ENABLED(CONFIG_IPV6) 848 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb, 849 struct net_device *dev, 850 struct geneve_sock *gs6, 851 struct flowi6 *fl6, 852 const struct ip_tunnel_info *info, 853 __be16 dport, __be16 sport) 854 { 855 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 856 struct geneve_dev *geneve = netdev_priv(dev); 857 struct dst_entry *dst = NULL; 858 struct dst_cache *dst_cache; 859 __u8 prio; 860 861 if (!gs6) 862 return ERR_PTR(-EIO); 863 864 memset(fl6, 0, sizeof(*fl6)); 865 fl6->flowi6_mark = skb->mark; 866 fl6->flowi6_proto = IPPROTO_UDP; 867 fl6->daddr = info->key.u.ipv6.dst; 868 fl6->saddr = info->key.u.ipv6.src; 869 fl6->fl6_dport = dport; 870 fl6->fl6_sport = sport; 871 872 prio = info->key.tos; 873 if ((prio == 1) && !geneve->cfg.collect_md) { 874 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb); 875 use_cache = false; 876 } 877 878 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio), 879 info->key.label); 880 dst_cache = (struct dst_cache *)&info->dst_cache; 881 if (use_cache) { 882 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr); 883 if (dst) 884 return dst; 885 } 886 dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6, 887 NULL); 888 if (IS_ERR(dst)) { 889 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr); 890 return ERR_PTR(-ENETUNREACH); 891 } 892 if (dst->dev == dev) { /* is this necessary? */ 893 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr); 894 dst_release(dst); 895 return ERR_PTR(-ELOOP); 896 } 897 898 if (use_cache) 899 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr); 900 return dst; 901 } 902 #endif 903 904 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev, 905 struct geneve_dev *geneve, 906 const struct ip_tunnel_info *info) 907 { 908 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 909 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 910 const struct ip_tunnel_key *key = &info->key; 911 struct rtable *rt; 912 struct flowi4 fl4; 913 __u8 tos, ttl; 914 __be16 df = 0; 915 __be16 sport; 916 int err; 917 918 if (!pskb_inet_may_pull(skb)) 919 return -EINVAL; 920 921 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 922 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info, 923 geneve->cfg.info.key.tp_dst, sport); 924 if (IS_ERR(rt)) 925 return PTR_ERR(rt); 926 927 err = skb_tunnel_check_pmtu(skb, &rt->dst, 928 GENEVE_IPV4_HLEN + info->options_len, 929 netif_is_any_bridge_port(dev)); 930 if (err < 0) { 931 dst_release(&rt->dst); 932 return err; 933 } else if (err) { 934 struct ip_tunnel_info *info; 935 936 info = skb_tunnel_info(skb); 937 if (info) { 938 struct ip_tunnel_info *unclone; 939 940 unclone = skb_tunnel_info_unclone(skb); 941 if (unlikely(!unclone)) { 942 dst_release(&rt->dst); 943 return -ENOMEM; 944 } 945 946 unclone->key.u.ipv4.dst = fl4.saddr; 947 unclone->key.u.ipv4.src = fl4.daddr; 948 } 949 950 if (!pskb_may_pull(skb, ETH_HLEN)) { 951 dst_release(&rt->dst); 952 return -EINVAL; 953 } 954 955 skb->protocol = eth_type_trans(skb, geneve->dev); 956 __netif_rx(skb); 957 dst_release(&rt->dst); 958 return -EMSGSIZE; 959 } 960 961 if (geneve->cfg.collect_md) { 962 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 963 ttl = key->ttl; 964 965 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0; 966 } else { 967 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb); 968 if (geneve->cfg.ttl_inherit) 969 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb); 970 else 971 ttl = key->ttl; 972 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst); 973 974 if (geneve->cfg.df == GENEVE_DF_SET) { 975 df = htons(IP_DF); 976 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) { 977 struct ethhdr *eth = eth_hdr(skb); 978 979 if (ntohs(eth->h_proto) == ETH_P_IPV6) { 980 df = htons(IP_DF); 981 } else if (ntohs(eth->h_proto) == ETH_P_IP) { 982 struct iphdr *iph = ip_hdr(skb); 983 984 if (iph->frag_off & htons(IP_DF)) 985 df = htons(IP_DF); 986 } 987 } 988 } 989 990 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr), 991 geneve->cfg.inner_proto_inherit); 992 if (unlikely(err)) 993 return err; 994 995 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr, 996 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst, 997 !net_eq(geneve->net, dev_net(geneve->dev)), 998 !(info->key.tun_flags & TUNNEL_CSUM)); 999 return 0; 1000 } 1001 1002 #if IS_ENABLED(CONFIG_IPV6) 1003 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev, 1004 struct geneve_dev *geneve, 1005 const struct ip_tunnel_info *info) 1006 { 1007 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 1008 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 1009 const struct ip_tunnel_key *key = &info->key; 1010 struct dst_entry *dst = NULL; 1011 struct flowi6 fl6; 1012 __u8 prio, ttl; 1013 __be16 sport; 1014 int err; 1015 1016 if (!pskb_inet_may_pull(skb)) 1017 return -EINVAL; 1018 1019 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 1020 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info, 1021 geneve->cfg.info.key.tp_dst, sport); 1022 if (IS_ERR(dst)) 1023 return PTR_ERR(dst); 1024 1025 err = skb_tunnel_check_pmtu(skb, dst, 1026 GENEVE_IPV6_HLEN + info->options_len, 1027 netif_is_any_bridge_port(dev)); 1028 if (err < 0) { 1029 dst_release(dst); 1030 return err; 1031 } else if (err) { 1032 struct ip_tunnel_info *info = skb_tunnel_info(skb); 1033 1034 if (info) { 1035 struct ip_tunnel_info *unclone; 1036 1037 unclone = skb_tunnel_info_unclone(skb); 1038 if (unlikely(!unclone)) { 1039 dst_release(dst); 1040 return -ENOMEM; 1041 } 1042 1043 unclone->key.u.ipv6.dst = fl6.saddr; 1044 unclone->key.u.ipv6.src = fl6.daddr; 1045 } 1046 1047 if (!pskb_may_pull(skb, ETH_HLEN)) { 1048 dst_release(dst); 1049 return -EINVAL; 1050 } 1051 1052 skb->protocol = eth_type_trans(skb, geneve->dev); 1053 __netif_rx(skb); 1054 dst_release(dst); 1055 return -EMSGSIZE; 1056 } 1057 1058 if (geneve->cfg.collect_md) { 1059 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 1060 ttl = key->ttl; 1061 } else { 1062 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel), 1063 ip_hdr(skb), skb); 1064 if (geneve->cfg.ttl_inherit) 1065 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb); 1066 else 1067 ttl = key->ttl; 1068 ttl = ttl ? : ip6_dst_hoplimit(dst); 1069 } 1070 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr), 1071 geneve->cfg.inner_proto_inherit); 1072 if (unlikely(err)) 1073 return err; 1074 1075 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev, 1076 &fl6.saddr, &fl6.daddr, prio, ttl, 1077 info->key.label, sport, geneve->cfg.info.key.tp_dst, 1078 !(info->key.tun_flags & TUNNEL_CSUM)); 1079 return 0; 1080 } 1081 #endif 1082 1083 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev) 1084 { 1085 struct geneve_dev *geneve = netdev_priv(dev); 1086 struct ip_tunnel_info *info = NULL; 1087 int err; 1088 1089 if (geneve->cfg.collect_md) { 1090 info = skb_tunnel_info(skb); 1091 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) { 1092 netdev_dbg(dev, "no tunnel metadata\n"); 1093 dev_kfree_skb(skb); 1094 dev->stats.tx_dropped++; 1095 return NETDEV_TX_OK; 1096 } 1097 } else { 1098 info = &geneve->cfg.info; 1099 } 1100 1101 rcu_read_lock(); 1102 #if IS_ENABLED(CONFIG_IPV6) 1103 if (info->mode & IP_TUNNEL_INFO_IPV6) 1104 err = geneve6_xmit_skb(skb, dev, geneve, info); 1105 else 1106 #endif 1107 err = geneve_xmit_skb(skb, dev, geneve, info); 1108 rcu_read_unlock(); 1109 1110 if (likely(!err)) 1111 return NETDEV_TX_OK; 1112 1113 if (err != -EMSGSIZE) 1114 dev_kfree_skb(skb); 1115 1116 if (err == -ELOOP) 1117 dev->stats.collisions++; 1118 else if (err == -ENETUNREACH) 1119 dev->stats.tx_carrier_errors++; 1120 1121 dev->stats.tx_errors++; 1122 return NETDEV_TX_OK; 1123 } 1124 1125 static int geneve_change_mtu(struct net_device *dev, int new_mtu) 1126 { 1127 if (new_mtu > dev->max_mtu) 1128 new_mtu = dev->max_mtu; 1129 else if (new_mtu < dev->min_mtu) 1130 new_mtu = dev->min_mtu; 1131 1132 dev->mtu = new_mtu; 1133 return 0; 1134 } 1135 1136 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 1137 { 1138 struct ip_tunnel_info *info = skb_tunnel_info(skb); 1139 struct geneve_dev *geneve = netdev_priv(dev); 1140 __be16 sport; 1141 1142 if (ip_tunnel_info_af(info) == AF_INET) { 1143 struct rtable *rt; 1144 struct flowi4 fl4; 1145 1146 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 1147 sport = udp_flow_src_port(geneve->net, skb, 1148 1, USHRT_MAX, true); 1149 1150 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info, 1151 geneve->cfg.info.key.tp_dst, sport); 1152 if (IS_ERR(rt)) 1153 return PTR_ERR(rt); 1154 1155 ip_rt_put(rt); 1156 info->key.u.ipv4.src = fl4.saddr; 1157 #if IS_ENABLED(CONFIG_IPV6) 1158 } else if (ip_tunnel_info_af(info) == AF_INET6) { 1159 struct dst_entry *dst; 1160 struct flowi6 fl6; 1161 1162 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 1163 sport = udp_flow_src_port(geneve->net, skb, 1164 1, USHRT_MAX, true); 1165 1166 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info, 1167 geneve->cfg.info.key.tp_dst, sport); 1168 if (IS_ERR(dst)) 1169 return PTR_ERR(dst); 1170 1171 dst_release(dst); 1172 info->key.u.ipv6.src = fl6.saddr; 1173 #endif 1174 } else { 1175 return -EINVAL; 1176 } 1177 1178 info->key.tp_src = sport; 1179 info->key.tp_dst = geneve->cfg.info.key.tp_dst; 1180 return 0; 1181 } 1182 1183 static const struct net_device_ops geneve_netdev_ops = { 1184 .ndo_init = geneve_init, 1185 .ndo_uninit = geneve_uninit, 1186 .ndo_open = geneve_open, 1187 .ndo_stop = geneve_stop, 1188 .ndo_start_xmit = geneve_xmit, 1189 .ndo_get_stats64 = dev_get_tstats64, 1190 .ndo_change_mtu = geneve_change_mtu, 1191 .ndo_validate_addr = eth_validate_addr, 1192 .ndo_set_mac_address = eth_mac_addr, 1193 .ndo_fill_metadata_dst = geneve_fill_metadata_dst, 1194 }; 1195 1196 static void geneve_get_drvinfo(struct net_device *dev, 1197 struct ethtool_drvinfo *drvinfo) 1198 { 1199 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version)); 1200 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver)); 1201 } 1202 1203 static const struct ethtool_ops geneve_ethtool_ops = { 1204 .get_drvinfo = geneve_get_drvinfo, 1205 .get_link = ethtool_op_get_link, 1206 }; 1207 1208 /* Info for udev, that this is a virtual tunnel endpoint */ 1209 static struct device_type geneve_type = { 1210 .name = "geneve", 1211 }; 1212 1213 /* Calls the ndo_udp_tunnel_add of the caller in order to 1214 * supply the listening GENEVE udp ports. Callers are expected 1215 * to implement the ndo_udp_tunnel_add. 1216 */ 1217 static void geneve_offload_rx_ports(struct net_device *dev, bool push) 1218 { 1219 struct net *net = dev_net(dev); 1220 struct geneve_net *gn = net_generic(net, geneve_net_id); 1221 struct geneve_sock *gs; 1222 1223 rcu_read_lock(); 1224 list_for_each_entry_rcu(gs, &gn->sock_list, list) { 1225 if (push) { 1226 udp_tunnel_push_rx_port(dev, gs->sock, 1227 UDP_TUNNEL_TYPE_GENEVE); 1228 } else { 1229 udp_tunnel_drop_rx_port(dev, gs->sock, 1230 UDP_TUNNEL_TYPE_GENEVE); 1231 } 1232 } 1233 rcu_read_unlock(); 1234 } 1235 1236 /* Initialize the device structure. */ 1237 static void geneve_setup(struct net_device *dev) 1238 { 1239 ether_setup(dev); 1240 1241 dev->netdev_ops = &geneve_netdev_ops; 1242 dev->ethtool_ops = &geneve_ethtool_ops; 1243 dev->needs_free_netdev = true; 1244 1245 SET_NETDEV_DEVTYPE(dev, &geneve_type); 1246 1247 dev->features |= NETIF_F_LLTX; 1248 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST; 1249 dev->features |= NETIF_F_RXCSUM; 1250 dev->features |= NETIF_F_GSO_SOFTWARE; 1251 1252 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST; 1253 dev->hw_features |= NETIF_F_RXCSUM; 1254 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 1255 1256 /* MTU range: 68 - (something less than 65535) */ 1257 dev->min_mtu = ETH_MIN_MTU; 1258 /* The max_mtu calculation does not take account of GENEVE 1259 * options, to avoid excluding potentially valid 1260 * configurations. This will be further reduced by IPvX hdr size. 1261 */ 1262 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len; 1263 1264 netif_keep_dst(dev); 1265 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1266 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE; 1267 eth_hw_addr_random(dev); 1268 } 1269 1270 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = { 1271 [IFLA_GENEVE_UNSPEC] = { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT }, 1272 [IFLA_GENEVE_ID] = { .type = NLA_U32 }, 1273 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) }, 1274 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) }, 1275 [IFLA_GENEVE_TTL] = { .type = NLA_U8 }, 1276 [IFLA_GENEVE_TOS] = { .type = NLA_U8 }, 1277 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 }, 1278 [IFLA_GENEVE_PORT] = { .type = NLA_U16 }, 1279 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1280 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 }, 1281 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 }, 1282 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 }, 1283 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 }, 1284 [IFLA_GENEVE_DF] = { .type = NLA_U8 }, 1285 [IFLA_GENEVE_INNER_PROTO_INHERIT] = { .type = NLA_FLAG }, 1286 }; 1287 1288 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[], 1289 struct netlink_ext_ack *extack) 1290 { 1291 if (tb[IFLA_ADDRESS]) { 1292 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1293 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1294 "Provided link layer address is not Ethernet"); 1295 return -EINVAL; 1296 } 1297 1298 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1299 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1300 "Provided Ethernet address is not unicast"); 1301 return -EADDRNOTAVAIL; 1302 } 1303 } 1304 1305 if (!data) { 1306 NL_SET_ERR_MSG(extack, 1307 "Not enough attributes provided to perform the operation"); 1308 return -EINVAL; 1309 } 1310 1311 if (data[IFLA_GENEVE_ID]) { 1312 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1313 1314 if (vni >= GENEVE_N_VID) { 1315 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID], 1316 "Geneve ID must be lower than 16777216"); 1317 return -ERANGE; 1318 } 1319 } 1320 1321 if (data[IFLA_GENEVE_DF]) { 1322 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]); 1323 1324 if (df < 0 || df > GENEVE_DF_MAX) { 1325 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF], 1326 "Invalid DF attribute"); 1327 return -EINVAL; 1328 } 1329 } 1330 1331 return 0; 1332 } 1333 1334 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn, 1335 const struct ip_tunnel_info *info, 1336 bool *tun_on_same_port, 1337 bool *tun_collect_md) 1338 { 1339 struct geneve_dev *geneve, *t = NULL; 1340 1341 *tun_on_same_port = false; 1342 *tun_collect_md = false; 1343 list_for_each_entry(geneve, &gn->geneve_list, next) { 1344 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) { 1345 *tun_collect_md = geneve->cfg.collect_md; 1346 *tun_on_same_port = true; 1347 } 1348 if (info->key.tun_id == geneve->cfg.info.key.tun_id && 1349 info->key.tp_dst == geneve->cfg.info.key.tp_dst && 1350 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u))) 1351 t = geneve; 1352 } 1353 return t; 1354 } 1355 1356 static bool is_tnl_info_zero(const struct ip_tunnel_info *info) 1357 { 1358 return !(info->key.tun_id || info->key.tun_flags || info->key.tos || 1359 info->key.ttl || info->key.label || info->key.tp_src || 1360 memchr_inv(&info->key.u, 0, sizeof(info->key.u))); 1361 } 1362 1363 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a, 1364 struct ip_tunnel_info *b) 1365 { 1366 if (ip_tunnel_info_af(a) == AF_INET) 1367 return a->key.u.ipv4.dst == b->key.u.ipv4.dst; 1368 else 1369 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst); 1370 } 1371 1372 static int geneve_configure(struct net *net, struct net_device *dev, 1373 struct netlink_ext_ack *extack, 1374 const struct geneve_config *cfg) 1375 { 1376 struct geneve_net *gn = net_generic(net, geneve_net_id); 1377 struct geneve_dev *t, *geneve = netdev_priv(dev); 1378 const struct ip_tunnel_info *info = &cfg->info; 1379 bool tun_collect_md, tun_on_same_port; 1380 int err, encap_len; 1381 1382 if (cfg->collect_md && !is_tnl_info_zero(info)) { 1383 NL_SET_ERR_MSG(extack, 1384 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified"); 1385 return -EINVAL; 1386 } 1387 1388 geneve->net = net; 1389 geneve->dev = dev; 1390 1391 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md); 1392 if (t) 1393 return -EBUSY; 1394 1395 /* make enough headroom for basic scenario */ 1396 encap_len = GENEVE_BASE_HLEN + ETH_HLEN; 1397 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) { 1398 encap_len += sizeof(struct iphdr); 1399 dev->max_mtu -= sizeof(struct iphdr); 1400 } else { 1401 encap_len += sizeof(struct ipv6hdr); 1402 dev->max_mtu -= sizeof(struct ipv6hdr); 1403 } 1404 dev->needed_headroom = encap_len + ETH_HLEN; 1405 1406 if (cfg->collect_md) { 1407 if (tun_on_same_port) { 1408 NL_SET_ERR_MSG(extack, 1409 "There can be only one externally controlled device on a destination port"); 1410 return -EPERM; 1411 } 1412 } else { 1413 if (tun_collect_md) { 1414 NL_SET_ERR_MSG(extack, 1415 "There already exists an externally controlled device on this destination port"); 1416 return -EPERM; 1417 } 1418 } 1419 1420 dst_cache_reset(&geneve->cfg.info.dst_cache); 1421 memcpy(&geneve->cfg, cfg, sizeof(*cfg)); 1422 1423 if (geneve->cfg.inner_proto_inherit) { 1424 dev->header_ops = NULL; 1425 dev->type = ARPHRD_NONE; 1426 dev->hard_header_len = 0; 1427 dev->addr_len = 0; 1428 dev->flags = IFF_NOARP; 1429 } 1430 1431 err = register_netdevice(dev); 1432 if (err) 1433 return err; 1434 1435 list_add(&geneve->next, &gn->geneve_list); 1436 return 0; 1437 } 1438 1439 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port) 1440 { 1441 memset(info, 0, sizeof(*info)); 1442 info->key.tp_dst = htons(dst_port); 1443 } 1444 1445 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[], 1446 struct netlink_ext_ack *extack, 1447 struct geneve_config *cfg, bool changelink) 1448 { 1449 struct ip_tunnel_info *info = &cfg->info; 1450 int attrtype; 1451 1452 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) { 1453 NL_SET_ERR_MSG(extack, 1454 "Cannot specify both IPv4 and IPv6 Remote addresses"); 1455 return -EINVAL; 1456 } 1457 1458 if (data[IFLA_GENEVE_REMOTE]) { 1459 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) { 1460 attrtype = IFLA_GENEVE_REMOTE; 1461 goto change_notsup; 1462 } 1463 1464 info->key.u.ipv4.dst = 1465 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]); 1466 1467 if (ipv4_is_multicast(info->key.u.ipv4.dst)) { 1468 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE], 1469 "Remote IPv4 address cannot be Multicast"); 1470 return -EINVAL; 1471 } 1472 } 1473 1474 if (data[IFLA_GENEVE_REMOTE6]) { 1475 #if IS_ENABLED(CONFIG_IPV6) 1476 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) { 1477 attrtype = IFLA_GENEVE_REMOTE6; 1478 goto change_notsup; 1479 } 1480 1481 info->mode = IP_TUNNEL_INFO_IPV6; 1482 info->key.u.ipv6.dst = 1483 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]); 1484 1485 if (ipv6_addr_type(&info->key.u.ipv6.dst) & 1486 IPV6_ADDR_LINKLOCAL) { 1487 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1488 "Remote IPv6 address cannot be link-local"); 1489 return -EINVAL; 1490 } 1491 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) { 1492 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1493 "Remote IPv6 address cannot be Multicast"); 1494 return -EINVAL; 1495 } 1496 info->key.tun_flags |= TUNNEL_CSUM; 1497 cfg->use_udp6_rx_checksums = true; 1498 #else 1499 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1500 "IPv6 support not enabled in the kernel"); 1501 return -EPFNOSUPPORT; 1502 #endif 1503 } 1504 1505 if (data[IFLA_GENEVE_ID]) { 1506 __u32 vni; 1507 __u8 tvni[3]; 1508 __be64 tunid; 1509 1510 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1511 tvni[0] = (vni & 0x00ff0000) >> 16; 1512 tvni[1] = (vni & 0x0000ff00) >> 8; 1513 tvni[2] = vni & 0x000000ff; 1514 1515 tunid = vni_to_tunnel_id(tvni); 1516 if (changelink && (tunid != info->key.tun_id)) { 1517 attrtype = IFLA_GENEVE_ID; 1518 goto change_notsup; 1519 } 1520 info->key.tun_id = tunid; 1521 } 1522 1523 if (data[IFLA_GENEVE_TTL_INHERIT]) { 1524 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT])) 1525 cfg->ttl_inherit = true; 1526 else 1527 cfg->ttl_inherit = false; 1528 } else if (data[IFLA_GENEVE_TTL]) { 1529 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]); 1530 cfg->ttl_inherit = false; 1531 } 1532 1533 if (data[IFLA_GENEVE_TOS]) 1534 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]); 1535 1536 if (data[IFLA_GENEVE_DF]) 1537 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]); 1538 1539 if (data[IFLA_GENEVE_LABEL]) { 1540 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) & 1541 IPV6_FLOWLABEL_MASK; 1542 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) { 1543 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL], 1544 "Label attribute only applies for IPv6 Geneve devices"); 1545 return -EINVAL; 1546 } 1547 } 1548 1549 if (data[IFLA_GENEVE_PORT]) { 1550 if (changelink) { 1551 attrtype = IFLA_GENEVE_PORT; 1552 goto change_notsup; 1553 } 1554 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]); 1555 } 1556 1557 if (data[IFLA_GENEVE_COLLECT_METADATA]) { 1558 if (changelink) { 1559 attrtype = IFLA_GENEVE_COLLECT_METADATA; 1560 goto change_notsup; 1561 } 1562 cfg->collect_md = true; 1563 } 1564 1565 if (data[IFLA_GENEVE_UDP_CSUM]) { 1566 if (changelink) { 1567 attrtype = IFLA_GENEVE_UDP_CSUM; 1568 goto change_notsup; 1569 } 1570 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM])) 1571 info->key.tun_flags |= TUNNEL_CSUM; 1572 } 1573 1574 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) { 1575 #if IS_ENABLED(CONFIG_IPV6) 1576 if (changelink) { 1577 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX; 1578 goto change_notsup; 1579 } 1580 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX])) 1581 info->key.tun_flags &= ~TUNNEL_CSUM; 1582 #else 1583 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX], 1584 "IPv6 support not enabled in the kernel"); 1585 return -EPFNOSUPPORT; 1586 #endif 1587 } 1588 1589 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) { 1590 #if IS_ENABLED(CONFIG_IPV6) 1591 if (changelink) { 1592 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX; 1593 goto change_notsup; 1594 } 1595 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX])) 1596 cfg->use_udp6_rx_checksums = false; 1597 #else 1598 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX], 1599 "IPv6 support not enabled in the kernel"); 1600 return -EPFNOSUPPORT; 1601 #endif 1602 } 1603 1604 if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) { 1605 if (changelink) { 1606 attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT; 1607 goto change_notsup; 1608 } 1609 cfg->inner_proto_inherit = true; 1610 } 1611 1612 return 0; 1613 change_notsup: 1614 NL_SET_ERR_MSG_ATTR(extack, data[attrtype], 1615 "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported"); 1616 return -EOPNOTSUPP; 1617 } 1618 1619 static void geneve_link_config(struct net_device *dev, 1620 struct ip_tunnel_info *info, struct nlattr *tb[]) 1621 { 1622 struct geneve_dev *geneve = netdev_priv(dev); 1623 int ldev_mtu = 0; 1624 1625 if (tb[IFLA_MTU]) { 1626 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU])); 1627 return; 1628 } 1629 1630 switch (ip_tunnel_info_af(info)) { 1631 case AF_INET: { 1632 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst }; 1633 struct rtable *rt = ip_route_output_key(geneve->net, &fl4); 1634 1635 if (!IS_ERR(rt) && rt->dst.dev) { 1636 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN; 1637 ip_rt_put(rt); 1638 } 1639 break; 1640 } 1641 #if IS_ENABLED(CONFIG_IPV6) 1642 case AF_INET6: { 1643 struct rt6_info *rt; 1644 1645 if (!__in6_dev_get(dev)) 1646 break; 1647 1648 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0, 1649 NULL, 0); 1650 1651 if (rt && rt->dst.dev) 1652 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN; 1653 ip6_rt_put(rt); 1654 break; 1655 } 1656 #endif 1657 } 1658 1659 if (ldev_mtu <= 0) 1660 return; 1661 1662 geneve_change_mtu(dev, ldev_mtu - info->options_len); 1663 } 1664 1665 static int geneve_newlink(struct net *net, struct net_device *dev, 1666 struct nlattr *tb[], struct nlattr *data[], 1667 struct netlink_ext_ack *extack) 1668 { 1669 struct geneve_config cfg = { 1670 .df = GENEVE_DF_UNSET, 1671 .use_udp6_rx_checksums = false, 1672 .ttl_inherit = false, 1673 .collect_md = false, 1674 }; 1675 int err; 1676 1677 init_tnl_info(&cfg.info, GENEVE_UDP_PORT); 1678 err = geneve_nl2info(tb, data, extack, &cfg, false); 1679 if (err) 1680 return err; 1681 1682 err = geneve_configure(net, dev, extack, &cfg); 1683 if (err) 1684 return err; 1685 1686 geneve_link_config(dev, &cfg.info, tb); 1687 1688 return 0; 1689 } 1690 1691 /* Quiesces the geneve device data path for both TX and RX. 1692 * 1693 * On transmit geneve checks for non-NULL geneve_sock before it proceeds. 1694 * So, if we set that socket to NULL under RCU and wait for synchronize_net() 1695 * to complete for the existing set of in-flight packets to be transmitted, 1696 * then we would have quiesced the transmit data path. All the future packets 1697 * will get dropped until we unquiesce the data path. 1698 * 1699 * On receive geneve dereference the geneve_sock stashed in the socket. So, 1700 * if we set that to NULL under RCU and wait for synchronize_net() to 1701 * complete, then we would have quiesced the receive data path. 1702 */ 1703 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4, 1704 struct geneve_sock **gs6) 1705 { 1706 *gs4 = rtnl_dereference(geneve->sock4); 1707 rcu_assign_pointer(geneve->sock4, NULL); 1708 if (*gs4) 1709 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL); 1710 #if IS_ENABLED(CONFIG_IPV6) 1711 *gs6 = rtnl_dereference(geneve->sock6); 1712 rcu_assign_pointer(geneve->sock6, NULL); 1713 if (*gs6) 1714 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL); 1715 #else 1716 *gs6 = NULL; 1717 #endif 1718 synchronize_net(); 1719 } 1720 1721 /* Resumes the geneve device data path for both TX and RX. */ 1722 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4, 1723 struct geneve_sock __maybe_unused *gs6) 1724 { 1725 rcu_assign_pointer(geneve->sock4, gs4); 1726 if (gs4) 1727 rcu_assign_sk_user_data(gs4->sock->sk, gs4); 1728 #if IS_ENABLED(CONFIG_IPV6) 1729 rcu_assign_pointer(geneve->sock6, gs6); 1730 if (gs6) 1731 rcu_assign_sk_user_data(gs6->sock->sk, gs6); 1732 #endif 1733 synchronize_net(); 1734 } 1735 1736 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[], 1737 struct nlattr *data[], 1738 struct netlink_ext_ack *extack) 1739 { 1740 struct geneve_dev *geneve = netdev_priv(dev); 1741 struct geneve_sock *gs4, *gs6; 1742 struct geneve_config cfg; 1743 int err; 1744 1745 /* If the geneve device is configured for metadata (or externally 1746 * controlled, for example, OVS), then nothing can be changed. 1747 */ 1748 if (geneve->cfg.collect_md) 1749 return -EOPNOTSUPP; 1750 1751 /* Start with the existing info. */ 1752 memcpy(&cfg, &geneve->cfg, sizeof(cfg)); 1753 err = geneve_nl2info(tb, data, extack, &cfg, true); 1754 if (err) 1755 return err; 1756 1757 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) { 1758 dst_cache_reset(&cfg.info.dst_cache); 1759 geneve_link_config(dev, &cfg.info, tb); 1760 } 1761 1762 geneve_quiesce(geneve, &gs4, &gs6); 1763 memcpy(&geneve->cfg, &cfg, sizeof(cfg)); 1764 geneve_unquiesce(geneve, gs4, gs6); 1765 1766 return 0; 1767 } 1768 1769 static void geneve_dellink(struct net_device *dev, struct list_head *head) 1770 { 1771 struct geneve_dev *geneve = netdev_priv(dev); 1772 1773 list_del(&geneve->next); 1774 unregister_netdevice_queue(dev, head); 1775 } 1776 1777 static size_t geneve_get_size(const struct net_device *dev) 1778 { 1779 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */ 1780 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */ 1781 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */ 1782 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */ 1783 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */ 1784 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */ 1785 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */ 1786 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */ 1787 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */ 1788 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */ 1789 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */ 1790 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */ 1791 nla_total_size(0) + /* IFLA_GENEVE_INNER_PROTO_INHERIT */ 1792 0; 1793 } 1794 1795 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev) 1796 { 1797 struct geneve_dev *geneve = netdev_priv(dev); 1798 struct ip_tunnel_info *info = &geneve->cfg.info; 1799 bool ttl_inherit = geneve->cfg.ttl_inherit; 1800 bool metadata = geneve->cfg.collect_md; 1801 __u8 tmp_vni[3]; 1802 __u32 vni; 1803 1804 tunnel_id_to_vni(info->key.tun_id, tmp_vni); 1805 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2]; 1806 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni)) 1807 goto nla_put_failure; 1808 1809 if (!metadata && ip_tunnel_info_af(info) == AF_INET) { 1810 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE, 1811 info->key.u.ipv4.dst)) 1812 goto nla_put_failure; 1813 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM, 1814 !!(info->key.tun_flags & TUNNEL_CSUM))) 1815 goto nla_put_failure; 1816 1817 #if IS_ENABLED(CONFIG_IPV6) 1818 } else if (!metadata) { 1819 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6, 1820 &info->key.u.ipv6.dst)) 1821 goto nla_put_failure; 1822 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX, 1823 !(info->key.tun_flags & TUNNEL_CSUM))) 1824 goto nla_put_failure; 1825 #endif 1826 } 1827 1828 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) || 1829 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) || 1830 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label)) 1831 goto nla_put_failure; 1832 1833 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df)) 1834 goto nla_put_failure; 1835 1836 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst)) 1837 goto nla_put_failure; 1838 1839 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA)) 1840 goto nla_put_failure; 1841 1842 #if IS_ENABLED(CONFIG_IPV6) 1843 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX, 1844 !geneve->cfg.use_udp6_rx_checksums)) 1845 goto nla_put_failure; 1846 #endif 1847 1848 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit)) 1849 goto nla_put_failure; 1850 1851 if (geneve->cfg.inner_proto_inherit && 1852 nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT)) 1853 goto nla_put_failure; 1854 1855 return 0; 1856 1857 nla_put_failure: 1858 return -EMSGSIZE; 1859 } 1860 1861 static struct rtnl_link_ops geneve_link_ops __read_mostly = { 1862 .kind = "geneve", 1863 .maxtype = IFLA_GENEVE_MAX, 1864 .policy = geneve_policy, 1865 .priv_size = sizeof(struct geneve_dev), 1866 .setup = geneve_setup, 1867 .validate = geneve_validate, 1868 .newlink = geneve_newlink, 1869 .changelink = geneve_changelink, 1870 .dellink = geneve_dellink, 1871 .get_size = geneve_get_size, 1872 .fill_info = geneve_fill_info, 1873 }; 1874 1875 struct net_device *geneve_dev_create_fb(struct net *net, const char *name, 1876 u8 name_assign_type, u16 dst_port) 1877 { 1878 struct nlattr *tb[IFLA_MAX + 1]; 1879 struct net_device *dev; 1880 LIST_HEAD(list_kill); 1881 int err; 1882 struct geneve_config cfg = { 1883 .df = GENEVE_DF_UNSET, 1884 .use_udp6_rx_checksums = true, 1885 .ttl_inherit = false, 1886 .collect_md = true, 1887 }; 1888 1889 memset(tb, 0, sizeof(tb)); 1890 dev = rtnl_create_link(net, name, name_assign_type, 1891 &geneve_link_ops, tb, NULL); 1892 if (IS_ERR(dev)) 1893 return dev; 1894 1895 init_tnl_info(&cfg.info, dst_port); 1896 err = geneve_configure(net, dev, NULL, &cfg); 1897 if (err) { 1898 free_netdev(dev); 1899 return ERR_PTR(err); 1900 } 1901 1902 /* openvswitch users expect packet sizes to be unrestricted, 1903 * so set the largest MTU we can. 1904 */ 1905 err = geneve_change_mtu(dev, IP_MAX_MTU); 1906 if (err) 1907 goto err; 1908 1909 err = rtnl_configure_link(dev, NULL); 1910 if (err < 0) 1911 goto err; 1912 1913 return dev; 1914 err: 1915 geneve_dellink(dev, &list_kill); 1916 unregister_netdevice_many(&list_kill); 1917 return ERR_PTR(err); 1918 } 1919 EXPORT_SYMBOL_GPL(geneve_dev_create_fb); 1920 1921 static int geneve_netdevice_event(struct notifier_block *unused, 1922 unsigned long event, void *ptr) 1923 { 1924 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1925 1926 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO) 1927 geneve_offload_rx_ports(dev, true); 1928 else if (event == NETDEV_UDP_TUNNEL_DROP_INFO) 1929 geneve_offload_rx_ports(dev, false); 1930 1931 return NOTIFY_DONE; 1932 } 1933 1934 static struct notifier_block geneve_notifier_block __read_mostly = { 1935 .notifier_call = geneve_netdevice_event, 1936 }; 1937 1938 static __net_init int geneve_init_net(struct net *net) 1939 { 1940 struct geneve_net *gn = net_generic(net, geneve_net_id); 1941 1942 INIT_LIST_HEAD(&gn->geneve_list); 1943 INIT_LIST_HEAD(&gn->sock_list); 1944 return 0; 1945 } 1946 1947 static void geneve_destroy_tunnels(struct net *net, struct list_head *head) 1948 { 1949 struct geneve_net *gn = net_generic(net, geneve_net_id); 1950 struct geneve_dev *geneve, *next; 1951 struct net_device *dev, *aux; 1952 1953 /* gather any geneve devices that were moved into this ns */ 1954 for_each_netdev_safe(net, dev, aux) 1955 if (dev->rtnl_link_ops == &geneve_link_ops) 1956 unregister_netdevice_queue(dev, head); 1957 1958 /* now gather any other geneve devices that were created in this ns */ 1959 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) { 1960 /* If geneve->dev is in the same netns, it was already added 1961 * to the list by the previous loop. 1962 */ 1963 if (!net_eq(dev_net(geneve->dev), net)) 1964 unregister_netdevice_queue(geneve->dev, head); 1965 } 1966 } 1967 1968 static void __net_exit geneve_exit_batch_net(struct list_head *net_list) 1969 { 1970 struct net *net; 1971 LIST_HEAD(list); 1972 1973 rtnl_lock(); 1974 list_for_each_entry(net, net_list, exit_list) 1975 geneve_destroy_tunnels(net, &list); 1976 1977 /* unregister the devices gathered above */ 1978 unregister_netdevice_many(&list); 1979 rtnl_unlock(); 1980 1981 list_for_each_entry(net, net_list, exit_list) { 1982 const struct geneve_net *gn = net_generic(net, geneve_net_id); 1983 1984 WARN_ON_ONCE(!list_empty(&gn->sock_list)); 1985 } 1986 } 1987 1988 static struct pernet_operations geneve_net_ops = { 1989 .init = geneve_init_net, 1990 .exit_batch = geneve_exit_batch_net, 1991 .id = &geneve_net_id, 1992 .size = sizeof(struct geneve_net), 1993 }; 1994 1995 static int __init geneve_init_module(void) 1996 { 1997 int rc; 1998 1999 rc = register_pernet_subsys(&geneve_net_ops); 2000 if (rc) 2001 goto out1; 2002 2003 rc = register_netdevice_notifier(&geneve_notifier_block); 2004 if (rc) 2005 goto out2; 2006 2007 rc = rtnl_link_register(&geneve_link_ops); 2008 if (rc) 2009 goto out3; 2010 2011 return 0; 2012 out3: 2013 unregister_netdevice_notifier(&geneve_notifier_block); 2014 out2: 2015 unregister_pernet_subsys(&geneve_net_ops); 2016 out1: 2017 return rc; 2018 } 2019 late_initcall(geneve_init_module); 2020 2021 static void __exit geneve_cleanup_module(void) 2022 { 2023 rtnl_link_unregister(&geneve_link_ops); 2024 unregister_netdevice_notifier(&geneve_notifier_block); 2025 unregister_pernet_subsys(&geneve_net_ops); 2026 } 2027 module_exit(geneve_cleanup_module); 2028 2029 MODULE_LICENSE("GPL"); 2030 MODULE_VERSION(GENEVE_NETDEV_VER); 2031 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>"); 2032 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic"); 2033 MODULE_ALIAS_RTNL_LINK("geneve"); 2034