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