1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * GENEVE: Generic Network Virtualization Encapsulation 4 * 5 * Copyright (c) 2015 Red Hat, Inc. 6 */ 7 8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 9 10 #include <linux/ethtool.h> 11 #include <linux/kernel.h> 12 #include <linux/module.h> 13 #include <linux/etherdevice.h> 14 #include <linux/hash.h> 15 #include <net/ipv6_stubs.h> 16 #include <net/dst_metadata.h> 17 #include <net/gro_cells.h> 18 #include <net/rtnetlink.h> 19 #include <net/geneve.h> 20 #include <net/gro.h> 21 #include <net/protocol.h> 22 23 #define GENEVE_NETDEV_VER "0.6" 24 25 #define GENEVE_N_VID (1u << 24) 26 #define GENEVE_VID_MASK (GENEVE_N_VID - 1) 27 28 #define VNI_HASH_BITS 10 29 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS) 30 31 static bool log_ecn_error = true; 32 module_param(log_ecn_error, bool, 0644); 33 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN"); 34 35 #define GENEVE_VER 0 36 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr)) 37 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN) 38 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN) 39 40 /* per-network namespace private data for this module */ 41 struct geneve_net { 42 struct list_head geneve_list; 43 struct list_head sock_list; 44 }; 45 46 static unsigned int geneve_net_id; 47 48 struct geneve_dev_node { 49 struct hlist_node hlist; 50 struct geneve_dev *geneve; 51 }; 52 53 struct geneve_config { 54 struct ip_tunnel_info info; 55 bool collect_md; 56 bool use_udp6_rx_checksums; 57 bool ttl_inherit; 58 enum ifla_geneve_df df; 59 bool inner_proto_inherit; 60 }; 61 62 /* Pseudo network device */ 63 struct geneve_dev { 64 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */ 65 #if IS_ENABLED(CONFIG_IPV6) 66 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */ 67 #endif 68 struct net *net; /* netns for packet i/o */ 69 struct net_device *dev; /* netdev for geneve tunnel */ 70 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */ 71 #if IS_ENABLED(CONFIG_IPV6) 72 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */ 73 #endif 74 struct list_head next; /* geneve's per namespace list */ 75 struct gro_cells gro_cells; 76 struct geneve_config cfg; 77 }; 78 79 struct geneve_sock { 80 bool collect_md; 81 struct list_head list; 82 struct socket *sock; 83 struct rcu_head rcu; 84 int refcnt; 85 struct hlist_head vni_list[VNI_HASH_SIZE]; 86 }; 87 88 static inline __u32 geneve_net_vni_hash(u8 vni[3]) 89 { 90 __u32 vnid; 91 92 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2]; 93 return hash_32(vnid, VNI_HASH_BITS); 94 } 95 96 static __be64 vni_to_tunnel_id(const __u8 *vni) 97 { 98 #ifdef __BIG_ENDIAN 99 return (vni[0] << 16) | (vni[1] << 8) | vni[2]; 100 #else 101 return (__force __be64)(((__force u64)vni[0] << 40) | 102 ((__force u64)vni[1] << 48) | 103 ((__force u64)vni[2] << 56)); 104 #endif 105 } 106 107 /* Convert 64 bit tunnel ID to 24 bit VNI. */ 108 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni) 109 { 110 #ifdef __BIG_ENDIAN 111 vni[0] = (__force __u8)(tun_id >> 16); 112 vni[1] = (__force __u8)(tun_id >> 8); 113 vni[2] = (__force __u8)tun_id; 114 #else 115 vni[0] = (__force __u8)((__force u64)tun_id >> 40); 116 vni[1] = (__force __u8)((__force u64)tun_id >> 48); 117 vni[2] = (__force __u8)((__force u64)tun_id >> 56); 118 #endif 119 } 120 121 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni) 122 { 123 return !memcmp(vni, &tun_id[5], 3); 124 } 125 126 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs) 127 { 128 return gs->sock->sk->sk_family; 129 } 130 131 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs, 132 __be32 addr, u8 vni[]) 133 { 134 struct hlist_head *vni_list_head; 135 struct geneve_dev_node *node; 136 __u32 hash; 137 138 /* Find the device for this VNI */ 139 hash = geneve_net_vni_hash(vni); 140 vni_list_head = &gs->vni_list[hash]; 141 hlist_for_each_entry_rcu(node, vni_list_head, hlist) { 142 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) && 143 addr == node->geneve->cfg.info.key.u.ipv4.dst) 144 return node->geneve; 145 } 146 return NULL; 147 } 148 149 #if IS_ENABLED(CONFIG_IPV6) 150 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs, 151 struct in6_addr addr6, u8 vni[]) 152 { 153 struct hlist_head *vni_list_head; 154 struct geneve_dev_node *node; 155 __u32 hash; 156 157 /* Find the device for this VNI */ 158 hash = geneve_net_vni_hash(vni); 159 vni_list_head = &gs->vni_list[hash]; 160 hlist_for_each_entry_rcu(node, vni_list_head, hlist) { 161 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) && 162 ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst)) 163 return node->geneve; 164 } 165 return NULL; 166 } 167 #endif 168 169 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb) 170 { 171 return (struct genevehdr *)(udp_hdr(skb) + 1); 172 } 173 174 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs, 175 struct sk_buff *skb) 176 { 177 static u8 zero_vni[3]; 178 u8 *vni; 179 180 if (geneve_get_sk_family(gs) == AF_INET) { 181 struct iphdr *iph; 182 __be32 addr; 183 184 iph = ip_hdr(skb); /* outer IP header... */ 185 186 if (gs->collect_md) { 187 vni = zero_vni; 188 addr = 0; 189 } else { 190 vni = geneve_hdr(skb)->vni; 191 addr = iph->saddr; 192 } 193 194 return geneve_lookup(gs, addr, vni); 195 #if IS_ENABLED(CONFIG_IPV6) 196 } else if (geneve_get_sk_family(gs) == AF_INET6) { 197 static struct in6_addr zero_addr6; 198 struct ipv6hdr *ip6h; 199 struct in6_addr addr6; 200 201 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */ 202 203 if (gs->collect_md) { 204 vni = zero_vni; 205 addr6 = zero_addr6; 206 } else { 207 vni = geneve_hdr(skb)->vni; 208 addr6 = ip6h->saddr; 209 } 210 211 return geneve6_lookup(gs, addr6, vni); 212 #endif 213 } 214 return NULL; 215 } 216 217 /* geneve receive/decap routine */ 218 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs, 219 struct sk_buff *skb) 220 { 221 struct genevehdr *gnvh = geneve_hdr(skb); 222 struct metadata_dst *tun_dst = NULL; 223 unsigned int len; 224 int err = 0; 225 void *oiph; 226 227 if (ip_tunnel_collect_metadata() || gs->collect_md) { 228 __be16 flags; 229 230 flags = TUNNEL_KEY | (gnvh->oam ? TUNNEL_OAM : 0) | 231 (gnvh->critical ? TUNNEL_CRIT_OPT : 0); 232 233 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags, 234 vni_to_tunnel_id(gnvh->vni), 235 gnvh->opt_len * 4); 236 if (!tun_dst) { 237 geneve->dev->stats.rx_dropped++; 238 goto drop; 239 } 240 /* Update tunnel dst according to Geneve options. */ 241 ip_tunnel_info_opts_set(&tun_dst->u.tun_info, 242 gnvh->options, gnvh->opt_len * 4, 243 TUNNEL_GENEVE_OPT); 244 } else { 245 /* Drop packets w/ critical options, 246 * since we don't support any... 247 */ 248 if (gnvh->critical) { 249 geneve->dev->stats.rx_frame_errors++; 250 geneve->dev->stats.rx_errors++; 251 goto drop; 252 } 253 } 254 255 if (tun_dst) 256 skb_dst_set(skb, &tun_dst->dst); 257 258 if (gnvh->proto_type == htons(ETH_P_TEB)) { 259 skb_reset_mac_header(skb); 260 skb->protocol = eth_type_trans(skb, geneve->dev); 261 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 262 263 /* Ignore packet loops (and multicast echo) */ 264 if (ether_addr_equal(eth_hdr(skb)->h_source, 265 geneve->dev->dev_addr)) { 266 geneve->dev->stats.rx_errors++; 267 goto drop; 268 } 269 } else { 270 skb_reset_mac_header(skb); 271 skb->dev = geneve->dev; 272 skb->pkt_type = PACKET_HOST; 273 } 274 275 oiph = skb_network_header(skb); 276 skb_reset_network_header(skb); 277 278 if (geneve_get_sk_family(gs) == AF_INET) 279 err = IP_ECN_decapsulate(oiph, skb); 280 #if IS_ENABLED(CONFIG_IPV6) 281 else 282 err = IP6_ECN_decapsulate(oiph, skb); 283 #endif 284 285 if (unlikely(err)) { 286 if (log_ecn_error) { 287 if (geneve_get_sk_family(gs) == AF_INET) 288 net_info_ratelimited("non-ECT from %pI4 " 289 "with TOS=%#x\n", 290 &((struct iphdr *)oiph)->saddr, 291 ((struct iphdr *)oiph)->tos); 292 #if IS_ENABLED(CONFIG_IPV6) 293 else 294 net_info_ratelimited("non-ECT from %pI6\n", 295 &((struct ipv6hdr *)oiph)->saddr); 296 #endif 297 } 298 if (err > 1) { 299 ++geneve->dev->stats.rx_frame_errors; 300 ++geneve->dev->stats.rx_errors; 301 goto drop; 302 } 303 } 304 305 len = skb->len; 306 err = gro_cells_receive(&geneve->gro_cells, skb); 307 if (likely(err == NET_RX_SUCCESS)) 308 dev_sw_netstats_rx_add(geneve->dev, len); 309 310 return; 311 drop: 312 /* Consume bad packet */ 313 kfree_skb(skb); 314 } 315 316 /* Setup stats when device is created */ 317 static int geneve_init(struct net_device *dev) 318 { 319 struct geneve_dev *geneve = netdev_priv(dev); 320 int err; 321 322 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 323 if (!dev->tstats) 324 return -ENOMEM; 325 326 err = gro_cells_init(&geneve->gro_cells, dev); 327 if (err) { 328 free_percpu(dev->tstats); 329 return err; 330 } 331 332 err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL); 333 if (err) { 334 free_percpu(dev->tstats); 335 gro_cells_destroy(&geneve->gro_cells); 336 return err; 337 } 338 return 0; 339 } 340 341 static void geneve_uninit(struct net_device *dev) 342 { 343 struct geneve_dev *geneve = netdev_priv(dev); 344 345 dst_cache_destroy(&geneve->cfg.info.dst_cache); 346 gro_cells_destroy(&geneve->gro_cells); 347 free_percpu(dev->tstats); 348 } 349 350 /* Callback from net/ipv4/udp.c to receive packets */ 351 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb) 352 { 353 struct genevehdr *geneveh; 354 struct geneve_dev *geneve; 355 struct geneve_sock *gs; 356 __be16 inner_proto; 357 int opts_len; 358 359 /* Need UDP and Geneve header to be present */ 360 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN))) 361 goto drop; 362 363 /* Return packets with reserved bits set */ 364 geneveh = geneve_hdr(skb); 365 if (unlikely(geneveh->ver != GENEVE_VER)) 366 goto drop; 367 368 gs = rcu_dereference_sk_user_data(sk); 369 if (!gs) 370 goto drop; 371 372 geneve = geneve_lookup_skb(gs, skb); 373 if (!geneve) 374 goto drop; 375 376 inner_proto = geneveh->proto_type; 377 378 if (unlikely((!geneve->cfg.inner_proto_inherit && 379 inner_proto != htons(ETH_P_TEB)))) { 380 geneve->dev->stats.rx_dropped++; 381 goto drop; 382 } 383 384 opts_len = geneveh->opt_len * 4; 385 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, inner_proto, 386 !net_eq(geneve->net, dev_net(geneve->dev)))) { 387 geneve->dev->stats.rx_dropped++; 388 goto drop; 389 } 390 391 geneve_rx(geneve, gs, skb); 392 return 0; 393 394 drop: 395 /* Consume bad packet */ 396 kfree_skb(skb); 397 return 0; 398 } 399 400 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */ 401 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb) 402 { 403 struct genevehdr *geneveh; 404 struct geneve_sock *gs; 405 u8 zero_vni[3] = { 0 }; 406 u8 *vni = zero_vni; 407 408 if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN)) 409 return -EINVAL; 410 411 geneveh = geneve_hdr(skb); 412 if (geneveh->ver != GENEVE_VER) 413 return -EINVAL; 414 415 if (geneveh->proto_type != htons(ETH_P_TEB)) 416 return -EINVAL; 417 418 gs = rcu_dereference_sk_user_data(sk); 419 if (!gs) 420 return -ENOENT; 421 422 if (geneve_get_sk_family(gs) == AF_INET) { 423 struct iphdr *iph = ip_hdr(skb); 424 __be32 addr4 = 0; 425 426 if (!gs->collect_md) { 427 vni = geneve_hdr(skb)->vni; 428 addr4 = iph->daddr; 429 } 430 431 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT; 432 } 433 434 #if IS_ENABLED(CONFIG_IPV6) 435 if (geneve_get_sk_family(gs) == AF_INET6) { 436 struct ipv6hdr *ip6h = ipv6_hdr(skb); 437 struct in6_addr addr6; 438 439 memset(&addr6, 0, sizeof(struct in6_addr)); 440 441 if (!gs->collect_md) { 442 vni = geneve_hdr(skb)->vni; 443 addr6 = ip6h->daddr; 444 } 445 446 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT; 447 } 448 #endif 449 450 return -EPFNOSUPPORT; 451 } 452 453 static struct socket *geneve_create_sock(struct net *net, bool ipv6, 454 __be16 port, bool ipv6_rx_csum) 455 { 456 struct socket *sock; 457 struct udp_port_cfg udp_conf; 458 int err; 459 460 memset(&udp_conf, 0, sizeof(udp_conf)); 461 462 if (ipv6) { 463 udp_conf.family = AF_INET6; 464 udp_conf.ipv6_v6only = 1; 465 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum; 466 } else { 467 udp_conf.family = AF_INET; 468 udp_conf.local_ip.s_addr = htonl(INADDR_ANY); 469 } 470 471 udp_conf.local_udp_port = port; 472 473 /* Open UDP socket */ 474 err = udp_sock_create(net, &udp_conf, &sock); 475 if (err < 0) 476 return ERR_PTR(err); 477 478 udp_allow_gso(sock->sk); 479 return sock; 480 } 481 482 static int geneve_hlen(struct genevehdr *gh) 483 { 484 return sizeof(*gh) + gh->opt_len * 4; 485 } 486 487 static struct sk_buff *geneve_gro_receive(struct sock *sk, 488 struct list_head *head, 489 struct sk_buff *skb) 490 { 491 struct sk_buff *pp = NULL; 492 struct sk_buff *p; 493 struct genevehdr *gh, *gh2; 494 unsigned int hlen, gh_len, off_gnv; 495 const struct packet_offload *ptype; 496 __be16 type; 497 int flush = 1; 498 499 off_gnv = skb_gro_offset(skb); 500 hlen = off_gnv + sizeof(*gh); 501 gh = skb_gro_header(skb, hlen, off_gnv); 502 if (unlikely(!gh)) 503 goto out; 504 505 if (gh->ver != GENEVE_VER || gh->oam) 506 goto out; 507 gh_len = geneve_hlen(gh); 508 509 hlen = off_gnv + gh_len; 510 if (skb_gro_header_hard(skb, hlen)) { 511 gh = skb_gro_header_slow(skb, hlen, off_gnv); 512 if (unlikely(!gh)) 513 goto out; 514 } 515 516 list_for_each_entry(p, head, list) { 517 if (!NAPI_GRO_CB(p)->same_flow) 518 continue; 519 520 gh2 = (struct genevehdr *)(p->data + off_gnv); 521 if (gh->opt_len != gh2->opt_len || 522 memcmp(gh, gh2, gh_len)) { 523 NAPI_GRO_CB(p)->same_flow = 0; 524 continue; 525 } 526 } 527 528 skb_gro_pull(skb, gh_len); 529 skb_gro_postpull_rcsum(skb, gh, gh_len); 530 type = gh->proto_type; 531 if (likely(type == htons(ETH_P_TEB))) 532 return call_gro_receive(eth_gro_receive, head, skb); 533 534 ptype = gro_find_receive_by_type(type); 535 if (!ptype) 536 goto out; 537 538 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb); 539 flush = 0; 540 541 out: 542 skb_gro_flush_final(skb, pp, flush); 543 544 return pp; 545 } 546 547 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb, 548 int nhoff) 549 { 550 struct genevehdr *gh; 551 struct packet_offload *ptype; 552 __be16 type; 553 int gh_len; 554 int err = -ENOSYS; 555 556 gh = (struct genevehdr *)(skb->data + nhoff); 557 gh_len = geneve_hlen(gh); 558 type = gh->proto_type; 559 560 /* since skb->encapsulation is set, eth_gro_complete() sets the inner mac header */ 561 if (likely(type == htons(ETH_P_TEB))) 562 return eth_gro_complete(skb, nhoff + gh_len); 563 564 ptype = gro_find_complete_by_type(type); 565 if (ptype) 566 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len); 567 568 skb_set_inner_mac_header(skb, nhoff + gh_len); 569 570 return err; 571 } 572 573 /* Create new listen socket if needed */ 574 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port, 575 bool ipv6, bool ipv6_rx_csum) 576 { 577 struct geneve_net *gn = net_generic(net, geneve_net_id); 578 struct geneve_sock *gs; 579 struct socket *sock; 580 struct udp_tunnel_sock_cfg tunnel_cfg; 581 int h; 582 583 gs = kzalloc(sizeof(*gs), GFP_KERNEL); 584 if (!gs) 585 return ERR_PTR(-ENOMEM); 586 587 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum); 588 if (IS_ERR(sock)) { 589 kfree(gs); 590 return ERR_CAST(sock); 591 } 592 593 gs->sock = sock; 594 gs->refcnt = 1; 595 for (h = 0; h < VNI_HASH_SIZE; ++h) 596 INIT_HLIST_HEAD(&gs->vni_list[h]); 597 598 /* Initialize the geneve udp offloads structure */ 599 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 600 601 /* Mark socket as an encapsulation socket */ 602 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg)); 603 tunnel_cfg.sk_user_data = gs; 604 tunnel_cfg.encap_type = 1; 605 tunnel_cfg.gro_receive = geneve_gro_receive; 606 tunnel_cfg.gro_complete = geneve_gro_complete; 607 tunnel_cfg.encap_rcv = geneve_udp_encap_recv; 608 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup; 609 tunnel_cfg.encap_destroy = NULL; 610 setup_udp_tunnel_sock(net, sock, &tunnel_cfg); 611 list_add(&gs->list, &gn->sock_list); 612 return gs; 613 } 614 615 static void __geneve_sock_release(struct geneve_sock *gs) 616 { 617 if (!gs || --gs->refcnt) 618 return; 619 620 list_del(&gs->list); 621 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 622 udp_tunnel_sock_release(gs->sock); 623 kfree_rcu(gs, rcu); 624 } 625 626 static void geneve_sock_release(struct geneve_dev *geneve) 627 { 628 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4); 629 #if IS_ENABLED(CONFIG_IPV6) 630 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6); 631 632 rcu_assign_pointer(geneve->sock6, NULL); 633 #endif 634 635 rcu_assign_pointer(geneve->sock4, NULL); 636 synchronize_net(); 637 638 __geneve_sock_release(gs4); 639 #if IS_ENABLED(CONFIG_IPV6) 640 __geneve_sock_release(gs6); 641 #endif 642 } 643 644 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn, 645 sa_family_t family, 646 __be16 dst_port) 647 { 648 struct geneve_sock *gs; 649 650 list_for_each_entry(gs, &gn->sock_list, list) { 651 if (inet_sk(gs->sock->sk)->inet_sport == dst_port && 652 geneve_get_sk_family(gs) == family) { 653 return gs; 654 } 655 } 656 return NULL; 657 } 658 659 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6) 660 { 661 struct net *net = geneve->net; 662 struct geneve_net *gn = net_generic(net, geneve_net_id); 663 struct geneve_dev_node *node; 664 struct geneve_sock *gs; 665 __u8 vni[3]; 666 __u32 hash; 667 668 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst); 669 if (gs) { 670 gs->refcnt++; 671 goto out; 672 } 673 674 gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6, 675 geneve->cfg.use_udp6_rx_checksums); 676 if (IS_ERR(gs)) 677 return PTR_ERR(gs); 678 679 out: 680 gs->collect_md = geneve->cfg.collect_md; 681 #if IS_ENABLED(CONFIG_IPV6) 682 if (ipv6) { 683 rcu_assign_pointer(geneve->sock6, gs); 684 node = &geneve->hlist6; 685 } else 686 #endif 687 { 688 rcu_assign_pointer(geneve->sock4, gs); 689 node = &geneve->hlist4; 690 } 691 node->geneve = geneve; 692 693 tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni); 694 hash = geneve_net_vni_hash(vni); 695 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]); 696 return 0; 697 } 698 699 static int geneve_open(struct net_device *dev) 700 { 701 struct geneve_dev *geneve = netdev_priv(dev); 702 bool metadata = geneve->cfg.collect_md; 703 bool ipv4, ipv6; 704 int ret = 0; 705 706 ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata; 707 ipv4 = !ipv6 || metadata; 708 #if IS_ENABLED(CONFIG_IPV6) 709 if (ipv6) { 710 ret = geneve_sock_add(geneve, true); 711 if (ret < 0 && ret != -EAFNOSUPPORT) 712 ipv4 = false; 713 } 714 #endif 715 if (ipv4) 716 ret = geneve_sock_add(geneve, false); 717 if (ret < 0) 718 geneve_sock_release(geneve); 719 720 return ret; 721 } 722 723 static int geneve_stop(struct net_device *dev) 724 { 725 struct geneve_dev *geneve = netdev_priv(dev); 726 727 hlist_del_init_rcu(&geneve->hlist4.hlist); 728 #if IS_ENABLED(CONFIG_IPV6) 729 hlist_del_init_rcu(&geneve->hlist6.hlist); 730 #endif 731 geneve_sock_release(geneve); 732 return 0; 733 } 734 735 static void geneve_build_header(struct genevehdr *geneveh, 736 const struct ip_tunnel_info *info, 737 __be16 inner_proto) 738 { 739 geneveh->ver = GENEVE_VER; 740 geneveh->opt_len = info->options_len / 4; 741 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM); 742 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT); 743 geneveh->rsvd1 = 0; 744 tunnel_id_to_vni(info->key.tun_id, geneveh->vni); 745 geneveh->proto_type = inner_proto; 746 geneveh->rsvd2 = 0; 747 748 if (info->key.tun_flags & TUNNEL_GENEVE_OPT) 749 ip_tunnel_info_opts_get(geneveh->options, info); 750 } 751 752 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb, 753 const struct ip_tunnel_info *info, 754 bool xnet, int ip_hdr_len, 755 bool inner_proto_inherit) 756 { 757 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM); 758 struct genevehdr *gnvh; 759 __be16 inner_proto; 760 int min_headroom; 761 int err; 762 763 skb_reset_mac_header(skb); 764 skb_scrub_packet(skb, xnet); 765 766 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len + 767 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len; 768 err = skb_cow_head(skb, min_headroom); 769 if (unlikely(err)) 770 goto free_dst; 771 772 err = udp_tunnel_handle_offloads(skb, udp_sum); 773 if (err) 774 goto free_dst; 775 776 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len); 777 inner_proto = inner_proto_inherit ? skb->protocol : htons(ETH_P_TEB); 778 geneve_build_header(gnvh, info, inner_proto); 779 skb_set_inner_protocol(skb, inner_proto); 780 return 0; 781 782 free_dst: 783 dst_release(dst); 784 return err; 785 } 786 787 static u8 geneve_get_dsfield(struct sk_buff *skb, struct net_device *dev, 788 const struct ip_tunnel_info *info, 789 bool *use_cache) 790 { 791 struct geneve_dev *geneve = netdev_priv(dev); 792 u8 dsfield; 793 794 dsfield = info->key.tos; 795 if (dsfield == 1 && !geneve->cfg.collect_md) { 796 dsfield = ip_tunnel_get_dsfield(ip_hdr(skb), skb); 797 *use_cache = false; 798 } 799 800 return dsfield; 801 } 802 803 #if IS_ENABLED(CONFIG_IPV6) 804 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb, 805 struct net_device *dev, 806 struct geneve_sock *gs6, 807 struct flowi6 *fl6, 808 const struct ip_tunnel_info *info, 809 __be16 dport, __be16 sport) 810 { 811 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 812 struct geneve_dev *geneve = netdev_priv(dev); 813 struct dst_entry *dst = NULL; 814 struct dst_cache *dst_cache; 815 __u8 prio; 816 817 if (!gs6) 818 return ERR_PTR(-EIO); 819 820 memset(fl6, 0, sizeof(*fl6)); 821 fl6->flowi6_mark = skb->mark; 822 fl6->flowi6_proto = IPPROTO_UDP; 823 fl6->daddr = info->key.u.ipv6.dst; 824 fl6->saddr = info->key.u.ipv6.src; 825 fl6->fl6_dport = dport; 826 fl6->fl6_sport = sport; 827 828 prio = geneve_get_dsfield(skb, dev, info, &use_cache); 829 fl6->flowlabel = ip6_make_flowinfo(prio, info->key.label); 830 dst_cache = (struct dst_cache *)&info->dst_cache; 831 if (use_cache) { 832 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr); 833 if (dst) 834 return dst; 835 } 836 dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6, 837 NULL); 838 if (IS_ERR(dst)) { 839 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr); 840 return ERR_PTR(-ENETUNREACH); 841 } 842 if (dst->dev == dev) { /* is this necessary? */ 843 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr); 844 dst_release(dst); 845 return ERR_PTR(-ELOOP); 846 } 847 848 if (use_cache) 849 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr); 850 return dst; 851 } 852 #endif 853 854 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev, 855 struct geneve_dev *geneve, 856 const struct ip_tunnel_info *info) 857 { 858 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 859 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 860 const struct ip_tunnel_key *key = &info->key; 861 struct rtable *rt; 862 bool use_cache; 863 __u8 tos, ttl; 864 __be16 df = 0; 865 __be32 saddr; 866 __be16 sport; 867 int err; 868 869 if (!pskb_inet_may_pull(skb)) 870 return -EINVAL; 871 872 if (!gs4) 873 return -EIO; 874 875 use_cache = ip_tunnel_dst_cache_usable(skb, info); 876 tos = geneve_get_dsfield(skb, dev, info, &use_cache); 877 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 878 879 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr, 880 &info->key, 881 sport, geneve->cfg.info.key.tp_dst, tos, 882 use_cache ? 883 (struct dst_cache *)&info->dst_cache : NULL); 884 if (IS_ERR(rt)) 885 return PTR_ERR(rt); 886 887 err = skb_tunnel_check_pmtu(skb, &rt->dst, 888 GENEVE_IPV4_HLEN + info->options_len, 889 netif_is_any_bridge_port(dev)); 890 if (err < 0) { 891 dst_release(&rt->dst); 892 return err; 893 } else if (err) { 894 struct ip_tunnel_info *info; 895 896 info = skb_tunnel_info(skb); 897 if (info) { 898 struct ip_tunnel_info *unclone; 899 900 unclone = skb_tunnel_info_unclone(skb); 901 if (unlikely(!unclone)) { 902 dst_release(&rt->dst); 903 return -ENOMEM; 904 } 905 906 unclone->key.u.ipv4.dst = saddr; 907 unclone->key.u.ipv4.src = info->key.u.ipv4.dst; 908 } 909 910 if (!pskb_may_pull(skb, ETH_HLEN)) { 911 dst_release(&rt->dst); 912 return -EINVAL; 913 } 914 915 skb->protocol = eth_type_trans(skb, geneve->dev); 916 __netif_rx(skb); 917 dst_release(&rt->dst); 918 return -EMSGSIZE; 919 } 920 921 tos = ip_tunnel_ecn_encap(tos, ip_hdr(skb), skb); 922 if (geneve->cfg.collect_md) { 923 ttl = key->ttl; 924 925 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0; 926 } else { 927 if (geneve->cfg.ttl_inherit) 928 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb); 929 else 930 ttl = key->ttl; 931 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst); 932 933 if (geneve->cfg.df == GENEVE_DF_SET) { 934 df = htons(IP_DF); 935 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) { 936 struct ethhdr *eth = eth_hdr(skb); 937 938 if (ntohs(eth->h_proto) == ETH_P_IPV6) { 939 df = htons(IP_DF); 940 } else if (ntohs(eth->h_proto) == ETH_P_IP) { 941 struct iphdr *iph = ip_hdr(skb); 942 943 if (iph->frag_off & htons(IP_DF)) 944 df = htons(IP_DF); 945 } 946 } 947 } 948 949 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr), 950 geneve->cfg.inner_proto_inherit); 951 if (unlikely(err)) 952 return err; 953 954 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, saddr, info->key.u.ipv4.dst, 955 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst, 956 !net_eq(geneve->net, dev_net(geneve->dev)), 957 !(info->key.tun_flags & TUNNEL_CSUM)); 958 return 0; 959 } 960 961 #if IS_ENABLED(CONFIG_IPV6) 962 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev, 963 struct geneve_dev *geneve, 964 const struct ip_tunnel_info *info) 965 { 966 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 967 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 968 const struct ip_tunnel_key *key = &info->key; 969 struct dst_entry *dst = NULL; 970 struct flowi6 fl6; 971 __u8 prio, ttl; 972 __be16 sport; 973 int err; 974 975 if (!pskb_inet_may_pull(skb)) 976 return -EINVAL; 977 978 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 979 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info, 980 geneve->cfg.info.key.tp_dst, sport); 981 if (IS_ERR(dst)) 982 return PTR_ERR(dst); 983 984 err = skb_tunnel_check_pmtu(skb, dst, 985 GENEVE_IPV6_HLEN + info->options_len, 986 netif_is_any_bridge_port(dev)); 987 if (err < 0) { 988 dst_release(dst); 989 return err; 990 } else if (err) { 991 struct ip_tunnel_info *info = skb_tunnel_info(skb); 992 993 if (info) { 994 struct ip_tunnel_info *unclone; 995 996 unclone = skb_tunnel_info_unclone(skb); 997 if (unlikely(!unclone)) { 998 dst_release(dst); 999 return -ENOMEM; 1000 } 1001 1002 unclone->key.u.ipv6.dst = fl6.saddr; 1003 unclone->key.u.ipv6.src = fl6.daddr; 1004 } 1005 1006 if (!pskb_may_pull(skb, ETH_HLEN)) { 1007 dst_release(dst); 1008 return -EINVAL; 1009 } 1010 1011 skb->protocol = eth_type_trans(skb, geneve->dev); 1012 __netif_rx(skb); 1013 dst_release(dst); 1014 return -EMSGSIZE; 1015 } 1016 1017 if (geneve->cfg.collect_md) { 1018 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 1019 ttl = key->ttl; 1020 } else { 1021 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel), 1022 ip_hdr(skb), skb); 1023 if (geneve->cfg.ttl_inherit) 1024 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb); 1025 else 1026 ttl = key->ttl; 1027 ttl = ttl ? : ip6_dst_hoplimit(dst); 1028 } 1029 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr), 1030 geneve->cfg.inner_proto_inherit); 1031 if (unlikely(err)) 1032 return err; 1033 1034 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev, 1035 &fl6.saddr, &fl6.daddr, prio, ttl, 1036 info->key.label, sport, geneve->cfg.info.key.tp_dst, 1037 !(info->key.tun_flags & TUNNEL_CSUM)); 1038 return 0; 1039 } 1040 #endif 1041 1042 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev) 1043 { 1044 struct geneve_dev *geneve = netdev_priv(dev); 1045 struct ip_tunnel_info *info = NULL; 1046 int err; 1047 1048 if (geneve->cfg.collect_md) { 1049 info = skb_tunnel_info(skb); 1050 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) { 1051 netdev_dbg(dev, "no tunnel metadata\n"); 1052 dev_kfree_skb(skb); 1053 dev->stats.tx_dropped++; 1054 return NETDEV_TX_OK; 1055 } 1056 } else { 1057 info = &geneve->cfg.info; 1058 } 1059 1060 rcu_read_lock(); 1061 #if IS_ENABLED(CONFIG_IPV6) 1062 if (info->mode & IP_TUNNEL_INFO_IPV6) 1063 err = geneve6_xmit_skb(skb, dev, geneve, info); 1064 else 1065 #endif 1066 err = geneve_xmit_skb(skb, dev, geneve, info); 1067 rcu_read_unlock(); 1068 1069 if (likely(!err)) 1070 return NETDEV_TX_OK; 1071 1072 if (err != -EMSGSIZE) 1073 dev_kfree_skb(skb); 1074 1075 if (err == -ELOOP) 1076 dev->stats.collisions++; 1077 else if (err == -ENETUNREACH) 1078 dev->stats.tx_carrier_errors++; 1079 1080 dev->stats.tx_errors++; 1081 return NETDEV_TX_OK; 1082 } 1083 1084 static int geneve_change_mtu(struct net_device *dev, int new_mtu) 1085 { 1086 if (new_mtu > dev->max_mtu) 1087 new_mtu = dev->max_mtu; 1088 else if (new_mtu < dev->min_mtu) 1089 new_mtu = dev->min_mtu; 1090 1091 dev->mtu = new_mtu; 1092 return 0; 1093 } 1094 1095 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 1096 { 1097 struct ip_tunnel_info *info = skb_tunnel_info(skb); 1098 struct geneve_dev *geneve = netdev_priv(dev); 1099 __be16 sport; 1100 1101 if (ip_tunnel_info_af(info) == AF_INET) { 1102 struct rtable *rt; 1103 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 1104 bool use_cache; 1105 __be32 saddr; 1106 u8 tos; 1107 1108 if (!gs4) 1109 return -EIO; 1110 1111 use_cache = ip_tunnel_dst_cache_usable(skb, info); 1112 tos = geneve_get_dsfield(skb, dev, info, &use_cache); 1113 sport = udp_flow_src_port(geneve->net, skb, 1114 1, USHRT_MAX, true); 1115 1116 rt = udp_tunnel_dst_lookup(skb, dev, geneve->net, 0, &saddr, 1117 &info->key, 1118 sport, geneve->cfg.info.key.tp_dst, 1119 tos, 1120 use_cache ? &info->dst_cache : NULL); 1121 if (IS_ERR(rt)) 1122 return PTR_ERR(rt); 1123 1124 ip_rt_put(rt); 1125 info->key.u.ipv4.src = saddr; 1126 #if IS_ENABLED(CONFIG_IPV6) 1127 } else if (ip_tunnel_info_af(info) == AF_INET6) { 1128 struct dst_entry *dst; 1129 struct flowi6 fl6; 1130 1131 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 1132 sport = udp_flow_src_port(geneve->net, skb, 1133 1, USHRT_MAX, true); 1134 1135 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info, 1136 geneve->cfg.info.key.tp_dst, sport); 1137 if (IS_ERR(dst)) 1138 return PTR_ERR(dst); 1139 1140 dst_release(dst); 1141 info->key.u.ipv6.src = fl6.saddr; 1142 #endif 1143 } else { 1144 return -EINVAL; 1145 } 1146 1147 info->key.tp_src = sport; 1148 info->key.tp_dst = geneve->cfg.info.key.tp_dst; 1149 return 0; 1150 } 1151 1152 static const struct net_device_ops geneve_netdev_ops = { 1153 .ndo_init = geneve_init, 1154 .ndo_uninit = geneve_uninit, 1155 .ndo_open = geneve_open, 1156 .ndo_stop = geneve_stop, 1157 .ndo_start_xmit = geneve_xmit, 1158 .ndo_get_stats64 = dev_get_tstats64, 1159 .ndo_change_mtu = geneve_change_mtu, 1160 .ndo_validate_addr = eth_validate_addr, 1161 .ndo_set_mac_address = eth_mac_addr, 1162 .ndo_fill_metadata_dst = geneve_fill_metadata_dst, 1163 }; 1164 1165 static void geneve_get_drvinfo(struct net_device *dev, 1166 struct ethtool_drvinfo *drvinfo) 1167 { 1168 strscpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version)); 1169 strscpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver)); 1170 } 1171 1172 static const struct ethtool_ops geneve_ethtool_ops = { 1173 .get_drvinfo = geneve_get_drvinfo, 1174 .get_link = ethtool_op_get_link, 1175 }; 1176 1177 /* Info for udev, that this is a virtual tunnel endpoint */ 1178 static struct device_type geneve_type = { 1179 .name = "geneve", 1180 }; 1181 1182 /* Calls the ndo_udp_tunnel_add of the caller in order to 1183 * supply the listening GENEVE udp ports. Callers are expected 1184 * to implement the ndo_udp_tunnel_add. 1185 */ 1186 static void geneve_offload_rx_ports(struct net_device *dev, bool push) 1187 { 1188 struct net *net = dev_net(dev); 1189 struct geneve_net *gn = net_generic(net, geneve_net_id); 1190 struct geneve_sock *gs; 1191 1192 rcu_read_lock(); 1193 list_for_each_entry_rcu(gs, &gn->sock_list, list) { 1194 if (push) { 1195 udp_tunnel_push_rx_port(dev, gs->sock, 1196 UDP_TUNNEL_TYPE_GENEVE); 1197 } else { 1198 udp_tunnel_drop_rx_port(dev, gs->sock, 1199 UDP_TUNNEL_TYPE_GENEVE); 1200 } 1201 } 1202 rcu_read_unlock(); 1203 } 1204 1205 /* Initialize the device structure. */ 1206 static void geneve_setup(struct net_device *dev) 1207 { 1208 ether_setup(dev); 1209 1210 dev->netdev_ops = &geneve_netdev_ops; 1211 dev->ethtool_ops = &geneve_ethtool_ops; 1212 dev->needs_free_netdev = true; 1213 1214 SET_NETDEV_DEVTYPE(dev, &geneve_type); 1215 1216 dev->features |= NETIF_F_LLTX; 1217 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST; 1218 dev->features |= NETIF_F_RXCSUM; 1219 dev->features |= NETIF_F_GSO_SOFTWARE; 1220 1221 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST; 1222 dev->hw_features |= NETIF_F_RXCSUM; 1223 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 1224 1225 /* MTU range: 68 - (something less than 65535) */ 1226 dev->min_mtu = ETH_MIN_MTU; 1227 /* The max_mtu calculation does not take account of GENEVE 1228 * options, to avoid excluding potentially valid 1229 * configurations. This will be further reduced by IPvX hdr size. 1230 */ 1231 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len; 1232 1233 netif_keep_dst(dev); 1234 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1235 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE; 1236 eth_hw_addr_random(dev); 1237 } 1238 1239 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = { 1240 [IFLA_GENEVE_UNSPEC] = { .strict_start_type = IFLA_GENEVE_INNER_PROTO_INHERIT }, 1241 [IFLA_GENEVE_ID] = { .type = NLA_U32 }, 1242 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) }, 1243 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) }, 1244 [IFLA_GENEVE_TTL] = { .type = NLA_U8 }, 1245 [IFLA_GENEVE_TOS] = { .type = NLA_U8 }, 1246 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 }, 1247 [IFLA_GENEVE_PORT] = { .type = NLA_U16 }, 1248 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1249 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 }, 1250 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 }, 1251 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 }, 1252 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 }, 1253 [IFLA_GENEVE_DF] = { .type = NLA_U8 }, 1254 [IFLA_GENEVE_INNER_PROTO_INHERIT] = { .type = NLA_FLAG }, 1255 }; 1256 1257 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[], 1258 struct netlink_ext_ack *extack) 1259 { 1260 if (tb[IFLA_ADDRESS]) { 1261 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1262 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1263 "Provided link layer address is not Ethernet"); 1264 return -EINVAL; 1265 } 1266 1267 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1268 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1269 "Provided Ethernet address is not unicast"); 1270 return -EADDRNOTAVAIL; 1271 } 1272 } 1273 1274 if (!data) { 1275 NL_SET_ERR_MSG(extack, 1276 "Not enough attributes provided to perform the operation"); 1277 return -EINVAL; 1278 } 1279 1280 if (data[IFLA_GENEVE_ID]) { 1281 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1282 1283 if (vni >= GENEVE_N_VID) { 1284 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID], 1285 "Geneve ID must be lower than 16777216"); 1286 return -ERANGE; 1287 } 1288 } 1289 1290 if (data[IFLA_GENEVE_DF]) { 1291 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]); 1292 1293 if (df < 0 || df > GENEVE_DF_MAX) { 1294 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF], 1295 "Invalid DF attribute"); 1296 return -EINVAL; 1297 } 1298 } 1299 1300 return 0; 1301 } 1302 1303 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn, 1304 const struct ip_tunnel_info *info, 1305 bool *tun_on_same_port, 1306 bool *tun_collect_md) 1307 { 1308 struct geneve_dev *geneve, *t = NULL; 1309 1310 *tun_on_same_port = false; 1311 *tun_collect_md = false; 1312 list_for_each_entry(geneve, &gn->geneve_list, next) { 1313 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) { 1314 *tun_collect_md = geneve->cfg.collect_md; 1315 *tun_on_same_port = true; 1316 } 1317 if (info->key.tun_id == geneve->cfg.info.key.tun_id && 1318 info->key.tp_dst == geneve->cfg.info.key.tp_dst && 1319 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u))) 1320 t = geneve; 1321 } 1322 return t; 1323 } 1324 1325 static bool is_tnl_info_zero(const struct ip_tunnel_info *info) 1326 { 1327 return !(info->key.tun_id || info->key.tun_flags || info->key.tos || 1328 info->key.ttl || info->key.label || info->key.tp_src || 1329 memchr_inv(&info->key.u, 0, sizeof(info->key.u))); 1330 } 1331 1332 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a, 1333 struct ip_tunnel_info *b) 1334 { 1335 if (ip_tunnel_info_af(a) == AF_INET) 1336 return a->key.u.ipv4.dst == b->key.u.ipv4.dst; 1337 else 1338 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst); 1339 } 1340 1341 static int geneve_configure(struct net *net, struct net_device *dev, 1342 struct netlink_ext_ack *extack, 1343 const struct geneve_config *cfg) 1344 { 1345 struct geneve_net *gn = net_generic(net, geneve_net_id); 1346 struct geneve_dev *t, *geneve = netdev_priv(dev); 1347 const struct ip_tunnel_info *info = &cfg->info; 1348 bool tun_collect_md, tun_on_same_port; 1349 int err, encap_len; 1350 1351 if (cfg->collect_md && !is_tnl_info_zero(info)) { 1352 NL_SET_ERR_MSG(extack, 1353 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified"); 1354 return -EINVAL; 1355 } 1356 1357 geneve->net = net; 1358 geneve->dev = dev; 1359 1360 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md); 1361 if (t) 1362 return -EBUSY; 1363 1364 /* make enough headroom for basic scenario */ 1365 encap_len = GENEVE_BASE_HLEN + ETH_HLEN; 1366 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) { 1367 encap_len += sizeof(struct iphdr); 1368 dev->max_mtu -= sizeof(struct iphdr); 1369 } else { 1370 encap_len += sizeof(struct ipv6hdr); 1371 dev->max_mtu -= sizeof(struct ipv6hdr); 1372 } 1373 dev->needed_headroom = encap_len + ETH_HLEN; 1374 1375 if (cfg->collect_md) { 1376 if (tun_on_same_port) { 1377 NL_SET_ERR_MSG(extack, 1378 "There can be only one externally controlled device on a destination port"); 1379 return -EPERM; 1380 } 1381 } else { 1382 if (tun_collect_md) { 1383 NL_SET_ERR_MSG(extack, 1384 "There already exists an externally controlled device on this destination port"); 1385 return -EPERM; 1386 } 1387 } 1388 1389 dst_cache_reset(&geneve->cfg.info.dst_cache); 1390 memcpy(&geneve->cfg, cfg, sizeof(*cfg)); 1391 1392 if (geneve->cfg.inner_proto_inherit) { 1393 dev->header_ops = NULL; 1394 dev->type = ARPHRD_NONE; 1395 dev->hard_header_len = 0; 1396 dev->addr_len = 0; 1397 dev->flags = IFF_POINTOPOINT | IFF_NOARP; 1398 } 1399 1400 err = register_netdevice(dev); 1401 if (err) 1402 return err; 1403 1404 list_add(&geneve->next, &gn->geneve_list); 1405 return 0; 1406 } 1407 1408 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port) 1409 { 1410 memset(info, 0, sizeof(*info)); 1411 info->key.tp_dst = htons(dst_port); 1412 } 1413 1414 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[], 1415 struct netlink_ext_ack *extack, 1416 struct geneve_config *cfg, bool changelink) 1417 { 1418 struct ip_tunnel_info *info = &cfg->info; 1419 int attrtype; 1420 1421 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) { 1422 NL_SET_ERR_MSG(extack, 1423 "Cannot specify both IPv4 and IPv6 Remote addresses"); 1424 return -EINVAL; 1425 } 1426 1427 if (data[IFLA_GENEVE_REMOTE]) { 1428 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) { 1429 attrtype = IFLA_GENEVE_REMOTE; 1430 goto change_notsup; 1431 } 1432 1433 info->key.u.ipv4.dst = 1434 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]); 1435 1436 if (ipv4_is_multicast(info->key.u.ipv4.dst)) { 1437 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE], 1438 "Remote IPv4 address cannot be Multicast"); 1439 return -EINVAL; 1440 } 1441 } 1442 1443 if (data[IFLA_GENEVE_REMOTE6]) { 1444 #if IS_ENABLED(CONFIG_IPV6) 1445 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) { 1446 attrtype = IFLA_GENEVE_REMOTE6; 1447 goto change_notsup; 1448 } 1449 1450 info->mode = IP_TUNNEL_INFO_IPV6; 1451 info->key.u.ipv6.dst = 1452 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]); 1453 1454 if (ipv6_addr_type(&info->key.u.ipv6.dst) & 1455 IPV6_ADDR_LINKLOCAL) { 1456 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1457 "Remote IPv6 address cannot be link-local"); 1458 return -EINVAL; 1459 } 1460 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) { 1461 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1462 "Remote IPv6 address cannot be Multicast"); 1463 return -EINVAL; 1464 } 1465 info->key.tun_flags |= TUNNEL_CSUM; 1466 cfg->use_udp6_rx_checksums = true; 1467 #else 1468 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1469 "IPv6 support not enabled in the kernel"); 1470 return -EPFNOSUPPORT; 1471 #endif 1472 } 1473 1474 if (data[IFLA_GENEVE_ID]) { 1475 __u32 vni; 1476 __u8 tvni[3]; 1477 __be64 tunid; 1478 1479 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1480 tvni[0] = (vni & 0x00ff0000) >> 16; 1481 tvni[1] = (vni & 0x0000ff00) >> 8; 1482 tvni[2] = vni & 0x000000ff; 1483 1484 tunid = vni_to_tunnel_id(tvni); 1485 if (changelink && (tunid != info->key.tun_id)) { 1486 attrtype = IFLA_GENEVE_ID; 1487 goto change_notsup; 1488 } 1489 info->key.tun_id = tunid; 1490 } 1491 1492 if (data[IFLA_GENEVE_TTL_INHERIT]) { 1493 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT])) 1494 cfg->ttl_inherit = true; 1495 else 1496 cfg->ttl_inherit = false; 1497 } else if (data[IFLA_GENEVE_TTL]) { 1498 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]); 1499 cfg->ttl_inherit = false; 1500 } 1501 1502 if (data[IFLA_GENEVE_TOS]) 1503 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]); 1504 1505 if (data[IFLA_GENEVE_DF]) 1506 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]); 1507 1508 if (data[IFLA_GENEVE_LABEL]) { 1509 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) & 1510 IPV6_FLOWLABEL_MASK; 1511 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) { 1512 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL], 1513 "Label attribute only applies for IPv6 Geneve devices"); 1514 return -EINVAL; 1515 } 1516 } 1517 1518 if (data[IFLA_GENEVE_PORT]) { 1519 if (changelink) { 1520 attrtype = IFLA_GENEVE_PORT; 1521 goto change_notsup; 1522 } 1523 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]); 1524 } 1525 1526 if (data[IFLA_GENEVE_COLLECT_METADATA]) { 1527 if (changelink) { 1528 attrtype = IFLA_GENEVE_COLLECT_METADATA; 1529 goto change_notsup; 1530 } 1531 cfg->collect_md = true; 1532 } 1533 1534 if (data[IFLA_GENEVE_UDP_CSUM]) { 1535 if (changelink) { 1536 attrtype = IFLA_GENEVE_UDP_CSUM; 1537 goto change_notsup; 1538 } 1539 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM])) 1540 info->key.tun_flags |= TUNNEL_CSUM; 1541 } 1542 1543 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) { 1544 #if IS_ENABLED(CONFIG_IPV6) 1545 if (changelink) { 1546 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX; 1547 goto change_notsup; 1548 } 1549 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX])) 1550 info->key.tun_flags &= ~TUNNEL_CSUM; 1551 #else 1552 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX], 1553 "IPv6 support not enabled in the kernel"); 1554 return -EPFNOSUPPORT; 1555 #endif 1556 } 1557 1558 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) { 1559 #if IS_ENABLED(CONFIG_IPV6) 1560 if (changelink) { 1561 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX; 1562 goto change_notsup; 1563 } 1564 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX])) 1565 cfg->use_udp6_rx_checksums = false; 1566 #else 1567 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX], 1568 "IPv6 support not enabled in the kernel"); 1569 return -EPFNOSUPPORT; 1570 #endif 1571 } 1572 1573 if (data[IFLA_GENEVE_INNER_PROTO_INHERIT]) { 1574 if (changelink) { 1575 attrtype = IFLA_GENEVE_INNER_PROTO_INHERIT; 1576 goto change_notsup; 1577 } 1578 cfg->inner_proto_inherit = true; 1579 } 1580 1581 return 0; 1582 change_notsup: 1583 NL_SET_ERR_MSG_ATTR(extack, data[attrtype], 1584 "Changing VNI, Port, endpoint IP address family, external, inner_proto_inherit, and UDP checksum attributes are not supported"); 1585 return -EOPNOTSUPP; 1586 } 1587 1588 static void geneve_link_config(struct net_device *dev, 1589 struct ip_tunnel_info *info, struct nlattr *tb[]) 1590 { 1591 struct geneve_dev *geneve = netdev_priv(dev); 1592 int ldev_mtu = 0; 1593 1594 if (tb[IFLA_MTU]) { 1595 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU])); 1596 return; 1597 } 1598 1599 switch (ip_tunnel_info_af(info)) { 1600 case AF_INET: { 1601 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst }; 1602 struct rtable *rt = ip_route_output_key(geneve->net, &fl4); 1603 1604 if (!IS_ERR(rt) && rt->dst.dev) { 1605 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN; 1606 ip_rt_put(rt); 1607 } 1608 break; 1609 } 1610 #if IS_ENABLED(CONFIG_IPV6) 1611 case AF_INET6: { 1612 struct rt6_info *rt; 1613 1614 if (!__in6_dev_get(dev)) 1615 break; 1616 1617 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0, 1618 NULL, 0); 1619 1620 if (rt && rt->dst.dev) 1621 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN; 1622 ip6_rt_put(rt); 1623 break; 1624 } 1625 #endif 1626 } 1627 1628 if (ldev_mtu <= 0) 1629 return; 1630 1631 geneve_change_mtu(dev, ldev_mtu - info->options_len); 1632 } 1633 1634 static int geneve_newlink(struct net *net, struct net_device *dev, 1635 struct nlattr *tb[], struct nlattr *data[], 1636 struct netlink_ext_ack *extack) 1637 { 1638 struct geneve_config cfg = { 1639 .df = GENEVE_DF_UNSET, 1640 .use_udp6_rx_checksums = false, 1641 .ttl_inherit = false, 1642 .collect_md = false, 1643 }; 1644 int err; 1645 1646 init_tnl_info(&cfg.info, GENEVE_UDP_PORT); 1647 err = geneve_nl2info(tb, data, extack, &cfg, false); 1648 if (err) 1649 return err; 1650 1651 err = geneve_configure(net, dev, extack, &cfg); 1652 if (err) 1653 return err; 1654 1655 geneve_link_config(dev, &cfg.info, tb); 1656 1657 return 0; 1658 } 1659 1660 /* Quiesces the geneve device data path for both TX and RX. 1661 * 1662 * On transmit geneve checks for non-NULL geneve_sock before it proceeds. 1663 * So, if we set that socket to NULL under RCU and wait for synchronize_net() 1664 * to complete for the existing set of in-flight packets to be transmitted, 1665 * then we would have quiesced the transmit data path. All the future packets 1666 * will get dropped until we unquiesce the data path. 1667 * 1668 * On receive geneve dereference the geneve_sock stashed in the socket. So, 1669 * if we set that to NULL under RCU and wait for synchronize_net() to 1670 * complete, then we would have quiesced the receive data path. 1671 */ 1672 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4, 1673 struct geneve_sock **gs6) 1674 { 1675 *gs4 = rtnl_dereference(geneve->sock4); 1676 rcu_assign_pointer(geneve->sock4, NULL); 1677 if (*gs4) 1678 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL); 1679 #if IS_ENABLED(CONFIG_IPV6) 1680 *gs6 = rtnl_dereference(geneve->sock6); 1681 rcu_assign_pointer(geneve->sock6, NULL); 1682 if (*gs6) 1683 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL); 1684 #else 1685 *gs6 = NULL; 1686 #endif 1687 synchronize_net(); 1688 } 1689 1690 /* Resumes the geneve device data path for both TX and RX. */ 1691 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4, 1692 struct geneve_sock __maybe_unused *gs6) 1693 { 1694 rcu_assign_pointer(geneve->sock4, gs4); 1695 if (gs4) 1696 rcu_assign_sk_user_data(gs4->sock->sk, gs4); 1697 #if IS_ENABLED(CONFIG_IPV6) 1698 rcu_assign_pointer(geneve->sock6, gs6); 1699 if (gs6) 1700 rcu_assign_sk_user_data(gs6->sock->sk, gs6); 1701 #endif 1702 synchronize_net(); 1703 } 1704 1705 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[], 1706 struct nlattr *data[], 1707 struct netlink_ext_ack *extack) 1708 { 1709 struct geneve_dev *geneve = netdev_priv(dev); 1710 struct geneve_sock *gs4, *gs6; 1711 struct geneve_config cfg; 1712 int err; 1713 1714 /* If the geneve device is configured for metadata (or externally 1715 * controlled, for example, OVS), then nothing can be changed. 1716 */ 1717 if (geneve->cfg.collect_md) 1718 return -EOPNOTSUPP; 1719 1720 /* Start with the existing info. */ 1721 memcpy(&cfg, &geneve->cfg, sizeof(cfg)); 1722 err = geneve_nl2info(tb, data, extack, &cfg, true); 1723 if (err) 1724 return err; 1725 1726 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) { 1727 dst_cache_reset(&cfg.info.dst_cache); 1728 geneve_link_config(dev, &cfg.info, tb); 1729 } 1730 1731 geneve_quiesce(geneve, &gs4, &gs6); 1732 memcpy(&geneve->cfg, &cfg, sizeof(cfg)); 1733 geneve_unquiesce(geneve, gs4, gs6); 1734 1735 return 0; 1736 } 1737 1738 static void geneve_dellink(struct net_device *dev, struct list_head *head) 1739 { 1740 struct geneve_dev *geneve = netdev_priv(dev); 1741 1742 list_del(&geneve->next); 1743 unregister_netdevice_queue(dev, head); 1744 } 1745 1746 static size_t geneve_get_size(const struct net_device *dev) 1747 { 1748 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */ 1749 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */ 1750 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */ 1751 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */ 1752 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */ 1753 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */ 1754 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */ 1755 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */ 1756 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */ 1757 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */ 1758 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */ 1759 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */ 1760 nla_total_size(0) + /* IFLA_GENEVE_INNER_PROTO_INHERIT */ 1761 0; 1762 } 1763 1764 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev) 1765 { 1766 struct geneve_dev *geneve = netdev_priv(dev); 1767 struct ip_tunnel_info *info = &geneve->cfg.info; 1768 bool ttl_inherit = geneve->cfg.ttl_inherit; 1769 bool metadata = geneve->cfg.collect_md; 1770 __u8 tmp_vni[3]; 1771 __u32 vni; 1772 1773 tunnel_id_to_vni(info->key.tun_id, tmp_vni); 1774 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2]; 1775 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni)) 1776 goto nla_put_failure; 1777 1778 if (!metadata && ip_tunnel_info_af(info) == AF_INET) { 1779 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE, 1780 info->key.u.ipv4.dst)) 1781 goto nla_put_failure; 1782 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM, 1783 !!(info->key.tun_flags & TUNNEL_CSUM))) 1784 goto nla_put_failure; 1785 1786 #if IS_ENABLED(CONFIG_IPV6) 1787 } else if (!metadata) { 1788 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6, 1789 &info->key.u.ipv6.dst)) 1790 goto nla_put_failure; 1791 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX, 1792 !(info->key.tun_flags & TUNNEL_CSUM))) 1793 goto nla_put_failure; 1794 #endif 1795 } 1796 1797 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) || 1798 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) || 1799 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label)) 1800 goto nla_put_failure; 1801 1802 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df)) 1803 goto nla_put_failure; 1804 1805 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst)) 1806 goto nla_put_failure; 1807 1808 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA)) 1809 goto nla_put_failure; 1810 1811 #if IS_ENABLED(CONFIG_IPV6) 1812 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX, 1813 !geneve->cfg.use_udp6_rx_checksums)) 1814 goto nla_put_failure; 1815 #endif 1816 1817 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit)) 1818 goto nla_put_failure; 1819 1820 if (geneve->cfg.inner_proto_inherit && 1821 nla_put_flag(skb, IFLA_GENEVE_INNER_PROTO_INHERIT)) 1822 goto nla_put_failure; 1823 1824 return 0; 1825 1826 nla_put_failure: 1827 return -EMSGSIZE; 1828 } 1829 1830 static struct rtnl_link_ops geneve_link_ops __read_mostly = { 1831 .kind = "geneve", 1832 .maxtype = IFLA_GENEVE_MAX, 1833 .policy = geneve_policy, 1834 .priv_size = sizeof(struct geneve_dev), 1835 .setup = geneve_setup, 1836 .validate = geneve_validate, 1837 .newlink = geneve_newlink, 1838 .changelink = geneve_changelink, 1839 .dellink = geneve_dellink, 1840 .get_size = geneve_get_size, 1841 .fill_info = geneve_fill_info, 1842 }; 1843 1844 struct net_device *geneve_dev_create_fb(struct net *net, const char *name, 1845 u8 name_assign_type, u16 dst_port) 1846 { 1847 struct nlattr *tb[IFLA_MAX + 1]; 1848 struct net_device *dev; 1849 LIST_HEAD(list_kill); 1850 int err; 1851 struct geneve_config cfg = { 1852 .df = GENEVE_DF_UNSET, 1853 .use_udp6_rx_checksums = true, 1854 .ttl_inherit = false, 1855 .collect_md = true, 1856 }; 1857 1858 memset(tb, 0, sizeof(tb)); 1859 dev = rtnl_create_link(net, name, name_assign_type, 1860 &geneve_link_ops, tb, NULL); 1861 if (IS_ERR(dev)) 1862 return dev; 1863 1864 init_tnl_info(&cfg.info, dst_port); 1865 err = geneve_configure(net, dev, NULL, &cfg); 1866 if (err) { 1867 free_netdev(dev); 1868 return ERR_PTR(err); 1869 } 1870 1871 /* openvswitch users expect packet sizes to be unrestricted, 1872 * so set the largest MTU we can. 1873 */ 1874 err = geneve_change_mtu(dev, IP_MAX_MTU); 1875 if (err) 1876 goto err; 1877 1878 err = rtnl_configure_link(dev, NULL, 0, NULL); 1879 if (err < 0) 1880 goto err; 1881 1882 return dev; 1883 err: 1884 geneve_dellink(dev, &list_kill); 1885 unregister_netdevice_many(&list_kill); 1886 return ERR_PTR(err); 1887 } 1888 EXPORT_SYMBOL_GPL(geneve_dev_create_fb); 1889 1890 static int geneve_netdevice_event(struct notifier_block *unused, 1891 unsigned long event, void *ptr) 1892 { 1893 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1894 1895 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO) 1896 geneve_offload_rx_ports(dev, true); 1897 else if (event == NETDEV_UDP_TUNNEL_DROP_INFO) 1898 geneve_offload_rx_ports(dev, false); 1899 1900 return NOTIFY_DONE; 1901 } 1902 1903 static struct notifier_block geneve_notifier_block __read_mostly = { 1904 .notifier_call = geneve_netdevice_event, 1905 }; 1906 1907 static __net_init int geneve_init_net(struct net *net) 1908 { 1909 struct geneve_net *gn = net_generic(net, geneve_net_id); 1910 1911 INIT_LIST_HEAD(&gn->geneve_list); 1912 INIT_LIST_HEAD(&gn->sock_list); 1913 return 0; 1914 } 1915 1916 static void geneve_destroy_tunnels(struct net *net, struct list_head *head) 1917 { 1918 struct geneve_net *gn = net_generic(net, geneve_net_id); 1919 struct geneve_dev *geneve, *next; 1920 struct net_device *dev, *aux; 1921 1922 /* gather any geneve devices that were moved into this ns */ 1923 for_each_netdev_safe(net, dev, aux) 1924 if (dev->rtnl_link_ops == &geneve_link_ops) 1925 unregister_netdevice_queue(dev, head); 1926 1927 /* now gather any other geneve devices that were created in this ns */ 1928 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) { 1929 /* If geneve->dev is in the same netns, it was already added 1930 * to the list by the previous loop. 1931 */ 1932 if (!net_eq(dev_net(geneve->dev), net)) 1933 unregister_netdevice_queue(geneve->dev, head); 1934 } 1935 } 1936 1937 static void __net_exit geneve_exit_batch_net(struct list_head *net_list) 1938 { 1939 struct net *net; 1940 LIST_HEAD(list); 1941 1942 rtnl_lock(); 1943 list_for_each_entry(net, net_list, exit_list) 1944 geneve_destroy_tunnels(net, &list); 1945 1946 /* unregister the devices gathered above */ 1947 unregister_netdevice_many(&list); 1948 rtnl_unlock(); 1949 1950 list_for_each_entry(net, net_list, exit_list) { 1951 const struct geneve_net *gn = net_generic(net, geneve_net_id); 1952 1953 WARN_ON_ONCE(!list_empty(&gn->sock_list)); 1954 } 1955 } 1956 1957 static struct pernet_operations geneve_net_ops = { 1958 .init = geneve_init_net, 1959 .exit_batch = geneve_exit_batch_net, 1960 .id = &geneve_net_id, 1961 .size = sizeof(struct geneve_net), 1962 }; 1963 1964 static int __init geneve_init_module(void) 1965 { 1966 int rc; 1967 1968 rc = register_pernet_subsys(&geneve_net_ops); 1969 if (rc) 1970 goto out1; 1971 1972 rc = register_netdevice_notifier(&geneve_notifier_block); 1973 if (rc) 1974 goto out2; 1975 1976 rc = rtnl_link_register(&geneve_link_ops); 1977 if (rc) 1978 goto out3; 1979 1980 return 0; 1981 out3: 1982 unregister_netdevice_notifier(&geneve_notifier_block); 1983 out2: 1984 unregister_pernet_subsys(&geneve_net_ops); 1985 out1: 1986 return rc; 1987 } 1988 late_initcall(geneve_init_module); 1989 1990 static void __exit geneve_cleanup_module(void) 1991 { 1992 rtnl_link_unregister(&geneve_link_ops); 1993 unregister_netdevice_notifier(&geneve_notifier_block); 1994 unregister_pernet_subsys(&geneve_net_ops); 1995 } 1996 module_exit(geneve_cleanup_module); 1997 1998 MODULE_LICENSE("GPL"); 1999 MODULE_VERSION(GENEVE_NETDEV_VER); 2000 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>"); 2001 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic"); 2002 MODULE_ALIAS_RTNL_LINK("geneve"); 2003