1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * GENEVE: Generic Network Virtualization Encapsulation 4 * 5 * Copyright (c) 2015 Red Hat, Inc. 6 */ 7 8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 9 10 #include <linux/ethtool.h> 11 #include <linux/kernel.h> 12 #include <linux/module.h> 13 #include <linux/etherdevice.h> 14 #include <linux/hash.h> 15 #include <net/ipv6_stubs.h> 16 #include <net/dst_metadata.h> 17 #include <net/gro_cells.h> 18 #include <net/rtnetlink.h> 19 #include <net/geneve.h> 20 #include <net/gro.h> 21 #include <net/protocol.h> 22 23 #define GENEVE_NETDEV_VER "0.6" 24 25 #define GENEVE_N_VID (1u << 24) 26 #define GENEVE_VID_MASK (GENEVE_N_VID - 1) 27 28 #define VNI_HASH_BITS 10 29 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS) 30 31 static bool log_ecn_error = true; 32 module_param(log_ecn_error, bool, 0644); 33 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN"); 34 35 #define GENEVE_VER 0 36 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr)) 37 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN) 38 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN) 39 40 /* per-network namespace private data for this module */ 41 struct geneve_net { 42 struct list_head geneve_list; 43 struct list_head sock_list; 44 }; 45 46 static unsigned int geneve_net_id; 47 48 struct geneve_dev_node { 49 struct hlist_node hlist; 50 struct geneve_dev *geneve; 51 }; 52 53 struct geneve_config { 54 struct ip_tunnel_info info; 55 bool collect_md; 56 bool use_udp6_rx_checksums; 57 bool ttl_inherit; 58 enum ifla_geneve_df df; 59 bool inner_proto_inherit; 60 }; 61 62 /* Pseudo network device */ 63 struct geneve_dev { 64 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */ 65 #if IS_ENABLED(CONFIG_IPV6) 66 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */ 67 #endif 68 struct net *net; /* netns for packet i/o */ 69 struct net_device *dev; /* netdev for geneve tunnel */ 70 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */ 71 #if IS_ENABLED(CONFIG_IPV6) 72 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */ 73 #endif 74 struct list_head next; /* geneve's per namespace list */ 75 struct gro_cells gro_cells; 76 struct geneve_config cfg; 77 }; 78 79 struct geneve_sock { 80 bool collect_md; 81 struct list_head list; 82 struct socket *sock; 83 struct rcu_head rcu; 84 int refcnt; 85 struct hlist_head vni_list[VNI_HASH_SIZE]; 86 }; 87 88 static inline __u32 geneve_net_vni_hash(u8 vni[3]) 89 { 90 __u32 vnid; 91 92 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2]; 93 return hash_32(vnid, VNI_HASH_BITS); 94 } 95 96 static __be64 vni_to_tunnel_id(const __u8 *vni) 97 { 98 #ifdef __BIG_ENDIAN 99 return (vni[0] << 16) | (vni[1] << 8) | vni[2]; 100 #else 101 return (__force __be64)(((__force u64)vni[0] << 40) | 102 ((__force u64)vni[1] << 48) | 103 ((__force u64)vni[2] << 56)); 104 #endif 105 } 106 107 /* Convert 64 bit tunnel ID to 24 bit VNI. */ 108 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni) 109 { 110 #ifdef __BIG_ENDIAN 111 vni[0] = (__force __u8)(tun_id >> 16); 112 vni[1] = (__force __u8)(tun_id >> 8); 113 vni[2] = (__force __u8)tun_id; 114 #else 115 vni[0] = (__force __u8)((__force u64)tun_id >> 40); 116 vni[1] = (__force __u8)((__force u64)tun_id >> 48); 117 vni[2] = (__force __u8)((__force u64)tun_id >> 56); 118 #endif 119 } 120 121 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni) 122 { 123 return !memcmp(vni, &tun_id[5], 3); 124 } 125 126 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs) 127 { 128 return gs->sock->sk->sk_family; 129 } 130 131 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs, 132 __be32 addr, u8 vni[]) 133 { 134 struct hlist_head *vni_list_head; 135 struct geneve_dev_node *node; 136 __u32 hash; 137 138 /* Find the device for this VNI */ 139 hash = geneve_net_vni_hash(vni); 140 vni_list_head = &gs->vni_list[hash]; 141 hlist_for_each_entry_rcu(node, vni_list_head, hlist) { 142 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) && 143 addr == node->geneve->cfg.info.key.u.ipv4.dst) 144 return node->geneve; 145 } 146 return NULL; 147 } 148 149 #if IS_ENABLED(CONFIG_IPV6) 150 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs, 151 struct in6_addr addr6, u8 vni[]) 152 { 153 struct hlist_head *vni_list_head; 154 struct geneve_dev_node *node; 155 __u32 hash; 156 157 /* Find the device for this VNI */ 158 hash = geneve_net_vni_hash(vni); 159 vni_list_head = &gs->vni_list[hash]; 160 hlist_for_each_entry_rcu(node, vni_list_head, hlist) { 161 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) && 162 ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst)) 163 return node->geneve; 164 } 165 return NULL; 166 } 167 #endif 168 169 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb) 170 { 171 return (struct genevehdr *)(udp_hdr(skb) + 1); 172 } 173 174 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs, 175 struct sk_buff *skb) 176 { 177 static u8 zero_vni[3]; 178 u8 *vni; 179 180 if (geneve_get_sk_family(gs) == AF_INET) { 181 struct iphdr *iph; 182 __be32 addr; 183 184 iph = ip_hdr(skb); /* outer IP header... */ 185 186 if (gs->collect_md) { 187 vni = zero_vni; 188 addr = 0; 189 } else { 190 vni = geneve_hdr(skb)->vni; 191 addr = iph->saddr; 192 } 193 194 return geneve_lookup(gs, addr, vni); 195 #if IS_ENABLED(CONFIG_IPV6) 196 } else if (geneve_get_sk_family(gs) == AF_INET6) { 197 static struct in6_addr zero_addr6; 198 struct ipv6hdr *ip6h; 199 struct in6_addr addr6; 200 201 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */ 202 203 if (gs->collect_md) { 204 vni = zero_vni; 205 addr6 = zero_addr6; 206 } else { 207 vni = geneve_hdr(skb)->vni; 208 addr6 = ip6h->saddr; 209 } 210 211 return geneve6_lookup(gs, addr6, vni); 212 #endif 213 } 214 return NULL; 215 } 216 217 /* geneve receive/decap routine */ 218 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs, 219 struct sk_buff *skb) 220 { 221 struct genevehdr *gnvh = geneve_hdr(skb); 222 struct metadata_dst *tun_dst = NULL; 223 unsigned int len; 224 int nh, err = 0; 225 void *oiph; 226 227 if (ip_tunnel_collect_metadata() || gs->collect_md) { 228 IP_TUNNEL_DECLARE_FLAGS(flags) = { }; 229 230 __set_bit(IP_TUNNEL_KEY_BIT, flags); 231 __assign_bit(IP_TUNNEL_OAM_BIT, flags, gnvh->oam); 232 __assign_bit(IP_TUNNEL_CRIT_OPT_BIT, flags, gnvh->critical); 233 234 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags, 235 vni_to_tunnel_id(gnvh->vni), 236 gnvh->opt_len * 4); 237 if (!tun_dst) { 238 DEV_STATS_INC(geneve->dev, rx_dropped); 239 goto drop; 240 } 241 /* Update tunnel dst according to Geneve options. */ 242 ip_tunnel_flags_zero(flags); 243 __set_bit(IP_TUNNEL_GENEVE_OPT_BIT, flags); 244 ip_tunnel_info_opts_set(&tun_dst->u.tun_info, 245 gnvh->options, gnvh->opt_len * 4, 246 flags); 247 } else { 248 /* Drop packets w/ critical options, 249 * since we don't support any... 250 */ 251 if (gnvh->critical) { 252 DEV_STATS_INC(geneve->dev, rx_frame_errors); 253 DEV_STATS_INC(geneve->dev, rx_errors); 254 goto drop; 255 } 256 } 257 258 if (tun_dst) 259 skb_dst_set(skb, &tun_dst->dst); 260 261 if (gnvh->proto_type == htons(ETH_P_TEB)) { 262 skb_reset_mac_header(skb); 263 skb->protocol = eth_type_trans(skb, geneve->dev); 264 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 265 266 /* Ignore packet loops (and multicast echo) */ 267 if (ether_addr_equal(eth_hdr(skb)->h_source, 268 geneve->dev->dev_addr)) { 269 DEV_STATS_INC(geneve->dev, rx_errors); 270 goto drop; 271 } 272 } else { 273 skb_reset_mac_header(skb); 274 skb->dev = geneve->dev; 275 skb->pkt_type = PACKET_HOST; 276 } 277 278 /* Save offset of outer header relative to skb->head, 279 * because we are going to reset the network header to the inner header 280 * and might change skb->head. 281 */ 282 nh = skb_network_header(skb) - skb->head; 283 284 skb_reset_network_header(skb); 285 286 if (!pskb_inet_may_pull(skb)) { 287 DEV_STATS_INC(geneve->dev, rx_length_errors); 288 DEV_STATS_INC(geneve->dev, rx_errors); 289 goto drop; 290 } 291 292 /* Get the outer header. */ 293 oiph = skb->head + nh; 294 295 if (geneve_get_sk_family(gs) == AF_INET) 296 err = IP_ECN_decapsulate(oiph, skb); 297 #if IS_ENABLED(CONFIG_IPV6) 298 else 299 err = IP6_ECN_decapsulate(oiph, skb); 300 #endif 301 302 if (unlikely(err)) { 303 if (log_ecn_error) { 304 if (geneve_get_sk_family(gs) == AF_INET) 305 net_info_ratelimited("non-ECT from %pI4 " 306 "with TOS=%#x\n", 307 &((struct iphdr *)oiph)->saddr, 308 ((struct iphdr *)oiph)->tos); 309 #if IS_ENABLED(CONFIG_IPV6) 310 else 311 net_info_ratelimited("non-ECT from %pI6\n", 312 &((struct ipv6hdr *)oiph)->saddr); 313 #endif 314 } 315 if (err > 1) { 316 DEV_STATS_INC(geneve->dev, rx_frame_errors); 317 DEV_STATS_INC(geneve->dev, rx_errors); 318 goto drop; 319 } 320 } 321 322 len = skb->len; 323 err = gro_cells_receive(&geneve->gro_cells, skb); 324 if (likely(err == NET_RX_SUCCESS)) 325 dev_sw_netstats_rx_add(geneve->dev, len); 326 327 return; 328 drop: 329 /* Consume bad packet */ 330 kfree_skb(skb); 331 } 332 333 /* Setup stats when device is created */ 334 static int geneve_init(struct net_device *dev) 335 { 336 struct geneve_dev *geneve = netdev_priv(dev); 337 int err; 338 339 err = gro_cells_init(&geneve->gro_cells, dev); 340 if (err) 341 return err; 342 343 err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL); 344 if (err) { 345 gro_cells_destroy(&geneve->gro_cells); 346 return err; 347 } 348 netdev_lockdep_set_classes(dev); 349 return 0; 350 } 351 352 static void geneve_uninit(struct net_device *dev) 353 { 354 struct geneve_dev *geneve = netdev_priv(dev); 355 356 dst_cache_destroy(&geneve->cfg.info.dst_cache); 357 gro_cells_destroy(&geneve->gro_cells); 358 } 359 360 /* Callback from net/ipv4/udp.c to receive packets */ 361 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb) 362 { 363 struct genevehdr *geneveh; 364 struct geneve_dev *geneve; 365 struct geneve_sock *gs; 366 __be16 inner_proto; 367 int opts_len; 368 369 /* Need UDP and Geneve header to be present */ 370 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN))) 371 goto drop; 372 373 /* Return packets with reserved bits set */ 374 geneveh = geneve_hdr(skb); 375 if (unlikely(geneveh->ver != GENEVE_VER)) 376 goto drop; 377 378 gs = rcu_dereference_sk_user_data(sk); 379 if (!gs) 380 goto drop; 381 382 geneve = geneve_lookup_skb(gs, skb); 383 if (!geneve) 384 goto drop; 385 386 inner_proto = geneveh->proto_type; 387 388 if (unlikely((!geneve->cfg.inner_proto_inherit && 389 inner_proto != htons(ETH_P_TEB)))) { 390 DEV_STATS_INC(geneve->dev, rx_dropped); 391 goto drop; 392 } 393 394 opts_len = geneveh->opt_len * 4; 395 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto, 396 !net_eq(geneve->net, dev_net(geneve->dev)))) { 397 DEV_STATS_INC(geneve->dev, rx_dropped); 398 goto drop; 399 } 400 401 geneve_rx(geneve, gs, skb); 402 return 0; 403 404 drop: 405 /* Consume bad packet */ 406 kfree_skb(skb); 407 return 0; 408 } 409 410 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */ 411 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb) 412 { 413 struct genevehdr *geneveh; 414 struct geneve_sock *gs; 415 u8 zero_vni[3] = { 0 }; 416 u8 *vni = zero_vni; 417 418 if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN)) 419 return -EINVAL; 420 421 geneveh = geneve_hdr(skb); 422 if (geneveh->ver != GENEVE_VER) 423 return -EINVAL; 424 425 if (geneveh->proto_type != htons(ETH_P_TEB)) 426 return -EINVAL; 427 428 gs = rcu_dereference_sk_user_data(sk); 429 if (!gs) 430 return -ENOENT; 431 432 if (geneve_get_sk_family(gs) == AF_INET) { 433 struct iphdr *iph = ip_hdr(skb); 434 __be32 addr4 = 0; 435 436 if (!gs->collect_md) { 437 vni = geneve_hdr(skb)->vni; 438 addr4 = iph->daddr; 439 } 440 441 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT; 442 } 443 444 #if IS_ENABLED(CONFIG_IPV6) 445 if (geneve_get_sk_family(gs) == AF_INET6) { 446 struct ipv6hdr *ip6h = ipv6_hdr(skb); 447 struct in6_addr addr6; 448 449 memset(&addr6, 0, sizeof(struct in6_addr)); 450 451 if (!gs->collect_md) { 452 vni = geneve_hdr(skb)->vni; 453 addr6 = ip6h->daddr; 454 } 455 456 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT; 457 } 458 #endif 459 460 return -EPFNOSUPPORT; 461 } 462 463 static struct socket *geneve_create_sock(struct net *net, bool ipv6, 464 __be16 port, bool ipv6_rx_csum) 465 { 466 struct socket *sock; 467 struct udp_port_cfg udp_conf; 468 int err; 469 470 memset(&udp_conf, 0, sizeof(udp_conf)); 471 472 if (ipv6) { 473 udp_conf.family = AF_INET6; 474 udp_conf.ipv6_v6only = 1; 475 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum; 476 } else { 477 udp_conf.family = AF_INET; 478 udp_conf.local_ip.s_addr = htonl(INADDR_ANY); 479 } 480 481 udp_conf.local_udp_port = port; 482 483 /* Open UDP socket */ 484 err = udp_sock_create(net, &udp_conf, &sock); 485 if (err < 0) 486 return ERR_PTR(err); 487 488 udp_allow_gso(sock->sk); 489 return sock; 490 } 491 492 static int geneve_hlen(struct genevehdr *gh) 493 { 494 return sizeof(*gh) + gh->opt_len * 4; 495 } 496 497 static struct sk_buff *geneve_gro_receive(struct sock *sk, 498 struct list_head *head, 499 struct sk_buff *skb) 500 { 501 struct sk_buff *pp = NULL; 502 struct sk_buff *p; 503 struct genevehdr *gh, *gh2; 504 unsigned int hlen, gh_len, off_gnv; 505 const struct packet_offload *ptype; 506 __be16 type; 507 int flush = 1; 508 509 off_gnv = skb_gro_offset(skb); 510 hlen = off_gnv + sizeof(*gh); 511 gh = skb_gro_header(skb, hlen, off_gnv); 512 if (unlikely(!gh)) 513 goto out; 514 515 if (gh->ver != GENEVE_VER || gh->oam) 516 goto out; 517 gh_len = geneve_hlen(gh); 518 519 hlen = off_gnv + gh_len; 520 if (!skb_gro_may_pull(skb, hlen)) { 521 gh = skb_gro_header_slow(skb, hlen, off_gnv); 522 if (unlikely(!gh)) 523 goto out; 524 } 525 526 list_for_each_entry(p, head, list) { 527 if (!NAPI_GRO_CB(p)->same_flow) 528 continue; 529 530 gh2 = (struct genevehdr *)(p->data + off_gnv); 531 if (gh->opt_len != gh2->opt_len || 532 memcmp(gh, gh2, gh_len)) { 533 NAPI_GRO_CB(p)->same_flow = 0; 534 continue; 535 } 536 } 537 538 skb_gro_pull(skb, gh_len); 539 skb_gro_postpull_rcsum(skb, gh, gh_len); 540 type = gh->proto_type; 541 if (likely(type == htons(ETH_P_TEB))) 542 return call_gro_receive(eth_gro_receive, head, skb); 543 544 ptype = gro_find_receive_by_type(type); 545 if (!ptype) 546 goto out; 547 548 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb); 549 flush = 0; 550 551 out: 552 skb_gro_flush_final(skb, pp, flush); 553 554 return pp; 555 } 556 557 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb, 558 int nhoff) 559 { 560 struct genevehdr *gh; 561 struct packet_offload *ptype; 562 __be16 type; 563 int gh_len; 564 int err = -ENOSYS; 565 566 gh = (struct genevehdr *)(skb->data + nhoff); 567 gh_len = geneve_hlen(gh); 568 type = gh->proto_type; 569 570 /* since skb->encapsulation is set, eth_gro_complete() sets the inner mac header */ 571 if (likely(type == htons(ETH_P_TEB))) 572 return eth_gro_complete(skb, nhoff + gh_len); 573 574 ptype = gro_find_complete_by_type(type); 575 if (ptype) 576 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len); 577 578 skb_set_inner_mac_header(skb, nhoff + gh_len); 579 580 return err; 581 } 582 583 /* Create new listen socket if needed */ 584 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port, 585 bool ipv6, bool ipv6_rx_csum) 586 { 587 struct geneve_net *gn = net_generic(net, geneve_net_id); 588 struct geneve_sock *gs; 589 struct socket *sock; 590 struct udp_tunnel_sock_cfg tunnel_cfg; 591 int h; 592 593 gs = kzalloc(sizeof(*gs), GFP_KERNEL); 594 if (!gs) 595 return ERR_PTR(-ENOMEM); 596 597 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum); 598 if (IS_ERR(sock)) { 599 kfree(gs); 600 return ERR_CAST(sock); 601 } 602 603 gs->sock = sock; 604 gs->refcnt = 1; 605 for (h = 0; h < VNI_HASH_SIZE; ++h) 606 INIT_HLIST_HEAD(&gs->vni_list[h]); 607 608 /* Initialize the geneve udp offloads structure */ 609 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 610 611 /* Mark socket as an encapsulation socket */ 612 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg)); 613 tunnel_cfg.sk_user_data = gs; 614 tunnel_cfg.encap_type = 1; 615 tunnel_cfg.gro_receive = geneve_gro_receive; 616 tunnel_cfg.gro_complete = geneve_gro_complete; 617 tunnel_cfg.encap_rcv = geneve_udp_encap_recv; 618 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup; 619 tunnel_cfg.encap_destroy = NULL; 620 setup_udp_tunnel_sock(net, sock, &tunnel_cfg); 621 list_add(&gs->list, &gn->sock_list); 622 return gs; 623 } 624 625 static void __geneve_sock_release(struct geneve_sock *gs) 626 { 627 if (!gs || --gs->refcnt) 628 return; 629 630 list_del(&gs->list); 631 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 632 udp_tunnel_sock_release(gs->sock); 633 kfree_rcu(gs, rcu); 634 } 635 636 static void geneve_sock_release(struct geneve_dev *geneve) 637 { 638 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4); 639 #if IS_ENABLED(CONFIG_IPV6) 640 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6); 641 642 rcu_assign_pointer(geneve->sock6, NULL); 643 #endif 644 645 rcu_assign_pointer(geneve->sock4, NULL); 646 synchronize_net(); 647 648 __geneve_sock_release(gs4); 649 #if IS_ENABLED(CONFIG_IPV6) 650 __geneve_sock_release(gs6); 651 #endif 652 } 653 654 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn, 655 sa_family_t family, 656 __be16 dst_port) 657 { 658 struct geneve_sock *gs; 659 660 list_for_each_entry(gs, &gn->sock_list, list) { 661 if (inet_sk(gs->sock->sk)->inet_sport == dst_port && 662 geneve_get_sk_family(gs) == family) { 663 return gs; 664 } 665 } 666 return NULL; 667 } 668 669 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6) 670 { 671 struct net *net = geneve->net; 672 struct geneve_net *gn = net_generic(net, geneve_net_id); 673 struct geneve_dev_node *node; 674 struct geneve_sock *gs; 675 __u8 vni[3]; 676 __u32 hash; 677 678 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst); 679 if (gs) { 680 gs->refcnt++; 681 goto out; 682 } 683 684 gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6, 685 geneve->cfg.use_udp6_rx_checksums); 686 if (IS_ERR(gs)) 687 return PTR_ERR(gs); 688 689 out: 690 gs->collect_md = geneve->cfg.collect_md; 691 #if IS_ENABLED(CONFIG_IPV6) 692 if (ipv6) { 693 rcu_assign_pointer(geneve->sock6, gs); 694 node = &geneve->hlist6; 695 } else 696 #endif 697 { 698 rcu_assign_pointer(geneve->sock4, gs); 699 node = &geneve->hlist4; 700 } 701 node->geneve = geneve; 702 703 tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni); 704 hash = geneve_net_vni_hash(vni); 705 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]); 706 return 0; 707 } 708 709 static int geneve_open(struct net_device *dev) 710 { 711 struct geneve_dev *geneve = netdev_priv(dev); 712 bool metadata = geneve->cfg.collect_md; 713 bool ipv4, ipv6; 714 int ret = 0; 715 716 ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata; 717 ipv4 = !ipv6 || metadata; 718 #if IS_ENABLED(CONFIG_IPV6) 719 if (ipv6) { 720 ret = geneve_sock_add(geneve, true); 721 if (ret < 0 && ret != -EAFNOSUPPORT) 722 ipv4 = false; 723 } 724 #endif 725 if (ipv4) 726 ret = geneve_sock_add(geneve, false); 727 if (ret < 0) 728 geneve_sock_release(geneve); 729 730 return ret; 731 } 732 733 static int geneve_stop(struct net_device *dev) 734 { 735 struct geneve_dev *geneve = netdev_priv(dev); 736 737 hlist_del_init_rcu(&geneve->hlist4.hlist); 738 #if IS_ENABLED(CONFIG_IPV6) 739 hlist_del_init_rcu(&geneve->hlist6.hlist); 740 #endif 741 geneve_sock_release(geneve); 742 return 0; 743 } 744 745 static void geneve_build_header(struct genevehdr *geneveh, 746 const struct ip_tunnel_info *info, 747 __be16 inner_proto) 748 { 749 geneveh->ver = GENEVE_VER; 750 geneveh->opt_len = info->options_len / 4; 751 geneveh->oam = test_bit(IP_TUNNEL_OAM_BIT, info->key.tun_flags); 752 geneveh->critical = test_bit(IP_TUNNEL_CRIT_OPT_BIT, 753 info->key.tun_flags); 754 geneveh->rsvd1 = 0; 755 tunnel_id_to_vni(info->key.tun_id, geneveh->vni); 756 geneveh->proto_type = inner_proto; 757 geneveh->rsvd2 = 0; 758 759 if (test_bit(IP_TUNNEL_GENEVE_OPT_BIT, info->key.tun_flags)) 760 ip_tunnel_info_opts_get(geneveh->options, info); 761 } 762 763 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb, 764 const struct ip_tunnel_info *info, 765 bool xnet, int ip_hdr_len, 766 bool inner_proto_inherit) 767 { 768 bool udp_sum = test_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags); 769 struct genevehdr *gnvh; 770 __be16 inner_proto; 771 int min_headroom; 772 int err; 773 774 skb_reset_mac_header(skb); 775 skb_scrub_packet(skb, xnet); 776 777 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len + 778 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len; 779 err = skb_cow_head(skb, min_headroom); 780 if (unlikely(err)) 781 goto free_dst; 782 783 err = udp_tunnel_handle_offloads(skb, udp_sum); 784 if (err) 785 goto free_dst; 786 787 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len); 788 inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB); 789 geneve_build_header(gnvh, info, inner_proto); 790 skb_set_inner_protocol(skb, inner_proto); 791 return 0; 792 793 free_dst: 794 dst_release(dst); 795 return err; 796 } 797 798 static u8 geneve_get_dsfield(struct sk_buff *skb, struct net_device *dev, 799 const struct ip_tunnel_info *info, 800 bool *use_cache) 801 { 802 struct geneve_dev *geneve = netdev_priv(dev); 803 u8 dsfield; 804 805 dsfield = info->key.tos; 806 if (dsfield == 1 && !geneve->cfg.collect_md) { 807 dsfield = ip_tunnel_get_dsfield(ip_hdr(skb), skb); 808 *use_cache = false; 809 } 810 811 return dsfield; 812 } 813 814 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev, 815 struct geneve_dev *geneve, 816 const struct ip_tunnel_info *info) 817 { 818 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 819 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 820 const struct ip_tunnel_key *key = &info->key; 821 struct rtable *rt; 822 bool use_cache; 823 __u8 tos, ttl; 824 __be16 df = 0; 825 __be32 saddr; 826 __be16 sport; 827 int err; 828 829 if (!skb_vlan_inet_prepare(skb)) 830 return -EINVAL; 831 832 if (!gs4) 833 return -EIO; 834 835 use_cache = ip_tunnel_dst_cache_usable(skb, info); 836 tos = geneve_get_dsfield(skb, dev, info, &use_cache); 837 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 838 839 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr, 840 &info->key, 841 sport, geneve->cfg.info.key.tp_dst, tos, 842 use_cache ? 843 (struct dst_cache *)&info->dst_cache : NULL); 844 if (IS_ERR(rt)) 845 return PTR_ERR(rt); 846 847 err = skb_tunnel_check_pmtu(skb, &rt->dst, 848 GENEVE_IPV4_HLEN + info->options_len, 849 netif_is_any_bridge_port(dev)); 850 if (err < 0) { 851 dst_release(&rt->dst); 852 return err; 853 } else if (err) { 854 struct ip_tunnel_info *info; 855 856 info = skb_tunnel_info(skb); 857 if (info) { 858 struct ip_tunnel_info *unclone; 859 860 unclone = skb_tunnel_info_unclone(skb); 861 if (unlikely(!unclone)) { 862 dst_release(&rt->dst); 863 return -ENOMEM; 864 } 865 866 unclone->key.u.ipv4.dst = saddr; 867 unclone->key.u.ipv4.src = info->key.u.ipv4.dst; 868 } 869 870 if (!pskb_may_pull(skb, ETH_HLEN)) { 871 dst_release(&rt->dst); 872 return -EINVAL; 873 } 874 875 skb->protocol = eth_type_trans(skb, geneve->dev); 876 __netif_rx(skb); 877 dst_release(&rt->dst); 878 return -EMSGSIZE; 879 } 880 881 tos = ip_tunnel_ecn_encap(tos, ip_hdr(skb), skb); 882 if (geneve->cfg.collect_md) { 883 ttl = key->ttl; 884 885 df = test_bit(IP_TUNNEL_DONT_FRAGMENT_BIT, key->tun_flags) ? 886 htons(IP_DF) : 0; 887 } else { 888 if (geneve->cfg.ttl_inherit) 889 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb); 890 else 891 ttl = key->ttl; 892 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst); 893 894 if (geneve->cfg.df == GENEVE_DF_SET) { 895 df = htons(IP_DF); 896 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) { 897 struct ethhdr *eth = eth_hdr(skb); 898 899 if (ntohs(eth->h_proto) == ETH_P_IPV6) { 900 df = htons(IP_DF); 901 } else if (ntohs(eth->h_proto) == ETH_P_IP) { 902 struct iphdr *iph = ip_hdr(skb); 903 904 if (iph->frag_off & htons(IP_DF)) 905 df = htons(IP_DF); 906 } 907 } 908 } 909 910 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr), 911 geneve->cfg.inner_proto_inherit); 912 if (unlikely(err)) 913 return err; 914 915 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, saddr, info->key.u.ipv4.dst, 916 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst, 917 !net_eq(geneve->net, dev_net(geneve->dev)), 918 !test_bit(IP_TUNNEL_CSUM_BIT, 919 info->key.tun_flags)); 920 return 0; 921 } 922 923 #if IS_ENABLED(CONFIG_IPV6) 924 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev, 925 struct geneve_dev *geneve, 926 const struct ip_tunnel_info *info) 927 { 928 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 929 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 930 const struct ip_tunnel_key *key = &info->key; 931 struct dst_entry *dst = NULL; 932 struct in6_addr saddr; 933 bool use_cache; 934 __u8 prio, ttl; 935 __be16 sport; 936 int err; 937 938 if (!skb_vlan_inet_prepare(skb)) 939 return -EINVAL; 940 941 if (!gs6) 942 return -EIO; 943 944 use_cache = ip_tunnel_dst_cache_usable(skb, info); 945 prio = geneve_get_dsfield(skb, dev, info, &use_cache); 946 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 947 948 dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0, 949 &saddr, key, sport, 950 geneve->cfg.info.key.tp_dst, prio, 951 use_cache ? 952 (struct dst_cache *)&info->dst_cache : NULL); 953 if (IS_ERR(dst)) 954 return PTR_ERR(dst); 955 956 err = skb_tunnel_check_pmtu(skb, dst, 957 GENEVE_IPV6_HLEN + info->options_len, 958 netif_is_any_bridge_port(dev)); 959 if (err < 0) { 960 dst_release(dst); 961 return err; 962 } else if (err) { 963 struct ip_tunnel_info *info = skb_tunnel_info(skb); 964 965 if (info) { 966 struct ip_tunnel_info *unclone; 967 968 unclone = skb_tunnel_info_unclone(skb); 969 if (unlikely(!unclone)) { 970 dst_release(dst); 971 return -ENOMEM; 972 } 973 974 unclone->key.u.ipv6.dst = saddr; 975 unclone->key.u.ipv6.src = info->key.u.ipv6.dst; 976 } 977 978 if (!pskb_may_pull(skb, ETH_HLEN)) { 979 dst_release(dst); 980 return -EINVAL; 981 } 982 983 skb->protocol = eth_type_trans(skb, geneve->dev); 984 __netif_rx(skb); 985 dst_release(dst); 986 return -EMSGSIZE; 987 } 988 989 prio = ip_tunnel_ecn_encap(prio, ip_hdr(skb), skb); 990 if (geneve->cfg.collect_md) { 991 ttl = key->ttl; 992 } else { 993 if (geneve->cfg.ttl_inherit) 994 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb); 995 else 996 ttl = key->ttl; 997 ttl = ttl ? : ip6_dst_hoplimit(dst); 998 } 999 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr), 1000 geneve->cfg.inner_proto_inherit); 1001 if (unlikely(err)) 1002 return err; 1003 1004 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev, 1005 &saddr, &key->u.ipv6.dst, prio, ttl, 1006 info->key.label, sport, geneve->cfg.info.key.tp_dst, 1007 !test_bit(IP_TUNNEL_CSUM_BIT, 1008 info->key.tun_flags)); 1009 return 0; 1010 } 1011 #endif 1012 1013 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev) 1014 { 1015 struct geneve_dev *geneve = netdev_priv(dev); 1016 struct ip_tunnel_info *info = NULL; 1017 int err; 1018 1019 if (geneve->cfg.collect_md) { 1020 info = skb_tunnel_info(skb); 1021 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) { 1022 netdev_dbg(dev, "no tunnel metadata\n"); 1023 dev_kfree_skb(skb); 1024 DEV_STATS_INC(dev, tx_dropped); 1025 return NETDEV_TX_OK; 1026 } 1027 } else { 1028 info = &geneve->cfg.info; 1029 } 1030 1031 rcu_read_lock(); 1032 #if IS_ENABLED(CONFIG_IPV6) 1033 if (info->mode & IP_TUNNEL_INFO_IPV6) 1034 err = geneve6_xmit_skb(skb, dev, geneve, info); 1035 else 1036 #endif 1037 err = geneve_xmit_skb(skb, dev, geneve, info); 1038 rcu_read_unlock(); 1039 1040 if (likely(!err)) 1041 return NETDEV_TX_OK; 1042 1043 if (err != -EMSGSIZE) 1044 dev_kfree_skb(skb); 1045 1046 if (err == -ELOOP) 1047 DEV_STATS_INC(dev, collisions); 1048 else if (err == -ENETUNREACH) 1049 DEV_STATS_INC(dev, tx_carrier_errors); 1050 1051 DEV_STATS_INC(dev, tx_errors); 1052 return NETDEV_TX_OK; 1053 } 1054 1055 static int geneve_change_mtu(struct net_device *dev, int new_mtu) 1056 { 1057 if (new_mtu > dev->max_mtu) 1058 new_mtu = dev->max_mtu; 1059 else if (new_mtu < dev->min_mtu) 1060 new_mtu = dev->min_mtu; 1061 1062 WRITE_ONCE(dev->mtu, new_mtu); 1063 return 0; 1064 } 1065 1066 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 1067 { 1068 struct ip_tunnel_info *info = skb_tunnel_info(skb); 1069 struct geneve_dev *geneve = netdev_priv(dev); 1070 __be16 sport; 1071 1072 if (ip_tunnel_info_af(info) == AF_INET) { 1073 struct rtable *rt; 1074 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 1075 bool use_cache; 1076 __be32 saddr; 1077 u8 tos; 1078 1079 if (!gs4) 1080 return -EIO; 1081 1082 use_cache = ip_tunnel_dst_cache_usable(skb, info); 1083 tos = geneve_get_dsfield(skb, dev, info, &use_cache); 1084 sport = udp_flow_src_port(geneve->net, skb, 1085 1, USHRT_MAX, true); 1086 1087 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr, 1088 &info->key, 1089 sport, geneve->cfg.info.key.tp_dst, 1090 tos, 1091 use_cache ? &info->dst_cache : NULL); 1092 if (IS_ERR(rt)) 1093 return PTR_ERR(rt); 1094 1095 ip_rt_put(rt); 1096 info->key.u.ipv4.src = saddr; 1097 #if IS_ENABLED(CONFIG_IPV6) 1098 } else if (ip_tunnel_info_af(info) == AF_INET6) { 1099 struct dst_entry *dst; 1100 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 1101 struct in6_addr saddr; 1102 bool use_cache; 1103 u8 prio; 1104 1105 if (!gs6) 1106 return -EIO; 1107 1108 use_cache = ip_tunnel_dst_cache_usable(skb, info); 1109 prio = geneve_get_dsfield(skb, dev, info, &use_cache); 1110 sport = udp_flow_src_port(geneve->net, skb, 1111 1, USHRT_MAX, true); 1112 1113 dst = udp_tunnel6_dst_lookup(skb, dev, geneve->net, gs6->sock, 0, 1114 &saddr, &info->key, sport, 1115 geneve->cfg.info.key.tp_dst, prio, 1116 use_cache ? &info->dst_cache : NULL); 1117 if (IS_ERR(dst)) 1118 return PTR_ERR(dst); 1119 1120 dst_release(dst); 1121 info->key.u.ipv6.src = saddr; 1122 #endif 1123 } else { 1124 return -EINVAL; 1125 } 1126 1127 info->key.tp_src = sport; 1128 info->key.tp_dst = geneve->cfg.info.key.tp_dst; 1129 return 0; 1130 } 1131 1132 static const struct net_device_ops geneve_netdev_ops = { 1133 .ndo_init = geneve_init, 1134 .ndo_uninit = geneve_uninit, 1135 .ndo_open = geneve_open, 1136 .ndo_stop = geneve_stop, 1137 .ndo_start_xmit = geneve_xmit, 1138 .ndo_change_mtu = geneve_change_mtu, 1139 .ndo_validate_addr = eth_validate_addr, 1140 .ndo_set_mac_address = eth_mac_addr, 1141 .ndo_fill_metadata_dst = geneve_fill_metadata_dst, 1142 }; 1143 1144 static void geneve_get_drvinfo(struct net_device *dev, 1145 struct ethtool_drvinfo *drvinfo) 1146 { 1147 strscpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version)); 1148 strscpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver)); 1149 } 1150 1151 static const struct ethtool_ops geneve_ethtool_ops = { 1152 .get_drvinfo = geneve_get_drvinfo, 1153 .get_link = ethtool_op_get_link, 1154 }; 1155 1156 /* Info for udev, that this is a virtual tunnel endpoint */ 1157 static const struct device_type geneve_type = { 1158 .name = "geneve", 1159 }; 1160 1161 /* Calls the ndo_udp_tunnel_add of the caller in order to 1162 * supply the listening GENEVE udp ports. Callers are expected 1163 * to implement the ndo_udp_tunnel_add. 1164 */ 1165 static void geneve_offload_rx_ports(struct net_device *dev, bool push) 1166 { 1167 struct net *net = dev_net(dev); 1168 struct geneve_net *gn = net_generic(net, geneve_net_id); 1169 struct geneve_sock *gs; 1170 1171 rcu_read_lock(); 1172 list_for_each_entry_rcu(gs, &gn->sock_list, list) { 1173 if (push) { 1174 udp_tunnel_push_rx_port(dev, gs->sock, 1175 UDP_TUNNEL_TYPE_GENEVE); 1176 } else { 1177 udp_tunnel_drop_rx_port(dev, gs->sock, 1178 UDP_TUNNEL_TYPE_GENEVE); 1179 } 1180 } 1181 rcu_read_unlock(); 1182 } 1183 1184 /* Initialize the device structure. */ 1185 static void geneve_setup(struct net_device *dev) 1186 { 1187 ether_setup(dev); 1188 1189 dev->netdev_ops = &geneve_netdev_ops; 1190 dev->ethtool_ops = &geneve_ethtool_ops; 1191 dev->needs_free_netdev = true; 1192 1193 SET_NETDEV_DEVTYPE(dev, &geneve_type); 1194 1195 dev->features |= NETIF_F_LLTX; 1196 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST; 1197 dev->features |= NETIF_F_RXCSUM; 1198 dev->features |= NETIF_F_GSO_SOFTWARE; 1199 1200 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST; 1201 dev->hw_features |= NETIF_F_RXCSUM; 1202 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 1203 1204 dev->pcpu_stat_type = NETDEV_PCPU_STAT_TSTATS; 1205 /* MTU range: 68 - (something less than 65535) */ 1206 dev->min_mtu = ETH_MIN_MTU; 1207 /* The max_mtu calculation does not take account of GENEVE 1208 * options, to avoid excluding potentially valid 1209 * configurations. This will be further reduced by IPvX hdr size. 1210 */ 1211 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len; 1212 1213 netif_keep_dst(dev); 1214 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1215 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE; 1216 eth_hw_addr_random(dev); 1217 } 1218 1219 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = { 1220 [IFLA_GENEVE_UNSPEC] = { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT }, 1221 [IFLA_GENEVE_ID] = { .type = NLA_U32 }, 1222 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) }, 1223 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) }, 1224 [IFLA_GENEVE_TTL] = { .type = NLA_U8 }, 1225 [IFLA_GENEVE_TOS] = { .type = NLA_U8 }, 1226 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 }, 1227 [IFLA_GENEVE_PORT] = { .type = NLA_U16 }, 1228 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1229 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 }, 1230 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 }, 1231 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 }, 1232 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 }, 1233 [IFLA_GENEVE_DF] = { .type = NLA_U8 }, 1234 [IFLA_GENEVE_INNER_PROTO_INHERIT] = { .type = NLA_FLAG }, 1235 }; 1236 1237 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[], 1238 struct netlink_ext_ack *extack) 1239 { 1240 if (tb[IFLA_ADDRESS]) { 1241 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1242 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1243 "Provided link layer address is not Ethernet"); 1244 return -EINVAL; 1245 } 1246 1247 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1248 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1249 "Provided Ethernet address is not unicast"); 1250 return -EADDRNOTAVAIL; 1251 } 1252 } 1253 1254 if (!data) { 1255 NL_SET_ERR_MSG(extack, 1256 "Not enough attributes provided to perform the operation"); 1257 return -EINVAL; 1258 } 1259 1260 if (data[IFLA_GENEVE_ID]) { 1261 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1262 1263 if (vni >= GENEVE_N_VID) { 1264 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID], 1265 "Geneve ID must be lower than 16777216"); 1266 return -ERANGE; 1267 } 1268 } 1269 1270 if (data[IFLA_GENEVE_DF]) { 1271 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]); 1272 1273 if (df < 0 || df > GENEVE_DF_MAX) { 1274 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF], 1275 "Invalid DF attribute"); 1276 return -EINVAL; 1277 } 1278 } 1279 1280 return 0; 1281 } 1282 1283 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn, 1284 const struct ip_tunnel_info *info, 1285 bool *tun_on_same_port, 1286 bool *tun_collect_md) 1287 { 1288 struct geneve_dev *geneve, *t = NULL; 1289 1290 *tun_on_same_port = false; 1291 *tun_collect_md = false; 1292 list_for_each_entry(geneve, &gn->geneve_list, next) { 1293 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) { 1294 *tun_collect_md = geneve->cfg.collect_md; 1295 *tun_on_same_port = true; 1296 } 1297 if (info->key.tun_id == geneve->cfg.info.key.tun_id && 1298 info->key.tp_dst == geneve->cfg.info.key.tp_dst && 1299 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u))) 1300 t = geneve; 1301 } 1302 return t; 1303 } 1304 1305 static bool is_tnl_info_zero(const struct ip_tunnel_info *info) 1306 { 1307 return !(info->key.tun_id || info->key.tos || 1308 !ip_tunnel_flags_empty(info->key.tun_flags) || 1309 info->key.ttl || info->key.label || info->key.tp_src || 1310 memchr_inv(&info->key.u, 0, sizeof(info->key.u))); 1311 } 1312 1313 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a, 1314 struct ip_tunnel_info *b) 1315 { 1316 if (ip_tunnel_info_af(a) == AF_INET) 1317 return a->key.u.ipv4.dst == b->key.u.ipv4.dst; 1318 else 1319 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst); 1320 } 1321 1322 static int geneve_configure(struct net *net, struct net_device *dev, 1323 struct netlink_ext_ack *extack, 1324 const struct geneve_config *cfg) 1325 { 1326 struct geneve_net *gn = net_generic(net, geneve_net_id); 1327 struct geneve_dev *t, *geneve = netdev_priv(dev); 1328 const struct ip_tunnel_info *info = &cfg->info; 1329 bool tun_collect_md, tun_on_same_port; 1330 int err, encap_len; 1331 1332 if (cfg->collect_md && !is_tnl_info_zero(info)) { 1333 NL_SET_ERR_MSG(extack, 1334 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified"); 1335 return -EINVAL; 1336 } 1337 1338 geneve->net = net; 1339 geneve->dev = dev; 1340 1341 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md); 1342 if (t) 1343 return -EBUSY; 1344 1345 /* make enough headroom for basic scenario */ 1346 encap_len = GENEVE_BASE_HLEN + ETH_HLEN; 1347 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) { 1348 encap_len += sizeof(struct iphdr); 1349 dev->max_mtu -= sizeof(struct iphdr); 1350 } else { 1351 encap_len += sizeof(struct ipv6hdr); 1352 dev->max_mtu -= sizeof(struct ipv6hdr); 1353 } 1354 dev->needed_headroom = encap_len + ETH_HLEN; 1355 1356 if (cfg->collect_md) { 1357 if (tun_on_same_port) { 1358 NL_SET_ERR_MSG(extack, 1359 "There can be only one externally controlled device on a destination port"); 1360 return -EPERM; 1361 } 1362 } else { 1363 if (tun_collect_md) { 1364 NL_SET_ERR_MSG(extack, 1365 "There already exists an externally controlled device on this destination port"); 1366 return -EPERM; 1367 } 1368 } 1369 1370 dst_cache_reset(&geneve->cfg.info.dst_cache); 1371 memcpy(&geneve->cfg, cfg, sizeof(*cfg)); 1372 1373 if (geneve->cfg.inner_proto_inherit) { 1374 dev->header_ops = NULL; 1375 dev->type = ARPHRD_NONE; 1376 dev->hard_header_len = 0; 1377 dev->addr_len = 0; 1378 dev->flags = IFF_POINTOPOINT | IFF_NOARP; 1379 } 1380 1381 err = register_netdevice(dev); 1382 if (err) 1383 return err; 1384 1385 list_add(&geneve->next, &gn->geneve_list); 1386 return 0; 1387 } 1388 1389 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port) 1390 { 1391 memset(info, 0, sizeof(*info)); 1392 info->key.tp_dst = htons(dst_port); 1393 } 1394 1395 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[], 1396 struct netlink_ext_ack *extack, 1397 struct geneve_config *cfg, bool changelink) 1398 { 1399 struct ip_tunnel_info *info = &cfg->info; 1400 int attrtype; 1401 1402 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) { 1403 NL_SET_ERR_MSG(extack, 1404 "Cannot specify both IPv4 and IPv6 Remote addresses"); 1405 return -EINVAL; 1406 } 1407 1408 if (data[IFLA_GENEVE_REMOTE]) { 1409 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) { 1410 attrtype = IFLA_GENEVE_REMOTE; 1411 goto change_notsup; 1412 } 1413 1414 info->key.u.ipv4.dst = 1415 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]); 1416 1417 if (ipv4_is_multicast(info->key.u.ipv4.dst)) { 1418 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE], 1419 "Remote IPv4 address cannot be Multicast"); 1420 return -EINVAL; 1421 } 1422 } 1423 1424 if (data[IFLA_GENEVE_REMOTE6]) { 1425 #if IS_ENABLED(CONFIG_IPV6) 1426 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) { 1427 attrtype = IFLA_GENEVE_REMOTE6; 1428 goto change_notsup; 1429 } 1430 1431 info->mode = IP_TUNNEL_INFO_IPV6; 1432 info->key.u.ipv6.dst = 1433 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]); 1434 1435 if (ipv6_addr_type(&info->key.u.ipv6.dst) & 1436 IPV6_ADDR_LINKLOCAL) { 1437 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1438 "Remote IPv6 address cannot be link-local"); 1439 return -EINVAL; 1440 } 1441 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) { 1442 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1443 "Remote IPv6 address cannot be Multicast"); 1444 return -EINVAL; 1445 } 1446 __set_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags); 1447 cfg->use_udp6_rx_checksums = true; 1448 #else 1449 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1450 "IPv6 support not enabled in the kernel"); 1451 return -EPFNOSUPPORT; 1452 #endif 1453 } 1454 1455 if (data[IFLA_GENEVE_ID]) { 1456 __u32 vni; 1457 __u8 tvni[3]; 1458 __be64 tunid; 1459 1460 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1461 tvni[0] = (vni & 0x00ff0000) >> 16; 1462 tvni[1] = (vni & 0x0000ff00) >> 8; 1463 tvni[2] = vni & 0x000000ff; 1464 1465 tunid = vni_to_tunnel_id(tvni); 1466 if (changelink && (tunid != info->key.tun_id)) { 1467 attrtype = IFLA_GENEVE_ID; 1468 goto change_notsup; 1469 } 1470 info->key.tun_id = tunid; 1471 } 1472 1473 if (data[IFLA_GENEVE_TTL_INHERIT]) { 1474 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT])) 1475 cfg->ttl_inherit = true; 1476 else 1477 cfg->ttl_inherit = false; 1478 } else if (data[IFLA_GENEVE_TTL]) { 1479 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]); 1480 cfg->ttl_inherit = false; 1481 } 1482 1483 if (data[IFLA_GENEVE_TOS]) 1484 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]); 1485 1486 if (data[IFLA_GENEVE_DF]) 1487 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]); 1488 1489 if (data[IFLA_GENEVE_LABEL]) { 1490 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) & 1491 IPV6_FLOWLABEL_MASK; 1492 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) { 1493 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL], 1494 "Label attribute only applies for IPv6 Geneve devices"); 1495 return -EINVAL; 1496 } 1497 } 1498 1499 if (data[IFLA_GENEVE_PORT]) { 1500 if (changelink) { 1501 attrtype = IFLA_GENEVE_PORT; 1502 goto change_notsup; 1503 } 1504 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]); 1505 } 1506 1507 if (data[IFLA_GENEVE_COLLECT_METADATA]) { 1508 if (changelink) { 1509 attrtype = IFLA_GENEVE_COLLECT_METADATA; 1510 goto change_notsup; 1511 } 1512 cfg->collect_md = true; 1513 } 1514 1515 if (data[IFLA_GENEVE_UDP_CSUM]) { 1516 if (changelink) { 1517 attrtype = IFLA_GENEVE_UDP_CSUM; 1518 goto change_notsup; 1519 } 1520 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM])) 1521 __set_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags); 1522 } 1523 1524 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) { 1525 #if IS_ENABLED(CONFIG_IPV6) 1526 if (changelink) { 1527 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX; 1528 goto change_notsup; 1529 } 1530 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX])) 1531 __clear_bit(IP_TUNNEL_CSUM_BIT, info->key.tun_flags); 1532 #else 1533 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX], 1534 "IPv6 support not enabled in the kernel"); 1535 return -EPFNOSUPPORT; 1536 #endif 1537 } 1538 1539 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) { 1540 #if IS_ENABLED(CONFIG_IPV6) 1541 if (changelink) { 1542 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX; 1543 goto change_notsup; 1544 } 1545 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX])) 1546 cfg->use_udp6_rx_checksums = false; 1547 #else 1548 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX], 1549 "IPv6 support not enabled in the kernel"); 1550 return -EPFNOSUPPORT; 1551 #endif 1552 } 1553 1554 if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) { 1555 if (changelink) { 1556 attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT; 1557 goto change_notsup; 1558 } 1559 cfg->inner_proto_inherit = true; 1560 } 1561 1562 return 0; 1563 change_notsup: 1564 NL_SET_ERR_MSG_ATTR(extack, data[attrtype], 1565 "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported"); 1566 return -EOPNOTSUPP; 1567 } 1568 1569 static void geneve_link_config(struct net_device *dev, 1570 struct ip_tunnel_info *info, struct nlattr *tb[]) 1571 { 1572 struct geneve_dev *geneve = netdev_priv(dev); 1573 int ldev_mtu = 0; 1574 1575 if (tb[IFLA_MTU]) { 1576 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU])); 1577 return; 1578 } 1579 1580 switch (ip_tunnel_info_af(info)) { 1581 case AF_INET: { 1582 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst }; 1583 struct rtable *rt = ip_route_output_key(geneve->net, &fl4); 1584 1585 if (!IS_ERR(rt) && rt->dst.dev) { 1586 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN; 1587 ip_rt_put(rt); 1588 } 1589 break; 1590 } 1591 #if IS_ENABLED(CONFIG_IPV6) 1592 case AF_INET6: { 1593 struct rt6_info *rt; 1594 1595 if (!__in6_dev_get(dev)) 1596 break; 1597 1598 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0, 1599 NULL, 0); 1600 1601 if (rt && rt->dst.dev) 1602 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN; 1603 ip6_rt_put(rt); 1604 break; 1605 } 1606 #endif 1607 } 1608 1609 if (ldev_mtu <= 0) 1610 return; 1611 1612 geneve_change_mtu(dev, ldev_mtu - info->options_len); 1613 } 1614 1615 static int geneve_newlink(struct net *net, struct net_device *dev, 1616 struct nlattr *tb[], struct nlattr *data[], 1617 struct netlink_ext_ack *extack) 1618 { 1619 struct geneve_config cfg = { 1620 .df = GENEVE_DF_UNSET, 1621 .use_udp6_rx_checksums = false, 1622 .ttl_inherit = false, 1623 .collect_md = false, 1624 }; 1625 int err; 1626 1627 init_tnl_info(&cfg.info, GENEVE_UDP_PORT); 1628 err = geneve_nl2info(tb, data, extack, &cfg, false); 1629 if (err) 1630 return err; 1631 1632 err = geneve_configure(net, dev, extack, &cfg); 1633 if (err) 1634 return err; 1635 1636 geneve_link_config(dev, &cfg.info, tb); 1637 1638 return 0; 1639 } 1640 1641 /* Quiesces the geneve device data path for both TX and RX. 1642 * 1643 * On transmit geneve checks for non-NULL geneve_sock before it proceeds. 1644 * So, if we set that socket to NULL under RCU and wait for synchronize_net() 1645 * to complete for the existing set of in-flight packets to be transmitted, 1646 * then we would have quiesced the transmit data path. All the future packets 1647 * will get dropped until we unquiesce the data path. 1648 * 1649 * On receive geneve dereference the geneve_sock stashed in the socket. So, 1650 * if we set that to NULL under RCU and wait for synchronize_net() to 1651 * complete, then we would have quiesced the receive data path. 1652 */ 1653 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4, 1654 struct geneve_sock **gs6) 1655 { 1656 *gs4 = rtnl_dereference(geneve->sock4); 1657 rcu_assign_pointer(geneve->sock4, NULL); 1658 if (*gs4) 1659 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL); 1660 #if IS_ENABLED(CONFIG_IPV6) 1661 *gs6 = rtnl_dereference(geneve->sock6); 1662 rcu_assign_pointer(geneve->sock6, NULL); 1663 if (*gs6) 1664 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL); 1665 #else 1666 *gs6 = NULL; 1667 #endif 1668 synchronize_net(); 1669 } 1670 1671 /* Resumes the geneve device data path for both TX and RX. */ 1672 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4, 1673 struct geneve_sock __maybe_unused *gs6) 1674 { 1675 rcu_assign_pointer(geneve->sock4, gs4); 1676 if (gs4) 1677 rcu_assign_sk_user_data(gs4->sock->sk, gs4); 1678 #if IS_ENABLED(CONFIG_IPV6) 1679 rcu_assign_pointer(geneve->sock6, gs6); 1680 if (gs6) 1681 rcu_assign_sk_user_data(gs6->sock->sk, gs6); 1682 #endif 1683 synchronize_net(); 1684 } 1685 1686 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[], 1687 struct nlattr *data[], 1688 struct netlink_ext_ack *extack) 1689 { 1690 struct geneve_dev *geneve = netdev_priv(dev); 1691 struct geneve_sock *gs4, *gs6; 1692 struct geneve_config cfg; 1693 int err; 1694 1695 /* If the geneve device is configured for metadata (or externally 1696 * controlled, for example, OVS), then nothing can be changed. 1697 */ 1698 if (geneve->cfg.collect_md) 1699 return -EOPNOTSUPP; 1700 1701 /* Start with the existing info. */ 1702 memcpy(&cfg, &geneve->cfg, sizeof(cfg)); 1703 err = geneve_nl2info(tb, data, extack, &cfg, true); 1704 if (err) 1705 return err; 1706 1707 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) { 1708 dst_cache_reset(&cfg.info.dst_cache); 1709 geneve_link_config(dev, &cfg.info, tb); 1710 } 1711 1712 geneve_quiesce(geneve, &gs4, &gs6); 1713 memcpy(&geneve->cfg, &cfg, sizeof(cfg)); 1714 geneve_unquiesce(geneve, gs4, gs6); 1715 1716 return 0; 1717 } 1718 1719 static void geneve_dellink(struct net_device *dev, struct list_head *head) 1720 { 1721 struct geneve_dev *geneve = netdev_priv(dev); 1722 1723 list_del(&geneve->next); 1724 unregister_netdevice_queue(dev, head); 1725 } 1726 1727 static size_t geneve_get_size(const struct net_device *dev) 1728 { 1729 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */ 1730 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */ 1731 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */ 1732 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */ 1733 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */ 1734 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */ 1735 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */ 1736 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */ 1737 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */ 1738 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */ 1739 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */ 1740 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */ 1741 nla_total_size(0) + /* IFLA_GENEVE_INNER_PROTO_INHERIT */ 1742 0; 1743 } 1744 1745 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev) 1746 { 1747 struct geneve_dev *geneve = netdev_priv(dev); 1748 struct ip_tunnel_info *info = &geneve->cfg.info; 1749 bool ttl_inherit = geneve->cfg.ttl_inherit; 1750 bool metadata = geneve->cfg.collect_md; 1751 __u8 tmp_vni[3]; 1752 __u32 vni; 1753 1754 tunnel_id_to_vni(info->key.tun_id, tmp_vni); 1755 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2]; 1756 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni)) 1757 goto nla_put_failure; 1758 1759 if (!metadata && ip_tunnel_info_af(info) == AF_INET) { 1760 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE, 1761 info->key.u.ipv4.dst)) 1762 goto nla_put_failure; 1763 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM, 1764 test_bit(IP_TUNNEL_CSUM_BIT, 1765 info->key.tun_flags))) 1766 goto nla_put_failure; 1767 1768 #if IS_ENABLED(CONFIG_IPV6) 1769 } else if (!metadata) { 1770 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6, 1771 &info->key.u.ipv6.dst)) 1772 goto nla_put_failure; 1773 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX, 1774 !test_bit(IP_TUNNEL_CSUM_BIT, 1775 info->key.tun_flags))) 1776 goto nla_put_failure; 1777 #endif 1778 } 1779 1780 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) || 1781 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) || 1782 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label)) 1783 goto nla_put_failure; 1784 1785 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df)) 1786 goto nla_put_failure; 1787 1788 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst)) 1789 goto nla_put_failure; 1790 1791 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA)) 1792 goto nla_put_failure; 1793 1794 #if IS_ENABLED(CONFIG_IPV6) 1795 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX, 1796 !geneve->cfg.use_udp6_rx_checksums)) 1797 goto nla_put_failure; 1798 #endif 1799 1800 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit)) 1801 goto nla_put_failure; 1802 1803 if (geneve->cfg.inner_proto_inherit && 1804 nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT)) 1805 goto nla_put_failure; 1806 1807 return 0; 1808 1809 nla_put_failure: 1810 return -EMSGSIZE; 1811 } 1812 1813 static struct rtnl_link_ops geneve_link_ops __read_mostly = { 1814 .kind = "geneve", 1815 .maxtype = IFLA_GENEVE_MAX, 1816 .policy = geneve_policy, 1817 .priv_size = sizeof(struct geneve_dev), 1818 .setup = geneve_setup, 1819 .validate = geneve_validate, 1820 .newlink = geneve_newlink, 1821 .changelink = geneve_changelink, 1822 .dellink = geneve_dellink, 1823 .get_size = geneve_get_size, 1824 .fill_info = geneve_fill_info, 1825 }; 1826 1827 struct net_device *geneve_dev_create_fb(struct net *net, const char *name, 1828 u8 name_assign_type, u16 dst_port) 1829 { 1830 struct nlattr *tb[IFLA_MAX + 1]; 1831 struct net_device *dev; 1832 LIST_HEAD(list_kill); 1833 int err; 1834 struct geneve_config cfg = { 1835 .df = GENEVE_DF_UNSET, 1836 .use_udp6_rx_checksums = true, 1837 .ttl_inherit = false, 1838 .collect_md = true, 1839 }; 1840 1841 memset(tb, 0, sizeof(tb)); 1842 dev = rtnl_create_link(net, name, name_assign_type, 1843 &geneve_link_ops, tb, NULL); 1844 if (IS_ERR(dev)) 1845 return dev; 1846 1847 init_tnl_info(&cfg.info, dst_port); 1848 err = geneve_configure(net, dev, NULL, &cfg); 1849 if (err) { 1850 free_netdev(dev); 1851 return ERR_PTR(err); 1852 } 1853 1854 /* openvswitch users expect packet sizes to be unrestricted, 1855 * so set the largest MTU we can. 1856 */ 1857 err = geneve_change_mtu(dev, IP_MAX_MTU); 1858 if (err) 1859 goto err; 1860 1861 err = rtnl_configure_link(dev, NULL, 0, NULL); 1862 if (err < 0) 1863 goto err; 1864 1865 return dev; 1866 err: 1867 geneve_dellink(dev, &list_kill); 1868 unregister_netdevice_many(&list_kill); 1869 return ERR_PTR(err); 1870 } 1871 EXPORT_SYMBOL_GPL(geneve_dev_create_fb); 1872 1873 static int geneve_netdevice_event(struct notifier_block *unused, 1874 unsigned long event, void *ptr) 1875 { 1876 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1877 1878 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO) 1879 geneve_offload_rx_ports(dev, true); 1880 else if (event == NETDEV_UDP_TUNNEL_DROP_INFO) 1881 geneve_offload_rx_ports(dev, false); 1882 1883 return NOTIFY_DONE; 1884 } 1885 1886 static struct notifier_block geneve_notifier_block __read_mostly = { 1887 .notifier_call = geneve_netdevice_event, 1888 }; 1889 1890 static __net_init int geneve_init_net(struct net *net) 1891 { 1892 struct geneve_net *gn = net_generic(net, geneve_net_id); 1893 1894 INIT_LIST_HEAD(&gn->geneve_list); 1895 INIT_LIST_HEAD(&gn->sock_list); 1896 return 0; 1897 } 1898 1899 static void geneve_destroy_tunnels(struct net *net, struct list_head *head) 1900 { 1901 struct geneve_net *gn = net_generic(net, geneve_net_id); 1902 struct geneve_dev *geneve, *next; 1903 struct net_device *dev, *aux; 1904 1905 /* gather any geneve devices that were moved into this ns */ 1906 for_each_netdev_safe(net, dev, aux) 1907 if (dev->rtnl_link_ops == &geneve_link_ops) 1908 unregister_netdevice_queue(dev, head); 1909 1910 /* now gather any other geneve devices that were created in this ns */ 1911 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) { 1912 /* If geneve->dev is in the same netns, it was already added 1913 * to the list by the previous loop. 1914 */ 1915 if (!net_eq(dev_net(geneve->dev), net)) 1916 unregister_netdevice_queue(geneve->dev, head); 1917 } 1918 } 1919 1920 static void __net_exit geneve_exit_batch_rtnl(struct list_head *net_list, 1921 struct list_head *dev_to_kill) 1922 { 1923 struct net *net; 1924 1925 list_for_each_entry(net, net_list, exit_list) 1926 geneve_destroy_tunnels(net, dev_to_kill); 1927 } 1928 1929 static void __net_exit geneve_exit_net(struct net *net) 1930 { 1931 const struct geneve_net *gn = net_generic(net, geneve_net_id); 1932 1933 WARN_ON_ONCE(!list_empty(&gn->sock_list)); 1934 } 1935 1936 static struct pernet_operations geneve_net_ops = { 1937 .init = geneve_init_net, 1938 .exit_batch_rtnl = geneve_exit_batch_rtnl, 1939 .exit = geneve_exit_net, 1940 .id = &geneve_net_id, 1941 .size = sizeof(struct geneve_net), 1942 }; 1943 1944 static int __init geneve_init_module(void) 1945 { 1946 int rc; 1947 1948 rc = register_pernet_subsys(&geneve_net_ops); 1949 if (rc) 1950 goto out1; 1951 1952 rc = register_netdevice_notifier(&geneve_notifier_block); 1953 if (rc) 1954 goto out2; 1955 1956 rc = rtnl_link_register(&geneve_link_ops); 1957 if (rc) 1958 goto out3; 1959 1960 return 0; 1961 out3: 1962 unregister_netdevice_notifier(&geneve_notifier_block); 1963 out2: 1964 unregister_pernet_subsys(&geneve_net_ops); 1965 out1: 1966 return rc; 1967 } 1968 late_initcall(geneve_init_module); 1969 1970 static void __exit geneve_cleanup_module(void) 1971 { 1972 rtnl_link_unregister(&geneve_link_ops); 1973 unregister_netdevice_notifier(&geneve_notifier_block); 1974 unregister_pernet_subsys(&geneve_net_ops); 1975 } 1976 module_exit(geneve_cleanup_module); 1977 1978 MODULE_LICENSE("GPL"); 1979 MODULE_VERSION(GENEVE_NETDEV_VER); 1980 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>"); 1981 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic"); 1982 MODULE_ALIAS_RTNL_LINK("geneve"); 1983