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 ptype = gro_find_receive_by_type(type); 521 if (!ptype) 522 goto out; 523 524 skb_gro_pull(skb, gh_len); 525 skb_gro_postpull_rcsum(skb, gh, gh_len); 526 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb); 527 flush = 0; 528 529 out: 530 skb_gro_flush_final(skb, pp, flush); 531 532 return pp; 533 } 534 535 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb, 536 int nhoff) 537 { 538 struct genevehdr *gh; 539 struct packet_offload *ptype; 540 __be16 type; 541 int gh_len; 542 int err = -ENOSYS; 543 544 gh = (struct genevehdr *)(skb->data + nhoff); 545 gh_len = geneve_hlen(gh); 546 type = gh->proto_type; 547 548 rcu_read_lock(); 549 ptype = gro_find_complete_by_type(type); 550 if (ptype) 551 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len); 552 553 rcu_read_unlock(); 554 555 skb_set_inner_mac_header(skb, nhoff + gh_len); 556 557 return err; 558 } 559 560 /* Create new listen socket if needed */ 561 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port, 562 bool ipv6, bool ipv6_rx_csum) 563 { 564 struct geneve_net *gn = net_generic(net, geneve_net_id); 565 struct geneve_sock *gs; 566 struct socket *sock; 567 struct udp_tunnel_sock_cfg tunnel_cfg; 568 int h; 569 570 gs = kzalloc(sizeof(*gs), GFP_KERNEL); 571 if (!gs) 572 return ERR_PTR(-ENOMEM); 573 574 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum); 575 if (IS_ERR(sock)) { 576 kfree(gs); 577 return ERR_CAST(sock); 578 } 579 580 gs->sock = sock; 581 gs->refcnt = 1; 582 for (h = 0; h < VNI_HASH_SIZE; ++h) 583 INIT_HLIST_HEAD(&gs->vni_list[h]); 584 585 /* Initialize the geneve udp offloads structure */ 586 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 587 588 /* Mark socket as an encapsulation socket */ 589 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg)); 590 tunnel_cfg.sk_user_data = gs; 591 tunnel_cfg.encap_type = 1; 592 tunnel_cfg.gro_receive = geneve_gro_receive; 593 tunnel_cfg.gro_complete = geneve_gro_complete; 594 tunnel_cfg.encap_rcv = geneve_udp_encap_recv; 595 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup; 596 tunnel_cfg.encap_destroy = NULL; 597 setup_udp_tunnel_sock(net, sock, &tunnel_cfg); 598 list_add(&gs->list, &gn->sock_list); 599 return gs; 600 } 601 602 static void __geneve_sock_release(struct geneve_sock *gs) 603 { 604 if (!gs || --gs->refcnt) 605 return; 606 607 list_del(&gs->list); 608 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 609 udp_tunnel_sock_release(gs->sock); 610 kfree_rcu(gs, rcu); 611 } 612 613 static void geneve_sock_release(struct geneve_dev *geneve) 614 { 615 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4); 616 #if IS_ENABLED(CONFIG_IPV6) 617 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6); 618 619 rcu_assign_pointer(geneve->sock6, NULL); 620 #endif 621 622 rcu_assign_pointer(geneve->sock4, NULL); 623 synchronize_net(); 624 625 __geneve_sock_release(gs4); 626 #if IS_ENABLED(CONFIG_IPV6) 627 __geneve_sock_release(gs6); 628 #endif 629 } 630 631 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn, 632 sa_family_t family, 633 __be16 dst_port) 634 { 635 struct geneve_sock *gs; 636 637 list_for_each_entry(gs, &gn->sock_list, list) { 638 if (inet_sk(gs->sock->sk)->inet_sport == dst_port && 639 geneve_get_sk_family(gs) == family) { 640 return gs; 641 } 642 } 643 return NULL; 644 } 645 646 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6) 647 { 648 struct net *net = geneve->net; 649 struct geneve_net *gn = net_generic(net, geneve_net_id); 650 struct geneve_dev_node *node; 651 struct geneve_sock *gs; 652 __u8 vni[3]; 653 __u32 hash; 654 655 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst); 656 if (gs) { 657 gs->refcnt++; 658 goto out; 659 } 660 661 gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6, 662 geneve->cfg.use_udp6_rx_checksums); 663 if (IS_ERR(gs)) 664 return PTR_ERR(gs); 665 666 out: 667 gs->collect_md = geneve->cfg.collect_md; 668 #if IS_ENABLED(CONFIG_IPV6) 669 if (ipv6) { 670 rcu_assign_pointer(geneve->sock6, gs); 671 node = &geneve->hlist6; 672 } else 673 #endif 674 { 675 rcu_assign_pointer(geneve->sock4, gs); 676 node = &geneve->hlist4; 677 } 678 node->geneve = geneve; 679 680 tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni); 681 hash = geneve_net_vni_hash(vni); 682 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]); 683 return 0; 684 } 685 686 static int geneve_open(struct net_device *dev) 687 { 688 struct geneve_dev *geneve = netdev_priv(dev); 689 bool metadata = geneve->cfg.collect_md; 690 bool ipv4, ipv6; 691 int ret = 0; 692 693 ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata; 694 ipv4 = !ipv6 || metadata; 695 #if IS_ENABLED(CONFIG_IPV6) 696 if (ipv6) { 697 ret = geneve_sock_add(geneve, true); 698 if (ret < 0 && ret != -EAFNOSUPPORT) 699 ipv4 = false; 700 } 701 #endif 702 if (ipv4) 703 ret = geneve_sock_add(geneve, false); 704 if (ret < 0) 705 geneve_sock_release(geneve); 706 707 return ret; 708 } 709 710 static int geneve_stop(struct net_device *dev) 711 { 712 struct geneve_dev *geneve = netdev_priv(dev); 713 714 hlist_del_init_rcu(&geneve->hlist4.hlist); 715 #if IS_ENABLED(CONFIG_IPV6) 716 hlist_del_init_rcu(&geneve->hlist6.hlist); 717 #endif 718 geneve_sock_release(geneve); 719 return 0; 720 } 721 722 static void geneve_build_header(struct genevehdr *geneveh, 723 const struct ip_tunnel_info *info) 724 { 725 geneveh->ver = GENEVE_VER; 726 geneveh->opt_len = info->options_len / 4; 727 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM); 728 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT); 729 geneveh->rsvd1 = 0; 730 tunnel_id_to_vni(info->key.tun_id, geneveh->vni); 731 geneveh->proto_type = htons(ETH_P_TEB); 732 geneveh->rsvd2 = 0; 733 734 if (info->key.tun_flags & TUNNEL_GENEVE_OPT) 735 ip_tunnel_info_opts_get(geneveh->options, info); 736 } 737 738 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb, 739 const struct ip_tunnel_info *info, 740 bool xnet, int ip_hdr_len) 741 { 742 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM); 743 struct genevehdr *gnvh; 744 int min_headroom; 745 int err; 746 747 skb_reset_mac_header(skb); 748 skb_scrub_packet(skb, xnet); 749 750 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len + 751 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len; 752 err = skb_cow_head(skb, min_headroom); 753 if (unlikely(err)) 754 goto free_dst; 755 756 err = udp_tunnel_handle_offloads(skb, udp_sum); 757 if (err) 758 goto free_dst; 759 760 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len); 761 geneve_build_header(gnvh, info); 762 skb_set_inner_protocol(skb, htons(ETH_P_TEB)); 763 return 0; 764 765 free_dst: 766 dst_release(dst); 767 return err; 768 } 769 770 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb, 771 struct net_device *dev, 772 struct geneve_sock *gs4, 773 struct flowi4 *fl4, 774 const struct ip_tunnel_info *info, 775 __be16 dport, __be16 sport) 776 { 777 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 778 struct geneve_dev *geneve = netdev_priv(dev); 779 struct dst_cache *dst_cache; 780 struct rtable *rt = NULL; 781 __u8 tos; 782 783 if (!gs4) 784 return ERR_PTR(-EIO); 785 786 memset(fl4, 0, sizeof(*fl4)); 787 fl4->flowi4_mark = skb->mark; 788 fl4->flowi4_proto = IPPROTO_UDP; 789 fl4->daddr = info->key.u.ipv4.dst; 790 fl4->saddr = info->key.u.ipv4.src; 791 fl4->fl4_dport = dport; 792 fl4->fl4_sport = sport; 793 794 tos = info->key.tos; 795 if ((tos == 1) && !geneve->cfg.collect_md) { 796 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb); 797 use_cache = false; 798 } 799 fl4->flowi4_tos = RT_TOS(tos); 800 801 dst_cache = (struct dst_cache *)&info->dst_cache; 802 if (use_cache) { 803 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr); 804 if (rt) 805 return rt; 806 } 807 rt = ip_route_output_key(geneve->net, fl4); 808 if (IS_ERR(rt)) { 809 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr); 810 return ERR_PTR(-ENETUNREACH); 811 } 812 if (rt->dst.dev == dev) { /* is this necessary? */ 813 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr); 814 ip_rt_put(rt); 815 return ERR_PTR(-ELOOP); 816 } 817 if (use_cache) 818 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr); 819 return rt; 820 } 821 822 #if IS_ENABLED(CONFIG_IPV6) 823 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb, 824 struct net_device *dev, 825 struct geneve_sock *gs6, 826 struct flowi6 *fl6, 827 const struct ip_tunnel_info *info, 828 __be16 dport, __be16 sport) 829 { 830 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 831 struct geneve_dev *geneve = netdev_priv(dev); 832 struct dst_entry *dst = NULL; 833 struct dst_cache *dst_cache; 834 __u8 prio; 835 836 if (!gs6) 837 return ERR_PTR(-EIO); 838 839 memset(fl6, 0, sizeof(*fl6)); 840 fl6->flowi6_mark = skb->mark; 841 fl6->flowi6_proto = IPPROTO_UDP; 842 fl6->daddr = info->key.u.ipv6.dst; 843 fl6->saddr = info->key.u.ipv6.src; 844 fl6->fl6_dport = dport; 845 fl6->fl6_sport = sport; 846 847 prio = info->key.tos; 848 if ((prio == 1) && !geneve->cfg.collect_md) { 849 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb); 850 use_cache = false; 851 } 852 853 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio), 854 info->key.label); 855 dst_cache = (struct dst_cache *)&info->dst_cache; 856 if (use_cache) { 857 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr); 858 if (dst) 859 return dst; 860 } 861 dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6, 862 NULL); 863 if (IS_ERR(dst)) { 864 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr); 865 return ERR_PTR(-ENETUNREACH); 866 } 867 if (dst->dev == dev) { /* is this necessary? */ 868 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr); 869 dst_release(dst); 870 return ERR_PTR(-ELOOP); 871 } 872 873 if (use_cache) 874 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr); 875 return dst; 876 } 877 #endif 878 879 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev, 880 struct geneve_dev *geneve, 881 const struct ip_tunnel_info *info) 882 { 883 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 884 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 885 const struct ip_tunnel_key *key = &info->key; 886 struct rtable *rt; 887 struct flowi4 fl4; 888 __u8 tos, ttl; 889 __be16 df = 0; 890 __be16 sport; 891 int err; 892 893 if (!pskb_inet_may_pull(skb)) 894 return -EINVAL; 895 896 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 897 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info, 898 geneve->cfg.info.key.tp_dst, sport); 899 if (IS_ERR(rt)) 900 return PTR_ERR(rt); 901 902 err = skb_tunnel_check_pmtu(skb, &rt->dst, 903 GENEVE_IPV4_HLEN + info->options_len, 904 netif_is_any_bridge_port(dev)); 905 if (err < 0) { 906 dst_release(&rt->dst); 907 return err; 908 } else if (err) { 909 struct ip_tunnel_info *info; 910 911 info = skb_tunnel_info(skb); 912 if (info) { 913 struct ip_tunnel_info *unclone; 914 915 unclone = skb_tunnel_info_unclone(skb); 916 if (unlikely(!unclone)) { 917 dst_release(&rt->dst); 918 return -ENOMEM; 919 } 920 921 unclone->key.u.ipv4.dst = fl4.saddr; 922 unclone->key.u.ipv4.src = fl4.daddr; 923 } 924 925 if (!pskb_may_pull(skb, ETH_HLEN)) { 926 dst_release(&rt->dst); 927 return -EINVAL; 928 } 929 930 skb->protocol = eth_type_trans(skb, geneve->dev); 931 netif_rx(skb); 932 dst_release(&rt->dst); 933 return -EMSGSIZE; 934 } 935 936 if (geneve->cfg.collect_md) { 937 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 938 ttl = key->ttl; 939 940 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0; 941 } else { 942 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb); 943 if (geneve->cfg.ttl_inherit) 944 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb); 945 else 946 ttl = key->ttl; 947 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst); 948 949 if (geneve->cfg.df == GENEVE_DF_SET) { 950 df = htons(IP_DF); 951 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) { 952 struct ethhdr *eth = eth_hdr(skb); 953 954 if (ntohs(eth->h_proto) == ETH_P_IPV6) { 955 df = htons(IP_DF); 956 } else if (ntohs(eth->h_proto) == ETH_P_IP) { 957 struct iphdr *iph = ip_hdr(skb); 958 959 if (iph->frag_off & htons(IP_DF)) 960 df = htons(IP_DF); 961 } 962 } 963 } 964 965 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr)); 966 if (unlikely(err)) 967 return err; 968 969 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr, 970 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst, 971 !net_eq(geneve->net, dev_net(geneve->dev)), 972 !(info->key.tun_flags & TUNNEL_CSUM)); 973 return 0; 974 } 975 976 #if IS_ENABLED(CONFIG_IPV6) 977 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev, 978 struct geneve_dev *geneve, 979 const struct ip_tunnel_info *info) 980 { 981 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 982 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 983 const struct ip_tunnel_key *key = &info->key; 984 struct dst_entry *dst = NULL; 985 struct flowi6 fl6; 986 __u8 prio, ttl; 987 __be16 sport; 988 int err; 989 990 if (!pskb_inet_may_pull(skb)) 991 return -EINVAL; 992 993 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 994 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info, 995 geneve->cfg.info.key.tp_dst, sport); 996 if (IS_ERR(dst)) 997 return PTR_ERR(dst); 998 999 err = skb_tunnel_check_pmtu(skb, dst, 1000 GENEVE_IPV6_HLEN + info->options_len, 1001 netif_is_any_bridge_port(dev)); 1002 if (err < 0) { 1003 dst_release(dst); 1004 return err; 1005 } else if (err) { 1006 struct ip_tunnel_info *info = skb_tunnel_info(skb); 1007 1008 if (info) { 1009 struct ip_tunnel_info *unclone; 1010 1011 unclone = skb_tunnel_info_unclone(skb); 1012 if (unlikely(!unclone)) { 1013 dst_release(dst); 1014 return -ENOMEM; 1015 } 1016 1017 unclone->key.u.ipv6.dst = fl6.saddr; 1018 unclone->key.u.ipv6.src = fl6.daddr; 1019 } 1020 1021 if (!pskb_may_pull(skb, ETH_HLEN)) { 1022 dst_release(dst); 1023 return -EINVAL; 1024 } 1025 1026 skb->protocol = eth_type_trans(skb, geneve->dev); 1027 netif_rx(skb); 1028 dst_release(dst); 1029 return -EMSGSIZE; 1030 } 1031 1032 if (geneve->cfg.collect_md) { 1033 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 1034 ttl = key->ttl; 1035 } else { 1036 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel), 1037 ip_hdr(skb), skb); 1038 if (geneve->cfg.ttl_inherit) 1039 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb); 1040 else 1041 ttl = key->ttl; 1042 ttl = ttl ? : ip6_dst_hoplimit(dst); 1043 } 1044 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr)); 1045 if (unlikely(err)) 1046 return err; 1047 1048 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev, 1049 &fl6.saddr, &fl6.daddr, prio, ttl, 1050 info->key.label, sport, geneve->cfg.info.key.tp_dst, 1051 !(info->key.tun_flags & TUNNEL_CSUM)); 1052 return 0; 1053 } 1054 #endif 1055 1056 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev) 1057 { 1058 struct geneve_dev *geneve = netdev_priv(dev); 1059 struct ip_tunnel_info *info = NULL; 1060 int err; 1061 1062 if (geneve->cfg.collect_md) { 1063 info = skb_tunnel_info(skb); 1064 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) { 1065 netdev_dbg(dev, "no tunnel metadata\n"); 1066 dev_kfree_skb(skb); 1067 dev->stats.tx_dropped++; 1068 return NETDEV_TX_OK; 1069 } 1070 } else { 1071 info = &geneve->cfg.info; 1072 } 1073 1074 rcu_read_lock(); 1075 #if IS_ENABLED(CONFIG_IPV6) 1076 if (info->mode & IP_TUNNEL_INFO_IPV6) 1077 err = geneve6_xmit_skb(skb, dev, geneve, info); 1078 else 1079 #endif 1080 err = geneve_xmit_skb(skb, dev, geneve, info); 1081 rcu_read_unlock(); 1082 1083 if (likely(!err)) 1084 return NETDEV_TX_OK; 1085 1086 if (err != -EMSGSIZE) 1087 dev_kfree_skb(skb); 1088 1089 if (err == -ELOOP) 1090 dev->stats.collisions++; 1091 else if (err == -ENETUNREACH) 1092 dev->stats.tx_carrier_errors++; 1093 1094 dev->stats.tx_errors++; 1095 return NETDEV_TX_OK; 1096 } 1097 1098 static int geneve_change_mtu(struct net_device *dev, int new_mtu) 1099 { 1100 if (new_mtu > dev->max_mtu) 1101 new_mtu = dev->max_mtu; 1102 else if (new_mtu < dev->min_mtu) 1103 new_mtu = dev->min_mtu; 1104 1105 dev->mtu = new_mtu; 1106 return 0; 1107 } 1108 1109 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 1110 { 1111 struct ip_tunnel_info *info = skb_tunnel_info(skb); 1112 struct geneve_dev *geneve = netdev_priv(dev); 1113 __be16 sport; 1114 1115 if (ip_tunnel_info_af(info) == AF_INET) { 1116 struct rtable *rt; 1117 struct flowi4 fl4; 1118 1119 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 1120 sport = udp_flow_src_port(geneve->net, skb, 1121 1, USHRT_MAX, true); 1122 1123 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info, 1124 geneve->cfg.info.key.tp_dst, sport); 1125 if (IS_ERR(rt)) 1126 return PTR_ERR(rt); 1127 1128 ip_rt_put(rt); 1129 info->key.u.ipv4.src = fl4.saddr; 1130 #if IS_ENABLED(CONFIG_IPV6) 1131 } else if (ip_tunnel_info_af(info) == AF_INET6) { 1132 struct dst_entry *dst; 1133 struct flowi6 fl6; 1134 1135 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 1136 sport = udp_flow_src_port(geneve->net, skb, 1137 1, USHRT_MAX, true); 1138 1139 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info, 1140 geneve->cfg.info.key.tp_dst, sport); 1141 if (IS_ERR(dst)) 1142 return PTR_ERR(dst); 1143 1144 dst_release(dst); 1145 info->key.u.ipv6.src = fl6.saddr; 1146 #endif 1147 } else { 1148 return -EINVAL; 1149 } 1150 1151 info->key.tp_src = sport; 1152 info->key.tp_dst = geneve->cfg.info.key.tp_dst; 1153 return 0; 1154 } 1155 1156 static const struct net_device_ops geneve_netdev_ops = { 1157 .ndo_init = geneve_init, 1158 .ndo_uninit = geneve_uninit, 1159 .ndo_open = geneve_open, 1160 .ndo_stop = geneve_stop, 1161 .ndo_start_xmit = geneve_xmit, 1162 .ndo_get_stats64 = dev_get_tstats64, 1163 .ndo_change_mtu = geneve_change_mtu, 1164 .ndo_validate_addr = eth_validate_addr, 1165 .ndo_set_mac_address = eth_mac_addr, 1166 .ndo_fill_metadata_dst = geneve_fill_metadata_dst, 1167 }; 1168 1169 static void geneve_get_drvinfo(struct net_device *dev, 1170 struct ethtool_drvinfo *drvinfo) 1171 { 1172 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version)); 1173 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver)); 1174 } 1175 1176 static const struct ethtool_ops geneve_ethtool_ops = { 1177 .get_drvinfo = geneve_get_drvinfo, 1178 .get_link = ethtool_op_get_link, 1179 }; 1180 1181 /* Info for udev, that this is a virtual tunnel endpoint */ 1182 static struct device_type geneve_type = { 1183 .name = "geneve", 1184 }; 1185 1186 /* Calls the ndo_udp_tunnel_add of the caller in order to 1187 * supply the listening GENEVE udp ports. Callers are expected 1188 * to implement the ndo_udp_tunnel_add. 1189 */ 1190 static void geneve_offload_rx_ports(struct net_device *dev, bool push) 1191 { 1192 struct net *net = dev_net(dev); 1193 struct geneve_net *gn = net_generic(net, geneve_net_id); 1194 struct geneve_sock *gs; 1195 1196 rcu_read_lock(); 1197 list_for_each_entry_rcu(gs, &gn->sock_list, list) { 1198 if (push) { 1199 udp_tunnel_push_rx_port(dev, gs->sock, 1200 UDP_TUNNEL_TYPE_GENEVE); 1201 } else { 1202 udp_tunnel_drop_rx_port(dev, gs->sock, 1203 UDP_TUNNEL_TYPE_GENEVE); 1204 } 1205 } 1206 rcu_read_unlock(); 1207 } 1208 1209 /* Initialize the device structure. */ 1210 static void geneve_setup(struct net_device *dev) 1211 { 1212 ether_setup(dev); 1213 1214 dev->netdev_ops = &geneve_netdev_ops; 1215 dev->ethtool_ops = &geneve_ethtool_ops; 1216 dev->needs_free_netdev = true; 1217 1218 SET_NETDEV_DEVTYPE(dev, &geneve_type); 1219 1220 dev->features |= NETIF_F_LLTX; 1221 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST; 1222 dev->features |= NETIF_F_RXCSUM; 1223 dev->features |= NETIF_F_GSO_SOFTWARE; 1224 1225 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST; 1226 dev->hw_features |= NETIF_F_RXCSUM; 1227 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 1228 1229 /* MTU range: 68 - (something less than 65535) */ 1230 dev->min_mtu = ETH_MIN_MTU; 1231 /* The max_mtu calculation does not take account of GENEVE 1232 * options, to avoid excluding potentially valid 1233 * configurations. This will be further reduced by IPvX hdr size. 1234 */ 1235 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len; 1236 1237 netif_keep_dst(dev); 1238 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1239 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE; 1240 eth_hw_addr_random(dev); 1241 } 1242 1243 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = { 1244 [IFLA_GENEVE_ID] = { .type = NLA_U32 }, 1245 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) }, 1246 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) }, 1247 [IFLA_GENEVE_TTL] = { .type = NLA_U8 }, 1248 [IFLA_GENEVE_TOS] = { .type = NLA_U8 }, 1249 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 }, 1250 [IFLA_GENEVE_PORT] = { .type = NLA_U16 }, 1251 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1252 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 }, 1253 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 }, 1254 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 }, 1255 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 }, 1256 [IFLA_GENEVE_DF] = { .type = NLA_U8 }, 1257 }; 1258 1259 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[], 1260 struct netlink_ext_ack *extack) 1261 { 1262 if (tb[IFLA_ADDRESS]) { 1263 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1264 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1265 "Provided link layer address is not Ethernet"); 1266 return -EINVAL; 1267 } 1268 1269 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1270 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1271 "Provided Ethernet address is not unicast"); 1272 return -EADDRNOTAVAIL; 1273 } 1274 } 1275 1276 if (!data) { 1277 NL_SET_ERR_MSG(extack, 1278 "Not enough attributes provided to perform the operation"); 1279 return -EINVAL; 1280 } 1281 1282 if (data[IFLA_GENEVE_ID]) { 1283 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1284 1285 if (vni >= GENEVE_N_VID) { 1286 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID], 1287 "Geneve ID must be lower than 16777216"); 1288 return -ERANGE; 1289 } 1290 } 1291 1292 if (data[IFLA_GENEVE_DF]) { 1293 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]); 1294 1295 if (df < 0 || df > GENEVE_DF_MAX) { 1296 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF], 1297 "Invalid DF attribute"); 1298 return -EINVAL; 1299 } 1300 } 1301 1302 return 0; 1303 } 1304 1305 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn, 1306 const struct ip_tunnel_info *info, 1307 bool *tun_on_same_port, 1308 bool *tun_collect_md) 1309 { 1310 struct geneve_dev *geneve, *t = NULL; 1311 1312 *tun_on_same_port = false; 1313 *tun_collect_md = false; 1314 list_for_each_entry(geneve, &gn->geneve_list, next) { 1315 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) { 1316 *tun_collect_md = geneve->cfg.collect_md; 1317 *tun_on_same_port = true; 1318 } 1319 if (info->key.tun_id == geneve->cfg.info.key.tun_id && 1320 info->key.tp_dst == geneve->cfg.info.key.tp_dst && 1321 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u))) 1322 t = geneve; 1323 } 1324 return t; 1325 } 1326 1327 static bool is_tnl_info_zero(const struct ip_tunnel_info *info) 1328 { 1329 return !(info->key.tun_id || info->key.tun_flags || info->key.tos || 1330 info->key.ttl || info->key.label || info->key.tp_src || 1331 memchr_inv(&info->key.u, 0, sizeof(info->key.u))); 1332 } 1333 1334 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a, 1335 struct ip_tunnel_info *b) 1336 { 1337 if (ip_tunnel_info_af(a) == AF_INET) 1338 return a->key.u.ipv4.dst == b->key.u.ipv4.dst; 1339 else 1340 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst); 1341 } 1342 1343 static int geneve_configure(struct net *net, struct net_device *dev, 1344 struct netlink_ext_ack *extack, 1345 const struct geneve_config *cfg) 1346 { 1347 struct geneve_net *gn = net_generic(net, geneve_net_id); 1348 struct geneve_dev *t, *geneve = netdev_priv(dev); 1349 const struct ip_tunnel_info *info = &cfg->info; 1350 bool tun_collect_md, tun_on_same_port; 1351 int err, encap_len; 1352 1353 if (cfg->collect_md && !is_tnl_info_zero(info)) { 1354 NL_SET_ERR_MSG(extack, 1355 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified"); 1356 return -EINVAL; 1357 } 1358 1359 geneve->net = net; 1360 geneve->dev = dev; 1361 1362 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md); 1363 if (t) 1364 return -EBUSY; 1365 1366 /* make enough headroom for basic scenario */ 1367 encap_len = GENEVE_BASE_HLEN + ETH_HLEN; 1368 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) { 1369 encap_len += sizeof(struct iphdr); 1370 dev->max_mtu -= sizeof(struct iphdr); 1371 } else { 1372 encap_len += sizeof(struct ipv6hdr); 1373 dev->max_mtu -= sizeof(struct ipv6hdr); 1374 } 1375 dev->needed_headroom = encap_len + ETH_HLEN; 1376 1377 if (cfg->collect_md) { 1378 if (tun_on_same_port) { 1379 NL_SET_ERR_MSG(extack, 1380 "There can be only one externally controlled device on a destination port"); 1381 return -EPERM; 1382 } 1383 } else { 1384 if (tun_collect_md) { 1385 NL_SET_ERR_MSG(extack, 1386 "There already exists an externally controlled device on this destination port"); 1387 return -EPERM; 1388 } 1389 } 1390 1391 dst_cache_reset(&geneve->cfg.info.dst_cache); 1392 memcpy(&geneve->cfg, cfg, sizeof(*cfg)); 1393 1394 err = register_netdevice(dev); 1395 if (err) 1396 return err; 1397 1398 list_add(&geneve->next, &gn->geneve_list); 1399 return 0; 1400 } 1401 1402 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port) 1403 { 1404 memset(info, 0, sizeof(*info)); 1405 info->key.tp_dst = htons(dst_port); 1406 } 1407 1408 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[], 1409 struct netlink_ext_ack *extack, 1410 struct geneve_config *cfg, bool changelink) 1411 { 1412 struct ip_tunnel_info *info = &cfg->info; 1413 int attrtype; 1414 1415 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) { 1416 NL_SET_ERR_MSG(extack, 1417 "Cannot specify both IPv4 and IPv6 Remote addresses"); 1418 return -EINVAL; 1419 } 1420 1421 if (data[IFLA_GENEVE_REMOTE]) { 1422 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) { 1423 attrtype = IFLA_GENEVE_REMOTE; 1424 goto change_notsup; 1425 } 1426 1427 info->key.u.ipv4.dst = 1428 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]); 1429 1430 if (ipv4_is_multicast(info->key.u.ipv4.dst)) { 1431 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE], 1432 "Remote IPv4 address cannot be Multicast"); 1433 return -EINVAL; 1434 } 1435 } 1436 1437 if (data[IFLA_GENEVE_REMOTE6]) { 1438 #if IS_ENABLED(CONFIG_IPV6) 1439 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) { 1440 attrtype = IFLA_GENEVE_REMOTE6; 1441 goto change_notsup; 1442 } 1443 1444 info->mode = IP_TUNNEL_INFO_IPV6; 1445 info->key.u.ipv6.dst = 1446 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]); 1447 1448 if (ipv6_addr_type(&info->key.u.ipv6.dst) & 1449 IPV6_ADDR_LINKLOCAL) { 1450 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1451 "Remote IPv6 address cannot be link-local"); 1452 return -EINVAL; 1453 } 1454 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) { 1455 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1456 "Remote IPv6 address cannot be Multicast"); 1457 return -EINVAL; 1458 } 1459 info->key.tun_flags |= TUNNEL_CSUM; 1460 cfg->use_udp6_rx_checksums = true; 1461 #else 1462 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1463 "IPv6 support not enabled in the kernel"); 1464 return -EPFNOSUPPORT; 1465 #endif 1466 } 1467 1468 if (data[IFLA_GENEVE_ID]) { 1469 __u32 vni; 1470 __u8 tvni[3]; 1471 __be64 tunid; 1472 1473 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1474 tvni[0] = (vni & 0x00ff0000) >> 16; 1475 tvni[1] = (vni & 0x0000ff00) >> 8; 1476 tvni[2] = vni & 0x000000ff; 1477 1478 tunid = vni_to_tunnel_id(tvni); 1479 if (changelink && (tunid != info->key.tun_id)) { 1480 attrtype = IFLA_GENEVE_ID; 1481 goto change_notsup; 1482 } 1483 info->key.tun_id = tunid; 1484 } 1485 1486 if (data[IFLA_GENEVE_TTL_INHERIT]) { 1487 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT])) 1488 cfg->ttl_inherit = true; 1489 else 1490 cfg->ttl_inherit = false; 1491 } else if (data[IFLA_GENEVE_TTL]) { 1492 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]); 1493 cfg->ttl_inherit = false; 1494 } 1495 1496 if (data[IFLA_GENEVE_TOS]) 1497 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]); 1498 1499 if (data[IFLA_GENEVE_DF]) 1500 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]); 1501 1502 if (data[IFLA_GENEVE_LABEL]) { 1503 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) & 1504 IPV6_FLOWLABEL_MASK; 1505 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) { 1506 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL], 1507 "Label attribute only applies for IPv6 Geneve devices"); 1508 return -EINVAL; 1509 } 1510 } 1511 1512 if (data[IFLA_GENEVE_PORT]) { 1513 if (changelink) { 1514 attrtype = IFLA_GENEVE_PORT; 1515 goto change_notsup; 1516 } 1517 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]); 1518 } 1519 1520 if (data[IFLA_GENEVE_COLLECT_METADATA]) { 1521 if (changelink) { 1522 attrtype = IFLA_GENEVE_COLLECT_METADATA; 1523 goto change_notsup; 1524 } 1525 cfg->collect_md = true; 1526 } 1527 1528 if (data[IFLA_GENEVE_UDP_CSUM]) { 1529 if (changelink) { 1530 attrtype = IFLA_GENEVE_UDP_CSUM; 1531 goto change_notsup; 1532 } 1533 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM])) 1534 info->key.tun_flags |= TUNNEL_CSUM; 1535 } 1536 1537 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) { 1538 #if IS_ENABLED(CONFIG_IPV6) 1539 if (changelink) { 1540 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX; 1541 goto change_notsup; 1542 } 1543 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX])) 1544 info->key.tun_flags &= ~TUNNEL_CSUM; 1545 #else 1546 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX], 1547 "IPv6 support not enabled in the kernel"); 1548 return -EPFNOSUPPORT; 1549 #endif 1550 } 1551 1552 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) { 1553 #if IS_ENABLED(CONFIG_IPV6) 1554 if (changelink) { 1555 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX; 1556 goto change_notsup; 1557 } 1558 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX])) 1559 cfg->use_udp6_rx_checksums = false; 1560 #else 1561 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX], 1562 "IPv6 support not enabled in the kernel"); 1563 return -EPFNOSUPPORT; 1564 #endif 1565 } 1566 1567 return 0; 1568 change_notsup: 1569 NL_SET_ERR_MSG_ATTR(extack, data[attrtype], 1570 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported"); 1571 return -EOPNOTSUPP; 1572 } 1573 1574 static void geneve_link_config(struct net_device *dev, 1575 struct ip_tunnel_info *info, struct nlattr *tb[]) 1576 { 1577 struct geneve_dev *geneve = netdev_priv(dev); 1578 int ldev_mtu = 0; 1579 1580 if (tb[IFLA_MTU]) { 1581 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU])); 1582 return; 1583 } 1584 1585 switch (ip_tunnel_info_af(info)) { 1586 case AF_INET: { 1587 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst }; 1588 struct rtable *rt = ip_route_output_key(geneve->net, &fl4); 1589 1590 if (!IS_ERR(rt) && rt->dst.dev) { 1591 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN; 1592 ip_rt_put(rt); 1593 } 1594 break; 1595 } 1596 #if IS_ENABLED(CONFIG_IPV6) 1597 case AF_INET6: { 1598 struct rt6_info *rt; 1599 1600 if (!__in6_dev_get(dev)) 1601 break; 1602 1603 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0, 1604 NULL, 0); 1605 1606 if (rt && rt->dst.dev) 1607 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN; 1608 ip6_rt_put(rt); 1609 break; 1610 } 1611 #endif 1612 } 1613 1614 if (ldev_mtu <= 0) 1615 return; 1616 1617 geneve_change_mtu(dev, ldev_mtu - info->options_len); 1618 } 1619 1620 static int geneve_newlink(struct net *net, struct net_device *dev, 1621 struct nlattr *tb[], struct nlattr *data[], 1622 struct netlink_ext_ack *extack) 1623 { 1624 struct geneve_config cfg = { 1625 .df = GENEVE_DF_UNSET, 1626 .use_udp6_rx_checksums = false, 1627 .ttl_inherit = false, 1628 .collect_md = false, 1629 }; 1630 int err; 1631 1632 init_tnl_info(&cfg.info, GENEVE_UDP_PORT); 1633 err = geneve_nl2info(tb, data, extack, &cfg, false); 1634 if (err) 1635 return err; 1636 1637 err = geneve_configure(net, dev, extack, &cfg); 1638 if (err) 1639 return err; 1640 1641 geneve_link_config(dev, &cfg.info, tb); 1642 1643 return 0; 1644 } 1645 1646 /* Quiesces the geneve device data path for both TX and RX. 1647 * 1648 * On transmit geneve checks for non-NULL geneve_sock before it proceeds. 1649 * So, if we set that socket to NULL under RCU and wait for synchronize_net() 1650 * to complete for the existing set of in-flight packets to be transmitted, 1651 * then we would have quiesced the transmit data path. All the future packets 1652 * will get dropped until we unquiesce the data path. 1653 * 1654 * On receive geneve dereference the geneve_sock stashed in the socket. So, 1655 * if we set that to NULL under RCU and wait for synchronize_net() to 1656 * complete, then we would have quiesced the receive data path. 1657 */ 1658 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4, 1659 struct geneve_sock **gs6) 1660 { 1661 *gs4 = rtnl_dereference(geneve->sock4); 1662 rcu_assign_pointer(geneve->sock4, NULL); 1663 if (*gs4) 1664 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL); 1665 #if IS_ENABLED(CONFIG_IPV6) 1666 *gs6 = rtnl_dereference(geneve->sock6); 1667 rcu_assign_pointer(geneve->sock6, NULL); 1668 if (*gs6) 1669 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL); 1670 #else 1671 *gs6 = NULL; 1672 #endif 1673 synchronize_net(); 1674 } 1675 1676 /* Resumes the geneve device data path for both TX and RX. */ 1677 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4, 1678 struct geneve_sock __maybe_unused *gs6) 1679 { 1680 rcu_assign_pointer(geneve->sock4, gs4); 1681 if (gs4) 1682 rcu_assign_sk_user_data(gs4->sock->sk, gs4); 1683 #if IS_ENABLED(CONFIG_IPV6) 1684 rcu_assign_pointer(geneve->sock6, gs6); 1685 if (gs6) 1686 rcu_assign_sk_user_data(gs6->sock->sk, gs6); 1687 #endif 1688 synchronize_net(); 1689 } 1690 1691 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[], 1692 struct nlattr *data[], 1693 struct netlink_ext_ack *extack) 1694 { 1695 struct geneve_dev *geneve = netdev_priv(dev); 1696 struct geneve_sock *gs4, *gs6; 1697 struct geneve_config cfg; 1698 int err; 1699 1700 /* If the geneve device is configured for metadata (or externally 1701 * controlled, for example, OVS), then nothing can be changed. 1702 */ 1703 if (geneve->cfg.collect_md) 1704 return -EOPNOTSUPP; 1705 1706 /* Start with the existing info. */ 1707 memcpy(&cfg, &geneve->cfg, sizeof(cfg)); 1708 err = geneve_nl2info(tb, data, extack, &cfg, true); 1709 if (err) 1710 return err; 1711 1712 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) { 1713 dst_cache_reset(&cfg.info.dst_cache); 1714 geneve_link_config(dev, &cfg.info, tb); 1715 } 1716 1717 geneve_quiesce(geneve, &gs4, &gs6); 1718 memcpy(&geneve->cfg, &cfg, sizeof(cfg)); 1719 geneve_unquiesce(geneve, gs4, gs6); 1720 1721 return 0; 1722 } 1723 1724 static void geneve_dellink(struct net_device *dev, struct list_head *head) 1725 { 1726 struct geneve_dev *geneve = netdev_priv(dev); 1727 1728 list_del(&geneve->next); 1729 unregister_netdevice_queue(dev, head); 1730 } 1731 1732 static size_t geneve_get_size(const struct net_device *dev) 1733 { 1734 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */ 1735 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */ 1736 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */ 1737 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */ 1738 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */ 1739 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */ 1740 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */ 1741 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */ 1742 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */ 1743 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */ 1744 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */ 1745 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */ 1746 0; 1747 } 1748 1749 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev) 1750 { 1751 struct geneve_dev *geneve = netdev_priv(dev); 1752 struct ip_tunnel_info *info = &geneve->cfg.info; 1753 bool ttl_inherit = geneve->cfg.ttl_inherit; 1754 bool metadata = geneve->cfg.collect_md; 1755 __u8 tmp_vni[3]; 1756 __u32 vni; 1757 1758 tunnel_id_to_vni(info->key.tun_id, tmp_vni); 1759 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2]; 1760 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni)) 1761 goto nla_put_failure; 1762 1763 if (!metadata && ip_tunnel_info_af(info) == AF_INET) { 1764 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE, 1765 info->key.u.ipv4.dst)) 1766 goto nla_put_failure; 1767 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM, 1768 !!(info->key.tun_flags & TUNNEL_CSUM))) 1769 goto nla_put_failure; 1770 1771 #if IS_ENABLED(CONFIG_IPV6) 1772 } else if (!metadata) { 1773 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6, 1774 &info->key.u.ipv6.dst)) 1775 goto nla_put_failure; 1776 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX, 1777 !(info->key.tun_flags & TUNNEL_CSUM))) 1778 goto nla_put_failure; 1779 #endif 1780 } 1781 1782 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) || 1783 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) || 1784 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label)) 1785 goto nla_put_failure; 1786 1787 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df)) 1788 goto nla_put_failure; 1789 1790 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst)) 1791 goto nla_put_failure; 1792 1793 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA)) 1794 goto nla_put_failure; 1795 1796 #if IS_ENABLED(CONFIG_IPV6) 1797 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX, 1798 !geneve->cfg.use_udp6_rx_checksums)) 1799 goto nla_put_failure; 1800 #endif 1801 1802 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit)) 1803 goto nla_put_failure; 1804 1805 return 0; 1806 1807 nla_put_failure: 1808 return -EMSGSIZE; 1809 } 1810 1811 static struct rtnl_link_ops geneve_link_ops __read_mostly = { 1812 .kind = "geneve", 1813 .maxtype = IFLA_GENEVE_MAX, 1814 .policy = geneve_policy, 1815 .priv_size = sizeof(struct geneve_dev), 1816 .setup = geneve_setup, 1817 .validate = geneve_validate, 1818 .newlink = geneve_newlink, 1819 .changelink = geneve_changelink, 1820 .dellink = geneve_dellink, 1821 .get_size = geneve_get_size, 1822 .fill_info = geneve_fill_info, 1823 }; 1824 1825 struct net_device *geneve_dev_create_fb(struct net *net, const char *name, 1826 u8 name_assign_type, u16 dst_port) 1827 { 1828 struct nlattr *tb[IFLA_MAX + 1]; 1829 struct net_device *dev; 1830 LIST_HEAD(list_kill); 1831 int err; 1832 struct geneve_config cfg = { 1833 .df = GENEVE_DF_UNSET, 1834 .use_udp6_rx_checksums = true, 1835 .ttl_inherit = false, 1836 .collect_md = true, 1837 }; 1838 1839 memset(tb, 0, sizeof(tb)); 1840 dev = rtnl_create_link(net, name, name_assign_type, 1841 &geneve_link_ops, tb, NULL); 1842 if (IS_ERR(dev)) 1843 return dev; 1844 1845 init_tnl_info(&cfg.info, dst_port); 1846 err = geneve_configure(net, dev, NULL, &cfg); 1847 if (err) { 1848 free_netdev(dev); 1849 return ERR_PTR(err); 1850 } 1851 1852 /* openvswitch users expect packet sizes to be unrestricted, 1853 * so set the largest MTU we can. 1854 */ 1855 err = geneve_change_mtu(dev, IP_MAX_MTU); 1856 if (err) 1857 goto err; 1858 1859 err = rtnl_configure_link(dev, NULL); 1860 if (err < 0) 1861 goto err; 1862 1863 return dev; 1864 err: 1865 geneve_dellink(dev, &list_kill); 1866 unregister_netdevice_many(&list_kill); 1867 return ERR_PTR(err); 1868 } 1869 EXPORT_SYMBOL_GPL(geneve_dev_create_fb); 1870 1871 static int geneve_netdevice_event(struct notifier_block *unused, 1872 unsigned long event, void *ptr) 1873 { 1874 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1875 1876 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO) 1877 geneve_offload_rx_ports(dev, true); 1878 else if (event == NETDEV_UDP_TUNNEL_DROP_INFO) 1879 geneve_offload_rx_ports(dev, false); 1880 1881 return NOTIFY_DONE; 1882 } 1883 1884 static struct notifier_block geneve_notifier_block __read_mostly = { 1885 .notifier_call = geneve_netdevice_event, 1886 }; 1887 1888 static __net_init int geneve_init_net(struct net *net) 1889 { 1890 struct geneve_net *gn = net_generic(net, geneve_net_id); 1891 1892 INIT_LIST_HEAD(&gn->geneve_list); 1893 INIT_LIST_HEAD(&gn->sock_list); 1894 return 0; 1895 } 1896 1897 static void geneve_destroy_tunnels(struct net *net, struct list_head *head) 1898 { 1899 struct geneve_net *gn = net_generic(net, geneve_net_id); 1900 struct geneve_dev *geneve, *next; 1901 struct net_device *dev, *aux; 1902 1903 /* gather any geneve devices that were moved into this ns */ 1904 for_each_netdev_safe(net, dev, aux) 1905 if (dev->rtnl_link_ops == &geneve_link_ops) 1906 unregister_netdevice_queue(dev, head); 1907 1908 /* now gather any other geneve devices that were created in this ns */ 1909 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) { 1910 /* If geneve->dev is in the same netns, it was already added 1911 * to the list by the previous loop. 1912 */ 1913 if (!net_eq(dev_net(geneve->dev), net)) 1914 unregister_netdevice_queue(geneve->dev, head); 1915 } 1916 } 1917 1918 static void __net_exit geneve_exit_batch_net(struct list_head *net_list) 1919 { 1920 struct net *net; 1921 LIST_HEAD(list); 1922 1923 rtnl_lock(); 1924 list_for_each_entry(net, net_list, exit_list) 1925 geneve_destroy_tunnels(net, &list); 1926 1927 /* unregister the devices gathered above */ 1928 unregister_netdevice_many(&list); 1929 rtnl_unlock(); 1930 1931 list_for_each_entry(net, net_list, exit_list) { 1932 const struct geneve_net *gn = net_generic(net, geneve_net_id); 1933 1934 WARN_ON_ONCE(!list_empty(&gn->sock_list)); 1935 } 1936 } 1937 1938 static struct pernet_operations geneve_net_ops = { 1939 .init = geneve_init_net, 1940 .exit_batch = geneve_exit_batch_net, 1941 .id = &geneve_net_id, 1942 .size = sizeof(struct geneve_net), 1943 }; 1944 1945 static int __init geneve_init_module(void) 1946 { 1947 int rc; 1948 1949 rc = register_pernet_subsys(&geneve_net_ops); 1950 if (rc) 1951 goto out1; 1952 1953 rc = register_netdevice_notifier(&geneve_notifier_block); 1954 if (rc) 1955 goto out2; 1956 1957 rc = rtnl_link_register(&geneve_link_ops); 1958 if (rc) 1959 goto out3; 1960 1961 return 0; 1962 out3: 1963 unregister_netdevice_notifier(&geneve_notifier_block); 1964 out2: 1965 unregister_pernet_subsys(&geneve_net_ops); 1966 out1: 1967 return rc; 1968 } 1969 late_initcall(geneve_init_module); 1970 1971 static void __exit geneve_cleanup_module(void) 1972 { 1973 rtnl_link_unregister(&geneve_link_ops); 1974 unregister_netdevice_notifier(&geneve_notifier_block); 1975 unregister_pernet_subsys(&geneve_net_ops); 1976 } 1977 module_exit(geneve_cleanup_module); 1978 1979 MODULE_LICENSE("GPL"); 1980 MODULE_VERSION(GENEVE_NETDEV_VER); 1981 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>"); 1982 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic"); 1983 MODULE_ALIAS_RTNL_LINK("geneve"); 1984