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/kernel.h> 11 #include <linux/module.h> 12 #include <linux/etherdevice.h> 13 #include <linux/hash.h> 14 #include <net/ipv6_stubs.h> 15 #include <net/dst_metadata.h> 16 #include <net/gro_cells.h> 17 #include <net/rtnetlink.h> 18 #include <net/geneve.h> 19 #include <net/protocol.h> 20 21 #define GENEVE_NETDEV_VER "0.6" 22 23 #define GENEVE_N_VID (1u << 24) 24 #define GENEVE_VID_MASK (GENEVE_N_VID - 1) 25 26 #define VNI_HASH_BITS 10 27 #define VNI_HASH_SIZE (1<<VNI_HASH_BITS) 28 29 static bool log_ecn_error = true; 30 module_param(log_ecn_error, bool, 0644); 31 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN"); 32 33 #define GENEVE_VER 0 34 #define GENEVE_BASE_HLEN (sizeof(struct udphdr) + sizeof(struct genevehdr)) 35 #define GENEVE_IPV4_HLEN (ETH_HLEN + sizeof(struct iphdr) + GENEVE_BASE_HLEN) 36 #define GENEVE_IPV6_HLEN (ETH_HLEN + sizeof(struct ipv6hdr) + GENEVE_BASE_HLEN) 37 38 /* per-network namespace private data for this module */ 39 struct geneve_net { 40 struct list_head geneve_list; 41 struct list_head sock_list; 42 }; 43 44 static unsigned int geneve_net_id; 45 46 struct geneve_dev_node { 47 struct hlist_node hlist; 48 struct geneve_dev *geneve; 49 }; 50 51 struct geneve_config { 52 struct ip_tunnel_info info; 53 bool collect_md; 54 bool use_udp6_rx_checksums; 55 bool ttl_inherit; 56 enum ifla_geneve_df df; 57 }; 58 59 /* Pseudo network device */ 60 struct geneve_dev { 61 struct geneve_dev_node hlist4; /* vni hash table for IPv4 socket */ 62 #if IS_ENABLED(CONFIG_IPV6) 63 struct geneve_dev_node hlist6; /* vni hash table for IPv6 socket */ 64 #endif 65 struct net *net; /* netns for packet i/o */ 66 struct net_device *dev; /* netdev for geneve tunnel */ 67 struct geneve_sock __rcu *sock4; /* IPv4 socket used for geneve tunnel */ 68 #if IS_ENABLED(CONFIG_IPV6) 69 struct geneve_sock __rcu *sock6; /* IPv6 socket used for geneve tunnel */ 70 #endif 71 struct list_head next; /* geneve's per namespace list */ 72 struct gro_cells gro_cells; 73 struct geneve_config cfg; 74 }; 75 76 struct geneve_sock { 77 bool collect_md; 78 struct list_head list; 79 struct socket *sock; 80 struct rcu_head rcu; 81 int refcnt; 82 struct hlist_head vni_list[VNI_HASH_SIZE]; 83 }; 84 85 static inline __u32 geneve_net_vni_hash(u8 vni[3]) 86 { 87 __u32 vnid; 88 89 vnid = (vni[0] << 16) | (vni[1] << 8) | vni[2]; 90 return hash_32(vnid, VNI_HASH_BITS); 91 } 92 93 static __be64 vni_to_tunnel_id(const __u8 *vni) 94 { 95 #ifdef __BIG_ENDIAN 96 return (vni[0] << 16) | (vni[1] << 8) | vni[2]; 97 #else 98 return (__force __be64)(((__force u64)vni[0] << 40) | 99 ((__force u64)vni[1] << 48) | 100 ((__force u64)vni[2] << 56)); 101 #endif 102 } 103 104 /* Convert 64 bit tunnel ID to 24 bit VNI. */ 105 static void tunnel_id_to_vni(__be64 tun_id, __u8 *vni) 106 { 107 #ifdef __BIG_ENDIAN 108 vni[0] = (__force __u8)(tun_id >> 16); 109 vni[1] = (__force __u8)(tun_id >> 8); 110 vni[2] = (__force __u8)tun_id; 111 #else 112 vni[0] = (__force __u8)((__force u64)tun_id >> 40); 113 vni[1] = (__force __u8)((__force u64)tun_id >> 48); 114 vni[2] = (__force __u8)((__force u64)tun_id >> 56); 115 #endif 116 } 117 118 static bool eq_tun_id_and_vni(u8 *tun_id, u8 *vni) 119 { 120 return !memcmp(vni, &tun_id[5], 3); 121 } 122 123 static sa_family_t geneve_get_sk_family(struct geneve_sock *gs) 124 { 125 return gs->sock->sk->sk_family; 126 } 127 128 static struct geneve_dev *geneve_lookup(struct geneve_sock *gs, 129 __be32 addr, u8 vni[]) 130 { 131 struct hlist_head *vni_list_head; 132 struct geneve_dev_node *node; 133 __u32 hash; 134 135 /* Find the device for this VNI */ 136 hash = geneve_net_vni_hash(vni); 137 vni_list_head = &gs->vni_list[hash]; 138 hlist_for_each_entry_rcu(node, vni_list_head, hlist) { 139 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) && 140 addr == node->geneve->cfg.info.key.u.ipv4.dst) 141 return node->geneve; 142 } 143 return NULL; 144 } 145 146 #if IS_ENABLED(CONFIG_IPV6) 147 static struct geneve_dev *geneve6_lookup(struct geneve_sock *gs, 148 struct in6_addr addr6, u8 vni[]) 149 { 150 struct hlist_head *vni_list_head; 151 struct geneve_dev_node *node; 152 __u32 hash; 153 154 /* Find the device for this VNI */ 155 hash = geneve_net_vni_hash(vni); 156 vni_list_head = &gs->vni_list[hash]; 157 hlist_for_each_entry_rcu(node, vni_list_head, hlist) { 158 if (eq_tun_id_and_vni((u8 *)&node->geneve->cfg.info.key.tun_id, vni) && 159 ipv6_addr_equal(&addr6, &node->geneve->cfg.info.key.u.ipv6.dst)) 160 return node->geneve; 161 } 162 return NULL; 163 } 164 #endif 165 166 static inline struct genevehdr *geneve_hdr(const struct sk_buff *skb) 167 { 168 return (struct genevehdr *)(udp_hdr(skb) + 1); 169 } 170 171 static struct geneve_dev *geneve_lookup_skb(struct geneve_sock *gs, 172 struct sk_buff *skb) 173 { 174 static u8 zero_vni[3]; 175 u8 *vni; 176 177 if (geneve_get_sk_family(gs) == AF_INET) { 178 struct iphdr *iph; 179 __be32 addr; 180 181 iph = ip_hdr(skb); /* outer IP header... */ 182 183 if (gs->collect_md) { 184 vni = zero_vni; 185 addr = 0; 186 } else { 187 vni = geneve_hdr(skb)->vni; 188 addr = iph->saddr; 189 } 190 191 return geneve_lookup(gs, addr, vni); 192 #if IS_ENABLED(CONFIG_IPV6) 193 } else if (geneve_get_sk_family(gs) == AF_INET6) { 194 static struct in6_addr zero_addr6; 195 struct ipv6hdr *ip6h; 196 struct in6_addr addr6; 197 198 ip6h = ipv6_hdr(skb); /* outer IPv6 header... */ 199 200 if (gs->collect_md) { 201 vni = zero_vni; 202 addr6 = zero_addr6; 203 } else { 204 vni = geneve_hdr(skb)->vni; 205 addr6 = ip6h->saddr; 206 } 207 208 return geneve6_lookup(gs, addr6, vni); 209 #endif 210 } 211 return NULL; 212 } 213 214 /* geneve receive/decap routine */ 215 static void geneve_rx(struct geneve_dev *geneve, struct geneve_sock *gs, 216 struct sk_buff *skb) 217 { 218 struct genevehdr *gnvh = geneve_hdr(skb); 219 struct metadata_dst *tun_dst = NULL; 220 struct pcpu_sw_netstats *stats; 221 unsigned int len; 222 int err = 0; 223 void *oiph; 224 225 if (ip_tunnel_collect_metadata() || gs->collect_md) { 226 __be16 flags; 227 228 flags = TUNNEL_KEY | TUNNEL_GENEVE_OPT | 229 (gnvh->oam ? TUNNEL_OAM : 0) | 230 (gnvh->critical ? TUNNEL_CRIT_OPT : 0); 231 232 tun_dst = udp_tun_rx_dst(skb, geneve_get_sk_family(gs), flags, 233 vni_to_tunnel_id(gnvh->vni), 234 gnvh->opt_len * 4); 235 if (!tun_dst) { 236 geneve->dev->stats.rx_dropped++; 237 goto drop; 238 } 239 /* Update tunnel dst according to Geneve options. */ 240 ip_tunnel_info_opts_set(&tun_dst->u.tun_info, 241 gnvh->options, gnvh->opt_len * 4, 242 TUNNEL_GENEVE_OPT); 243 } else { 244 /* Drop packets w/ critical options, 245 * since we don't support any... 246 */ 247 if (gnvh->critical) { 248 geneve->dev->stats.rx_frame_errors++; 249 geneve->dev->stats.rx_errors++; 250 goto drop; 251 } 252 } 253 254 skb_reset_mac_header(skb); 255 skb->protocol = eth_type_trans(skb, geneve->dev); 256 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 257 258 if (tun_dst) 259 skb_dst_set(skb, &tun_dst->dst); 260 261 /* Ignore packet loops (and multicast echo) */ 262 if (ether_addr_equal(eth_hdr(skb)->h_source, geneve->dev->dev_addr)) { 263 geneve->dev->stats.rx_errors++; 264 goto drop; 265 } 266 267 oiph = skb_network_header(skb); 268 skb_reset_network_header(skb); 269 270 if (geneve_get_sk_family(gs) == AF_INET) 271 err = IP_ECN_decapsulate(oiph, skb); 272 #if IS_ENABLED(CONFIG_IPV6) 273 else 274 err = IP6_ECN_decapsulate(oiph, skb); 275 #endif 276 277 if (unlikely(err)) { 278 if (log_ecn_error) { 279 if (geneve_get_sk_family(gs) == AF_INET) 280 net_info_ratelimited("non-ECT from %pI4 " 281 "with TOS=%#x\n", 282 &((struct iphdr *)oiph)->saddr, 283 ((struct iphdr *)oiph)->tos); 284 #if IS_ENABLED(CONFIG_IPV6) 285 else 286 net_info_ratelimited("non-ECT from %pI6\n", 287 &((struct ipv6hdr *)oiph)->saddr); 288 #endif 289 } 290 if (err > 1) { 291 ++geneve->dev->stats.rx_frame_errors; 292 ++geneve->dev->stats.rx_errors; 293 goto drop; 294 } 295 } 296 297 len = skb->len; 298 err = gro_cells_receive(&geneve->gro_cells, skb); 299 if (likely(err == NET_RX_SUCCESS)) { 300 stats = this_cpu_ptr(geneve->dev->tstats); 301 u64_stats_update_begin(&stats->syncp); 302 stats->rx_packets++; 303 stats->rx_bytes += len; 304 u64_stats_update_end(&stats->syncp); 305 } 306 return; 307 drop: 308 /* Consume bad packet */ 309 kfree_skb(skb); 310 } 311 312 /* Setup stats when device is created */ 313 static int geneve_init(struct net_device *dev) 314 { 315 struct geneve_dev *geneve = netdev_priv(dev); 316 int err; 317 318 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 319 if (!dev->tstats) 320 return -ENOMEM; 321 322 err = gro_cells_init(&geneve->gro_cells, dev); 323 if (err) { 324 free_percpu(dev->tstats); 325 return err; 326 } 327 328 err = dst_cache_init(&geneve->cfg.info.dst_cache, GFP_KERNEL); 329 if (err) { 330 free_percpu(dev->tstats); 331 gro_cells_destroy(&geneve->gro_cells); 332 return err; 333 } 334 return 0; 335 } 336 337 static void geneve_uninit(struct net_device *dev) 338 { 339 struct geneve_dev *geneve = netdev_priv(dev); 340 341 dst_cache_destroy(&geneve->cfg.info.dst_cache); 342 gro_cells_destroy(&geneve->gro_cells); 343 free_percpu(dev->tstats); 344 } 345 346 /* Callback from net/ipv4/udp.c to receive packets */ 347 static int geneve_udp_encap_recv(struct sock *sk, struct sk_buff *skb) 348 { 349 struct genevehdr *geneveh; 350 struct geneve_dev *geneve; 351 struct geneve_sock *gs; 352 int opts_len; 353 354 /* Need UDP and Geneve header to be present */ 355 if (unlikely(!pskb_may_pull(skb, GENEVE_BASE_HLEN))) 356 goto drop; 357 358 /* Return packets with reserved bits set */ 359 geneveh = geneve_hdr(skb); 360 if (unlikely(geneveh->ver != GENEVE_VER)) 361 goto drop; 362 363 if (unlikely(geneveh->proto_type != htons(ETH_P_TEB))) 364 goto drop; 365 366 gs = rcu_dereference_sk_user_data(sk); 367 if (!gs) 368 goto drop; 369 370 geneve = geneve_lookup_skb(gs, skb); 371 if (!geneve) 372 goto drop; 373 374 opts_len = geneveh->opt_len * 4; 375 if (iptunnel_pull_header(skb, GENEVE_BASE_HLEN + opts_len, 376 htons(ETH_P_TEB), 377 !net_eq(geneve->net, dev_net(geneve->dev)))) { 378 geneve->dev->stats.rx_dropped++; 379 goto drop; 380 } 381 382 geneve_rx(geneve, gs, skb); 383 return 0; 384 385 drop: 386 /* Consume bad packet */ 387 kfree_skb(skb); 388 return 0; 389 } 390 391 /* Callback from net/ipv{4,6}/udp.c to check that we have a tunnel for errors */ 392 static int geneve_udp_encap_err_lookup(struct sock *sk, struct sk_buff *skb) 393 { 394 struct genevehdr *geneveh; 395 struct geneve_sock *gs; 396 u8 zero_vni[3] = { 0 }; 397 u8 *vni = zero_vni; 398 399 if (!pskb_may_pull(skb, skb_transport_offset(skb) + GENEVE_BASE_HLEN)) 400 return -EINVAL; 401 402 geneveh = geneve_hdr(skb); 403 if (geneveh->ver != GENEVE_VER) 404 return -EINVAL; 405 406 if (geneveh->proto_type != htons(ETH_P_TEB)) 407 return -EINVAL; 408 409 gs = rcu_dereference_sk_user_data(sk); 410 if (!gs) 411 return -ENOENT; 412 413 if (geneve_get_sk_family(gs) == AF_INET) { 414 struct iphdr *iph = ip_hdr(skb); 415 __be32 addr4 = 0; 416 417 if (!gs->collect_md) { 418 vni = geneve_hdr(skb)->vni; 419 addr4 = iph->daddr; 420 } 421 422 return geneve_lookup(gs, addr4, vni) ? 0 : -ENOENT; 423 } 424 425 #if IS_ENABLED(CONFIG_IPV6) 426 if (geneve_get_sk_family(gs) == AF_INET6) { 427 struct ipv6hdr *ip6h = ipv6_hdr(skb); 428 struct in6_addr addr6; 429 430 memset(&addr6, 0, sizeof(struct in6_addr)); 431 432 if (!gs->collect_md) { 433 vni = geneve_hdr(skb)->vni; 434 addr6 = ip6h->daddr; 435 } 436 437 return geneve6_lookup(gs, addr6, vni) ? 0 : -ENOENT; 438 } 439 #endif 440 441 return -EPFNOSUPPORT; 442 } 443 444 static struct socket *geneve_create_sock(struct net *net, bool ipv6, 445 __be16 port, bool ipv6_rx_csum) 446 { 447 struct socket *sock; 448 struct udp_port_cfg udp_conf; 449 int err; 450 451 memset(&udp_conf, 0, sizeof(udp_conf)); 452 453 if (ipv6) { 454 udp_conf.family = AF_INET6; 455 udp_conf.ipv6_v6only = 1; 456 udp_conf.use_udp6_rx_checksums = ipv6_rx_csum; 457 } else { 458 udp_conf.family = AF_INET; 459 udp_conf.local_ip.s_addr = htonl(INADDR_ANY); 460 } 461 462 udp_conf.local_udp_port = port; 463 464 /* Open UDP socket */ 465 err = udp_sock_create(net, &udp_conf, &sock); 466 if (err < 0) 467 return ERR_PTR(err); 468 469 return sock; 470 } 471 472 static int geneve_hlen(struct genevehdr *gh) 473 { 474 return sizeof(*gh) + gh->opt_len * 4; 475 } 476 477 static struct sk_buff *geneve_gro_receive(struct sock *sk, 478 struct list_head *head, 479 struct sk_buff *skb) 480 { 481 struct sk_buff *pp = NULL; 482 struct sk_buff *p; 483 struct genevehdr *gh, *gh2; 484 unsigned int hlen, gh_len, off_gnv; 485 const struct packet_offload *ptype; 486 __be16 type; 487 int flush = 1; 488 489 off_gnv = skb_gro_offset(skb); 490 hlen = off_gnv + sizeof(*gh); 491 gh = skb_gro_header_fast(skb, off_gnv); 492 if (skb_gro_header_hard(skb, hlen)) { 493 gh = skb_gro_header_slow(skb, hlen, off_gnv); 494 if (unlikely(!gh)) 495 goto out; 496 } 497 498 if (gh->ver != GENEVE_VER || gh->oam) 499 goto out; 500 gh_len = geneve_hlen(gh); 501 502 hlen = off_gnv + gh_len; 503 if (skb_gro_header_hard(skb, hlen)) { 504 gh = skb_gro_header_slow(skb, hlen, off_gnv); 505 if (unlikely(!gh)) 506 goto out; 507 } 508 509 list_for_each_entry(p, head, list) { 510 if (!NAPI_GRO_CB(p)->same_flow) 511 continue; 512 513 gh2 = (struct genevehdr *)(p->data + off_gnv); 514 if (gh->opt_len != gh2->opt_len || 515 memcmp(gh, gh2, gh_len)) { 516 NAPI_GRO_CB(p)->same_flow = 0; 517 continue; 518 } 519 } 520 521 type = gh->proto_type; 522 523 rcu_read_lock(); 524 ptype = gro_find_receive_by_type(type); 525 if (!ptype) 526 goto out_unlock; 527 528 skb_gro_pull(skb, gh_len); 529 skb_gro_postpull_rcsum(skb, gh, gh_len); 530 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb); 531 flush = 0; 532 533 out_unlock: 534 rcu_read_unlock(); 535 out: 536 skb_gro_flush_final(skb, pp, flush); 537 538 return pp; 539 } 540 541 static int geneve_gro_complete(struct sock *sk, struct sk_buff *skb, 542 int nhoff) 543 { 544 struct genevehdr *gh; 545 struct packet_offload *ptype; 546 __be16 type; 547 int gh_len; 548 int err = -ENOSYS; 549 550 gh = (struct genevehdr *)(skb->data + nhoff); 551 gh_len = geneve_hlen(gh); 552 type = gh->proto_type; 553 554 rcu_read_lock(); 555 ptype = gro_find_complete_by_type(type); 556 if (ptype) 557 err = ptype->callbacks.gro_complete(skb, nhoff + gh_len); 558 559 rcu_read_unlock(); 560 561 skb_set_inner_mac_header(skb, nhoff + gh_len); 562 563 return err; 564 } 565 566 /* Create new listen socket if needed */ 567 static struct geneve_sock *geneve_socket_create(struct net *net, __be16 port, 568 bool ipv6, bool ipv6_rx_csum) 569 { 570 struct geneve_net *gn = net_generic(net, geneve_net_id); 571 struct geneve_sock *gs; 572 struct socket *sock; 573 struct udp_tunnel_sock_cfg tunnel_cfg; 574 int h; 575 576 gs = kzalloc(sizeof(*gs), GFP_KERNEL); 577 if (!gs) 578 return ERR_PTR(-ENOMEM); 579 580 sock = geneve_create_sock(net, ipv6, port, ipv6_rx_csum); 581 if (IS_ERR(sock)) { 582 kfree(gs); 583 return ERR_CAST(sock); 584 } 585 586 gs->sock = sock; 587 gs->refcnt = 1; 588 for (h = 0; h < VNI_HASH_SIZE; ++h) 589 INIT_HLIST_HEAD(&gs->vni_list[h]); 590 591 /* Initialize the geneve udp offloads structure */ 592 udp_tunnel_notify_add_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 593 594 /* Mark socket as an encapsulation socket */ 595 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg)); 596 tunnel_cfg.sk_user_data = gs; 597 tunnel_cfg.encap_type = 1; 598 tunnel_cfg.gro_receive = geneve_gro_receive; 599 tunnel_cfg.gro_complete = geneve_gro_complete; 600 tunnel_cfg.encap_rcv = geneve_udp_encap_recv; 601 tunnel_cfg.encap_err_lookup = geneve_udp_encap_err_lookup; 602 tunnel_cfg.encap_destroy = NULL; 603 setup_udp_tunnel_sock(net, sock, &tunnel_cfg); 604 list_add(&gs->list, &gn->sock_list); 605 return gs; 606 } 607 608 static void __geneve_sock_release(struct geneve_sock *gs) 609 { 610 if (!gs || --gs->refcnt) 611 return; 612 613 list_del(&gs->list); 614 udp_tunnel_notify_del_rx_port(gs->sock, UDP_TUNNEL_TYPE_GENEVE); 615 udp_tunnel_sock_release(gs->sock); 616 kfree_rcu(gs, rcu); 617 } 618 619 static void geneve_sock_release(struct geneve_dev *geneve) 620 { 621 struct geneve_sock *gs4 = rtnl_dereference(geneve->sock4); 622 #if IS_ENABLED(CONFIG_IPV6) 623 struct geneve_sock *gs6 = rtnl_dereference(geneve->sock6); 624 625 rcu_assign_pointer(geneve->sock6, NULL); 626 #endif 627 628 rcu_assign_pointer(geneve->sock4, NULL); 629 synchronize_net(); 630 631 __geneve_sock_release(gs4); 632 #if IS_ENABLED(CONFIG_IPV6) 633 __geneve_sock_release(gs6); 634 #endif 635 } 636 637 static struct geneve_sock *geneve_find_sock(struct geneve_net *gn, 638 sa_family_t family, 639 __be16 dst_port) 640 { 641 struct geneve_sock *gs; 642 643 list_for_each_entry(gs, &gn->sock_list, list) { 644 if (inet_sk(gs->sock->sk)->inet_sport == dst_port && 645 geneve_get_sk_family(gs) == family) { 646 return gs; 647 } 648 } 649 return NULL; 650 } 651 652 static int geneve_sock_add(struct geneve_dev *geneve, bool ipv6) 653 { 654 struct net *net = geneve->net; 655 struct geneve_net *gn = net_generic(net, geneve_net_id); 656 struct geneve_dev_node *node; 657 struct geneve_sock *gs; 658 __u8 vni[3]; 659 __u32 hash; 660 661 gs = geneve_find_sock(gn, ipv6 ? AF_INET6 : AF_INET, geneve->cfg.info.key.tp_dst); 662 if (gs) { 663 gs->refcnt++; 664 goto out; 665 } 666 667 gs = geneve_socket_create(net, geneve->cfg.info.key.tp_dst, ipv6, 668 geneve->cfg.use_udp6_rx_checksums); 669 if (IS_ERR(gs)) 670 return PTR_ERR(gs); 671 672 out: 673 gs->collect_md = geneve->cfg.collect_md; 674 #if IS_ENABLED(CONFIG_IPV6) 675 if (ipv6) { 676 rcu_assign_pointer(geneve->sock6, gs); 677 node = &geneve->hlist6; 678 } else 679 #endif 680 { 681 rcu_assign_pointer(geneve->sock4, gs); 682 node = &geneve->hlist4; 683 } 684 node->geneve = geneve; 685 686 tunnel_id_to_vni(geneve->cfg.info.key.tun_id, vni); 687 hash = geneve_net_vni_hash(vni); 688 hlist_add_head_rcu(&node->hlist, &gs->vni_list[hash]); 689 return 0; 690 } 691 692 static int geneve_open(struct net_device *dev) 693 { 694 struct geneve_dev *geneve = netdev_priv(dev); 695 bool metadata = geneve->cfg.collect_md; 696 bool ipv4, ipv6; 697 int ret = 0; 698 699 ipv6 = geneve->cfg.info.mode & IP_TUNNEL_INFO_IPV6 || metadata; 700 ipv4 = !ipv6 || metadata; 701 #if IS_ENABLED(CONFIG_IPV6) 702 if (ipv6) { 703 ret = geneve_sock_add(geneve, true); 704 if (ret < 0 && ret != -EAFNOSUPPORT) 705 ipv4 = false; 706 } 707 #endif 708 if (ipv4) 709 ret = geneve_sock_add(geneve, false); 710 if (ret < 0) 711 geneve_sock_release(geneve); 712 713 return ret; 714 } 715 716 static int geneve_stop(struct net_device *dev) 717 { 718 struct geneve_dev *geneve = netdev_priv(dev); 719 720 hlist_del_init_rcu(&geneve->hlist4.hlist); 721 #if IS_ENABLED(CONFIG_IPV6) 722 hlist_del_init_rcu(&geneve->hlist6.hlist); 723 #endif 724 geneve_sock_release(geneve); 725 return 0; 726 } 727 728 static void geneve_build_header(struct genevehdr *geneveh, 729 const struct ip_tunnel_info *info) 730 { 731 geneveh->ver = GENEVE_VER; 732 geneveh->opt_len = info->options_len / 4; 733 geneveh->oam = !!(info->key.tun_flags & TUNNEL_OAM); 734 geneveh->critical = !!(info->key.tun_flags & TUNNEL_CRIT_OPT); 735 geneveh->rsvd1 = 0; 736 tunnel_id_to_vni(info->key.tun_id, geneveh->vni); 737 geneveh->proto_type = htons(ETH_P_TEB); 738 geneveh->rsvd2 = 0; 739 740 if (info->key.tun_flags & TUNNEL_GENEVE_OPT) 741 ip_tunnel_info_opts_get(geneveh->options, info); 742 } 743 744 static int geneve_build_skb(struct dst_entry *dst, struct sk_buff *skb, 745 const struct ip_tunnel_info *info, 746 bool xnet, int ip_hdr_len) 747 { 748 bool udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM); 749 struct genevehdr *gnvh; 750 int min_headroom; 751 int err; 752 753 skb_reset_mac_header(skb); 754 skb_scrub_packet(skb, xnet); 755 756 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len + 757 GENEVE_BASE_HLEN + info->options_len + ip_hdr_len; 758 err = skb_cow_head(skb, min_headroom); 759 if (unlikely(err)) 760 goto free_dst; 761 762 err = udp_tunnel_handle_offloads(skb, udp_sum); 763 if (err) 764 goto free_dst; 765 766 gnvh = __skb_push(skb, sizeof(*gnvh) + info->options_len); 767 geneve_build_header(gnvh, info); 768 skb_set_inner_protocol(skb, htons(ETH_P_TEB)); 769 return 0; 770 771 free_dst: 772 dst_release(dst); 773 return err; 774 } 775 776 static struct rtable *geneve_get_v4_rt(struct sk_buff *skb, 777 struct net_device *dev, 778 struct geneve_sock *gs4, 779 struct flowi4 *fl4, 780 const struct ip_tunnel_info *info) 781 { 782 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 783 struct geneve_dev *geneve = netdev_priv(dev); 784 struct dst_cache *dst_cache; 785 struct rtable *rt = NULL; 786 __u8 tos; 787 788 if (!gs4) 789 return ERR_PTR(-EIO); 790 791 memset(fl4, 0, sizeof(*fl4)); 792 fl4->flowi4_mark = skb->mark; 793 fl4->flowi4_proto = IPPROTO_UDP; 794 fl4->daddr = info->key.u.ipv4.dst; 795 fl4->saddr = info->key.u.ipv4.src; 796 797 tos = info->key.tos; 798 if ((tos == 1) && !geneve->cfg.collect_md) { 799 tos = ip_tunnel_get_dsfield(ip_hdr(skb), skb); 800 use_cache = false; 801 } 802 fl4->flowi4_tos = RT_TOS(tos); 803 804 dst_cache = (struct dst_cache *)&info->dst_cache; 805 if (use_cache) { 806 rt = dst_cache_get_ip4(dst_cache, &fl4->saddr); 807 if (rt) 808 return rt; 809 } 810 rt = ip_route_output_key(geneve->net, fl4); 811 if (IS_ERR(rt)) { 812 netdev_dbg(dev, "no route to %pI4\n", &fl4->daddr); 813 return ERR_PTR(-ENETUNREACH); 814 } 815 if (rt->dst.dev == dev) { /* is this necessary? */ 816 netdev_dbg(dev, "circular route to %pI4\n", &fl4->daddr); 817 ip_rt_put(rt); 818 return ERR_PTR(-ELOOP); 819 } 820 if (use_cache) 821 dst_cache_set_ip4(dst_cache, &rt->dst, fl4->saddr); 822 return rt; 823 } 824 825 #if IS_ENABLED(CONFIG_IPV6) 826 static struct dst_entry *geneve_get_v6_dst(struct sk_buff *skb, 827 struct net_device *dev, 828 struct geneve_sock *gs6, 829 struct flowi6 *fl6, 830 const struct ip_tunnel_info *info) 831 { 832 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 833 struct geneve_dev *geneve = netdev_priv(dev); 834 struct dst_entry *dst = NULL; 835 struct dst_cache *dst_cache; 836 __u8 prio; 837 838 if (!gs6) 839 return ERR_PTR(-EIO); 840 841 memset(fl6, 0, sizeof(*fl6)); 842 fl6->flowi6_mark = skb->mark; 843 fl6->flowi6_proto = IPPROTO_UDP; 844 fl6->daddr = info->key.u.ipv6.dst; 845 fl6->saddr = info->key.u.ipv6.src; 846 prio = info->key.tos; 847 if ((prio == 1) && !geneve->cfg.collect_md) { 848 prio = ip_tunnel_get_dsfield(ip_hdr(skb), skb); 849 use_cache = false; 850 } 851 852 fl6->flowlabel = ip6_make_flowinfo(RT_TOS(prio), 853 info->key.label); 854 dst_cache = (struct dst_cache *)&info->dst_cache; 855 if (use_cache) { 856 dst = dst_cache_get_ip6(dst_cache, &fl6->saddr); 857 if (dst) 858 return dst; 859 } 860 dst = ipv6_stub->ipv6_dst_lookup_flow(geneve->net, gs6->sock->sk, fl6, 861 NULL); 862 if (IS_ERR(dst)) { 863 netdev_dbg(dev, "no route to %pI6\n", &fl6->daddr); 864 return ERR_PTR(-ENETUNREACH); 865 } 866 if (dst->dev == dev) { /* is this necessary? */ 867 netdev_dbg(dev, "circular route to %pI6\n", &fl6->daddr); 868 dst_release(dst); 869 return ERR_PTR(-ELOOP); 870 } 871 872 if (use_cache) 873 dst_cache_set_ip6(dst_cache, dst, &fl6->saddr); 874 return dst; 875 } 876 #endif 877 878 static int geneve_xmit_skb(struct sk_buff *skb, struct net_device *dev, 879 struct geneve_dev *geneve, 880 const struct ip_tunnel_info *info) 881 { 882 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 883 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 884 const struct ip_tunnel_key *key = &info->key; 885 struct rtable *rt; 886 struct flowi4 fl4; 887 __u8 tos, ttl; 888 __be16 df = 0; 889 __be16 sport; 890 int err; 891 892 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info); 893 if (IS_ERR(rt)) 894 return PTR_ERR(rt); 895 896 skb_tunnel_check_pmtu(skb, &rt->dst, 897 GENEVE_IPV4_HLEN + info->options_len); 898 899 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 900 if (geneve->cfg.collect_md) { 901 tos = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 902 ttl = key->ttl; 903 904 df = key->tun_flags & TUNNEL_DONT_FRAGMENT ? htons(IP_DF) : 0; 905 } else { 906 tos = ip_tunnel_ecn_encap(fl4.flowi4_tos, ip_hdr(skb), skb); 907 if (geneve->cfg.ttl_inherit) 908 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb); 909 else 910 ttl = key->ttl; 911 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst); 912 913 if (geneve->cfg.df == GENEVE_DF_SET) { 914 df = htons(IP_DF); 915 } else if (geneve->cfg.df == GENEVE_DF_INHERIT) { 916 struct ethhdr *eth = eth_hdr(skb); 917 918 if (ntohs(eth->h_proto) == ETH_P_IPV6) { 919 df = htons(IP_DF); 920 } else if (ntohs(eth->h_proto) == ETH_P_IP) { 921 struct iphdr *iph = ip_hdr(skb); 922 923 if (iph->frag_off & htons(IP_DF)) 924 df = htons(IP_DF); 925 } 926 } 927 } 928 929 err = geneve_build_skb(&rt->dst, skb, info, xnet, sizeof(struct iphdr)); 930 if (unlikely(err)) 931 return err; 932 933 udp_tunnel_xmit_skb(rt, gs4->sock->sk, skb, fl4.saddr, fl4.daddr, 934 tos, ttl, df, sport, geneve->cfg.info.key.tp_dst, 935 !net_eq(geneve->net, dev_net(geneve->dev)), 936 !(info->key.tun_flags & TUNNEL_CSUM)); 937 return 0; 938 } 939 940 #if IS_ENABLED(CONFIG_IPV6) 941 static int geneve6_xmit_skb(struct sk_buff *skb, struct net_device *dev, 942 struct geneve_dev *geneve, 943 const struct ip_tunnel_info *info) 944 { 945 bool xnet = !net_eq(geneve->net, dev_net(geneve->dev)); 946 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 947 const struct ip_tunnel_key *key = &info->key; 948 struct dst_entry *dst = NULL; 949 struct flowi6 fl6; 950 __u8 prio, ttl; 951 __be16 sport; 952 int err; 953 954 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info); 955 if (IS_ERR(dst)) 956 return PTR_ERR(dst); 957 958 skb_tunnel_check_pmtu(skb, dst, GENEVE_IPV6_HLEN + info->options_len); 959 960 sport = udp_flow_src_port(geneve->net, skb, 1, USHRT_MAX, true); 961 if (geneve->cfg.collect_md) { 962 prio = ip_tunnel_ecn_encap(key->tos, ip_hdr(skb), skb); 963 ttl = key->ttl; 964 } else { 965 prio = ip_tunnel_ecn_encap(ip6_tclass(fl6.flowlabel), 966 ip_hdr(skb), skb); 967 if (geneve->cfg.ttl_inherit) 968 ttl = ip_tunnel_get_ttl(ip_hdr(skb), skb); 969 else 970 ttl = key->ttl; 971 ttl = ttl ? : ip6_dst_hoplimit(dst); 972 } 973 err = geneve_build_skb(dst, skb, info, xnet, sizeof(struct ipv6hdr)); 974 if (unlikely(err)) 975 return err; 976 977 udp_tunnel6_xmit_skb(dst, gs6->sock->sk, skb, dev, 978 &fl6.saddr, &fl6.daddr, prio, ttl, 979 info->key.label, sport, geneve->cfg.info.key.tp_dst, 980 !(info->key.tun_flags & TUNNEL_CSUM)); 981 return 0; 982 } 983 #endif 984 985 static netdev_tx_t geneve_xmit(struct sk_buff *skb, struct net_device *dev) 986 { 987 struct geneve_dev *geneve = netdev_priv(dev); 988 struct ip_tunnel_info *info = NULL; 989 int err; 990 991 if (geneve->cfg.collect_md) { 992 info = skb_tunnel_info(skb); 993 if (unlikely(!info || !(info->mode & IP_TUNNEL_INFO_TX))) { 994 netdev_dbg(dev, "no tunnel metadata\n"); 995 dev_kfree_skb(skb); 996 dev->stats.tx_dropped++; 997 return NETDEV_TX_OK; 998 } 999 } else { 1000 info = &geneve->cfg.info; 1001 } 1002 1003 rcu_read_lock(); 1004 #if IS_ENABLED(CONFIG_IPV6) 1005 if (info->mode & IP_TUNNEL_INFO_IPV6) 1006 err = geneve6_xmit_skb(skb, dev, geneve, info); 1007 else 1008 #endif 1009 err = geneve_xmit_skb(skb, dev, geneve, info); 1010 rcu_read_unlock(); 1011 1012 if (likely(!err)) 1013 return NETDEV_TX_OK; 1014 1015 dev_kfree_skb(skb); 1016 1017 if (err == -ELOOP) 1018 dev->stats.collisions++; 1019 else if (err == -ENETUNREACH) 1020 dev->stats.tx_carrier_errors++; 1021 1022 dev->stats.tx_errors++; 1023 return NETDEV_TX_OK; 1024 } 1025 1026 static int geneve_change_mtu(struct net_device *dev, int new_mtu) 1027 { 1028 if (new_mtu > dev->max_mtu) 1029 new_mtu = dev->max_mtu; 1030 else if (new_mtu < dev->min_mtu) 1031 new_mtu = dev->min_mtu; 1032 1033 dev->mtu = new_mtu; 1034 return 0; 1035 } 1036 1037 static int geneve_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 1038 { 1039 struct ip_tunnel_info *info = skb_tunnel_info(skb); 1040 struct geneve_dev *geneve = netdev_priv(dev); 1041 1042 if (ip_tunnel_info_af(info) == AF_INET) { 1043 struct rtable *rt; 1044 struct flowi4 fl4; 1045 struct geneve_sock *gs4 = rcu_dereference(geneve->sock4); 1046 1047 rt = geneve_get_v4_rt(skb, dev, gs4, &fl4, info); 1048 if (IS_ERR(rt)) 1049 return PTR_ERR(rt); 1050 1051 ip_rt_put(rt); 1052 info->key.u.ipv4.src = fl4.saddr; 1053 #if IS_ENABLED(CONFIG_IPV6) 1054 } else if (ip_tunnel_info_af(info) == AF_INET6) { 1055 struct dst_entry *dst; 1056 struct flowi6 fl6; 1057 struct geneve_sock *gs6 = rcu_dereference(geneve->sock6); 1058 1059 dst = geneve_get_v6_dst(skb, dev, gs6, &fl6, info); 1060 if (IS_ERR(dst)) 1061 return PTR_ERR(dst); 1062 1063 dst_release(dst); 1064 info->key.u.ipv6.src = fl6.saddr; 1065 #endif 1066 } else { 1067 return -EINVAL; 1068 } 1069 1070 info->key.tp_src = udp_flow_src_port(geneve->net, skb, 1071 1, USHRT_MAX, true); 1072 info->key.tp_dst = geneve->cfg.info.key.tp_dst; 1073 return 0; 1074 } 1075 1076 static const struct net_device_ops geneve_netdev_ops = { 1077 .ndo_init = geneve_init, 1078 .ndo_uninit = geneve_uninit, 1079 .ndo_open = geneve_open, 1080 .ndo_stop = geneve_stop, 1081 .ndo_start_xmit = geneve_xmit, 1082 .ndo_get_stats64 = ip_tunnel_get_stats64, 1083 .ndo_change_mtu = geneve_change_mtu, 1084 .ndo_validate_addr = eth_validate_addr, 1085 .ndo_set_mac_address = eth_mac_addr, 1086 .ndo_fill_metadata_dst = geneve_fill_metadata_dst, 1087 }; 1088 1089 static void geneve_get_drvinfo(struct net_device *dev, 1090 struct ethtool_drvinfo *drvinfo) 1091 { 1092 strlcpy(drvinfo->version, GENEVE_NETDEV_VER, sizeof(drvinfo->version)); 1093 strlcpy(drvinfo->driver, "geneve", sizeof(drvinfo->driver)); 1094 } 1095 1096 static const struct ethtool_ops geneve_ethtool_ops = { 1097 .get_drvinfo = geneve_get_drvinfo, 1098 .get_link = ethtool_op_get_link, 1099 }; 1100 1101 /* Info for udev, that this is a virtual tunnel endpoint */ 1102 static struct device_type geneve_type = { 1103 .name = "geneve", 1104 }; 1105 1106 /* Calls the ndo_udp_tunnel_add of the caller in order to 1107 * supply the listening GENEVE udp ports. Callers are expected 1108 * to implement the ndo_udp_tunnel_add. 1109 */ 1110 static void geneve_offload_rx_ports(struct net_device *dev, bool push) 1111 { 1112 struct net *net = dev_net(dev); 1113 struct geneve_net *gn = net_generic(net, geneve_net_id); 1114 struct geneve_sock *gs; 1115 1116 rcu_read_lock(); 1117 list_for_each_entry_rcu(gs, &gn->sock_list, list) { 1118 if (push) { 1119 udp_tunnel_push_rx_port(dev, gs->sock, 1120 UDP_TUNNEL_TYPE_GENEVE); 1121 } else { 1122 udp_tunnel_drop_rx_port(dev, gs->sock, 1123 UDP_TUNNEL_TYPE_GENEVE); 1124 } 1125 } 1126 rcu_read_unlock(); 1127 } 1128 1129 /* Initialize the device structure. */ 1130 static void geneve_setup(struct net_device *dev) 1131 { 1132 ether_setup(dev); 1133 1134 dev->netdev_ops = &geneve_netdev_ops; 1135 dev->ethtool_ops = &geneve_ethtool_ops; 1136 dev->needs_free_netdev = true; 1137 1138 SET_NETDEV_DEVTYPE(dev, &geneve_type); 1139 1140 dev->features |= NETIF_F_LLTX; 1141 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM; 1142 dev->features |= NETIF_F_RXCSUM; 1143 dev->features |= NETIF_F_GSO_SOFTWARE; 1144 1145 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM; 1146 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 1147 1148 /* MTU range: 68 - (something less than 65535) */ 1149 dev->min_mtu = ETH_MIN_MTU; 1150 /* The max_mtu calculation does not take account of GENEVE 1151 * options, to avoid excluding potentially valid 1152 * configurations. This will be further reduced by IPvX hdr size. 1153 */ 1154 dev->max_mtu = IP_MAX_MTU - GENEVE_BASE_HLEN - dev->hard_header_len; 1155 1156 netif_keep_dst(dev); 1157 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1158 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE | IFF_NO_QUEUE; 1159 eth_hw_addr_random(dev); 1160 } 1161 1162 static const struct nla_policy geneve_policy[IFLA_GENEVE_MAX + 1] = { 1163 [IFLA_GENEVE_ID] = { .type = NLA_U32 }, 1164 [IFLA_GENEVE_REMOTE] = { .len = sizeof_field(struct iphdr, daddr) }, 1165 [IFLA_GENEVE_REMOTE6] = { .len = sizeof(struct in6_addr) }, 1166 [IFLA_GENEVE_TTL] = { .type = NLA_U8 }, 1167 [IFLA_GENEVE_TOS] = { .type = NLA_U8 }, 1168 [IFLA_GENEVE_LABEL] = { .type = NLA_U32 }, 1169 [IFLA_GENEVE_PORT] = { .type = NLA_U16 }, 1170 [IFLA_GENEVE_COLLECT_METADATA] = { .type = NLA_FLAG }, 1171 [IFLA_GENEVE_UDP_CSUM] = { .type = NLA_U8 }, 1172 [IFLA_GENEVE_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 }, 1173 [IFLA_GENEVE_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 }, 1174 [IFLA_GENEVE_TTL_INHERIT] = { .type = NLA_U8 }, 1175 [IFLA_GENEVE_DF] = { .type = NLA_U8 }, 1176 }; 1177 1178 static int geneve_validate(struct nlattr *tb[], struct nlattr *data[], 1179 struct netlink_ext_ack *extack) 1180 { 1181 if (tb[IFLA_ADDRESS]) { 1182 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 1183 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1184 "Provided link layer address is not Ethernet"); 1185 return -EINVAL; 1186 } 1187 1188 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 1189 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 1190 "Provided Ethernet address is not unicast"); 1191 return -EADDRNOTAVAIL; 1192 } 1193 } 1194 1195 if (!data) { 1196 NL_SET_ERR_MSG(extack, 1197 "Not enough attributes provided to perform the operation"); 1198 return -EINVAL; 1199 } 1200 1201 if (data[IFLA_GENEVE_ID]) { 1202 __u32 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1203 1204 if (vni >= GENEVE_N_VID) { 1205 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_ID], 1206 "Geneve ID must be lower than 16777216"); 1207 return -ERANGE; 1208 } 1209 } 1210 1211 if (data[IFLA_GENEVE_DF]) { 1212 enum ifla_geneve_df df = nla_get_u8(data[IFLA_GENEVE_DF]); 1213 1214 if (df < 0 || df > GENEVE_DF_MAX) { 1215 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_DF], 1216 "Invalid DF attribute"); 1217 return -EINVAL; 1218 } 1219 } 1220 1221 return 0; 1222 } 1223 1224 static struct geneve_dev *geneve_find_dev(struct geneve_net *gn, 1225 const struct ip_tunnel_info *info, 1226 bool *tun_on_same_port, 1227 bool *tun_collect_md) 1228 { 1229 struct geneve_dev *geneve, *t = NULL; 1230 1231 *tun_on_same_port = false; 1232 *tun_collect_md = false; 1233 list_for_each_entry(geneve, &gn->geneve_list, next) { 1234 if (info->key.tp_dst == geneve->cfg.info.key.tp_dst) { 1235 *tun_collect_md = geneve->cfg.collect_md; 1236 *tun_on_same_port = true; 1237 } 1238 if (info->key.tun_id == geneve->cfg.info.key.tun_id && 1239 info->key.tp_dst == geneve->cfg.info.key.tp_dst && 1240 !memcmp(&info->key.u, &geneve->cfg.info.key.u, sizeof(info->key.u))) 1241 t = geneve; 1242 } 1243 return t; 1244 } 1245 1246 static bool is_tnl_info_zero(const struct ip_tunnel_info *info) 1247 { 1248 return !(info->key.tun_id || info->key.tun_flags || info->key.tos || 1249 info->key.ttl || info->key.label || info->key.tp_src || 1250 memchr_inv(&info->key.u, 0, sizeof(info->key.u))); 1251 } 1252 1253 static bool geneve_dst_addr_equal(struct ip_tunnel_info *a, 1254 struct ip_tunnel_info *b) 1255 { 1256 if (ip_tunnel_info_af(a) == AF_INET) 1257 return a->key.u.ipv4.dst == b->key.u.ipv4.dst; 1258 else 1259 return ipv6_addr_equal(&a->key.u.ipv6.dst, &b->key.u.ipv6.dst); 1260 } 1261 1262 static int geneve_configure(struct net *net, struct net_device *dev, 1263 struct netlink_ext_ack *extack, 1264 const struct geneve_config *cfg) 1265 { 1266 struct geneve_net *gn = net_generic(net, geneve_net_id); 1267 struct geneve_dev *t, *geneve = netdev_priv(dev); 1268 const struct ip_tunnel_info *info = &cfg->info; 1269 bool tun_collect_md, tun_on_same_port; 1270 int err, encap_len; 1271 1272 if (cfg->collect_md && !is_tnl_info_zero(info)) { 1273 NL_SET_ERR_MSG(extack, 1274 "Device is externally controlled, so attributes (VNI, Port, and so on) must not be specified"); 1275 return -EINVAL; 1276 } 1277 1278 geneve->net = net; 1279 geneve->dev = dev; 1280 1281 t = geneve_find_dev(gn, info, &tun_on_same_port, &tun_collect_md); 1282 if (t) 1283 return -EBUSY; 1284 1285 /* make enough headroom for basic scenario */ 1286 encap_len = GENEVE_BASE_HLEN + ETH_HLEN; 1287 if (!cfg->collect_md && ip_tunnel_info_af(info) == AF_INET) { 1288 encap_len += sizeof(struct iphdr); 1289 dev->max_mtu -= sizeof(struct iphdr); 1290 } else { 1291 encap_len += sizeof(struct ipv6hdr); 1292 dev->max_mtu -= sizeof(struct ipv6hdr); 1293 } 1294 dev->needed_headroom = encap_len + ETH_HLEN; 1295 1296 if (cfg->collect_md) { 1297 if (tun_on_same_port) { 1298 NL_SET_ERR_MSG(extack, 1299 "There can be only one externally controlled device on a destination port"); 1300 return -EPERM; 1301 } 1302 } else { 1303 if (tun_collect_md) { 1304 NL_SET_ERR_MSG(extack, 1305 "There already exists an externally controlled device on this destination port"); 1306 return -EPERM; 1307 } 1308 } 1309 1310 dst_cache_reset(&geneve->cfg.info.dst_cache); 1311 memcpy(&geneve->cfg, cfg, sizeof(*cfg)); 1312 1313 err = register_netdevice(dev); 1314 if (err) 1315 return err; 1316 1317 list_add(&geneve->next, &gn->geneve_list); 1318 return 0; 1319 } 1320 1321 static void init_tnl_info(struct ip_tunnel_info *info, __u16 dst_port) 1322 { 1323 memset(info, 0, sizeof(*info)); 1324 info->key.tp_dst = htons(dst_port); 1325 } 1326 1327 static int geneve_nl2info(struct nlattr *tb[], struct nlattr *data[], 1328 struct netlink_ext_ack *extack, 1329 struct geneve_config *cfg, bool changelink) 1330 { 1331 struct ip_tunnel_info *info = &cfg->info; 1332 int attrtype; 1333 1334 if (data[IFLA_GENEVE_REMOTE] && data[IFLA_GENEVE_REMOTE6]) { 1335 NL_SET_ERR_MSG(extack, 1336 "Cannot specify both IPv4 and IPv6 Remote addresses"); 1337 return -EINVAL; 1338 } 1339 1340 if (data[IFLA_GENEVE_REMOTE]) { 1341 if (changelink && (ip_tunnel_info_af(info) == AF_INET6)) { 1342 attrtype = IFLA_GENEVE_REMOTE; 1343 goto change_notsup; 1344 } 1345 1346 info->key.u.ipv4.dst = 1347 nla_get_in_addr(data[IFLA_GENEVE_REMOTE]); 1348 1349 if (ipv4_is_multicast(info->key.u.ipv4.dst)) { 1350 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE], 1351 "Remote IPv4 address cannot be Multicast"); 1352 return -EINVAL; 1353 } 1354 } 1355 1356 if (data[IFLA_GENEVE_REMOTE6]) { 1357 #if IS_ENABLED(CONFIG_IPV6) 1358 if (changelink && (ip_tunnel_info_af(info) == AF_INET)) { 1359 attrtype = IFLA_GENEVE_REMOTE6; 1360 goto change_notsup; 1361 } 1362 1363 info->mode = IP_TUNNEL_INFO_IPV6; 1364 info->key.u.ipv6.dst = 1365 nla_get_in6_addr(data[IFLA_GENEVE_REMOTE6]); 1366 1367 if (ipv6_addr_type(&info->key.u.ipv6.dst) & 1368 IPV6_ADDR_LINKLOCAL) { 1369 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1370 "Remote IPv6 address cannot be link-local"); 1371 return -EINVAL; 1372 } 1373 if (ipv6_addr_is_multicast(&info->key.u.ipv6.dst)) { 1374 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1375 "Remote IPv6 address cannot be Multicast"); 1376 return -EINVAL; 1377 } 1378 info->key.tun_flags |= TUNNEL_CSUM; 1379 cfg->use_udp6_rx_checksums = true; 1380 #else 1381 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_REMOTE6], 1382 "IPv6 support not enabled in the kernel"); 1383 return -EPFNOSUPPORT; 1384 #endif 1385 } 1386 1387 if (data[IFLA_GENEVE_ID]) { 1388 __u32 vni; 1389 __u8 tvni[3]; 1390 __be64 tunid; 1391 1392 vni = nla_get_u32(data[IFLA_GENEVE_ID]); 1393 tvni[0] = (vni & 0x00ff0000) >> 16; 1394 tvni[1] = (vni & 0x0000ff00) >> 8; 1395 tvni[2] = vni & 0x000000ff; 1396 1397 tunid = vni_to_tunnel_id(tvni); 1398 if (changelink && (tunid != info->key.tun_id)) { 1399 attrtype = IFLA_GENEVE_ID; 1400 goto change_notsup; 1401 } 1402 info->key.tun_id = tunid; 1403 } 1404 1405 if (data[IFLA_GENEVE_TTL_INHERIT]) { 1406 if (nla_get_u8(data[IFLA_GENEVE_TTL_INHERIT])) 1407 cfg->ttl_inherit = true; 1408 else 1409 cfg->ttl_inherit = false; 1410 } else if (data[IFLA_GENEVE_TTL]) { 1411 info->key.ttl = nla_get_u8(data[IFLA_GENEVE_TTL]); 1412 cfg->ttl_inherit = false; 1413 } 1414 1415 if (data[IFLA_GENEVE_TOS]) 1416 info->key.tos = nla_get_u8(data[IFLA_GENEVE_TOS]); 1417 1418 if (data[IFLA_GENEVE_DF]) 1419 cfg->df = nla_get_u8(data[IFLA_GENEVE_DF]); 1420 1421 if (data[IFLA_GENEVE_LABEL]) { 1422 info->key.label = nla_get_be32(data[IFLA_GENEVE_LABEL]) & 1423 IPV6_FLOWLABEL_MASK; 1424 if (info->key.label && (!(info->mode & IP_TUNNEL_INFO_IPV6))) { 1425 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_LABEL], 1426 "Label attribute only applies for IPv6 Geneve devices"); 1427 return -EINVAL; 1428 } 1429 } 1430 1431 if (data[IFLA_GENEVE_PORT]) { 1432 if (changelink) { 1433 attrtype = IFLA_GENEVE_PORT; 1434 goto change_notsup; 1435 } 1436 info->key.tp_dst = nla_get_be16(data[IFLA_GENEVE_PORT]); 1437 } 1438 1439 if (data[IFLA_GENEVE_COLLECT_METADATA]) { 1440 if (changelink) { 1441 attrtype = IFLA_GENEVE_COLLECT_METADATA; 1442 goto change_notsup; 1443 } 1444 cfg->collect_md = true; 1445 } 1446 1447 if (data[IFLA_GENEVE_UDP_CSUM]) { 1448 if (changelink) { 1449 attrtype = IFLA_GENEVE_UDP_CSUM; 1450 goto change_notsup; 1451 } 1452 if (nla_get_u8(data[IFLA_GENEVE_UDP_CSUM])) 1453 info->key.tun_flags |= TUNNEL_CSUM; 1454 } 1455 1456 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX]) { 1457 #if IS_ENABLED(CONFIG_IPV6) 1458 if (changelink) { 1459 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_TX; 1460 goto change_notsup; 1461 } 1462 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX])) 1463 info->key.tun_flags &= ~TUNNEL_CSUM; 1464 #else 1465 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_TX], 1466 "IPv6 support not enabled in the kernel"); 1467 return -EPFNOSUPPORT; 1468 #endif 1469 } 1470 1471 if (data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX]) { 1472 #if IS_ENABLED(CONFIG_IPV6) 1473 if (changelink) { 1474 attrtype = IFLA_GENEVE_UDP_ZERO_CSUM6_RX; 1475 goto change_notsup; 1476 } 1477 if (nla_get_u8(data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX])) 1478 cfg->use_udp6_rx_checksums = false; 1479 #else 1480 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_GENEVE_UDP_ZERO_CSUM6_RX], 1481 "IPv6 support not enabled in the kernel"); 1482 return -EPFNOSUPPORT; 1483 #endif 1484 } 1485 1486 return 0; 1487 change_notsup: 1488 NL_SET_ERR_MSG_ATTR(extack, data[attrtype], 1489 "Changing VNI, Port, endpoint IP address family, external, and UDP checksum attributes are not supported"); 1490 return -EOPNOTSUPP; 1491 } 1492 1493 static void geneve_link_config(struct net_device *dev, 1494 struct ip_tunnel_info *info, struct nlattr *tb[]) 1495 { 1496 struct geneve_dev *geneve = netdev_priv(dev); 1497 int ldev_mtu = 0; 1498 1499 if (tb[IFLA_MTU]) { 1500 geneve_change_mtu(dev, nla_get_u32(tb[IFLA_MTU])); 1501 return; 1502 } 1503 1504 switch (ip_tunnel_info_af(info)) { 1505 case AF_INET: { 1506 struct flowi4 fl4 = { .daddr = info->key.u.ipv4.dst }; 1507 struct rtable *rt = ip_route_output_key(geneve->net, &fl4); 1508 1509 if (!IS_ERR(rt) && rt->dst.dev) { 1510 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV4_HLEN; 1511 ip_rt_put(rt); 1512 } 1513 break; 1514 } 1515 #if IS_ENABLED(CONFIG_IPV6) 1516 case AF_INET6: { 1517 struct rt6_info *rt; 1518 1519 if (!__in6_dev_get(dev)) 1520 break; 1521 1522 rt = rt6_lookup(geneve->net, &info->key.u.ipv6.dst, NULL, 0, 1523 NULL, 0); 1524 1525 if (rt && rt->dst.dev) 1526 ldev_mtu = rt->dst.dev->mtu - GENEVE_IPV6_HLEN; 1527 ip6_rt_put(rt); 1528 break; 1529 } 1530 #endif 1531 } 1532 1533 if (ldev_mtu <= 0) 1534 return; 1535 1536 geneve_change_mtu(dev, ldev_mtu - info->options_len); 1537 } 1538 1539 static int geneve_newlink(struct net *net, struct net_device *dev, 1540 struct nlattr *tb[], struct nlattr *data[], 1541 struct netlink_ext_ack *extack) 1542 { 1543 struct geneve_config cfg = { 1544 .df = GENEVE_DF_UNSET, 1545 .use_udp6_rx_checksums = false, 1546 .ttl_inherit = false, 1547 .collect_md = false, 1548 }; 1549 int err; 1550 1551 init_tnl_info(&cfg.info, GENEVE_UDP_PORT); 1552 err = geneve_nl2info(tb, data, extack, &cfg, false); 1553 if (err) 1554 return err; 1555 1556 err = geneve_configure(net, dev, extack, &cfg); 1557 if (err) 1558 return err; 1559 1560 geneve_link_config(dev, &cfg.info, tb); 1561 1562 return 0; 1563 } 1564 1565 /* Quiesces the geneve device data path for both TX and RX. 1566 * 1567 * On transmit geneve checks for non-NULL geneve_sock before it proceeds. 1568 * So, if we set that socket to NULL under RCU and wait for synchronize_net() 1569 * to complete for the existing set of in-flight packets to be transmitted, 1570 * then we would have quiesced the transmit data path. All the future packets 1571 * will get dropped until we unquiesce the data path. 1572 * 1573 * On receive geneve dereference the geneve_sock stashed in the socket. So, 1574 * if we set that to NULL under RCU and wait for synchronize_net() to 1575 * complete, then we would have quiesced the receive data path. 1576 */ 1577 static void geneve_quiesce(struct geneve_dev *geneve, struct geneve_sock **gs4, 1578 struct geneve_sock **gs6) 1579 { 1580 *gs4 = rtnl_dereference(geneve->sock4); 1581 rcu_assign_pointer(geneve->sock4, NULL); 1582 if (*gs4) 1583 rcu_assign_sk_user_data((*gs4)->sock->sk, NULL); 1584 #if IS_ENABLED(CONFIG_IPV6) 1585 *gs6 = rtnl_dereference(geneve->sock6); 1586 rcu_assign_pointer(geneve->sock6, NULL); 1587 if (*gs6) 1588 rcu_assign_sk_user_data((*gs6)->sock->sk, NULL); 1589 #else 1590 *gs6 = NULL; 1591 #endif 1592 synchronize_net(); 1593 } 1594 1595 /* Resumes the geneve device data path for both TX and RX. */ 1596 static void geneve_unquiesce(struct geneve_dev *geneve, struct geneve_sock *gs4, 1597 struct geneve_sock __maybe_unused *gs6) 1598 { 1599 rcu_assign_pointer(geneve->sock4, gs4); 1600 if (gs4) 1601 rcu_assign_sk_user_data(gs4->sock->sk, gs4); 1602 #if IS_ENABLED(CONFIG_IPV6) 1603 rcu_assign_pointer(geneve->sock6, gs6); 1604 if (gs6) 1605 rcu_assign_sk_user_data(gs6->sock->sk, gs6); 1606 #endif 1607 synchronize_net(); 1608 } 1609 1610 static int geneve_changelink(struct net_device *dev, struct nlattr *tb[], 1611 struct nlattr *data[], 1612 struct netlink_ext_ack *extack) 1613 { 1614 struct geneve_dev *geneve = netdev_priv(dev); 1615 struct geneve_sock *gs4, *gs6; 1616 struct geneve_config cfg; 1617 int err; 1618 1619 /* If the geneve device is configured for metadata (or externally 1620 * controlled, for example, OVS), then nothing can be changed. 1621 */ 1622 if (geneve->cfg.collect_md) 1623 return -EOPNOTSUPP; 1624 1625 /* Start with the existing info. */ 1626 memcpy(&cfg, &geneve->cfg, sizeof(cfg)); 1627 err = geneve_nl2info(tb, data, extack, &cfg, true); 1628 if (err) 1629 return err; 1630 1631 if (!geneve_dst_addr_equal(&geneve->cfg.info, &cfg.info)) { 1632 dst_cache_reset(&cfg.info.dst_cache); 1633 geneve_link_config(dev, &cfg.info, tb); 1634 } 1635 1636 geneve_quiesce(geneve, &gs4, &gs6); 1637 memcpy(&geneve->cfg, &cfg, sizeof(cfg)); 1638 geneve_unquiesce(geneve, gs4, gs6); 1639 1640 return 0; 1641 } 1642 1643 static void geneve_dellink(struct net_device *dev, struct list_head *head) 1644 { 1645 struct geneve_dev *geneve = netdev_priv(dev); 1646 1647 list_del(&geneve->next); 1648 unregister_netdevice_queue(dev, head); 1649 } 1650 1651 static size_t geneve_get_size(const struct net_device *dev) 1652 { 1653 return nla_total_size(sizeof(__u32)) + /* IFLA_GENEVE_ID */ 1654 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_GENEVE_REMOTE{6} */ 1655 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL */ 1656 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TOS */ 1657 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_DF */ 1658 nla_total_size(sizeof(__be32)) + /* IFLA_GENEVE_LABEL */ 1659 nla_total_size(sizeof(__be16)) + /* IFLA_GENEVE_PORT */ 1660 nla_total_size(0) + /* IFLA_GENEVE_COLLECT_METADATA */ 1661 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_CSUM */ 1662 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_TX */ 1663 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_UDP_ZERO_CSUM6_RX */ 1664 nla_total_size(sizeof(__u8)) + /* IFLA_GENEVE_TTL_INHERIT */ 1665 0; 1666 } 1667 1668 static int geneve_fill_info(struct sk_buff *skb, const struct net_device *dev) 1669 { 1670 struct geneve_dev *geneve = netdev_priv(dev); 1671 struct ip_tunnel_info *info = &geneve->cfg.info; 1672 bool ttl_inherit = geneve->cfg.ttl_inherit; 1673 bool metadata = geneve->cfg.collect_md; 1674 __u8 tmp_vni[3]; 1675 __u32 vni; 1676 1677 tunnel_id_to_vni(info->key.tun_id, tmp_vni); 1678 vni = (tmp_vni[0] << 16) | (tmp_vni[1] << 8) | tmp_vni[2]; 1679 if (nla_put_u32(skb, IFLA_GENEVE_ID, vni)) 1680 goto nla_put_failure; 1681 1682 if (!metadata && ip_tunnel_info_af(info) == AF_INET) { 1683 if (nla_put_in_addr(skb, IFLA_GENEVE_REMOTE, 1684 info->key.u.ipv4.dst)) 1685 goto nla_put_failure; 1686 if (nla_put_u8(skb, IFLA_GENEVE_UDP_CSUM, 1687 !!(info->key.tun_flags & TUNNEL_CSUM))) 1688 goto nla_put_failure; 1689 1690 #if IS_ENABLED(CONFIG_IPV6) 1691 } else if (!metadata) { 1692 if (nla_put_in6_addr(skb, IFLA_GENEVE_REMOTE6, 1693 &info->key.u.ipv6.dst)) 1694 goto nla_put_failure; 1695 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_TX, 1696 !(info->key.tun_flags & TUNNEL_CSUM))) 1697 goto nla_put_failure; 1698 #endif 1699 } 1700 1701 if (nla_put_u8(skb, IFLA_GENEVE_TTL, info->key.ttl) || 1702 nla_put_u8(skb, IFLA_GENEVE_TOS, info->key.tos) || 1703 nla_put_be32(skb, IFLA_GENEVE_LABEL, info->key.label)) 1704 goto nla_put_failure; 1705 1706 if (nla_put_u8(skb, IFLA_GENEVE_DF, geneve->cfg.df)) 1707 goto nla_put_failure; 1708 1709 if (nla_put_be16(skb, IFLA_GENEVE_PORT, info->key.tp_dst)) 1710 goto nla_put_failure; 1711 1712 if (metadata && nla_put_flag(skb, IFLA_GENEVE_COLLECT_METADATA)) 1713 goto nla_put_failure; 1714 1715 #if IS_ENABLED(CONFIG_IPV6) 1716 if (nla_put_u8(skb, IFLA_GENEVE_UDP_ZERO_CSUM6_RX, 1717 !geneve->cfg.use_udp6_rx_checksums)) 1718 goto nla_put_failure; 1719 #endif 1720 1721 if (nla_put_u8(skb, IFLA_GENEVE_TTL_INHERIT, ttl_inherit)) 1722 goto nla_put_failure; 1723 1724 return 0; 1725 1726 nla_put_failure: 1727 return -EMSGSIZE; 1728 } 1729 1730 static struct rtnl_link_ops geneve_link_ops __read_mostly = { 1731 .kind = "geneve", 1732 .maxtype = IFLA_GENEVE_MAX, 1733 .policy = geneve_policy, 1734 .priv_size = sizeof(struct geneve_dev), 1735 .setup = geneve_setup, 1736 .validate = geneve_validate, 1737 .newlink = geneve_newlink, 1738 .changelink = geneve_changelink, 1739 .dellink = geneve_dellink, 1740 .get_size = geneve_get_size, 1741 .fill_info = geneve_fill_info, 1742 }; 1743 1744 struct net_device *geneve_dev_create_fb(struct net *net, const char *name, 1745 u8 name_assign_type, u16 dst_port) 1746 { 1747 struct nlattr *tb[IFLA_MAX + 1]; 1748 struct net_device *dev; 1749 LIST_HEAD(list_kill); 1750 int err; 1751 struct geneve_config cfg = { 1752 .df = GENEVE_DF_UNSET, 1753 .use_udp6_rx_checksums = true, 1754 .ttl_inherit = false, 1755 .collect_md = true, 1756 }; 1757 1758 memset(tb, 0, sizeof(tb)); 1759 dev = rtnl_create_link(net, name, name_assign_type, 1760 &geneve_link_ops, tb, NULL); 1761 if (IS_ERR(dev)) 1762 return dev; 1763 1764 init_tnl_info(&cfg.info, dst_port); 1765 err = geneve_configure(net, dev, NULL, &cfg); 1766 if (err) { 1767 free_netdev(dev); 1768 return ERR_PTR(err); 1769 } 1770 1771 /* openvswitch users expect packet sizes to be unrestricted, 1772 * so set the largest MTU we can. 1773 */ 1774 err = geneve_change_mtu(dev, IP_MAX_MTU); 1775 if (err) 1776 goto err; 1777 1778 err = rtnl_configure_link(dev, NULL); 1779 if (err < 0) 1780 goto err; 1781 1782 return dev; 1783 err: 1784 geneve_dellink(dev, &list_kill); 1785 unregister_netdevice_many(&list_kill); 1786 return ERR_PTR(err); 1787 } 1788 EXPORT_SYMBOL_GPL(geneve_dev_create_fb); 1789 1790 static int geneve_netdevice_event(struct notifier_block *unused, 1791 unsigned long event, void *ptr) 1792 { 1793 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 1794 1795 if (event == NETDEV_UDP_TUNNEL_PUSH_INFO || 1796 event == NETDEV_UDP_TUNNEL_DROP_INFO) { 1797 geneve_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO); 1798 } else if (event == NETDEV_UNREGISTER) { 1799 if (!dev->udp_tunnel_nic_info) 1800 geneve_offload_rx_ports(dev, false); 1801 } else if (event == NETDEV_REGISTER) { 1802 if (!dev->udp_tunnel_nic_info) 1803 geneve_offload_rx_ports(dev, true); 1804 } 1805 1806 return NOTIFY_DONE; 1807 } 1808 1809 static struct notifier_block geneve_notifier_block __read_mostly = { 1810 .notifier_call = geneve_netdevice_event, 1811 }; 1812 1813 static __net_init int geneve_init_net(struct net *net) 1814 { 1815 struct geneve_net *gn = net_generic(net, geneve_net_id); 1816 1817 INIT_LIST_HEAD(&gn->geneve_list); 1818 INIT_LIST_HEAD(&gn->sock_list); 1819 return 0; 1820 } 1821 1822 static void geneve_destroy_tunnels(struct net *net, struct list_head *head) 1823 { 1824 struct geneve_net *gn = net_generic(net, geneve_net_id); 1825 struct geneve_dev *geneve, *next; 1826 struct net_device *dev, *aux; 1827 1828 /* gather any geneve devices that were moved into this ns */ 1829 for_each_netdev_safe(net, dev, aux) 1830 if (dev->rtnl_link_ops == &geneve_link_ops) 1831 unregister_netdevice_queue(dev, head); 1832 1833 /* now gather any other geneve devices that were created in this ns */ 1834 list_for_each_entry_safe(geneve, next, &gn->geneve_list, next) { 1835 /* If geneve->dev is in the same netns, it was already added 1836 * to the list by the previous loop. 1837 */ 1838 if (!net_eq(dev_net(geneve->dev), net)) 1839 unregister_netdevice_queue(geneve->dev, head); 1840 } 1841 } 1842 1843 static void __net_exit geneve_exit_batch_net(struct list_head *net_list) 1844 { 1845 struct net *net; 1846 LIST_HEAD(list); 1847 1848 rtnl_lock(); 1849 list_for_each_entry(net, net_list, exit_list) 1850 geneve_destroy_tunnels(net, &list); 1851 1852 /* unregister the devices gathered above */ 1853 unregister_netdevice_many(&list); 1854 rtnl_unlock(); 1855 1856 list_for_each_entry(net, net_list, exit_list) { 1857 const struct geneve_net *gn = net_generic(net, geneve_net_id); 1858 1859 WARN_ON_ONCE(!list_empty(&gn->sock_list)); 1860 } 1861 } 1862 1863 static struct pernet_operations geneve_net_ops = { 1864 .init = geneve_init_net, 1865 .exit_batch = geneve_exit_batch_net, 1866 .id = &geneve_net_id, 1867 .size = sizeof(struct geneve_net), 1868 }; 1869 1870 static int __init geneve_init_module(void) 1871 { 1872 int rc; 1873 1874 rc = register_pernet_subsys(&geneve_net_ops); 1875 if (rc) 1876 goto out1; 1877 1878 rc = register_netdevice_notifier(&geneve_notifier_block); 1879 if (rc) 1880 goto out2; 1881 1882 rc = rtnl_link_register(&geneve_link_ops); 1883 if (rc) 1884 goto out3; 1885 1886 return 0; 1887 out3: 1888 unregister_netdevice_notifier(&geneve_notifier_block); 1889 out2: 1890 unregister_pernet_subsys(&geneve_net_ops); 1891 out1: 1892 return rc; 1893 } 1894 late_initcall(geneve_init_module); 1895 1896 static void __exit geneve_cleanup_module(void) 1897 { 1898 rtnl_link_unregister(&geneve_link_ops); 1899 unregister_netdevice_notifier(&geneve_notifier_block); 1900 unregister_pernet_subsys(&geneve_net_ops); 1901 } 1902 module_exit(geneve_cleanup_module); 1903 1904 MODULE_LICENSE("GPL"); 1905 MODULE_VERSION(GENEVE_NETDEV_VER); 1906 MODULE_AUTHOR("John W. Linville <linville@tuxdriver.com>"); 1907 MODULE_DESCRIPTION("Interface driver for GENEVE encapsulated traffic"); 1908 MODULE_ALIAS_RTNL_LINK("geneve"); 1909