1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * VXLAN: Virtual eXtensible Local Area Network 4 * 5 * Copyright (c) 2012-2013 Vyatta 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/errno.h> 13 #include <linux/slab.h> 14 #include <linux/udp.h> 15 #include <linux/igmp.h> 16 #include <linux/if_ether.h> 17 #include <linux/ethtool.h> 18 #include <net/arp.h> 19 #include <net/ndisc.h> 20 #include <net/gro.h> 21 #include <net/ipv6_stubs.h> 22 #include <net/ip.h> 23 #include <net/icmp.h> 24 #include <net/rtnetlink.h> 25 #include <net/inet_ecn.h> 26 #include <net/net_namespace.h> 27 #include <net/netns/generic.h> 28 #include <net/tun_proto.h> 29 #include <net/vxlan.h> 30 #include <net/nexthop.h> 31 32 #if IS_ENABLED(CONFIG_IPV6) 33 #include <net/ip6_tunnel.h> 34 #include <net/ip6_checksum.h> 35 #endif 36 37 #include "vxlan_private.h" 38 39 #define VXLAN_VERSION "0.1" 40 41 #define FDB_AGE_DEFAULT 300 /* 5 min */ 42 #define FDB_AGE_INTERVAL (10 * HZ) /* rescan interval */ 43 44 /* UDP port for VXLAN traffic. 45 * The IANA assigned port is 4789, but the Linux default is 8472 46 * for compatibility with early adopters. 47 */ 48 static unsigned short vxlan_port __read_mostly = 8472; 49 module_param_named(udp_port, vxlan_port, ushort, 0444); 50 MODULE_PARM_DESC(udp_port, "Destination UDP port"); 51 52 static bool log_ecn_error = true; 53 module_param(log_ecn_error, bool, 0644); 54 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN"); 55 56 unsigned int vxlan_net_id; 57 58 const u8 all_zeros_mac[ETH_ALEN + 2]; 59 static struct rtnl_link_ops vxlan_link_ops; 60 61 static int vxlan_sock_add(struct vxlan_dev *vxlan); 62 63 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan); 64 65 /* salt for hash table */ 66 static u32 vxlan_salt __read_mostly; 67 68 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs) 69 { 70 return vs->flags & VXLAN_F_COLLECT_METADATA || 71 ip_tunnel_collect_metadata(); 72 } 73 74 #if IS_ENABLED(CONFIG_IPV6) 75 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla) 76 { 77 if (nla_len(nla) >= sizeof(struct in6_addr)) { 78 ip->sin6.sin6_addr = nla_get_in6_addr(nla); 79 ip->sa.sa_family = AF_INET6; 80 return 0; 81 } else if (nla_len(nla) >= sizeof(__be32)) { 82 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla); 83 ip->sa.sa_family = AF_INET; 84 return 0; 85 } else { 86 return -EAFNOSUPPORT; 87 } 88 } 89 90 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr, 91 const union vxlan_addr *ip) 92 { 93 if (ip->sa.sa_family == AF_INET6) 94 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr); 95 else 96 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr); 97 } 98 99 #else /* !CONFIG_IPV6 */ 100 101 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla) 102 { 103 if (nla_len(nla) >= sizeof(struct in6_addr)) { 104 return -EAFNOSUPPORT; 105 } else if (nla_len(nla) >= sizeof(__be32)) { 106 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla); 107 ip->sa.sa_family = AF_INET; 108 return 0; 109 } else { 110 return -EAFNOSUPPORT; 111 } 112 } 113 114 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr, 115 const union vxlan_addr *ip) 116 { 117 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr); 118 } 119 #endif 120 121 /* Find VXLAN socket based on network namespace, address family, UDP port, 122 * enabled unshareable flags and socket device binding (see l3mdev with 123 * non-default VRF). 124 */ 125 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family, 126 __be16 port, u32 flags, int ifindex) 127 { 128 struct vxlan_sock *vs; 129 130 flags &= VXLAN_F_RCV_FLAGS; 131 132 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) { 133 if (inet_sk(vs->sock->sk)->inet_sport == port && 134 vxlan_get_sk_family(vs) == family && 135 vs->flags == flags && 136 vs->sock->sk->sk_bound_dev_if == ifindex) 137 return vs; 138 } 139 return NULL; 140 } 141 142 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, 143 int ifindex, __be32 vni, 144 struct vxlan_vni_node **vninode) 145 { 146 struct vxlan_vni_node *vnode; 147 struct vxlan_dev_node *node; 148 149 /* For flow based devices, map all packets to VNI 0 */ 150 if (vs->flags & VXLAN_F_COLLECT_METADATA && 151 !(vs->flags & VXLAN_F_VNIFILTER)) 152 vni = 0; 153 154 hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) { 155 if (!node->vxlan) 156 continue; 157 vnode = NULL; 158 if (node->vxlan->cfg.flags & VXLAN_F_VNIFILTER) { 159 vnode = vxlan_vnifilter_lookup(node->vxlan, vni); 160 if (!vnode) 161 continue; 162 } else if (node->vxlan->default_dst.remote_vni != vni) { 163 continue; 164 } 165 166 if (IS_ENABLED(CONFIG_IPV6)) { 167 const struct vxlan_config *cfg = &node->vxlan->cfg; 168 169 if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) && 170 cfg->remote_ifindex != ifindex) 171 continue; 172 } 173 174 if (vninode) 175 *vninode = vnode; 176 return node->vxlan; 177 } 178 179 return NULL; 180 } 181 182 /* Look up VNI in a per net namespace table */ 183 static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex, 184 __be32 vni, sa_family_t family, 185 __be16 port, u32 flags) 186 { 187 struct vxlan_sock *vs; 188 189 vs = vxlan_find_sock(net, family, port, flags, ifindex); 190 if (!vs) 191 return NULL; 192 193 return vxlan_vs_find_vni(vs, ifindex, vni, NULL); 194 } 195 196 /* Fill in neighbour message in skbuff. */ 197 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan, 198 const struct vxlan_fdb *fdb, 199 u32 portid, u32 seq, int type, unsigned int flags, 200 const struct vxlan_rdst *rdst) 201 { 202 unsigned long now = jiffies; 203 struct nda_cacheinfo ci; 204 bool send_ip, send_eth; 205 struct nlmsghdr *nlh; 206 struct nexthop *nh; 207 struct ndmsg *ndm; 208 int nh_family; 209 u32 nh_id; 210 211 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags); 212 if (nlh == NULL) 213 return -EMSGSIZE; 214 215 ndm = nlmsg_data(nlh); 216 memset(ndm, 0, sizeof(*ndm)); 217 218 send_eth = send_ip = true; 219 220 rcu_read_lock(); 221 nh = rcu_dereference(fdb->nh); 222 if (nh) { 223 nh_family = nexthop_get_family(nh); 224 nh_id = nh->id; 225 } 226 rcu_read_unlock(); 227 228 if (type == RTM_GETNEIGH) { 229 if (rdst) { 230 send_ip = !vxlan_addr_any(&rdst->remote_ip); 231 ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET; 232 } else if (nh) { 233 ndm->ndm_family = nh_family; 234 } 235 send_eth = !is_zero_ether_addr(fdb->eth_addr); 236 } else 237 ndm->ndm_family = AF_BRIDGE; 238 ndm->ndm_state = fdb->state; 239 ndm->ndm_ifindex = vxlan->dev->ifindex; 240 ndm->ndm_flags = fdb->flags; 241 if (rdst && rdst->offloaded) 242 ndm->ndm_flags |= NTF_OFFLOADED; 243 ndm->ndm_type = RTN_UNICAST; 244 245 if (!net_eq(dev_net(vxlan->dev), vxlan->net) && 246 nla_put_s32(skb, NDA_LINK_NETNSID, 247 peernet2id(dev_net(vxlan->dev), vxlan->net))) 248 goto nla_put_failure; 249 250 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr)) 251 goto nla_put_failure; 252 if (nh) { 253 if (nla_put_u32(skb, NDA_NH_ID, nh_id)) 254 goto nla_put_failure; 255 } else if (rdst) { 256 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, 257 &rdst->remote_ip)) 258 goto nla_put_failure; 259 260 if (rdst->remote_port && 261 rdst->remote_port != vxlan->cfg.dst_port && 262 nla_put_be16(skb, NDA_PORT, rdst->remote_port)) 263 goto nla_put_failure; 264 if (rdst->remote_vni != vxlan->default_dst.remote_vni && 265 nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni))) 266 goto nla_put_failure; 267 if (rdst->remote_ifindex && 268 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex)) 269 goto nla_put_failure; 270 } 271 272 if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni && 273 nla_put_u32(skb, NDA_SRC_VNI, 274 be32_to_cpu(fdb->vni))) 275 goto nla_put_failure; 276 277 ci.ndm_used = jiffies_to_clock_t(now - fdb->used); 278 ci.ndm_confirmed = 0; 279 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated); 280 ci.ndm_refcnt = 0; 281 282 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci)) 283 goto nla_put_failure; 284 285 nlmsg_end(skb, nlh); 286 return 0; 287 288 nla_put_failure: 289 nlmsg_cancel(skb, nlh); 290 return -EMSGSIZE; 291 } 292 293 static inline size_t vxlan_nlmsg_size(void) 294 { 295 return NLMSG_ALIGN(sizeof(struct ndmsg)) 296 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */ 297 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */ 298 + nla_total_size(sizeof(__be16)) /* NDA_PORT */ 299 + nla_total_size(sizeof(__be32)) /* NDA_VNI */ 300 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */ 301 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */ 302 + nla_total_size(sizeof(struct nda_cacheinfo)); 303 } 304 305 static void __vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb, 306 struct vxlan_rdst *rd, int type) 307 { 308 struct net *net = dev_net(vxlan->dev); 309 struct sk_buff *skb; 310 int err = -ENOBUFS; 311 312 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC); 313 if (skb == NULL) 314 goto errout; 315 316 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd); 317 if (err < 0) { 318 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */ 319 WARN_ON(err == -EMSGSIZE); 320 kfree_skb(skb); 321 goto errout; 322 } 323 324 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC); 325 return; 326 errout: 327 if (err < 0) 328 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err); 329 } 330 331 static void vxlan_fdb_switchdev_notifier_info(const struct vxlan_dev *vxlan, 332 const struct vxlan_fdb *fdb, 333 const struct vxlan_rdst *rd, 334 struct netlink_ext_ack *extack, 335 struct switchdev_notifier_vxlan_fdb_info *fdb_info) 336 { 337 fdb_info->info.dev = vxlan->dev; 338 fdb_info->info.extack = extack; 339 fdb_info->remote_ip = rd->remote_ip; 340 fdb_info->remote_port = rd->remote_port; 341 fdb_info->remote_vni = rd->remote_vni; 342 fdb_info->remote_ifindex = rd->remote_ifindex; 343 memcpy(fdb_info->eth_addr, fdb->eth_addr, ETH_ALEN); 344 fdb_info->vni = fdb->vni; 345 fdb_info->offloaded = rd->offloaded; 346 fdb_info->added_by_user = fdb->flags & NTF_VXLAN_ADDED_BY_USER; 347 } 348 349 static int vxlan_fdb_switchdev_call_notifiers(struct vxlan_dev *vxlan, 350 struct vxlan_fdb *fdb, 351 struct vxlan_rdst *rd, 352 bool adding, 353 struct netlink_ext_ack *extack) 354 { 355 struct switchdev_notifier_vxlan_fdb_info info; 356 enum switchdev_notifier_type notifier_type; 357 int ret; 358 359 if (WARN_ON(!rd)) 360 return 0; 361 362 notifier_type = adding ? SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE 363 : SWITCHDEV_VXLAN_FDB_DEL_TO_DEVICE; 364 vxlan_fdb_switchdev_notifier_info(vxlan, fdb, rd, NULL, &info); 365 ret = call_switchdev_notifiers(notifier_type, vxlan->dev, 366 &info.info, extack); 367 return notifier_to_errno(ret); 368 } 369 370 static int vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb, 371 struct vxlan_rdst *rd, int type, bool swdev_notify, 372 struct netlink_ext_ack *extack) 373 { 374 int err; 375 376 if (swdev_notify && rd) { 377 switch (type) { 378 case RTM_NEWNEIGH: 379 err = vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd, 380 true, extack); 381 if (err) 382 return err; 383 break; 384 case RTM_DELNEIGH: 385 vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd, 386 false, extack); 387 break; 388 } 389 } 390 391 __vxlan_fdb_notify(vxlan, fdb, rd, type); 392 return 0; 393 } 394 395 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa) 396 { 397 struct vxlan_dev *vxlan = netdev_priv(dev); 398 struct vxlan_fdb f = { 399 .state = NUD_STALE, 400 }; 401 struct vxlan_rdst remote = { 402 .remote_ip = *ipa, /* goes to NDA_DST */ 403 .remote_vni = cpu_to_be32(VXLAN_N_VID), 404 }; 405 406 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL); 407 } 408 409 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN]) 410 { 411 struct vxlan_fdb f = { 412 .state = NUD_STALE, 413 }; 414 struct vxlan_rdst remote = { }; 415 416 memcpy(f.eth_addr, eth_addr, ETH_ALEN); 417 418 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL); 419 } 420 421 /* Hash Ethernet address */ 422 static u32 eth_hash(const unsigned char *addr) 423 { 424 u64 value = get_unaligned((u64 *)addr); 425 426 /* only want 6 bytes */ 427 #ifdef __BIG_ENDIAN 428 value >>= 16; 429 #else 430 value <<= 16; 431 #endif 432 return hash_64(value, FDB_HASH_BITS); 433 } 434 435 u32 eth_vni_hash(const unsigned char *addr, __be32 vni) 436 { 437 /* use 1 byte of OUI and 3 bytes of NIC */ 438 u32 key = get_unaligned((u32 *)(addr + 2)); 439 440 return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1); 441 } 442 443 u32 fdb_head_index(struct vxlan_dev *vxlan, const u8 *mac, __be32 vni) 444 { 445 if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) 446 return eth_vni_hash(mac, vni); 447 else 448 return eth_hash(mac); 449 } 450 451 /* Hash chain to use given mac address */ 452 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan, 453 const u8 *mac, __be32 vni) 454 { 455 return &vxlan->fdb_head[fdb_head_index(vxlan, mac, vni)]; 456 } 457 458 /* Look up Ethernet address in forwarding table */ 459 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan, 460 const u8 *mac, __be32 vni) 461 { 462 struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni); 463 struct vxlan_fdb *f; 464 465 hlist_for_each_entry_rcu(f, head, hlist) { 466 if (ether_addr_equal(mac, f->eth_addr)) { 467 if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) { 468 if (vni == f->vni) 469 return f; 470 } else { 471 return f; 472 } 473 } 474 } 475 476 return NULL; 477 } 478 479 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan, 480 const u8 *mac, __be32 vni) 481 { 482 struct vxlan_fdb *f; 483 484 f = __vxlan_find_mac(vxlan, mac, vni); 485 if (f && f->used != jiffies) 486 f->used = jiffies; 487 488 return f; 489 } 490 491 /* caller should hold vxlan->hash_lock */ 492 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f, 493 union vxlan_addr *ip, __be16 port, 494 __be32 vni, __u32 ifindex) 495 { 496 struct vxlan_rdst *rd; 497 498 list_for_each_entry(rd, &f->remotes, list) { 499 if (vxlan_addr_equal(&rd->remote_ip, ip) && 500 rd->remote_port == port && 501 rd->remote_vni == vni && 502 rd->remote_ifindex == ifindex) 503 return rd; 504 } 505 506 return NULL; 507 } 508 509 int vxlan_fdb_find_uc(struct net_device *dev, const u8 *mac, __be32 vni, 510 struct switchdev_notifier_vxlan_fdb_info *fdb_info) 511 { 512 struct vxlan_dev *vxlan = netdev_priv(dev); 513 u8 eth_addr[ETH_ALEN + 2] = { 0 }; 514 struct vxlan_rdst *rdst; 515 struct vxlan_fdb *f; 516 int rc = 0; 517 518 if (is_multicast_ether_addr(mac) || 519 is_zero_ether_addr(mac)) 520 return -EINVAL; 521 522 ether_addr_copy(eth_addr, mac); 523 524 rcu_read_lock(); 525 526 f = __vxlan_find_mac(vxlan, eth_addr, vni); 527 if (!f) { 528 rc = -ENOENT; 529 goto out; 530 } 531 532 rdst = first_remote_rcu(f); 533 vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, NULL, fdb_info); 534 535 out: 536 rcu_read_unlock(); 537 return rc; 538 } 539 EXPORT_SYMBOL_GPL(vxlan_fdb_find_uc); 540 541 static int vxlan_fdb_notify_one(struct notifier_block *nb, 542 const struct vxlan_dev *vxlan, 543 const struct vxlan_fdb *f, 544 const struct vxlan_rdst *rdst, 545 struct netlink_ext_ack *extack) 546 { 547 struct switchdev_notifier_vxlan_fdb_info fdb_info; 548 int rc; 549 550 vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, extack, &fdb_info); 551 rc = nb->notifier_call(nb, SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE, 552 &fdb_info); 553 return notifier_to_errno(rc); 554 } 555 556 int vxlan_fdb_replay(const struct net_device *dev, __be32 vni, 557 struct notifier_block *nb, 558 struct netlink_ext_ack *extack) 559 { 560 struct vxlan_dev *vxlan; 561 struct vxlan_rdst *rdst; 562 struct vxlan_fdb *f; 563 unsigned int h; 564 int rc = 0; 565 566 if (!netif_is_vxlan(dev)) 567 return -EINVAL; 568 vxlan = netdev_priv(dev); 569 570 for (h = 0; h < FDB_HASH_SIZE; ++h) { 571 spin_lock_bh(&vxlan->hash_lock[h]); 572 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) { 573 if (f->vni == vni) { 574 list_for_each_entry(rdst, &f->remotes, list) { 575 rc = vxlan_fdb_notify_one(nb, vxlan, 576 f, rdst, 577 extack); 578 if (rc) 579 goto unlock; 580 } 581 } 582 } 583 spin_unlock_bh(&vxlan->hash_lock[h]); 584 } 585 return 0; 586 587 unlock: 588 spin_unlock_bh(&vxlan->hash_lock[h]); 589 return rc; 590 } 591 EXPORT_SYMBOL_GPL(vxlan_fdb_replay); 592 593 void vxlan_fdb_clear_offload(const struct net_device *dev, __be32 vni) 594 { 595 struct vxlan_dev *vxlan; 596 struct vxlan_rdst *rdst; 597 struct vxlan_fdb *f; 598 unsigned int h; 599 600 if (!netif_is_vxlan(dev)) 601 return; 602 vxlan = netdev_priv(dev); 603 604 for (h = 0; h < FDB_HASH_SIZE; ++h) { 605 spin_lock_bh(&vxlan->hash_lock[h]); 606 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) 607 if (f->vni == vni) 608 list_for_each_entry(rdst, &f->remotes, list) 609 rdst->offloaded = false; 610 spin_unlock_bh(&vxlan->hash_lock[h]); 611 } 612 613 } 614 EXPORT_SYMBOL_GPL(vxlan_fdb_clear_offload); 615 616 /* Replace destination of unicast mac */ 617 static int vxlan_fdb_replace(struct vxlan_fdb *f, 618 union vxlan_addr *ip, __be16 port, __be32 vni, 619 __u32 ifindex, struct vxlan_rdst *oldrd) 620 { 621 struct vxlan_rdst *rd; 622 623 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex); 624 if (rd) 625 return 0; 626 627 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list); 628 if (!rd) 629 return 0; 630 631 *oldrd = *rd; 632 dst_cache_reset(&rd->dst_cache); 633 rd->remote_ip = *ip; 634 rd->remote_port = port; 635 rd->remote_vni = vni; 636 rd->remote_ifindex = ifindex; 637 rd->offloaded = false; 638 return 1; 639 } 640 641 /* Add/update destinations for multicast */ 642 static int vxlan_fdb_append(struct vxlan_fdb *f, 643 union vxlan_addr *ip, __be16 port, __be32 vni, 644 __u32 ifindex, struct vxlan_rdst **rdp) 645 { 646 struct vxlan_rdst *rd; 647 648 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex); 649 if (rd) 650 return 0; 651 652 rd = kmalloc(sizeof(*rd), GFP_ATOMIC); 653 if (rd == NULL) 654 return -ENOBUFS; 655 656 if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) { 657 kfree(rd); 658 return -ENOBUFS; 659 } 660 661 rd->remote_ip = *ip; 662 rd->remote_port = port; 663 rd->offloaded = false; 664 rd->remote_vni = vni; 665 rd->remote_ifindex = ifindex; 666 667 list_add_tail_rcu(&rd->list, &f->remotes); 668 669 *rdp = rd; 670 return 1; 671 } 672 673 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb, 674 unsigned int off, 675 struct vxlanhdr *vh, size_t hdrlen, 676 __be32 vni_field, 677 struct gro_remcsum *grc, 678 bool nopartial) 679 { 680 size_t start, offset; 681 682 if (skb->remcsum_offload) 683 return vh; 684 685 if (!NAPI_GRO_CB(skb)->csum_valid) 686 return NULL; 687 688 start = vxlan_rco_start(vni_field); 689 offset = start + vxlan_rco_offset(vni_field); 690 691 vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen, 692 start, offset, grc, nopartial); 693 694 skb->remcsum_offload = 1; 695 696 return vh; 697 } 698 699 static struct sk_buff *vxlan_gro_receive(struct sock *sk, 700 struct list_head *head, 701 struct sk_buff *skb) 702 { 703 struct sk_buff *pp = NULL; 704 struct sk_buff *p; 705 struct vxlanhdr *vh, *vh2; 706 unsigned int hlen, off_vx; 707 int flush = 1; 708 struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk); 709 __be32 flags; 710 struct gro_remcsum grc; 711 712 skb_gro_remcsum_init(&grc); 713 714 off_vx = skb_gro_offset(skb); 715 hlen = off_vx + sizeof(*vh); 716 vh = skb_gro_header_fast(skb, off_vx); 717 if (skb_gro_header_hard(skb, hlen)) { 718 vh = skb_gro_header_slow(skb, hlen, off_vx); 719 if (unlikely(!vh)) 720 goto out; 721 } 722 723 skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr)); 724 725 flags = vh->vx_flags; 726 727 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) { 728 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr), 729 vh->vx_vni, &grc, 730 !!(vs->flags & 731 VXLAN_F_REMCSUM_NOPARTIAL)); 732 733 if (!vh) 734 goto out; 735 } 736 737 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */ 738 739 list_for_each_entry(p, head, list) { 740 if (!NAPI_GRO_CB(p)->same_flow) 741 continue; 742 743 vh2 = (struct vxlanhdr *)(p->data + off_vx); 744 if (vh->vx_flags != vh2->vx_flags || 745 vh->vx_vni != vh2->vx_vni) { 746 NAPI_GRO_CB(p)->same_flow = 0; 747 continue; 748 } 749 } 750 751 pp = call_gro_receive(eth_gro_receive, head, skb); 752 flush = 0; 753 754 out: 755 skb_gro_flush_final_remcsum(skb, pp, flush, &grc); 756 757 return pp; 758 } 759 760 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff) 761 { 762 /* Sets 'skb->inner_mac_header' since we are always called with 763 * 'skb->encapsulation' set. 764 */ 765 return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr)); 766 } 767 768 static struct vxlan_fdb *vxlan_fdb_alloc(struct vxlan_dev *vxlan, const u8 *mac, 769 __u16 state, __be32 src_vni, 770 __u16 ndm_flags) 771 { 772 struct vxlan_fdb *f; 773 774 f = kmalloc(sizeof(*f), GFP_ATOMIC); 775 if (!f) 776 return NULL; 777 f->state = state; 778 f->flags = ndm_flags; 779 f->updated = f->used = jiffies; 780 f->vni = src_vni; 781 f->nh = NULL; 782 RCU_INIT_POINTER(f->vdev, vxlan); 783 INIT_LIST_HEAD(&f->nh_list); 784 INIT_LIST_HEAD(&f->remotes); 785 memcpy(f->eth_addr, mac, ETH_ALEN); 786 787 return f; 788 } 789 790 static void vxlan_fdb_insert(struct vxlan_dev *vxlan, const u8 *mac, 791 __be32 src_vni, struct vxlan_fdb *f) 792 { 793 ++vxlan->addrcnt; 794 hlist_add_head_rcu(&f->hlist, 795 vxlan_fdb_head(vxlan, mac, src_vni)); 796 } 797 798 static int vxlan_fdb_nh_update(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb, 799 u32 nhid, struct netlink_ext_ack *extack) 800 { 801 struct nexthop *old_nh = rtnl_dereference(fdb->nh); 802 struct nexthop *nh; 803 int err = -EINVAL; 804 805 if (old_nh && old_nh->id == nhid) 806 return 0; 807 808 nh = nexthop_find_by_id(vxlan->net, nhid); 809 if (!nh) { 810 NL_SET_ERR_MSG(extack, "Nexthop id does not exist"); 811 goto err_inval; 812 } 813 814 if (nh) { 815 if (!nexthop_get(nh)) { 816 NL_SET_ERR_MSG(extack, "Nexthop has been deleted"); 817 nh = NULL; 818 goto err_inval; 819 } 820 if (!nexthop_is_fdb(nh)) { 821 NL_SET_ERR_MSG(extack, "Nexthop is not a fdb nexthop"); 822 goto err_inval; 823 } 824 825 if (!nexthop_is_multipath(nh)) { 826 NL_SET_ERR_MSG(extack, "Nexthop is not a multipath group"); 827 goto err_inval; 828 } 829 830 /* check nexthop group family */ 831 switch (vxlan->default_dst.remote_ip.sa.sa_family) { 832 case AF_INET: 833 if (!nexthop_has_v4(nh)) { 834 err = -EAFNOSUPPORT; 835 NL_SET_ERR_MSG(extack, "Nexthop group family not supported"); 836 goto err_inval; 837 } 838 break; 839 case AF_INET6: 840 if (nexthop_has_v4(nh)) { 841 err = -EAFNOSUPPORT; 842 NL_SET_ERR_MSG(extack, "Nexthop group family not supported"); 843 goto err_inval; 844 } 845 } 846 } 847 848 if (old_nh) { 849 list_del_rcu(&fdb->nh_list); 850 nexthop_put(old_nh); 851 } 852 rcu_assign_pointer(fdb->nh, nh); 853 list_add_tail_rcu(&fdb->nh_list, &nh->fdb_list); 854 return 1; 855 856 err_inval: 857 if (nh) 858 nexthop_put(nh); 859 return err; 860 } 861 862 int vxlan_fdb_create(struct vxlan_dev *vxlan, 863 const u8 *mac, union vxlan_addr *ip, 864 __u16 state, __be16 port, __be32 src_vni, 865 __be32 vni, __u32 ifindex, __u16 ndm_flags, 866 u32 nhid, struct vxlan_fdb **fdb, 867 struct netlink_ext_ack *extack) 868 { 869 struct vxlan_rdst *rd = NULL; 870 struct vxlan_fdb *f; 871 int rc; 872 873 if (vxlan->cfg.addrmax && 874 vxlan->addrcnt >= vxlan->cfg.addrmax) 875 return -ENOSPC; 876 877 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip); 878 f = vxlan_fdb_alloc(vxlan, mac, state, src_vni, ndm_flags); 879 if (!f) 880 return -ENOMEM; 881 882 if (nhid) 883 rc = vxlan_fdb_nh_update(vxlan, f, nhid, extack); 884 else 885 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd); 886 if (rc < 0) 887 goto errout; 888 889 *fdb = f; 890 891 return 0; 892 893 errout: 894 kfree(f); 895 return rc; 896 } 897 898 static void __vxlan_fdb_free(struct vxlan_fdb *f) 899 { 900 struct vxlan_rdst *rd, *nd; 901 struct nexthop *nh; 902 903 nh = rcu_dereference_raw(f->nh); 904 if (nh) { 905 rcu_assign_pointer(f->nh, NULL); 906 rcu_assign_pointer(f->vdev, NULL); 907 nexthop_put(nh); 908 } 909 910 list_for_each_entry_safe(rd, nd, &f->remotes, list) { 911 dst_cache_destroy(&rd->dst_cache); 912 kfree(rd); 913 } 914 kfree(f); 915 } 916 917 static void vxlan_fdb_free(struct rcu_head *head) 918 { 919 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu); 920 921 __vxlan_fdb_free(f); 922 } 923 924 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f, 925 bool do_notify, bool swdev_notify) 926 { 927 struct vxlan_rdst *rd; 928 929 netdev_dbg(vxlan->dev, "delete %pM\n", f->eth_addr); 930 931 --vxlan->addrcnt; 932 if (do_notify) { 933 if (rcu_access_pointer(f->nh)) 934 vxlan_fdb_notify(vxlan, f, NULL, RTM_DELNEIGH, 935 swdev_notify, NULL); 936 else 937 list_for_each_entry(rd, &f->remotes, list) 938 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, 939 swdev_notify, NULL); 940 } 941 942 hlist_del_rcu(&f->hlist); 943 list_del_rcu(&f->nh_list); 944 call_rcu(&f->rcu, vxlan_fdb_free); 945 } 946 947 static void vxlan_dst_free(struct rcu_head *head) 948 { 949 struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu); 950 951 dst_cache_destroy(&rd->dst_cache); 952 kfree(rd); 953 } 954 955 static int vxlan_fdb_update_existing(struct vxlan_dev *vxlan, 956 union vxlan_addr *ip, 957 __u16 state, __u16 flags, 958 __be16 port, __be32 vni, 959 __u32 ifindex, __u16 ndm_flags, 960 struct vxlan_fdb *f, u32 nhid, 961 bool swdev_notify, 962 struct netlink_ext_ack *extack) 963 { 964 __u16 fdb_flags = (ndm_flags & ~NTF_USE); 965 struct vxlan_rdst *rd = NULL; 966 struct vxlan_rdst oldrd; 967 int notify = 0; 968 int rc = 0; 969 int err; 970 971 if (nhid && !rcu_access_pointer(f->nh)) { 972 NL_SET_ERR_MSG(extack, 973 "Cannot replace an existing non nexthop fdb with a nexthop"); 974 return -EOPNOTSUPP; 975 } 976 977 if (nhid && (flags & NLM_F_APPEND)) { 978 NL_SET_ERR_MSG(extack, 979 "Cannot append to a nexthop fdb"); 980 return -EOPNOTSUPP; 981 } 982 983 /* Do not allow an externally learned entry to take over an entry added 984 * by the user. 985 */ 986 if (!(fdb_flags & NTF_EXT_LEARNED) || 987 !(f->flags & NTF_VXLAN_ADDED_BY_USER)) { 988 if (f->state != state) { 989 f->state = state; 990 f->updated = jiffies; 991 notify = 1; 992 } 993 if (f->flags != fdb_flags) { 994 f->flags = fdb_flags; 995 f->updated = jiffies; 996 notify = 1; 997 } 998 } 999 1000 if ((flags & NLM_F_REPLACE)) { 1001 /* Only change unicasts */ 1002 if (!(is_multicast_ether_addr(f->eth_addr) || 1003 is_zero_ether_addr(f->eth_addr))) { 1004 if (nhid) { 1005 rc = vxlan_fdb_nh_update(vxlan, f, nhid, extack); 1006 if (rc < 0) 1007 return rc; 1008 } else { 1009 rc = vxlan_fdb_replace(f, ip, port, vni, 1010 ifindex, &oldrd); 1011 } 1012 notify |= rc; 1013 } else { 1014 NL_SET_ERR_MSG(extack, "Cannot replace non-unicast fdb entries"); 1015 return -EOPNOTSUPP; 1016 } 1017 } 1018 if ((flags & NLM_F_APPEND) && 1019 (is_multicast_ether_addr(f->eth_addr) || 1020 is_zero_ether_addr(f->eth_addr))) { 1021 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd); 1022 1023 if (rc < 0) 1024 return rc; 1025 notify |= rc; 1026 } 1027 1028 if (ndm_flags & NTF_USE) 1029 f->used = jiffies; 1030 1031 if (notify) { 1032 if (rd == NULL) 1033 rd = first_remote_rtnl(f); 1034 1035 err = vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH, 1036 swdev_notify, extack); 1037 if (err) 1038 goto err_notify; 1039 } 1040 1041 return 0; 1042 1043 err_notify: 1044 if (nhid) 1045 return err; 1046 if ((flags & NLM_F_REPLACE) && rc) 1047 *rd = oldrd; 1048 else if ((flags & NLM_F_APPEND) && rc) { 1049 list_del_rcu(&rd->list); 1050 call_rcu(&rd->rcu, vxlan_dst_free); 1051 } 1052 return err; 1053 } 1054 1055 static int vxlan_fdb_update_create(struct vxlan_dev *vxlan, 1056 const u8 *mac, union vxlan_addr *ip, 1057 __u16 state, __u16 flags, 1058 __be16 port, __be32 src_vni, __be32 vni, 1059 __u32 ifindex, __u16 ndm_flags, u32 nhid, 1060 bool swdev_notify, 1061 struct netlink_ext_ack *extack) 1062 { 1063 __u16 fdb_flags = (ndm_flags & ~NTF_USE); 1064 struct vxlan_fdb *f; 1065 int rc; 1066 1067 /* Disallow replace to add a multicast entry */ 1068 if ((flags & NLM_F_REPLACE) && 1069 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac))) 1070 return -EOPNOTSUPP; 1071 1072 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip); 1073 rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni, 1074 vni, ifindex, fdb_flags, nhid, &f, extack); 1075 if (rc < 0) 1076 return rc; 1077 1078 vxlan_fdb_insert(vxlan, mac, src_vni, f); 1079 rc = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH, 1080 swdev_notify, extack); 1081 if (rc) 1082 goto err_notify; 1083 1084 return 0; 1085 1086 err_notify: 1087 vxlan_fdb_destroy(vxlan, f, false, false); 1088 return rc; 1089 } 1090 1091 /* Add new entry to forwarding table -- assumes lock held */ 1092 int vxlan_fdb_update(struct vxlan_dev *vxlan, 1093 const u8 *mac, union vxlan_addr *ip, 1094 __u16 state, __u16 flags, 1095 __be16 port, __be32 src_vni, __be32 vni, 1096 __u32 ifindex, __u16 ndm_flags, u32 nhid, 1097 bool swdev_notify, 1098 struct netlink_ext_ack *extack) 1099 { 1100 struct vxlan_fdb *f; 1101 1102 f = __vxlan_find_mac(vxlan, mac, src_vni); 1103 if (f) { 1104 if (flags & NLM_F_EXCL) { 1105 netdev_dbg(vxlan->dev, 1106 "lost race to create %pM\n", mac); 1107 return -EEXIST; 1108 } 1109 1110 return vxlan_fdb_update_existing(vxlan, ip, state, flags, port, 1111 vni, ifindex, ndm_flags, f, 1112 nhid, swdev_notify, extack); 1113 } else { 1114 if (!(flags & NLM_F_CREATE)) 1115 return -ENOENT; 1116 1117 return vxlan_fdb_update_create(vxlan, mac, ip, state, flags, 1118 port, src_vni, vni, ifindex, 1119 ndm_flags, nhid, swdev_notify, 1120 extack); 1121 } 1122 } 1123 1124 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f, 1125 struct vxlan_rdst *rd, bool swdev_notify) 1126 { 1127 list_del_rcu(&rd->list); 1128 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, swdev_notify, NULL); 1129 call_rcu(&rd->rcu, vxlan_dst_free); 1130 } 1131 1132 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan, 1133 union vxlan_addr *ip, __be16 *port, __be32 *src_vni, 1134 __be32 *vni, u32 *ifindex, u32 *nhid) 1135 { 1136 struct net *net = dev_net(vxlan->dev); 1137 int err; 1138 1139 if (tb[NDA_NH_ID] && (tb[NDA_DST] || tb[NDA_VNI] || tb[NDA_IFINDEX] || 1140 tb[NDA_PORT])) 1141 return -EINVAL; 1142 1143 if (tb[NDA_DST]) { 1144 err = vxlan_nla_get_addr(ip, tb[NDA_DST]); 1145 if (err) 1146 return err; 1147 } else { 1148 union vxlan_addr *remote = &vxlan->default_dst.remote_ip; 1149 1150 if (remote->sa.sa_family == AF_INET) { 1151 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY); 1152 ip->sa.sa_family = AF_INET; 1153 #if IS_ENABLED(CONFIG_IPV6) 1154 } else { 1155 ip->sin6.sin6_addr = in6addr_any; 1156 ip->sa.sa_family = AF_INET6; 1157 #endif 1158 } 1159 } 1160 1161 if (tb[NDA_PORT]) { 1162 if (nla_len(tb[NDA_PORT]) != sizeof(__be16)) 1163 return -EINVAL; 1164 *port = nla_get_be16(tb[NDA_PORT]); 1165 } else { 1166 *port = vxlan->cfg.dst_port; 1167 } 1168 1169 if (tb[NDA_VNI]) { 1170 if (nla_len(tb[NDA_VNI]) != sizeof(u32)) 1171 return -EINVAL; 1172 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI])); 1173 } else { 1174 *vni = vxlan->default_dst.remote_vni; 1175 } 1176 1177 if (tb[NDA_SRC_VNI]) { 1178 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32)) 1179 return -EINVAL; 1180 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI])); 1181 } else { 1182 *src_vni = vxlan->default_dst.remote_vni; 1183 } 1184 1185 if (tb[NDA_IFINDEX]) { 1186 struct net_device *tdev; 1187 1188 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32)) 1189 return -EINVAL; 1190 *ifindex = nla_get_u32(tb[NDA_IFINDEX]); 1191 tdev = __dev_get_by_index(net, *ifindex); 1192 if (!tdev) 1193 return -EADDRNOTAVAIL; 1194 } else { 1195 *ifindex = 0; 1196 } 1197 1198 if (tb[NDA_NH_ID]) 1199 *nhid = nla_get_u32(tb[NDA_NH_ID]); 1200 else 1201 *nhid = 0; 1202 1203 return 0; 1204 } 1205 1206 /* Add static entry (via netlink) */ 1207 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[], 1208 struct net_device *dev, 1209 const unsigned char *addr, u16 vid, u16 flags, 1210 struct netlink_ext_ack *extack) 1211 { 1212 struct vxlan_dev *vxlan = netdev_priv(dev); 1213 /* struct net *net = dev_net(vxlan->dev); */ 1214 union vxlan_addr ip; 1215 __be16 port; 1216 __be32 src_vni, vni; 1217 u32 ifindex, nhid; 1218 u32 hash_index; 1219 int err; 1220 1221 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) { 1222 pr_info("RTM_NEWNEIGH with invalid state %#x\n", 1223 ndm->ndm_state); 1224 return -EINVAL; 1225 } 1226 1227 if (!tb || (!tb[NDA_DST] && !tb[NDA_NH_ID])) 1228 return -EINVAL; 1229 1230 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex, 1231 &nhid); 1232 if (err) 1233 return err; 1234 1235 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family) 1236 return -EAFNOSUPPORT; 1237 1238 hash_index = fdb_head_index(vxlan, addr, src_vni); 1239 spin_lock_bh(&vxlan->hash_lock[hash_index]); 1240 err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags, 1241 port, src_vni, vni, ifindex, 1242 ndm->ndm_flags | NTF_VXLAN_ADDED_BY_USER, 1243 nhid, true, extack); 1244 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 1245 1246 return err; 1247 } 1248 1249 int __vxlan_fdb_delete(struct vxlan_dev *vxlan, 1250 const unsigned char *addr, union vxlan_addr ip, 1251 __be16 port, __be32 src_vni, __be32 vni, 1252 u32 ifindex, bool swdev_notify) 1253 { 1254 struct vxlan_rdst *rd = NULL; 1255 struct vxlan_fdb *f; 1256 int err = -ENOENT; 1257 1258 f = vxlan_find_mac(vxlan, addr, src_vni); 1259 if (!f) 1260 return err; 1261 1262 if (!vxlan_addr_any(&ip)) { 1263 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex); 1264 if (!rd) 1265 goto out; 1266 } 1267 1268 /* remove a destination if it's not the only one on the list, 1269 * otherwise destroy the fdb entry 1270 */ 1271 if (rd && !list_is_singular(&f->remotes)) { 1272 vxlan_fdb_dst_destroy(vxlan, f, rd, swdev_notify); 1273 goto out; 1274 } 1275 1276 vxlan_fdb_destroy(vxlan, f, true, swdev_notify); 1277 1278 out: 1279 return 0; 1280 } 1281 1282 /* Delete entry (via netlink) */ 1283 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[], 1284 struct net_device *dev, 1285 const unsigned char *addr, u16 vid) 1286 { 1287 struct vxlan_dev *vxlan = netdev_priv(dev); 1288 union vxlan_addr ip; 1289 __be32 src_vni, vni; 1290 u32 ifindex, nhid; 1291 u32 hash_index; 1292 __be16 port; 1293 int err; 1294 1295 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex, 1296 &nhid); 1297 if (err) 1298 return err; 1299 1300 hash_index = fdb_head_index(vxlan, addr, src_vni); 1301 spin_lock_bh(&vxlan->hash_lock[hash_index]); 1302 err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex, 1303 true); 1304 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 1305 1306 return err; 1307 } 1308 1309 /* Dump forwarding table */ 1310 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb, 1311 struct net_device *dev, 1312 struct net_device *filter_dev, int *idx) 1313 { 1314 struct vxlan_dev *vxlan = netdev_priv(dev); 1315 unsigned int h; 1316 int err = 0; 1317 1318 for (h = 0; h < FDB_HASH_SIZE; ++h) { 1319 struct vxlan_fdb *f; 1320 1321 rcu_read_lock(); 1322 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) { 1323 struct vxlan_rdst *rd; 1324 1325 if (rcu_access_pointer(f->nh)) { 1326 if (*idx < cb->args[2]) 1327 goto skip_nh; 1328 err = vxlan_fdb_info(skb, vxlan, f, 1329 NETLINK_CB(cb->skb).portid, 1330 cb->nlh->nlmsg_seq, 1331 RTM_NEWNEIGH, 1332 NLM_F_MULTI, NULL); 1333 if (err < 0) { 1334 rcu_read_unlock(); 1335 goto out; 1336 } 1337 skip_nh: 1338 *idx += 1; 1339 continue; 1340 } 1341 1342 list_for_each_entry_rcu(rd, &f->remotes, list) { 1343 if (*idx < cb->args[2]) 1344 goto skip; 1345 1346 err = vxlan_fdb_info(skb, vxlan, f, 1347 NETLINK_CB(cb->skb).portid, 1348 cb->nlh->nlmsg_seq, 1349 RTM_NEWNEIGH, 1350 NLM_F_MULTI, rd); 1351 if (err < 0) { 1352 rcu_read_unlock(); 1353 goto out; 1354 } 1355 skip: 1356 *idx += 1; 1357 } 1358 } 1359 rcu_read_unlock(); 1360 } 1361 out: 1362 return err; 1363 } 1364 1365 static int vxlan_fdb_get(struct sk_buff *skb, 1366 struct nlattr *tb[], 1367 struct net_device *dev, 1368 const unsigned char *addr, 1369 u16 vid, u32 portid, u32 seq, 1370 struct netlink_ext_ack *extack) 1371 { 1372 struct vxlan_dev *vxlan = netdev_priv(dev); 1373 struct vxlan_fdb *f; 1374 __be32 vni; 1375 int err; 1376 1377 if (tb[NDA_VNI]) 1378 vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI])); 1379 else 1380 vni = vxlan->default_dst.remote_vni; 1381 1382 rcu_read_lock(); 1383 1384 f = __vxlan_find_mac(vxlan, addr, vni); 1385 if (!f) { 1386 NL_SET_ERR_MSG(extack, "Fdb entry not found"); 1387 err = -ENOENT; 1388 goto errout; 1389 } 1390 1391 err = vxlan_fdb_info(skb, vxlan, f, portid, seq, 1392 RTM_NEWNEIGH, 0, first_remote_rcu(f)); 1393 errout: 1394 rcu_read_unlock(); 1395 return err; 1396 } 1397 1398 /* Watch incoming packets to learn mapping between Ethernet address 1399 * and Tunnel endpoint. 1400 * Return true if packet is bogus and should be dropped. 1401 */ 1402 static bool vxlan_snoop(struct net_device *dev, 1403 union vxlan_addr *src_ip, const u8 *src_mac, 1404 u32 src_ifindex, __be32 vni) 1405 { 1406 struct vxlan_dev *vxlan = netdev_priv(dev); 1407 struct vxlan_fdb *f; 1408 u32 ifindex = 0; 1409 1410 #if IS_ENABLED(CONFIG_IPV6) 1411 if (src_ip->sa.sa_family == AF_INET6 && 1412 (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)) 1413 ifindex = src_ifindex; 1414 #endif 1415 1416 f = vxlan_find_mac(vxlan, src_mac, vni); 1417 if (likely(f)) { 1418 struct vxlan_rdst *rdst = first_remote_rcu(f); 1419 1420 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) && 1421 rdst->remote_ifindex == ifindex)) 1422 return false; 1423 1424 /* Don't migrate static entries, drop packets */ 1425 if (f->state & (NUD_PERMANENT | NUD_NOARP)) 1426 return true; 1427 1428 /* Don't override an fdb with nexthop with a learnt entry */ 1429 if (rcu_access_pointer(f->nh)) 1430 return true; 1431 1432 if (net_ratelimit()) 1433 netdev_info(dev, 1434 "%pM migrated from %pIS to %pIS\n", 1435 src_mac, &rdst->remote_ip.sa, &src_ip->sa); 1436 1437 rdst->remote_ip = *src_ip; 1438 f->updated = jiffies; 1439 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH, true, NULL); 1440 } else { 1441 u32 hash_index = fdb_head_index(vxlan, src_mac, vni); 1442 1443 /* learned new entry */ 1444 spin_lock(&vxlan->hash_lock[hash_index]); 1445 1446 /* close off race between vxlan_flush and incoming packets */ 1447 if (netif_running(dev)) 1448 vxlan_fdb_update(vxlan, src_mac, src_ip, 1449 NUD_REACHABLE, 1450 NLM_F_EXCL|NLM_F_CREATE, 1451 vxlan->cfg.dst_port, 1452 vni, 1453 vxlan->default_dst.remote_vni, 1454 ifindex, NTF_SELF, 0, true, NULL); 1455 spin_unlock(&vxlan->hash_lock[hash_index]); 1456 } 1457 1458 return false; 1459 } 1460 1461 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs) 1462 { 1463 struct vxlan_net *vn; 1464 1465 if (!vs) 1466 return false; 1467 if (!refcount_dec_and_test(&vs->refcnt)) 1468 return false; 1469 1470 vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id); 1471 spin_lock(&vn->sock_lock); 1472 hlist_del_rcu(&vs->hlist); 1473 udp_tunnel_notify_del_rx_port(vs->sock, 1474 (vs->flags & VXLAN_F_GPE) ? 1475 UDP_TUNNEL_TYPE_VXLAN_GPE : 1476 UDP_TUNNEL_TYPE_VXLAN); 1477 spin_unlock(&vn->sock_lock); 1478 1479 return true; 1480 } 1481 1482 static void vxlan_sock_release(struct vxlan_dev *vxlan) 1483 { 1484 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock); 1485 #if IS_ENABLED(CONFIG_IPV6) 1486 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock); 1487 1488 RCU_INIT_POINTER(vxlan->vn6_sock, NULL); 1489 #endif 1490 1491 RCU_INIT_POINTER(vxlan->vn4_sock, NULL); 1492 synchronize_net(); 1493 1494 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER) 1495 vxlan_vs_del_vnigrp(vxlan); 1496 else 1497 vxlan_vs_del_dev(vxlan); 1498 1499 if (__vxlan_sock_release_prep(sock4)) { 1500 udp_tunnel_sock_release(sock4->sock); 1501 kfree(sock4); 1502 } 1503 1504 #if IS_ENABLED(CONFIG_IPV6) 1505 if (__vxlan_sock_release_prep(sock6)) { 1506 udp_tunnel_sock_release(sock6->sock); 1507 kfree(sock6); 1508 } 1509 #endif 1510 } 1511 1512 static bool vxlan_remcsum(struct vxlanhdr *unparsed, 1513 struct sk_buff *skb, u32 vxflags) 1514 { 1515 size_t start, offset; 1516 1517 if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload) 1518 goto out; 1519 1520 start = vxlan_rco_start(unparsed->vx_vni); 1521 offset = start + vxlan_rco_offset(unparsed->vx_vni); 1522 1523 if (!pskb_may_pull(skb, offset + sizeof(u16))) 1524 return false; 1525 1526 skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset, 1527 !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL)); 1528 out: 1529 unparsed->vx_flags &= ~VXLAN_HF_RCO; 1530 unparsed->vx_vni &= VXLAN_VNI_MASK; 1531 return true; 1532 } 1533 1534 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed, 1535 struct sk_buff *skb, u32 vxflags, 1536 struct vxlan_metadata *md) 1537 { 1538 struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed; 1539 struct metadata_dst *tun_dst; 1540 1541 if (!(unparsed->vx_flags & VXLAN_HF_GBP)) 1542 goto out; 1543 1544 md->gbp = ntohs(gbp->policy_id); 1545 1546 tun_dst = (struct metadata_dst *)skb_dst(skb); 1547 if (tun_dst) { 1548 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT; 1549 tun_dst->u.tun_info.options_len = sizeof(*md); 1550 } 1551 if (gbp->dont_learn) 1552 md->gbp |= VXLAN_GBP_DONT_LEARN; 1553 1554 if (gbp->policy_applied) 1555 md->gbp |= VXLAN_GBP_POLICY_APPLIED; 1556 1557 /* In flow-based mode, GBP is carried in dst_metadata */ 1558 if (!(vxflags & VXLAN_F_COLLECT_METADATA)) 1559 skb->mark = md->gbp; 1560 out: 1561 unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS; 1562 } 1563 1564 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed, 1565 __be16 *protocol, 1566 struct sk_buff *skb, u32 vxflags) 1567 { 1568 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed; 1569 1570 /* Need to have Next Protocol set for interfaces in GPE mode. */ 1571 if (!gpe->np_applied) 1572 return false; 1573 /* "The initial version is 0. If a receiver does not support the 1574 * version indicated it MUST drop the packet. 1575 */ 1576 if (gpe->version != 0) 1577 return false; 1578 /* "When the O bit is set to 1, the packet is an OAM packet and OAM 1579 * processing MUST occur." However, we don't implement OAM 1580 * processing, thus drop the packet. 1581 */ 1582 if (gpe->oam_flag) 1583 return false; 1584 1585 *protocol = tun_p_to_eth_p(gpe->next_protocol); 1586 if (!*protocol) 1587 return false; 1588 1589 unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS; 1590 return true; 1591 } 1592 1593 static bool vxlan_set_mac(struct vxlan_dev *vxlan, 1594 struct vxlan_sock *vs, 1595 struct sk_buff *skb, __be32 vni) 1596 { 1597 union vxlan_addr saddr; 1598 u32 ifindex = skb->dev->ifindex; 1599 1600 skb_reset_mac_header(skb); 1601 skb->protocol = eth_type_trans(skb, vxlan->dev); 1602 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 1603 1604 /* Ignore packet loops (and multicast echo) */ 1605 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr)) 1606 return false; 1607 1608 /* Get address from the outer IP header */ 1609 if (vxlan_get_sk_family(vs) == AF_INET) { 1610 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr; 1611 saddr.sa.sa_family = AF_INET; 1612 #if IS_ENABLED(CONFIG_IPV6) 1613 } else { 1614 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr; 1615 saddr.sa.sa_family = AF_INET6; 1616 #endif 1617 } 1618 1619 if ((vxlan->cfg.flags & VXLAN_F_LEARN) && 1620 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni)) 1621 return false; 1622 1623 return true; 1624 } 1625 1626 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph, 1627 struct sk_buff *skb) 1628 { 1629 int err = 0; 1630 1631 if (vxlan_get_sk_family(vs) == AF_INET) 1632 err = IP_ECN_decapsulate(oiph, skb); 1633 #if IS_ENABLED(CONFIG_IPV6) 1634 else 1635 err = IP6_ECN_decapsulate(oiph, skb); 1636 #endif 1637 1638 if (unlikely(err) && log_ecn_error) { 1639 if (vxlan_get_sk_family(vs) == AF_INET) 1640 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n", 1641 &((struct iphdr *)oiph)->saddr, 1642 ((struct iphdr *)oiph)->tos); 1643 else 1644 net_info_ratelimited("non-ECT from %pI6\n", 1645 &((struct ipv6hdr *)oiph)->saddr); 1646 } 1647 return err <= 1; 1648 } 1649 1650 /* Callback from net/ipv4/udp.c to receive packets */ 1651 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb) 1652 { 1653 struct vxlan_vni_node *vninode = NULL; 1654 struct vxlan_dev *vxlan; 1655 struct vxlan_sock *vs; 1656 struct vxlanhdr unparsed; 1657 struct vxlan_metadata _md; 1658 struct vxlan_metadata *md = &_md; 1659 __be16 protocol = htons(ETH_P_TEB); 1660 bool raw_proto = false; 1661 void *oiph; 1662 __be32 vni = 0; 1663 1664 /* Need UDP and VXLAN header to be present */ 1665 if (!pskb_may_pull(skb, VXLAN_HLEN)) 1666 goto drop; 1667 1668 unparsed = *vxlan_hdr(skb); 1669 /* VNI flag always required to be set */ 1670 if (!(unparsed.vx_flags & VXLAN_HF_VNI)) { 1671 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n", 1672 ntohl(vxlan_hdr(skb)->vx_flags), 1673 ntohl(vxlan_hdr(skb)->vx_vni)); 1674 /* Return non vxlan pkt */ 1675 goto drop; 1676 } 1677 unparsed.vx_flags &= ~VXLAN_HF_VNI; 1678 unparsed.vx_vni &= ~VXLAN_VNI_MASK; 1679 1680 vs = rcu_dereference_sk_user_data(sk); 1681 if (!vs) 1682 goto drop; 1683 1684 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni); 1685 1686 vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni, &vninode); 1687 if (!vxlan) 1688 goto drop; 1689 1690 /* For backwards compatibility, only allow reserved fields to be 1691 * used by VXLAN extensions if explicitly requested. 1692 */ 1693 if (vs->flags & VXLAN_F_GPE) { 1694 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags)) 1695 goto drop; 1696 raw_proto = true; 1697 } 1698 1699 if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto, 1700 !net_eq(vxlan->net, dev_net(vxlan->dev)))) 1701 goto drop; 1702 1703 if (vs->flags & VXLAN_F_REMCSUM_RX) 1704 if (unlikely(!vxlan_remcsum(&unparsed, skb, vs->flags))) 1705 goto drop; 1706 1707 if (vxlan_collect_metadata(vs)) { 1708 struct metadata_dst *tun_dst; 1709 1710 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY, 1711 key32_to_tunnel_id(vni), sizeof(*md)); 1712 1713 if (!tun_dst) 1714 goto drop; 1715 1716 md = ip_tunnel_info_opts(&tun_dst->u.tun_info); 1717 1718 skb_dst_set(skb, (struct dst_entry *)tun_dst); 1719 } else { 1720 memset(md, 0, sizeof(*md)); 1721 } 1722 1723 if (vs->flags & VXLAN_F_GBP) 1724 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md); 1725 /* Note that GBP and GPE can never be active together. This is 1726 * ensured in vxlan_dev_configure. 1727 */ 1728 1729 if (unparsed.vx_flags || unparsed.vx_vni) { 1730 /* If there are any unprocessed flags remaining treat 1731 * this as a malformed packet. This behavior diverges from 1732 * VXLAN RFC (RFC7348) which stipulates that bits in reserved 1733 * in reserved fields are to be ignored. The approach here 1734 * maintains compatibility with previous stack code, and also 1735 * is more robust and provides a little more security in 1736 * adding extensions to VXLAN. 1737 */ 1738 goto drop; 1739 } 1740 1741 if (!raw_proto) { 1742 if (!vxlan_set_mac(vxlan, vs, skb, vni)) 1743 goto drop; 1744 } else { 1745 skb_reset_mac_header(skb); 1746 skb->dev = vxlan->dev; 1747 skb->pkt_type = PACKET_HOST; 1748 } 1749 1750 oiph = skb_network_header(skb); 1751 skb_reset_network_header(skb); 1752 1753 if (!vxlan_ecn_decapsulate(vs, oiph, skb)) { 1754 ++vxlan->dev->stats.rx_frame_errors; 1755 ++vxlan->dev->stats.rx_errors; 1756 vxlan_vnifilter_count(vxlan, vni, vninode, 1757 VXLAN_VNI_STATS_RX_ERRORS, 0); 1758 goto drop; 1759 } 1760 1761 rcu_read_lock(); 1762 1763 if (unlikely(!(vxlan->dev->flags & IFF_UP))) { 1764 rcu_read_unlock(); 1765 atomic_long_inc(&vxlan->dev->rx_dropped); 1766 vxlan_vnifilter_count(vxlan, vni, vninode, 1767 VXLAN_VNI_STATS_RX_DROPS, 0); 1768 goto drop; 1769 } 1770 1771 dev_sw_netstats_rx_add(vxlan->dev, skb->len); 1772 vxlan_vnifilter_count(vxlan, vni, vninode, VXLAN_VNI_STATS_RX, skb->len); 1773 gro_cells_receive(&vxlan->gro_cells, skb); 1774 1775 rcu_read_unlock(); 1776 1777 return 0; 1778 1779 drop: 1780 /* Consume bad packet */ 1781 kfree_skb(skb); 1782 return 0; 1783 } 1784 1785 /* Callback from net/ipv{4,6}/udp.c to check that we have a VNI for errors */ 1786 static int vxlan_err_lookup(struct sock *sk, struct sk_buff *skb) 1787 { 1788 struct vxlan_dev *vxlan; 1789 struct vxlan_sock *vs; 1790 struct vxlanhdr *hdr; 1791 __be32 vni; 1792 1793 if (!pskb_may_pull(skb, skb_transport_offset(skb) + VXLAN_HLEN)) 1794 return -EINVAL; 1795 1796 hdr = vxlan_hdr(skb); 1797 1798 if (!(hdr->vx_flags & VXLAN_HF_VNI)) 1799 return -EINVAL; 1800 1801 vs = rcu_dereference_sk_user_data(sk); 1802 if (!vs) 1803 return -ENOENT; 1804 1805 vni = vxlan_vni(hdr->vx_vni); 1806 vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni, NULL); 1807 if (!vxlan) 1808 return -ENOENT; 1809 1810 return 0; 1811 } 1812 1813 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni) 1814 { 1815 struct vxlan_dev *vxlan = netdev_priv(dev); 1816 struct arphdr *parp; 1817 u8 *arpptr, *sha; 1818 __be32 sip, tip; 1819 struct neighbour *n; 1820 1821 if (dev->flags & IFF_NOARP) 1822 goto out; 1823 1824 if (!pskb_may_pull(skb, arp_hdr_len(dev))) { 1825 dev->stats.tx_dropped++; 1826 goto out; 1827 } 1828 parp = arp_hdr(skb); 1829 1830 if ((parp->ar_hrd != htons(ARPHRD_ETHER) && 1831 parp->ar_hrd != htons(ARPHRD_IEEE802)) || 1832 parp->ar_pro != htons(ETH_P_IP) || 1833 parp->ar_op != htons(ARPOP_REQUEST) || 1834 parp->ar_hln != dev->addr_len || 1835 parp->ar_pln != 4) 1836 goto out; 1837 arpptr = (u8 *)parp + sizeof(struct arphdr); 1838 sha = arpptr; 1839 arpptr += dev->addr_len; /* sha */ 1840 memcpy(&sip, arpptr, sizeof(sip)); 1841 arpptr += sizeof(sip); 1842 arpptr += dev->addr_len; /* tha */ 1843 memcpy(&tip, arpptr, sizeof(tip)); 1844 1845 if (ipv4_is_loopback(tip) || 1846 ipv4_is_multicast(tip)) 1847 goto out; 1848 1849 n = neigh_lookup(&arp_tbl, &tip, dev); 1850 1851 if (n) { 1852 struct vxlan_fdb *f; 1853 struct sk_buff *reply; 1854 1855 if (!(n->nud_state & NUD_CONNECTED)) { 1856 neigh_release(n); 1857 goto out; 1858 } 1859 1860 f = vxlan_find_mac(vxlan, n->ha, vni); 1861 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) { 1862 /* bridge-local neighbor */ 1863 neigh_release(n); 1864 goto out; 1865 } 1866 1867 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha, 1868 n->ha, sha); 1869 1870 neigh_release(n); 1871 1872 if (reply == NULL) 1873 goto out; 1874 1875 skb_reset_mac_header(reply); 1876 __skb_pull(reply, skb_network_offset(reply)); 1877 reply->ip_summed = CHECKSUM_UNNECESSARY; 1878 reply->pkt_type = PACKET_HOST; 1879 1880 if (netif_rx_ni(reply) == NET_RX_DROP) { 1881 dev->stats.rx_dropped++; 1882 vxlan_vnifilter_count(vxlan, vni, NULL, 1883 VXLAN_VNI_STATS_RX_DROPS, 0); 1884 } 1885 1886 } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) { 1887 union vxlan_addr ipa = { 1888 .sin.sin_addr.s_addr = tip, 1889 .sin.sin_family = AF_INET, 1890 }; 1891 1892 vxlan_ip_miss(dev, &ipa); 1893 } 1894 out: 1895 consume_skb(skb); 1896 return NETDEV_TX_OK; 1897 } 1898 1899 #if IS_ENABLED(CONFIG_IPV6) 1900 static struct sk_buff *vxlan_na_create(struct sk_buff *request, 1901 struct neighbour *n, bool isrouter) 1902 { 1903 struct net_device *dev = request->dev; 1904 struct sk_buff *reply; 1905 struct nd_msg *ns, *na; 1906 struct ipv6hdr *pip6; 1907 u8 *daddr; 1908 int na_olen = 8; /* opt hdr + ETH_ALEN for target */ 1909 int ns_olen; 1910 int i, len; 1911 1912 if (dev == NULL || !pskb_may_pull(request, request->len)) 1913 return NULL; 1914 1915 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) + 1916 sizeof(*na) + na_olen + dev->needed_tailroom; 1917 reply = alloc_skb(len, GFP_ATOMIC); 1918 if (reply == NULL) 1919 return NULL; 1920 1921 reply->protocol = htons(ETH_P_IPV6); 1922 reply->dev = dev; 1923 skb_reserve(reply, LL_RESERVED_SPACE(request->dev)); 1924 skb_push(reply, sizeof(struct ethhdr)); 1925 skb_reset_mac_header(reply); 1926 1927 ns = (struct nd_msg *)(ipv6_hdr(request) + 1); 1928 1929 daddr = eth_hdr(request)->h_source; 1930 ns_olen = request->len - skb_network_offset(request) - 1931 sizeof(struct ipv6hdr) - sizeof(*ns); 1932 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) { 1933 if (!ns->opt[i + 1]) { 1934 kfree_skb(reply); 1935 return NULL; 1936 } 1937 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) { 1938 daddr = ns->opt + i + sizeof(struct nd_opt_hdr); 1939 break; 1940 } 1941 } 1942 1943 /* Ethernet header */ 1944 ether_addr_copy(eth_hdr(reply)->h_dest, daddr); 1945 ether_addr_copy(eth_hdr(reply)->h_source, n->ha); 1946 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6); 1947 reply->protocol = htons(ETH_P_IPV6); 1948 1949 skb_pull(reply, sizeof(struct ethhdr)); 1950 skb_reset_network_header(reply); 1951 skb_put(reply, sizeof(struct ipv6hdr)); 1952 1953 /* IPv6 header */ 1954 1955 pip6 = ipv6_hdr(reply); 1956 memset(pip6, 0, sizeof(struct ipv6hdr)); 1957 pip6->version = 6; 1958 pip6->priority = ipv6_hdr(request)->priority; 1959 pip6->nexthdr = IPPROTO_ICMPV6; 1960 pip6->hop_limit = 255; 1961 pip6->daddr = ipv6_hdr(request)->saddr; 1962 pip6->saddr = *(struct in6_addr *)n->primary_key; 1963 1964 skb_pull(reply, sizeof(struct ipv6hdr)); 1965 skb_reset_transport_header(reply); 1966 1967 /* Neighbor Advertisement */ 1968 na = skb_put_zero(reply, sizeof(*na) + na_olen); 1969 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT; 1970 na->icmph.icmp6_router = isrouter; 1971 na->icmph.icmp6_override = 1; 1972 na->icmph.icmp6_solicited = 1; 1973 na->target = ns->target; 1974 ether_addr_copy(&na->opt[2], n->ha); 1975 na->opt[0] = ND_OPT_TARGET_LL_ADDR; 1976 na->opt[1] = na_olen >> 3; 1977 1978 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr, 1979 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6, 1980 csum_partial(na, sizeof(*na)+na_olen, 0)); 1981 1982 pip6->payload_len = htons(sizeof(*na)+na_olen); 1983 1984 skb_push(reply, sizeof(struct ipv6hdr)); 1985 1986 reply->ip_summed = CHECKSUM_UNNECESSARY; 1987 1988 return reply; 1989 } 1990 1991 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni) 1992 { 1993 struct vxlan_dev *vxlan = netdev_priv(dev); 1994 const struct in6_addr *daddr; 1995 const struct ipv6hdr *iphdr; 1996 struct inet6_dev *in6_dev; 1997 struct neighbour *n; 1998 struct nd_msg *msg; 1999 2000 rcu_read_lock(); 2001 in6_dev = __in6_dev_get(dev); 2002 if (!in6_dev) 2003 goto out; 2004 2005 iphdr = ipv6_hdr(skb); 2006 daddr = &iphdr->daddr; 2007 msg = (struct nd_msg *)(iphdr + 1); 2008 2009 if (ipv6_addr_loopback(daddr) || 2010 ipv6_addr_is_multicast(&msg->target)) 2011 goto out; 2012 2013 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev); 2014 2015 if (n) { 2016 struct vxlan_fdb *f; 2017 struct sk_buff *reply; 2018 2019 if (!(n->nud_state & NUD_CONNECTED)) { 2020 neigh_release(n); 2021 goto out; 2022 } 2023 2024 f = vxlan_find_mac(vxlan, n->ha, vni); 2025 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) { 2026 /* bridge-local neighbor */ 2027 neigh_release(n); 2028 goto out; 2029 } 2030 2031 reply = vxlan_na_create(skb, n, 2032 !!(f ? f->flags & NTF_ROUTER : 0)); 2033 2034 neigh_release(n); 2035 2036 if (reply == NULL) 2037 goto out; 2038 2039 if (netif_rx_ni(reply) == NET_RX_DROP) { 2040 dev->stats.rx_dropped++; 2041 vxlan_vnifilter_count(vxlan, vni, NULL, 2042 VXLAN_VNI_STATS_RX_DROPS, 0); 2043 } 2044 } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) { 2045 union vxlan_addr ipa = { 2046 .sin6.sin6_addr = msg->target, 2047 .sin6.sin6_family = AF_INET6, 2048 }; 2049 2050 vxlan_ip_miss(dev, &ipa); 2051 } 2052 2053 out: 2054 rcu_read_unlock(); 2055 consume_skb(skb); 2056 return NETDEV_TX_OK; 2057 } 2058 #endif 2059 2060 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb) 2061 { 2062 struct vxlan_dev *vxlan = netdev_priv(dev); 2063 struct neighbour *n; 2064 2065 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest)) 2066 return false; 2067 2068 n = NULL; 2069 switch (ntohs(eth_hdr(skb)->h_proto)) { 2070 case ETH_P_IP: 2071 { 2072 struct iphdr *pip; 2073 2074 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 2075 return false; 2076 pip = ip_hdr(skb); 2077 n = neigh_lookup(&arp_tbl, &pip->daddr, dev); 2078 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) { 2079 union vxlan_addr ipa = { 2080 .sin.sin_addr.s_addr = pip->daddr, 2081 .sin.sin_family = AF_INET, 2082 }; 2083 2084 vxlan_ip_miss(dev, &ipa); 2085 return false; 2086 } 2087 2088 break; 2089 } 2090 #if IS_ENABLED(CONFIG_IPV6) 2091 case ETH_P_IPV6: 2092 { 2093 struct ipv6hdr *pip6; 2094 2095 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr))) 2096 return false; 2097 pip6 = ipv6_hdr(skb); 2098 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev); 2099 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) { 2100 union vxlan_addr ipa = { 2101 .sin6.sin6_addr = pip6->daddr, 2102 .sin6.sin6_family = AF_INET6, 2103 }; 2104 2105 vxlan_ip_miss(dev, &ipa); 2106 return false; 2107 } 2108 2109 break; 2110 } 2111 #endif 2112 default: 2113 return false; 2114 } 2115 2116 if (n) { 2117 bool diff; 2118 2119 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha); 2120 if (diff) { 2121 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest, 2122 dev->addr_len); 2123 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len); 2124 } 2125 neigh_release(n); 2126 return diff; 2127 } 2128 2129 return false; 2130 } 2131 2132 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags, 2133 struct vxlan_metadata *md) 2134 { 2135 struct vxlanhdr_gbp *gbp; 2136 2137 if (!md->gbp) 2138 return; 2139 2140 gbp = (struct vxlanhdr_gbp *)vxh; 2141 vxh->vx_flags |= VXLAN_HF_GBP; 2142 2143 if (md->gbp & VXLAN_GBP_DONT_LEARN) 2144 gbp->dont_learn = 1; 2145 2146 if (md->gbp & VXLAN_GBP_POLICY_APPLIED) 2147 gbp->policy_applied = 1; 2148 2149 gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK); 2150 } 2151 2152 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags, 2153 __be16 protocol) 2154 { 2155 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh; 2156 2157 gpe->np_applied = 1; 2158 gpe->next_protocol = tun_p_from_eth_p(protocol); 2159 if (!gpe->next_protocol) 2160 return -EPFNOSUPPORT; 2161 return 0; 2162 } 2163 2164 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst, 2165 int iphdr_len, __be32 vni, 2166 struct vxlan_metadata *md, u32 vxflags, 2167 bool udp_sum) 2168 { 2169 struct vxlanhdr *vxh; 2170 int min_headroom; 2171 int err; 2172 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL; 2173 __be16 inner_protocol = htons(ETH_P_TEB); 2174 2175 if ((vxflags & VXLAN_F_REMCSUM_TX) && 2176 skb->ip_summed == CHECKSUM_PARTIAL) { 2177 int csum_start = skb_checksum_start_offset(skb); 2178 2179 if (csum_start <= VXLAN_MAX_REMCSUM_START && 2180 !(csum_start & VXLAN_RCO_SHIFT_MASK) && 2181 (skb->csum_offset == offsetof(struct udphdr, check) || 2182 skb->csum_offset == offsetof(struct tcphdr, check))) 2183 type |= SKB_GSO_TUNNEL_REMCSUM; 2184 } 2185 2186 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len 2187 + VXLAN_HLEN + iphdr_len; 2188 2189 /* Need space for new headers (invalidates iph ptr) */ 2190 err = skb_cow_head(skb, min_headroom); 2191 if (unlikely(err)) 2192 return err; 2193 2194 err = iptunnel_handle_offloads(skb, type); 2195 if (err) 2196 return err; 2197 2198 vxh = __skb_push(skb, sizeof(*vxh)); 2199 vxh->vx_flags = VXLAN_HF_VNI; 2200 vxh->vx_vni = vxlan_vni_field(vni); 2201 2202 if (type & SKB_GSO_TUNNEL_REMCSUM) { 2203 unsigned int start; 2204 2205 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr); 2206 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset); 2207 vxh->vx_flags |= VXLAN_HF_RCO; 2208 2209 if (!skb_is_gso(skb)) { 2210 skb->ip_summed = CHECKSUM_NONE; 2211 skb->encapsulation = 0; 2212 } 2213 } 2214 2215 if (vxflags & VXLAN_F_GBP) 2216 vxlan_build_gbp_hdr(vxh, vxflags, md); 2217 if (vxflags & VXLAN_F_GPE) { 2218 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol); 2219 if (err < 0) 2220 return err; 2221 inner_protocol = skb->protocol; 2222 } 2223 2224 skb_set_inner_protocol(skb, inner_protocol); 2225 return 0; 2226 } 2227 2228 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev, 2229 struct vxlan_sock *sock4, 2230 struct sk_buff *skb, int oif, u8 tos, 2231 __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport, 2232 struct dst_cache *dst_cache, 2233 const struct ip_tunnel_info *info) 2234 { 2235 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 2236 struct rtable *rt = NULL; 2237 struct flowi4 fl4; 2238 2239 if (!sock4) 2240 return ERR_PTR(-EIO); 2241 2242 if (tos && !info) 2243 use_cache = false; 2244 if (use_cache) { 2245 rt = dst_cache_get_ip4(dst_cache, saddr); 2246 if (rt) 2247 return rt; 2248 } 2249 2250 memset(&fl4, 0, sizeof(fl4)); 2251 fl4.flowi4_oif = oif; 2252 fl4.flowi4_tos = RT_TOS(tos); 2253 fl4.flowi4_mark = skb->mark; 2254 fl4.flowi4_proto = IPPROTO_UDP; 2255 fl4.daddr = daddr; 2256 fl4.saddr = *saddr; 2257 fl4.fl4_dport = dport; 2258 fl4.fl4_sport = sport; 2259 2260 rt = ip_route_output_key(vxlan->net, &fl4); 2261 if (!IS_ERR(rt)) { 2262 if (rt->dst.dev == dev) { 2263 netdev_dbg(dev, "circular route to %pI4\n", &daddr); 2264 ip_rt_put(rt); 2265 return ERR_PTR(-ELOOP); 2266 } 2267 2268 *saddr = fl4.saddr; 2269 if (use_cache) 2270 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr); 2271 } else { 2272 netdev_dbg(dev, "no route to %pI4\n", &daddr); 2273 return ERR_PTR(-ENETUNREACH); 2274 } 2275 return rt; 2276 } 2277 2278 #if IS_ENABLED(CONFIG_IPV6) 2279 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan, 2280 struct net_device *dev, 2281 struct vxlan_sock *sock6, 2282 struct sk_buff *skb, int oif, u8 tos, 2283 __be32 label, 2284 const struct in6_addr *daddr, 2285 struct in6_addr *saddr, 2286 __be16 dport, __be16 sport, 2287 struct dst_cache *dst_cache, 2288 const struct ip_tunnel_info *info) 2289 { 2290 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 2291 struct dst_entry *ndst; 2292 struct flowi6 fl6; 2293 2294 if (!sock6) 2295 return ERR_PTR(-EIO); 2296 2297 if (tos && !info) 2298 use_cache = false; 2299 if (use_cache) { 2300 ndst = dst_cache_get_ip6(dst_cache, saddr); 2301 if (ndst) 2302 return ndst; 2303 } 2304 2305 memset(&fl6, 0, sizeof(fl6)); 2306 fl6.flowi6_oif = oif; 2307 fl6.daddr = *daddr; 2308 fl6.saddr = *saddr; 2309 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label); 2310 fl6.flowi6_mark = skb->mark; 2311 fl6.flowi6_proto = IPPROTO_UDP; 2312 fl6.fl6_dport = dport; 2313 fl6.fl6_sport = sport; 2314 2315 ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk, 2316 &fl6, NULL); 2317 if (IS_ERR(ndst)) { 2318 netdev_dbg(dev, "no route to %pI6\n", daddr); 2319 return ERR_PTR(-ENETUNREACH); 2320 } 2321 2322 if (unlikely(ndst->dev == dev)) { 2323 netdev_dbg(dev, "circular route to %pI6\n", daddr); 2324 dst_release(ndst); 2325 return ERR_PTR(-ELOOP); 2326 } 2327 2328 *saddr = fl6.saddr; 2329 if (use_cache) 2330 dst_cache_set_ip6(dst_cache, ndst, saddr); 2331 return ndst; 2332 } 2333 #endif 2334 2335 /* Bypass encapsulation if the destination is local */ 2336 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan, 2337 struct vxlan_dev *dst_vxlan, __be32 vni, 2338 bool snoop) 2339 { 2340 struct pcpu_sw_netstats *tx_stats, *rx_stats; 2341 union vxlan_addr loopback; 2342 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip; 2343 struct net_device *dev; 2344 int len = skb->len; 2345 2346 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats); 2347 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats); 2348 skb->pkt_type = PACKET_HOST; 2349 skb->encapsulation = 0; 2350 skb->dev = dst_vxlan->dev; 2351 __skb_pull(skb, skb_network_offset(skb)); 2352 2353 if (remote_ip->sa.sa_family == AF_INET) { 2354 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK); 2355 loopback.sa.sa_family = AF_INET; 2356 #if IS_ENABLED(CONFIG_IPV6) 2357 } else { 2358 loopback.sin6.sin6_addr = in6addr_loopback; 2359 loopback.sa.sa_family = AF_INET6; 2360 #endif 2361 } 2362 2363 rcu_read_lock(); 2364 dev = skb->dev; 2365 if (unlikely(!(dev->flags & IFF_UP))) { 2366 kfree_skb(skb); 2367 goto drop; 2368 } 2369 2370 if ((dst_vxlan->cfg.flags & VXLAN_F_LEARN) && snoop) 2371 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni); 2372 2373 u64_stats_update_begin(&tx_stats->syncp); 2374 tx_stats->tx_packets++; 2375 tx_stats->tx_bytes += len; 2376 u64_stats_update_end(&tx_stats->syncp); 2377 vxlan_vnifilter_count(src_vxlan, vni, NULL, VXLAN_VNI_STATS_TX, len); 2378 2379 if (__netif_rx(skb) == NET_RX_SUCCESS) { 2380 u64_stats_update_begin(&rx_stats->syncp); 2381 rx_stats->rx_packets++; 2382 rx_stats->rx_bytes += len; 2383 u64_stats_update_end(&rx_stats->syncp); 2384 vxlan_vnifilter_count(dst_vxlan, vni, NULL, VXLAN_VNI_STATS_RX, 2385 len); 2386 } else { 2387 drop: 2388 dev->stats.rx_dropped++; 2389 vxlan_vnifilter_count(dst_vxlan, vni, NULL, 2390 VXLAN_VNI_STATS_RX_DROPS, 0); 2391 } 2392 rcu_read_unlock(); 2393 } 2394 2395 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev, 2396 struct vxlan_dev *vxlan, 2397 union vxlan_addr *daddr, 2398 __be16 dst_port, int dst_ifindex, __be32 vni, 2399 struct dst_entry *dst, 2400 u32 rt_flags) 2401 { 2402 #if IS_ENABLED(CONFIG_IPV6) 2403 /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of 2404 * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple 2405 * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry. 2406 */ 2407 BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL); 2408 #endif 2409 /* Bypass encapsulation if the destination is local */ 2410 if (rt_flags & RTCF_LOCAL && 2411 !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) { 2412 struct vxlan_dev *dst_vxlan; 2413 2414 dst_release(dst); 2415 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni, 2416 daddr->sa.sa_family, dst_port, 2417 vxlan->cfg.flags); 2418 if (!dst_vxlan) { 2419 dev->stats.tx_errors++; 2420 vxlan_vnifilter_count(vxlan, vni, NULL, 2421 VXLAN_VNI_STATS_TX_ERRORS, 0); 2422 kfree_skb(skb); 2423 2424 return -ENOENT; 2425 } 2426 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni, true); 2427 return 1; 2428 } 2429 2430 return 0; 2431 } 2432 2433 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev, 2434 __be32 default_vni, struct vxlan_rdst *rdst, 2435 bool did_rsc) 2436 { 2437 struct dst_cache *dst_cache; 2438 struct ip_tunnel_info *info; 2439 struct vxlan_dev *vxlan = netdev_priv(dev); 2440 const struct iphdr *old_iph = ip_hdr(skb); 2441 union vxlan_addr *dst; 2442 union vxlan_addr remote_ip, local_ip; 2443 struct vxlan_metadata _md; 2444 struct vxlan_metadata *md = &_md; 2445 unsigned int pkt_len = skb->len; 2446 __be16 src_port = 0, dst_port; 2447 struct dst_entry *ndst = NULL; 2448 __u8 tos, ttl; 2449 int ifindex; 2450 int err; 2451 u32 flags = vxlan->cfg.flags; 2452 bool udp_sum = false; 2453 bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev)); 2454 __be32 vni = 0; 2455 #if IS_ENABLED(CONFIG_IPV6) 2456 __be32 label; 2457 #endif 2458 2459 info = skb_tunnel_info(skb); 2460 2461 if (rdst) { 2462 dst = &rdst->remote_ip; 2463 if (vxlan_addr_any(dst)) { 2464 if (did_rsc) { 2465 /* short-circuited back to local bridge */ 2466 vxlan_encap_bypass(skb, vxlan, vxlan, 2467 default_vni, true); 2468 return; 2469 } 2470 goto drop; 2471 } 2472 2473 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port; 2474 vni = (rdst->remote_vni) ? : default_vni; 2475 ifindex = rdst->remote_ifindex; 2476 local_ip = vxlan->cfg.saddr; 2477 dst_cache = &rdst->dst_cache; 2478 md->gbp = skb->mark; 2479 if (flags & VXLAN_F_TTL_INHERIT) { 2480 ttl = ip_tunnel_get_ttl(old_iph, skb); 2481 } else { 2482 ttl = vxlan->cfg.ttl; 2483 if (!ttl && vxlan_addr_multicast(dst)) 2484 ttl = 1; 2485 } 2486 2487 tos = vxlan->cfg.tos; 2488 if (tos == 1) 2489 tos = ip_tunnel_get_dsfield(old_iph, skb); 2490 2491 if (dst->sa.sa_family == AF_INET) 2492 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX); 2493 else 2494 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX); 2495 #if IS_ENABLED(CONFIG_IPV6) 2496 label = vxlan->cfg.label; 2497 #endif 2498 } else { 2499 if (!info) { 2500 WARN_ONCE(1, "%s: Missing encapsulation instructions\n", 2501 dev->name); 2502 goto drop; 2503 } 2504 remote_ip.sa.sa_family = ip_tunnel_info_af(info); 2505 if (remote_ip.sa.sa_family == AF_INET) { 2506 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst; 2507 local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src; 2508 } else { 2509 remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst; 2510 local_ip.sin6.sin6_addr = info->key.u.ipv6.src; 2511 } 2512 dst = &remote_ip; 2513 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port; 2514 vni = tunnel_id_to_key32(info->key.tun_id); 2515 ifindex = 0; 2516 dst_cache = &info->dst_cache; 2517 if (info->key.tun_flags & TUNNEL_VXLAN_OPT) { 2518 if (info->options_len < sizeof(*md)) 2519 goto drop; 2520 md = ip_tunnel_info_opts(info); 2521 } 2522 ttl = info->key.ttl; 2523 tos = info->key.tos; 2524 #if IS_ENABLED(CONFIG_IPV6) 2525 label = info->key.label; 2526 #endif 2527 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM); 2528 } 2529 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min, 2530 vxlan->cfg.port_max, true); 2531 2532 rcu_read_lock(); 2533 if (dst->sa.sa_family == AF_INET) { 2534 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock); 2535 struct rtable *rt; 2536 __be16 df = 0; 2537 2538 if (!ifindex) 2539 ifindex = sock4->sock->sk->sk_bound_dev_if; 2540 2541 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos, 2542 dst->sin.sin_addr.s_addr, 2543 &local_ip.sin.sin_addr.s_addr, 2544 dst_port, src_port, 2545 dst_cache, info); 2546 if (IS_ERR(rt)) { 2547 err = PTR_ERR(rt); 2548 goto tx_error; 2549 } 2550 2551 if (!info) { 2552 /* Bypass encapsulation if the destination is local */ 2553 err = encap_bypass_if_local(skb, dev, vxlan, dst, 2554 dst_port, ifindex, vni, 2555 &rt->dst, rt->rt_flags); 2556 if (err) 2557 goto out_unlock; 2558 2559 if (vxlan->cfg.df == VXLAN_DF_SET) { 2560 df = htons(IP_DF); 2561 } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) { 2562 struct ethhdr *eth = eth_hdr(skb); 2563 2564 if (ntohs(eth->h_proto) == ETH_P_IPV6 || 2565 (ntohs(eth->h_proto) == ETH_P_IP && 2566 old_iph->frag_off & htons(IP_DF))) 2567 df = htons(IP_DF); 2568 } 2569 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) { 2570 df = htons(IP_DF); 2571 } 2572 2573 ndst = &rt->dst; 2574 err = skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM, 2575 netif_is_any_bridge_port(dev)); 2576 if (err < 0) { 2577 goto tx_error; 2578 } else if (err) { 2579 if (info) { 2580 struct ip_tunnel_info *unclone; 2581 struct in_addr src, dst; 2582 2583 unclone = skb_tunnel_info_unclone(skb); 2584 if (unlikely(!unclone)) 2585 goto tx_error; 2586 2587 src = remote_ip.sin.sin_addr; 2588 dst = local_ip.sin.sin_addr; 2589 unclone->key.u.ipv4.src = src.s_addr; 2590 unclone->key.u.ipv4.dst = dst.s_addr; 2591 } 2592 vxlan_encap_bypass(skb, vxlan, vxlan, vni, false); 2593 dst_release(ndst); 2594 goto out_unlock; 2595 } 2596 2597 tos = ip_tunnel_ecn_encap(tos, old_iph, skb); 2598 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst); 2599 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr), 2600 vni, md, flags, udp_sum); 2601 if (err < 0) 2602 goto tx_error; 2603 2604 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr, 2605 dst->sin.sin_addr.s_addr, tos, ttl, df, 2606 src_port, dst_port, xnet, !udp_sum); 2607 #if IS_ENABLED(CONFIG_IPV6) 2608 } else { 2609 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock); 2610 2611 if (!ifindex) 2612 ifindex = sock6->sock->sk->sk_bound_dev_if; 2613 2614 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos, 2615 label, &dst->sin6.sin6_addr, 2616 &local_ip.sin6.sin6_addr, 2617 dst_port, src_port, 2618 dst_cache, info); 2619 if (IS_ERR(ndst)) { 2620 err = PTR_ERR(ndst); 2621 ndst = NULL; 2622 goto tx_error; 2623 } 2624 2625 if (!info) { 2626 u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags; 2627 2628 err = encap_bypass_if_local(skb, dev, vxlan, dst, 2629 dst_port, ifindex, vni, 2630 ndst, rt6i_flags); 2631 if (err) 2632 goto out_unlock; 2633 } 2634 2635 err = skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM, 2636 netif_is_any_bridge_port(dev)); 2637 if (err < 0) { 2638 goto tx_error; 2639 } else if (err) { 2640 if (info) { 2641 struct ip_tunnel_info *unclone; 2642 struct in6_addr src, dst; 2643 2644 unclone = skb_tunnel_info_unclone(skb); 2645 if (unlikely(!unclone)) 2646 goto tx_error; 2647 2648 src = remote_ip.sin6.sin6_addr; 2649 dst = local_ip.sin6.sin6_addr; 2650 unclone->key.u.ipv6.src = src; 2651 unclone->key.u.ipv6.dst = dst; 2652 } 2653 2654 vxlan_encap_bypass(skb, vxlan, vxlan, vni, false); 2655 dst_release(ndst); 2656 goto out_unlock; 2657 } 2658 2659 tos = ip_tunnel_ecn_encap(tos, old_iph, skb); 2660 ttl = ttl ? : ip6_dst_hoplimit(ndst); 2661 skb_scrub_packet(skb, xnet); 2662 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr), 2663 vni, md, flags, udp_sum); 2664 if (err < 0) 2665 goto tx_error; 2666 2667 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev, 2668 &local_ip.sin6.sin6_addr, 2669 &dst->sin6.sin6_addr, tos, ttl, 2670 label, src_port, dst_port, !udp_sum); 2671 #endif 2672 } 2673 vxlan_vnifilter_count(vxlan, vni, NULL, VXLAN_VNI_STATS_TX, pkt_len); 2674 out_unlock: 2675 rcu_read_unlock(); 2676 return; 2677 2678 drop: 2679 dev->stats.tx_dropped++; 2680 vxlan_vnifilter_count(vxlan, vni, NULL, VXLAN_VNI_STATS_TX_DROPS, 0); 2681 dev_kfree_skb(skb); 2682 return; 2683 2684 tx_error: 2685 rcu_read_unlock(); 2686 if (err == -ELOOP) 2687 dev->stats.collisions++; 2688 else if (err == -ENETUNREACH) 2689 dev->stats.tx_carrier_errors++; 2690 dst_release(ndst); 2691 dev->stats.tx_errors++; 2692 vxlan_vnifilter_count(vxlan, vni, NULL, VXLAN_VNI_STATS_TX_ERRORS, 0); 2693 kfree_skb(skb); 2694 } 2695 2696 static void vxlan_xmit_nh(struct sk_buff *skb, struct net_device *dev, 2697 struct vxlan_fdb *f, __be32 vni, bool did_rsc) 2698 { 2699 struct vxlan_rdst nh_rdst; 2700 struct nexthop *nh; 2701 bool do_xmit; 2702 u32 hash; 2703 2704 memset(&nh_rdst, 0, sizeof(struct vxlan_rdst)); 2705 hash = skb_get_hash(skb); 2706 2707 rcu_read_lock(); 2708 nh = rcu_dereference(f->nh); 2709 if (!nh) { 2710 rcu_read_unlock(); 2711 goto drop; 2712 } 2713 do_xmit = vxlan_fdb_nh_path_select(nh, hash, &nh_rdst); 2714 rcu_read_unlock(); 2715 2716 if (likely(do_xmit)) 2717 vxlan_xmit_one(skb, dev, vni, &nh_rdst, did_rsc); 2718 else 2719 goto drop; 2720 2721 return; 2722 2723 drop: 2724 dev->stats.tx_dropped++; 2725 vxlan_vnifilter_count(netdev_priv(dev), vni, NULL, 2726 VXLAN_VNI_STATS_TX_DROPS, 0); 2727 dev_kfree_skb(skb); 2728 } 2729 2730 /* Transmit local packets over Vxlan 2731 * 2732 * Outer IP header inherits ECN and DF from inner header. 2733 * Outer UDP destination is the VXLAN assigned port. 2734 * source port is based on hash of flow 2735 */ 2736 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev) 2737 { 2738 struct vxlan_dev *vxlan = netdev_priv(dev); 2739 struct vxlan_rdst *rdst, *fdst = NULL; 2740 const struct ip_tunnel_info *info; 2741 bool did_rsc = false; 2742 struct vxlan_fdb *f; 2743 struct ethhdr *eth; 2744 __be32 vni = 0; 2745 2746 info = skb_tunnel_info(skb); 2747 2748 skb_reset_mac_header(skb); 2749 2750 if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) { 2751 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE && 2752 info->mode & IP_TUNNEL_INFO_TX) { 2753 vni = tunnel_id_to_key32(info->key.tun_id); 2754 } else { 2755 if (info && info->mode & IP_TUNNEL_INFO_TX) 2756 vxlan_xmit_one(skb, dev, vni, NULL, false); 2757 else 2758 kfree_skb(skb); 2759 return NETDEV_TX_OK; 2760 } 2761 } 2762 2763 if (vxlan->cfg.flags & VXLAN_F_PROXY) { 2764 eth = eth_hdr(skb); 2765 if (ntohs(eth->h_proto) == ETH_P_ARP) 2766 return arp_reduce(dev, skb, vni); 2767 #if IS_ENABLED(CONFIG_IPV6) 2768 else if (ntohs(eth->h_proto) == ETH_P_IPV6 && 2769 pskb_may_pull(skb, sizeof(struct ipv6hdr) + 2770 sizeof(struct nd_msg)) && 2771 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) { 2772 struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1); 2773 2774 if (m->icmph.icmp6_code == 0 && 2775 m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) 2776 return neigh_reduce(dev, skb, vni); 2777 } 2778 #endif 2779 } 2780 2781 eth = eth_hdr(skb); 2782 f = vxlan_find_mac(vxlan, eth->h_dest, vni); 2783 did_rsc = false; 2784 2785 if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) && 2786 (ntohs(eth->h_proto) == ETH_P_IP || 2787 ntohs(eth->h_proto) == ETH_P_IPV6)) { 2788 did_rsc = route_shortcircuit(dev, skb); 2789 if (did_rsc) 2790 f = vxlan_find_mac(vxlan, eth->h_dest, vni); 2791 } 2792 2793 if (f == NULL) { 2794 f = vxlan_find_mac(vxlan, all_zeros_mac, vni); 2795 if (f == NULL) { 2796 if ((vxlan->cfg.flags & VXLAN_F_L2MISS) && 2797 !is_multicast_ether_addr(eth->h_dest)) 2798 vxlan_fdb_miss(vxlan, eth->h_dest); 2799 2800 dev->stats.tx_dropped++; 2801 vxlan_vnifilter_count(vxlan, vni, NULL, 2802 VXLAN_VNI_STATS_TX_DROPS, 0); 2803 kfree_skb(skb); 2804 return NETDEV_TX_OK; 2805 } 2806 } 2807 2808 if (rcu_access_pointer(f->nh)) { 2809 vxlan_xmit_nh(skb, dev, f, 2810 (vni ? : vxlan->default_dst.remote_vni), did_rsc); 2811 } else { 2812 list_for_each_entry_rcu(rdst, &f->remotes, list) { 2813 struct sk_buff *skb1; 2814 2815 if (!fdst) { 2816 fdst = rdst; 2817 continue; 2818 } 2819 skb1 = skb_clone(skb, GFP_ATOMIC); 2820 if (skb1) 2821 vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc); 2822 } 2823 if (fdst) 2824 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc); 2825 else 2826 kfree_skb(skb); 2827 } 2828 2829 return NETDEV_TX_OK; 2830 } 2831 2832 /* Walk the forwarding table and purge stale entries */ 2833 static void vxlan_cleanup(struct timer_list *t) 2834 { 2835 struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer); 2836 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL; 2837 unsigned int h; 2838 2839 if (!netif_running(vxlan->dev)) 2840 return; 2841 2842 for (h = 0; h < FDB_HASH_SIZE; ++h) { 2843 struct hlist_node *p, *n; 2844 2845 spin_lock(&vxlan->hash_lock[h]); 2846 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) { 2847 struct vxlan_fdb *f 2848 = container_of(p, struct vxlan_fdb, hlist); 2849 unsigned long timeout; 2850 2851 if (f->state & (NUD_PERMANENT | NUD_NOARP)) 2852 continue; 2853 2854 if (f->flags & NTF_EXT_LEARNED) 2855 continue; 2856 2857 timeout = f->used + vxlan->cfg.age_interval * HZ; 2858 if (time_before_eq(timeout, jiffies)) { 2859 netdev_dbg(vxlan->dev, 2860 "garbage collect %pM\n", 2861 f->eth_addr); 2862 f->state = NUD_STALE; 2863 vxlan_fdb_destroy(vxlan, f, true, true); 2864 } else if (time_before(timeout, next_timer)) 2865 next_timer = timeout; 2866 } 2867 spin_unlock(&vxlan->hash_lock[h]); 2868 } 2869 2870 mod_timer(&vxlan->age_timer, next_timer); 2871 } 2872 2873 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan) 2874 { 2875 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id); 2876 2877 spin_lock(&vn->sock_lock); 2878 hlist_del_init_rcu(&vxlan->hlist4.hlist); 2879 #if IS_ENABLED(CONFIG_IPV6) 2880 hlist_del_init_rcu(&vxlan->hlist6.hlist); 2881 #endif 2882 spin_unlock(&vn->sock_lock); 2883 } 2884 2885 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan, 2886 struct vxlan_dev_node *node) 2887 { 2888 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id); 2889 __be32 vni = vxlan->default_dst.remote_vni; 2890 2891 node->vxlan = vxlan; 2892 spin_lock(&vn->sock_lock); 2893 hlist_add_head_rcu(&node->hlist, vni_head(vs, vni)); 2894 spin_unlock(&vn->sock_lock); 2895 } 2896 2897 /* Setup stats when device is created */ 2898 static int vxlan_init(struct net_device *dev) 2899 { 2900 struct vxlan_dev *vxlan = netdev_priv(dev); 2901 int err; 2902 2903 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER) 2904 vxlan_vnigroup_init(vxlan); 2905 2906 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 2907 if (!dev->tstats) 2908 return -ENOMEM; 2909 2910 err = gro_cells_init(&vxlan->gro_cells, dev); 2911 if (err) { 2912 free_percpu(dev->tstats); 2913 return err; 2914 } 2915 2916 return 0; 2917 } 2918 2919 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni) 2920 { 2921 struct vxlan_fdb *f; 2922 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, vni); 2923 2924 spin_lock_bh(&vxlan->hash_lock[hash_index]); 2925 f = __vxlan_find_mac(vxlan, all_zeros_mac, vni); 2926 if (f) 2927 vxlan_fdb_destroy(vxlan, f, true, true); 2928 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 2929 } 2930 2931 static void vxlan_uninit(struct net_device *dev) 2932 { 2933 struct vxlan_dev *vxlan = netdev_priv(dev); 2934 2935 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER) 2936 vxlan_vnigroup_uninit(vxlan); 2937 2938 gro_cells_destroy(&vxlan->gro_cells); 2939 2940 vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni); 2941 2942 free_percpu(dev->tstats); 2943 } 2944 2945 /* Start ageing timer and join group when device is brought up */ 2946 static int vxlan_open(struct net_device *dev) 2947 { 2948 struct vxlan_dev *vxlan = netdev_priv(dev); 2949 int ret; 2950 2951 ret = vxlan_sock_add(vxlan); 2952 if (ret < 0) 2953 return ret; 2954 2955 ret = vxlan_multicast_join(vxlan); 2956 if (ret) { 2957 vxlan_sock_release(vxlan); 2958 return ret; 2959 } 2960 2961 if (vxlan->cfg.age_interval) 2962 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL); 2963 2964 return ret; 2965 } 2966 2967 /* Purge the forwarding table */ 2968 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all) 2969 { 2970 unsigned int h; 2971 2972 for (h = 0; h < FDB_HASH_SIZE; ++h) { 2973 struct hlist_node *p, *n; 2974 2975 spin_lock_bh(&vxlan->hash_lock[h]); 2976 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) { 2977 struct vxlan_fdb *f 2978 = container_of(p, struct vxlan_fdb, hlist); 2979 if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP))) 2980 continue; 2981 /* the all_zeros_mac entry is deleted at vxlan_uninit */ 2982 if (is_zero_ether_addr(f->eth_addr) && 2983 f->vni == vxlan->cfg.vni) 2984 continue; 2985 vxlan_fdb_destroy(vxlan, f, true, true); 2986 } 2987 spin_unlock_bh(&vxlan->hash_lock[h]); 2988 } 2989 } 2990 2991 /* Cleanup timer and forwarding table on shutdown */ 2992 static int vxlan_stop(struct net_device *dev) 2993 { 2994 struct vxlan_dev *vxlan = netdev_priv(dev); 2995 int ret = 0; 2996 2997 vxlan_multicast_leave(vxlan); 2998 2999 del_timer_sync(&vxlan->age_timer); 3000 3001 vxlan_flush(vxlan, false); 3002 vxlan_sock_release(vxlan); 3003 3004 return ret; 3005 } 3006 3007 /* Stub, nothing needs to be done. */ 3008 static void vxlan_set_multicast_list(struct net_device *dev) 3009 { 3010 } 3011 3012 static int vxlan_change_mtu(struct net_device *dev, int new_mtu) 3013 { 3014 struct vxlan_dev *vxlan = netdev_priv(dev); 3015 struct vxlan_rdst *dst = &vxlan->default_dst; 3016 struct net_device *lowerdev = __dev_get_by_index(vxlan->net, 3017 dst->remote_ifindex); 3018 bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6); 3019 3020 /* This check is different than dev->max_mtu, because it looks at 3021 * the lowerdev->mtu, rather than the static dev->max_mtu 3022 */ 3023 if (lowerdev) { 3024 int max_mtu = lowerdev->mtu - 3025 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM); 3026 if (new_mtu > max_mtu) 3027 return -EINVAL; 3028 } 3029 3030 dev->mtu = new_mtu; 3031 return 0; 3032 } 3033 3034 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 3035 { 3036 struct vxlan_dev *vxlan = netdev_priv(dev); 3037 struct ip_tunnel_info *info = skb_tunnel_info(skb); 3038 __be16 sport, dport; 3039 3040 sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min, 3041 vxlan->cfg.port_max, true); 3042 dport = info->key.tp_dst ? : vxlan->cfg.dst_port; 3043 3044 if (ip_tunnel_info_af(info) == AF_INET) { 3045 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock); 3046 struct rtable *rt; 3047 3048 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos, 3049 info->key.u.ipv4.dst, 3050 &info->key.u.ipv4.src, dport, sport, 3051 &info->dst_cache, info); 3052 if (IS_ERR(rt)) 3053 return PTR_ERR(rt); 3054 ip_rt_put(rt); 3055 } else { 3056 #if IS_ENABLED(CONFIG_IPV6) 3057 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock); 3058 struct dst_entry *ndst; 3059 3060 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos, 3061 info->key.label, &info->key.u.ipv6.dst, 3062 &info->key.u.ipv6.src, dport, sport, 3063 &info->dst_cache, info); 3064 if (IS_ERR(ndst)) 3065 return PTR_ERR(ndst); 3066 dst_release(ndst); 3067 #else /* !CONFIG_IPV6 */ 3068 return -EPFNOSUPPORT; 3069 #endif 3070 } 3071 info->key.tp_src = sport; 3072 info->key.tp_dst = dport; 3073 return 0; 3074 } 3075 3076 static const struct net_device_ops vxlan_netdev_ether_ops = { 3077 .ndo_init = vxlan_init, 3078 .ndo_uninit = vxlan_uninit, 3079 .ndo_open = vxlan_open, 3080 .ndo_stop = vxlan_stop, 3081 .ndo_start_xmit = vxlan_xmit, 3082 .ndo_get_stats64 = dev_get_tstats64, 3083 .ndo_set_rx_mode = vxlan_set_multicast_list, 3084 .ndo_change_mtu = vxlan_change_mtu, 3085 .ndo_validate_addr = eth_validate_addr, 3086 .ndo_set_mac_address = eth_mac_addr, 3087 .ndo_fdb_add = vxlan_fdb_add, 3088 .ndo_fdb_del = vxlan_fdb_delete, 3089 .ndo_fdb_dump = vxlan_fdb_dump, 3090 .ndo_fdb_get = vxlan_fdb_get, 3091 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst, 3092 }; 3093 3094 static const struct net_device_ops vxlan_netdev_raw_ops = { 3095 .ndo_init = vxlan_init, 3096 .ndo_uninit = vxlan_uninit, 3097 .ndo_open = vxlan_open, 3098 .ndo_stop = vxlan_stop, 3099 .ndo_start_xmit = vxlan_xmit, 3100 .ndo_get_stats64 = dev_get_tstats64, 3101 .ndo_change_mtu = vxlan_change_mtu, 3102 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst, 3103 }; 3104 3105 /* Info for udev, that this is a virtual tunnel endpoint */ 3106 static struct device_type vxlan_type = { 3107 .name = "vxlan", 3108 }; 3109 3110 /* Calls the ndo_udp_tunnel_add of the caller in order to 3111 * supply the listening VXLAN udp ports. Callers are expected 3112 * to implement the ndo_udp_tunnel_add. 3113 */ 3114 static void vxlan_offload_rx_ports(struct net_device *dev, bool push) 3115 { 3116 struct vxlan_sock *vs; 3117 struct net *net = dev_net(dev); 3118 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 3119 unsigned int i; 3120 3121 spin_lock(&vn->sock_lock); 3122 for (i = 0; i < PORT_HASH_SIZE; ++i) { 3123 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) { 3124 unsigned short type; 3125 3126 if (vs->flags & VXLAN_F_GPE) 3127 type = UDP_TUNNEL_TYPE_VXLAN_GPE; 3128 else 3129 type = UDP_TUNNEL_TYPE_VXLAN; 3130 3131 if (push) 3132 udp_tunnel_push_rx_port(dev, vs->sock, type); 3133 else 3134 udp_tunnel_drop_rx_port(dev, vs->sock, type); 3135 } 3136 } 3137 spin_unlock(&vn->sock_lock); 3138 } 3139 3140 /* Initialize the device structure. */ 3141 static void vxlan_setup(struct net_device *dev) 3142 { 3143 struct vxlan_dev *vxlan = netdev_priv(dev); 3144 unsigned int h; 3145 3146 eth_hw_addr_random(dev); 3147 ether_setup(dev); 3148 3149 dev->needs_free_netdev = true; 3150 SET_NETDEV_DEVTYPE(dev, &vxlan_type); 3151 3152 dev->features |= NETIF_F_LLTX; 3153 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST; 3154 dev->features |= NETIF_F_RXCSUM; 3155 dev->features |= NETIF_F_GSO_SOFTWARE; 3156 3157 dev->vlan_features = dev->features; 3158 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST; 3159 dev->hw_features |= NETIF_F_RXCSUM; 3160 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 3161 netif_keep_dst(dev); 3162 dev->priv_flags |= IFF_NO_QUEUE | IFF_CHANGE_PROTO_DOWN; 3163 3164 /* MTU range: 68 - 65535 */ 3165 dev->min_mtu = ETH_MIN_MTU; 3166 dev->max_mtu = ETH_MAX_MTU; 3167 3168 INIT_LIST_HEAD(&vxlan->next); 3169 3170 timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE); 3171 3172 vxlan->dev = dev; 3173 3174 for (h = 0; h < FDB_HASH_SIZE; ++h) { 3175 spin_lock_init(&vxlan->hash_lock[h]); 3176 INIT_HLIST_HEAD(&vxlan->fdb_head[h]); 3177 } 3178 } 3179 3180 static void vxlan_ether_setup(struct net_device *dev) 3181 { 3182 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 3183 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 3184 dev->netdev_ops = &vxlan_netdev_ether_ops; 3185 } 3186 3187 static void vxlan_raw_setup(struct net_device *dev) 3188 { 3189 dev->header_ops = NULL; 3190 dev->type = ARPHRD_NONE; 3191 dev->hard_header_len = 0; 3192 dev->addr_len = 0; 3193 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST; 3194 dev->netdev_ops = &vxlan_netdev_raw_ops; 3195 } 3196 3197 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = { 3198 [IFLA_VXLAN_ID] = { .type = NLA_U32 }, 3199 [IFLA_VXLAN_GROUP] = { .len = sizeof_field(struct iphdr, daddr) }, 3200 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) }, 3201 [IFLA_VXLAN_LINK] = { .type = NLA_U32 }, 3202 [IFLA_VXLAN_LOCAL] = { .len = sizeof_field(struct iphdr, saddr) }, 3203 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) }, 3204 [IFLA_VXLAN_TOS] = { .type = NLA_U8 }, 3205 [IFLA_VXLAN_TTL] = { .type = NLA_U8 }, 3206 [IFLA_VXLAN_LABEL] = { .type = NLA_U32 }, 3207 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 }, 3208 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 }, 3209 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 }, 3210 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) }, 3211 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 }, 3212 [IFLA_VXLAN_RSC] = { .type = NLA_U8 }, 3213 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 }, 3214 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 }, 3215 [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 }, 3216 [IFLA_VXLAN_PORT] = { .type = NLA_U16 }, 3217 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 }, 3218 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 }, 3219 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 }, 3220 [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 }, 3221 [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 }, 3222 [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, }, 3223 [IFLA_VXLAN_GPE] = { .type = NLA_FLAG, }, 3224 [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG }, 3225 [IFLA_VXLAN_TTL_INHERIT] = { .type = NLA_FLAG }, 3226 [IFLA_VXLAN_DF] = { .type = NLA_U8 }, 3227 [IFLA_VXLAN_VNIFILTER] = { .type = NLA_U8 }, 3228 }; 3229 3230 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[], 3231 struct netlink_ext_ack *extack) 3232 { 3233 if (tb[IFLA_ADDRESS]) { 3234 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 3235 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 3236 "Provided link layer address is not Ethernet"); 3237 return -EINVAL; 3238 } 3239 3240 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 3241 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 3242 "Provided Ethernet address is not unicast"); 3243 return -EADDRNOTAVAIL; 3244 } 3245 } 3246 3247 if (tb[IFLA_MTU]) { 3248 u32 mtu = nla_get_u32(tb[IFLA_MTU]); 3249 3250 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) { 3251 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU], 3252 "MTU must be between 68 and 65535"); 3253 return -EINVAL; 3254 } 3255 } 3256 3257 if (!data) { 3258 NL_SET_ERR_MSG(extack, 3259 "Required attributes not provided to perform the operation"); 3260 return -EINVAL; 3261 } 3262 3263 if (data[IFLA_VXLAN_ID]) { 3264 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]); 3265 3266 if (id >= VXLAN_N_VID) { 3267 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_ID], 3268 "VXLAN ID must be lower than 16777216"); 3269 return -ERANGE; 3270 } 3271 } 3272 3273 if (data[IFLA_VXLAN_PORT_RANGE]) { 3274 const struct ifla_vxlan_port_range *p 3275 = nla_data(data[IFLA_VXLAN_PORT_RANGE]); 3276 3277 if (ntohs(p->high) < ntohs(p->low)) { 3278 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_PORT_RANGE], 3279 "Invalid source port range"); 3280 return -EINVAL; 3281 } 3282 } 3283 3284 if (data[IFLA_VXLAN_DF]) { 3285 enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]); 3286 3287 if (df < 0 || df > VXLAN_DF_MAX) { 3288 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_DF], 3289 "Invalid DF attribute"); 3290 return -EINVAL; 3291 } 3292 } 3293 3294 return 0; 3295 } 3296 3297 static void vxlan_get_drvinfo(struct net_device *netdev, 3298 struct ethtool_drvinfo *drvinfo) 3299 { 3300 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version)); 3301 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver)); 3302 } 3303 3304 static int vxlan_get_link_ksettings(struct net_device *dev, 3305 struct ethtool_link_ksettings *cmd) 3306 { 3307 struct vxlan_dev *vxlan = netdev_priv(dev); 3308 struct vxlan_rdst *dst = &vxlan->default_dst; 3309 struct net_device *lowerdev = __dev_get_by_index(vxlan->net, 3310 dst->remote_ifindex); 3311 3312 if (!lowerdev) { 3313 cmd->base.duplex = DUPLEX_UNKNOWN; 3314 cmd->base.port = PORT_OTHER; 3315 cmd->base.speed = SPEED_UNKNOWN; 3316 3317 return 0; 3318 } 3319 3320 return __ethtool_get_link_ksettings(lowerdev, cmd); 3321 } 3322 3323 static const struct ethtool_ops vxlan_ethtool_ops = { 3324 .get_drvinfo = vxlan_get_drvinfo, 3325 .get_link = ethtool_op_get_link, 3326 .get_link_ksettings = vxlan_get_link_ksettings, 3327 }; 3328 3329 static struct socket *vxlan_create_sock(struct net *net, bool ipv6, 3330 __be16 port, u32 flags, int ifindex) 3331 { 3332 struct socket *sock; 3333 struct udp_port_cfg udp_conf; 3334 int err; 3335 3336 memset(&udp_conf, 0, sizeof(udp_conf)); 3337 3338 if (ipv6) { 3339 udp_conf.family = AF_INET6; 3340 udp_conf.use_udp6_rx_checksums = 3341 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX); 3342 udp_conf.ipv6_v6only = 1; 3343 } else { 3344 udp_conf.family = AF_INET; 3345 } 3346 3347 udp_conf.local_udp_port = port; 3348 udp_conf.bind_ifindex = ifindex; 3349 3350 /* Open UDP socket */ 3351 err = udp_sock_create(net, &udp_conf, &sock); 3352 if (err < 0) 3353 return ERR_PTR(err); 3354 3355 udp_allow_gso(sock->sk); 3356 return sock; 3357 } 3358 3359 /* Create new listen socket if needed */ 3360 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6, 3361 __be16 port, u32 flags, 3362 int ifindex) 3363 { 3364 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 3365 struct vxlan_sock *vs; 3366 struct socket *sock; 3367 unsigned int h; 3368 struct udp_tunnel_sock_cfg tunnel_cfg; 3369 3370 vs = kzalloc(sizeof(*vs), GFP_KERNEL); 3371 if (!vs) 3372 return ERR_PTR(-ENOMEM); 3373 3374 for (h = 0; h < VNI_HASH_SIZE; ++h) 3375 INIT_HLIST_HEAD(&vs->vni_list[h]); 3376 3377 sock = vxlan_create_sock(net, ipv6, port, flags, ifindex); 3378 if (IS_ERR(sock)) { 3379 kfree(vs); 3380 return ERR_CAST(sock); 3381 } 3382 3383 vs->sock = sock; 3384 refcount_set(&vs->refcnt, 1); 3385 vs->flags = (flags & VXLAN_F_RCV_FLAGS); 3386 3387 spin_lock(&vn->sock_lock); 3388 hlist_add_head_rcu(&vs->hlist, vs_head(net, port)); 3389 udp_tunnel_notify_add_rx_port(sock, 3390 (vs->flags & VXLAN_F_GPE) ? 3391 UDP_TUNNEL_TYPE_VXLAN_GPE : 3392 UDP_TUNNEL_TYPE_VXLAN); 3393 spin_unlock(&vn->sock_lock); 3394 3395 /* Mark socket as an encapsulation socket. */ 3396 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg)); 3397 tunnel_cfg.sk_user_data = vs; 3398 tunnel_cfg.encap_type = 1; 3399 tunnel_cfg.encap_rcv = vxlan_rcv; 3400 tunnel_cfg.encap_err_lookup = vxlan_err_lookup; 3401 tunnel_cfg.encap_destroy = NULL; 3402 tunnel_cfg.gro_receive = vxlan_gro_receive; 3403 tunnel_cfg.gro_complete = vxlan_gro_complete; 3404 3405 setup_udp_tunnel_sock(net, sock, &tunnel_cfg); 3406 3407 return vs; 3408 } 3409 3410 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6) 3411 { 3412 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id); 3413 bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA; 3414 struct vxlan_sock *vs = NULL; 3415 struct vxlan_dev_node *node; 3416 int l3mdev_index = 0; 3417 3418 if (vxlan->cfg.remote_ifindex) 3419 l3mdev_index = l3mdev_master_upper_ifindex_by_index( 3420 vxlan->net, vxlan->cfg.remote_ifindex); 3421 3422 if (!vxlan->cfg.no_share) { 3423 spin_lock(&vn->sock_lock); 3424 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET, 3425 vxlan->cfg.dst_port, vxlan->cfg.flags, 3426 l3mdev_index); 3427 if (vs && !refcount_inc_not_zero(&vs->refcnt)) { 3428 spin_unlock(&vn->sock_lock); 3429 return -EBUSY; 3430 } 3431 spin_unlock(&vn->sock_lock); 3432 } 3433 if (!vs) 3434 vs = vxlan_socket_create(vxlan->net, ipv6, 3435 vxlan->cfg.dst_port, vxlan->cfg.flags, 3436 l3mdev_index); 3437 if (IS_ERR(vs)) 3438 return PTR_ERR(vs); 3439 #if IS_ENABLED(CONFIG_IPV6) 3440 if (ipv6) { 3441 rcu_assign_pointer(vxlan->vn6_sock, vs); 3442 node = &vxlan->hlist6; 3443 } else 3444 #endif 3445 { 3446 rcu_assign_pointer(vxlan->vn4_sock, vs); 3447 node = &vxlan->hlist4; 3448 } 3449 3450 if (metadata && (vxlan->cfg.flags & VXLAN_F_VNIFILTER)) 3451 vxlan_vs_add_vnigrp(vxlan, vs, ipv6); 3452 else 3453 vxlan_vs_add_dev(vs, vxlan, node); 3454 3455 return 0; 3456 } 3457 3458 static int vxlan_sock_add(struct vxlan_dev *vxlan) 3459 { 3460 bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA; 3461 bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata; 3462 bool ipv4 = !ipv6 || metadata; 3463 int ret = 0; 3464 3465 RCU_INIT_POINTER(vxlan->vn4_sock, NULL); 3466 #if IS_ENABLED(CONFIG_IPV6) 3467 RCU_INIT_POINTER(vxlan->vn6_sock, NULL); 3468 if (ipv6) { 3469 ret = __vxlan_sock_add(vxlan, true); 3470 if (ret < 0 && ret != -EAFNOSUPPORT) 3471 ipv4 = false; 3472 } 3473 #endif 3474 if (ipv4) 3475 ret = __vxlan_sock_add(vxlan, false); 3476 if (ret < 0) 3477 vxlan_sock_release(vxlan); 3478 return ret; 3479 } 3480 3481 int vxlan_vni_in_use(struct net *src_net, struct vxlan_dev *vxlan, 3482 struct vxlan_config *conf, __be32 vni) 3483 { 3484 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id); 3485 struct vxlan_dev *tmp; 3486 3487 list_for_each_entry(tmp, &vn->vxlan_list, next) { 3488 if (tmp == vxlan) 3489 continue; 3490 if (tmp->cfg.flags & VXLAN_F_VNIFILTER) { 3491 if (!vxlan_vnifilter_lookup(tmp, vni)) 3492 continue; 3493 } else if (tmp->cfg.vni != vni) { 3494 continue; 3495 } 3496 if (tmp->cfg.dst_port != conf->dst_port) 3497 continue; 3498 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) != 3499 (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6))) 3500 continue; 3501 3502 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) && 3503 tmp->cfg.remote_ifindex != conf->remote_ifindex) 3504 continue; 3505 3506 return -EEXIST; 3507 } 3508 3509 return 0; 3510 } 3511 3512 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf, 3513 struct net_device **lower, 3514 struct vxlan_dev *old, 3515 struct netlink_ext_ack *extack) 3516 { 3517 bool use_ipv6 = false; 3518 3519 if (conf->flags & VXLAN_F_GPE) { 3520 /* For now, allow GPE only together with 3521 * COLLECT_METADATA. This can be relaxed later; in such 3522 * case, the other side of the PtP link will have to be 3523 * provided. 3524 */ 3525 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) || 3526 !(conf->flags & VXLAN_F_COLLECT_METADATA)) { 3527 NL_SET_ERR_MSG(extack, 3528 "VXLAN GPE does not support this combination of attributes"); 3529 return -EINVAL; 3530 } 3531 } 3532 3533 if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) { 3534 /* Unless IPv6 is explicitly requested, assume IPv4 */ 3535 conf->remote_ip.sa.sa_family = AF_INET; 3536 conf->saddr.sa.sa_family = AF_INET; 3537 } else if (!conf->remote_ip.sa.sa_family) { 3538 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family; 3539 } else if (!conf->saddr.sa.sa_family) { 3540 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family; 3541 } 3542 3543 if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) { 3544 NL_SET_ERR_MSG(extack, 3545 "Local and remote address must be from the same family"); 3546 return -EINVAL; 3547 } 3548 3549 if (vxlan_addr_multicast(&conf->saddr)) { 3550 NL_SET_ERR_MSG(extack, "Local address cannot be multicast"); 3551 return -EINVAL; 3552 } 3553 3554 if (conf->saddr.sa.sa_family == AF_INET6) { 3555 if (!IS_ENABLED(CONFIG_IPV6)) { 3556 NL_SET_ERR_MSG(extack, 3557 "IPv6 support not enabled in the kernel"); 3558 return -EPFNOSUPPORT; 3559 } 3560 use_ipv6 = true; 3561 conf->flags |= VXLAN_F_IPV6; 3562 3563 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) { 3564 int local_type = 3565 ipv6_addr_type(&conf->saddr.sin6.sin6_addr); 3566 int remote_type = 3567 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr); 3568 3569 if (local_type & IPV6_ADDR_LINKLOCAL) { 3570 if (!(remote_type & IPV6_ADDR_LINKLOCAL) && 3571 (remote_type != IPV6_ADDR_ANY)) { 3572 NL_SET_ERR_MSG(extack, 3573 "Invalid combination of local and remote address scopes"); 3574 return -EINVAL; 3575 } 3576 3577 conf->flags |= VXLAN_F_IPV6_LINKLOCAL; 3578 } else { 3579 if (remote_type == 3580 (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) { 3581 NL_SET_ERR_MSG(extack, 3582 "Invalid combination of local and remote address scopes"); 3583 return -EINVAL; 3584 } 3585 3586 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL; 3587 } 3588 } 3589 } 3590 3591 if (conf->label && !use_ipv6) { 3592 NL_SET_ERR_MSG(extack, 3593 "Label attribute only applies to IPv6 VXLAN devices"); 3594 return -EINVAL; 3595 } 3596 3597 if (conf->remote_ifindex) { 3598 struct net_device *lowerdev; 3599 3600 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex); 3601 if (!lowerdev) { 3602 NL_SET_ERR_MSG(extack, 3603 "Invalid local interface, device not found"); 3604 return -ENODEV; 3605 } 3606 3607 #if IS_ENABLED(CONFIG_IPV6) 3608 if (use_ipv6) { 3609 struct inet6_dev *idev = __in6_dev_get(lowerdev); 3610 3611 if (idev && idev->cnf.disable_ipv6) { 3612 NL_SET_ERR_MSG(extack, 3613 "IPv6 support disabled by administrator"); 3614 return -EPERM; 3615 } 3616 } 3617 #endif 3618 3619 *lower = lowerdev; 3620 } else { 3621 if (vxlan_addr_multicast(&conf->remote_ip)) { 3622 NL_SET_ERR_MSG(extack, 3623 "Local interface required for multicast remote destination"); 3624 3625 return -EINVAL; 3626 } 3627 3628 #if IS_ENABLED(CONFIG_IPV6) 3629 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) { 3630 NL_SET_ERR_MSG(extack, 3631 "Local interface required for link-local local/remote addresses"); 3632 return -EINVAL; 3633 } 3634 #endif 3635 3636 *lower = NULL; 3637 } 3638 3639 if (!conf->dst_port) { 3640 if (conf->flags & VXLAN_F_GPE) 3641 conf->dst_port = htons(IANA_VXLAN_GPE_UDP_PORT); 3642 else 3643 conf->dst_port = htons(vxlan_port); 3644 } 3645 3646 if (!conf->age_interval) 3647 conf->age_interval = FDB_AGE_DEFAULT; 3648 3649 if (vxlan_vni_in_use(src_net, old, conf, conf->vni)) { 3650 NL_SET_ERR_MSG(extack, 3651 "A VXLAN device with the specified VNI already exists"); 3652 return -EEXIST; 3653 } 3654 3655 return 0; 3656 } 3657 3658 static void vxlan_config_apply(struct net_device *dev, 3659 struct vxlan_config *conf, 3660 struct net_device *lowerdev, 3661 struct net *src_net, 3662 bool changelink) 3663 { 3664 struct vxlan_dev *vxlan = netdev_priv(dev); 3665 struct vxlan_rdst *dst = &vxlan->default_dst; 3666 unsigned short needed_headroom = ETH_HLEN; 3667 bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6); 3668 int max_mtu = ETH_MAX_MTU; 3669 3670 if (!changelink) { 3671 if (conf->flags & VXLAN_F_GPE) 3672 vxlan_raw_setup(dev); 3673 else 3674 vxlan_ether_setup(dev); 3675 3676 if (conf->mtu) 3677 dev->mtu = conf->mtu; 3678 3679 vxlan->net = src_net; 3680 } 3681 3682 dst->remote_vni = conf->vni; 3683 3684 memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip)); 3685 3686 if (lowerdev) { 3687 dst->remote_ifindex = conf->remote_ifindex; 3688 3689 netif_set_gso_max_size(dev, lowerdev->gso_max_size); 3690 netif_set_gso_max_segs(dev, lowerdev->gso_max_segs); 3691 3692 needed_headroom = lowerdev->hard_header_len; 3693 needed_headroom += lowerdev->needed_headroom; 3694 3695 dev->needed_tailroom = lowerdev->needed_tailroom; 3696 3697 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : 3698 VXLAN_HEADROOM); 3699 if (max_mtu < ETH_MIN_MTU) 3700 max_mtu = ETH_MIN_MTU; 3701 3702 if (!changelink && !conf->mtu) 3703 dev->mtu = max_mtu; 3704 } 3705 3706 if (dev->mtu > max_mtu) 3707 dev->mtu = max_mtu; 3708 3709 if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA) 3710 needed_headroom += VXLAN6_HEADROOM; 3711 else 3712 needed_headroom += VXLAN_HEADROOM; 3713 dev->needed_headroom = needed_headroom; 3714 3715 memcpy(&vxlan->cfg, conf, sizeof(*conf)); 3716 } 3717 3718 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev, 3719 struct vxlan_config *conf, bool changelink, 3720 struct netlink_ext_ack *extack) 3721 { 3722 struct vxlan_dev *vxlan = netdev_priv(dev); 3723 struct net_device *lowerdev; 3724 int ret; 3725 3726 ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack); 3727 if (ret) 3728 return ret; 3729 3730 vxlan_config_apply(dev, conf, lowerdev, src_net, changelink); 3731 3732 return 0; 3733 } 3734 3735 static int __vxlan_dev_create(struct net *net, struct net_device *dev, 3736 struct vxlan_config *conf, 3737 struct netlink_ext_ack *extack) 3738 { 3739 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 3740 struct vxlan_dev *vxlan = netdev_priv(dev); 3741 struct net_device *remote_dev = NULL; 3742 struct vxlan_fdb *f = NULL; 3743 bool unregister = false; 3744 struct vxlan_rdst *dst; 3745 int err; 3746 3747 dst = &vxlan->default_dst; 3748 err = vxlan_dev_configure(net, dev, conf, false, extack); 3749 if (err) 3750 return err; 3751 3752 dev->ethtool_ops = &vxlan_ethtool_ops; 3753 3754 /* create an fdb entry for a valid default destination */ 3755 if (!vxlan_addr_any(&dst->remote_ip)) { 3756 err = vxlan_fdb_create(vxlan, all_zeros_mac, 3757 &dst->remote_ip, 3758 NUD_REACHABLE | NUD_PERMANENT, 3759 vxlan->cfg.dst_port, 3760 dst->remote_vni, 3761 dst->remote_vni, 3762 dst->remote_ifindex, 3763 NTF_SELF, 0, &f, extack); 3764 if (err) 3765 return err; 3766 } 3767 3768 err = register_netdevice(dev); 3769 if (err) 3770 goto errout; 3771 unregister = true; 3772 3773 if (dst->remote_ifindex) { 3774 remote_dev = __dev_get_by_index(net, dst->remote_ifindex); 3775 if (!remote_dev) { 3776 err = -ENODEV; 3777 goto errout; 3778 } 3779 3780 err = netdev_upper_dev_link(remote_dev, dev, extack); 3781 if (err) 3782 goto errout; 3783 } 3784 3785 err = rtnl_configure_link(dev, NULL); 3786 if (err < 0) 3787 goto unlink; 3788 3789 if (f) { 3790 vxlan_fdb_insert(vxlan, all_zeros_mac, dst->remote_vni, f); 3791 3792 /* notify default fdb entry */ 3793 err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), 3794 RTM_NEWNEIGH, true, extack); 3795 if (err) { 3796 vxlan_fdb_destroy(vxlan, f, false, false); 3797 if (remote_dev) 3798 netdev_upper_dev_unlink(remote_dev, dev); 3799 goto unregister; 3800 } 3801 } 3802 3803 list_add(&vxlan->next, &vn->vxlan_list); 3804 if (remote_dev) 3805 dst->remote_dev = remote_dev; 3806 return 0; 3807 unlink: 3808 if (remote_dev) 3809 netdev_upper_dev_unlink(remote_dev, dev); 3810 errout: 3811 /* unregister_netdevice() destroys the default FDB entry with deletion 3812 * notification. But the addition notification was not sent yet, so 3813 * destroy the entry by hand here. 3814 */ 3815 if (f) 3816 __vxlan_fdb_free(f); 3817 unregister: 3818 if (unregister) 3819 unregister_netdevice(dev); 3820 return err; 3821 } 3822 3823 /* Set/clear flags based on attribute */ 3824 static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[], 3825 int attrtype, unsigned long mask, bool changelink, 3826 bool changelink_supported, 3827 struct netlink_ext_ack *extack) 3828 { 3829 unsigned long flags; 3830 3831 if (!tb[attrtype]) 3832 return 0; 3833 3834 if (changelink && !changelink_supported) { 3835 vxlan_flag_attr_error(attrtype, extack); 3836 return -EOPNOTSUPP; 3837 } 3838 3839 if (vxlan_policy[attrtype].type == NLA_FLAG) 3840 flags = conf->flags | mask; 3841 else if (nla_get_u8(tb[attrtype])) 3842 flags = conf->flags | mask; 3843 else 3844 flags = conf->flags & ~mask; 3845 3846 conf->flags = flags; 3847 3848 return 0; 3849 } 3850 3851 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[], 3852 struct net_device *dev, struct vxlan_config *conf, 3853 bool changelink, struct netlink_ext_ack *extack) 3854 { 3855 struct vxlan_dev *vxlan = netdev_priv(dev); 3856 int err = 0; 3857 3858 memset(conf, 0, sizeof(*conf)); 3859 3860 /* if changelink operation, start with old existing cfg */ 3861 if (changelink) 3862 memcpy(conf, &vxlan->cfg, sizeof(*conf)); 3863 3864 if (data[IFLA_VXLAN_ID]) { 3865 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID])); 3866 3867 if (changelink && (vni != conf->vni)) { 3868 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI"); 3869 return -EOPNOTSUPP; 3870 } 3871 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID])); 3872 } 3873 3874 if (data[IFLA_VXLAN_GROUP]) { 3875 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) { 3876 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group"); 3877 return -EOPNOTSUPP; 3878 } 3879 3880 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]); 3881 conf->remote_ip.sa.sa_family = AF_INET; 3882 } else if (data[IFLA_VXLAN_GROUP6]) { 3883 if (!IS_ENABLED(CONFIG_IPV6)) { 3884 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel"); 3885 return -EPFNOSUPPORT; 3886 } 3887 3888 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) { 3889 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group"); 3890 return -EOPNOTSUPP; 3891 } 3892 3893 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]); 3894 conf->remote_ip.sa.sa_family = AF_INET6; 3895 } 3896 3897 if (data[IFLA_VXLAN_LOCAL]) { 3898 if (changelink && (conf->saddr.sa.sa_family != AF_INET)) { 3899 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old"); 3900 return -EOPNOTSUPP; 3901 } 3902 3903 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]); 3904 conf->saddr.sa.sa_family = AF_INET; 3905 } else if (data[IFLA_VXLAN_LOCAL6]) { 3906 if (!IS_ENABLED(CONFIG_IPV6)) { 3907 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel"); 3908 return -EPFNOSUPPORT; 3909 } 3910 3911 if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) { 3912 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old"); 3913 return -EOPNOTSUPP; 3914 } 3915 3916 /* TODO: respect scope id */ 3917 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]); 3918 conf->saddr.sa.sa_family = AF_INET6; 3919 } 3920 3921 if (data[IFLA_VXLAN_LINK]) 3922 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]); 3923 3924 if (data[IFLA_VXLAN_TOS]) 3925 conf->tos = nla_get_u8(data[IFLA_VXLAN_TOS]); 3926 3927 if (data[IFLA_VXLAN_TTL]) 3928 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]); 3929 3930 if (data[IFLA_VXLAN_TTL_INHERIT]) { 3931 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT, 3932 VXLAN_F_TTL_INHERIT, changelink, false, 3933 extack); 3934 if (err) 3935 return err; 3936 3937 } 3938 3939 if (data[IFLA_VXLAN_LABEL]) 3940 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) & 3941 IPV6_FLOWLABEL_MASK; 3942 3943 if (data[IFLA_VXLAN_LEARNING]) { 3944 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING, 3945 VXLAN_F_LEARN, changelink, true, 3946 extack); 3947 if (err) 3948 return err; 3949 } else if (!changelink) { 3950 /* default to learn on a new device */ 3951 conf->flags |= VXLAN_F_LEARN; 3952 } 3953 3954 if (data[IFLA_VXLAN_AGEING]) 3955 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]); 3956 3957 if (data[IFLA_VXLAN_PROXY]) { 3958 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY, 3959 VXLAN_F_PROXY, changelink, false, 3960 extack); 3961 if (err) 3962 return err; 3963 } 3964 3965 if (data[IFLA_VXLAN_RSC]) { 3966 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC, 3967 VXLAN_F_RSC, changelink, false, 3968 extack); 3969 if (err) 3970 return err; 3971 } 3972 3973 if (data[IFLA_VXLAN_L2MISS]) { 3974 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS, 3975 VXLAN_F_L2MISS, changelink, false, 3976 extack); 3977 if (err) 3978 return err; 3979 } 3980 3981 if (data[IFLA_VXLAN_L3MISS]) { 3982 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS, 3983 VXLAN_F_L3MISS, changelink, false, 3984 extack); 3985 if (err) 3986 return err; 3987 } 3988 3989 if (data[IFLA_VXLAN_LIMIT]) { 3990 if (changelink) { 3991 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT], 3992 "Cannot change limit"); 3993 return -EOPNOTSUPP; 3994 } 3995 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]); 3996 } 3997 3998 if (data[IFLA_VXLAN_COLLECT_METADATA]) { 3999 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA, 4000 VXLAN_F_COLLECT_METADATA, changelink, false, 4001 extack); 4002 if (err) 4003 return err; 4004 } 4005 4006 if (data[IFLA_VXLAN_PORT_RANGE]) { 4007 if (!changelink) { 4008 const struct ifla_vxlan_port_range *p 4009 = nla_data(data[IFLA_VXLAN_PORT_RANGE]); 4010 conf->port_min = ntohs(p->low); 4011 conf->port_max = ntohs(p->high); 4012 } else { 4013 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE], 4014 "Cannot change port range"); 4015 return -EOPNOTSUPP; 4016 } 4017 } 4018 4019 if (data[IFLA_VXLAN_PORT]) { 4020 if (changelink) { 4021 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT], 4022 "Cannot change port"); 4023 return -EOPNOTSUPP; 4024 } 4025 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]); 4026 } 4027 4028 if (data[IFLA_VXLAN_UDP_CSUM]) { 4029 if (changelink) { 4030 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM], 4031 "Cannot change UDP_CSUM flag"); 4032 return -EOPNOTSUPP; 4033 } 4034 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM])) 4035 conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX; 4036 } 4037 4038 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) { 4039 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX, 4040 VXLAN_F_UDP_ZERO_CSUM6_TX, changelink, 4041 false, extack); 4042 if (err) 4043 return err; 4044 } 4045 4046 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) { 4047 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX, 4048 VXLAN_F_UDP_ZERO_CSUM6_RX, changelink, 4049 false, extack); 4050 if (err) 4051 return err; 4052 } 4053 4054 if (data[IFLA_VXLAN_REMCSUM_TX]) { 4055 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX, 4056 VXLAN_F_REMCSUM_TX, changelink, false, 4057 extack); 4058 if (err) 4059 return err; 4060 } 4061 4062 if (data[IFLA_VXLAN_REMCSUM_RX]) { 4063 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX, 4064 VXLAN_F_REMCSUM_RX, changelink, false, 4065 extack); 4066 if (err) 4067 return err; 4068 } 4069 4070 if (data[IFLA_VXLAN_GBP]) { 4071 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP, 4072 VXLAN_F_GBP, changelink, false, extack); 4073 if (err) 4074 return err; 4075 } 4076 4077 if (data[IFLA_VXLAN_GPE]) { 4078 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE, 4079 VXLAN_F_GPE, changelink, false, 4080 extack); 4081 if (err) 4082 return err; 4083 } 4084 4085 if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) { 4086 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL, 4087 VXLAN_F_REMCSUM_NOPARTIAL, changelink, 4088 false, extack); 4089 if (err) 4090 return err; 4091 } 4092 4093 if (tb[IFLA_MTU]) { 4094 if (changelink) { 4095 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU], 4096 "Cannot change mtu"); 4097 return -EOPNOTSUPP; 4098 } 4099 conf->mtu = nla_get_u32(tb[IFLA_MTU]); 4100 } 4101 4102 if (data[IFLA_VXLAN_DF]) 4103 conf->df = nla_get_u8(data[IFLA_VXLAN_DF]); 4104 4105 if (data[IFLA_VXLAN_VNIFILTER]) { 4106 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_VNIFILTER, 4107 VXLAN_F_VNIFILTER, changelink, false, 4108 extack); 4109 if (err) 4110 return err; 4111 4112 if ((conf->flags & VXLAN_F_VNIFILTER) && 4113 !(conf->flags & VXLAN_F_COLLECT_METADATA)) { 4114 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_VNIFILTER], 4115 "vxlan vnifilter only valid in collect metadata mode"); 4116 return -EINVAL; 4117 } 4118 } 4119 4120 return 0; 4121 } 4122 4123 static int vxlan_newlink(struct net *src_net, struct net_device *dev, 4124 struct nlattr *tb[], struct nlattr *data[], 4125 struct netlink_ext_ack *extack) 4126 { 4127 struct vxlan_config conf; 4128 int err; 4129 4130 err = vxlan_nl2conf(tb, data, dev, &conf, false, extack); 4131 if (err) 4132 return err; 4133 4134 return __vxlan_dev_create(src_net, dev, &conf, extack); 4135 } 4136 4137 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[], 4138 struct nlattr *data[], 4139 struct netlink_ext_ack *extack) 4140 { 4141 struct vxlan_dev *vxlan = netdev_priv(dev); 4142 struct net_device *lowerdev; 4143 struct vxlan_config conf; 4144 struct vxlan_rdst *dst; 4145 int err; 4146 4147 dst = &vxlan->default_dst; 4148 err = vxlan_nl2conf(tb, data, dev, &conf, true, extack); 4149 if (err) 4150 return err; 4151 4152 err = vxlan_config_validate(vxlan->net, &conf, &lowerdev, 4153 vxlan, extack); 4154 if (err) 4155 return err; 4156 4157 if (dst->remote_dev == lowerdev) 4158 lowerdev = NULL; 4159 4160 err = netdev_adjacent_change_prepare(dst->remote_dev, lowerdev, dev, 4161 extack); 4162 if (err) 4163 return err; 4164 4165 /* handle default dst entry */ 4166 if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) { 4167 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, conf.vni); 4168 4169 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4170 if (!vxlan_addr_any(&conf.remote_ip)) { 4171 err = vxlan_fdb_update(vxlan, all_zeros_mac, 4172 &conf.remote_ip, 4173 NUD_REACHABLE | NUD_PERMANENT, 4174 NLM_F_APPEND | NLM_F_CREATE, 4175 vxlan->cfg.dst_port, 4176 conf.vni, conf.vni, 4177 conf.remote_ifindex, 4178 NTF_SELF, 0, true, extack); 4179 if (err) { 4180 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4181 netdev_adjacent_change_abort(dst->remote_dev, 4182 lowerdev, dev); 4183 return err; 4184 } 4185 } 4186 if (!vxlan_addr_any(&dst->remote_ip)) 4187 __vxlan_fdb_delete(vxlan, all_zeros_mac, 4188 dst->remote_ip, 4189 vxlan->cfg.dst_port, 4190 dst->remote_vni, 4191 dst->remote_vni, 4192 dst->remote_ifindex, 4193 true); 4194 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4195 4196 /* If vni filtering device, also update fdb entries of 4197 * all vnis that were using default remote ip 4198 */ 4199 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER) { 4200 err = vxlan_vnilist_update_group(vxlan, &dst->remote_ip, 4201 &conf.remote_ip, extack); 4202 if (err) { 4203 netdev_adjacent_change_abort(dst->remote_dev, 4204 lowerdev, dev); 4205 return err; 4206 } 4207 } 4208 } 4209 4210 if (conf.age_interval != vxlan->cfg.age_interval) 4211 mod_timer(&vxlan->age_timer, jiffies); 4212 4213 netdev_adjacent_change_commit(dst->remote_dev, lowerdev, dev); 4214 if (lowerdev && lowerdev != dst->remote_dev) 4215 dst->remote_dev = lowerdev; 4216 vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true); 4217 return 0; 4218 } 4219 4220 static void vxlan_dellink(struct net_device *dev, struct list_head *head) 4221 { 4222 struct vxlan_dev *vxlan = netdev_priv(dev); 4223 4224 vxlan_flush(vxlan, true); 4225 4226 list_del(&vxlan->next); 4227 unregister_netdevice_queue(dev, head); 4228 if (vxlan->default_dst.remote_dev) 4229 netdev_upper_dev_unlink(vxlan->default_dst.remote_dev, dev); 4230 } 4231 4232 static size_t vxlan_get_size(const struct net_device *dev) 4233 { 4234 4235 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */ 4236 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */ 4237 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */ 4238 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */ 4239 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */ 4240 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL_INHERIT */ 4241 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */ 4242 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_DF */ 4243 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */ 4244 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */ 4245 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */ 4246 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */ 4247 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */ 4248 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */ 4249 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */ 4250 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */ 4251 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */ 4252 nla_total_size(sizeof(struct ifla_vxlan_port_range)) + 4253 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */ 4254 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */ 4255 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */ 4256 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */ 4257 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */ 4258 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */ 4259 0; 4260 } 4261 4262 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev) 4263 { 4264 const struct vxlan_dev *vxlan = netdev_priv(dev); 4265 const struct vxlan_rdst *dst = &vxlan->default_dst; 4266 struct ifla_vxlan_port_range ports = { 4267 .low = htons(vxlan->cfg.port_min), 4268 .high = htons(vxlan->cfg.port_max), 4269 }; 4270 4271 if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni))) 4272 goto nla_put_failure; 4273 4274 if (!vxlan_addr_any(&dst->remote_ip)) { 4275 if (dst->remote_ip.sa.sa_family == AF_INET) { 4276 if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP, 4277 dst->remote_ip.sin.sin_addr.s_addr)) 4278 goto nla_put_failure; 4279 #if IS_ENABLED(CONFIG_IPV6) 4280 } else { 4281 if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6, 4282 &dst->remote_ip.sin6.sin6_addr)) 4283 goto nla_put_failure; 4284 #endif 4285 } 4286 } 4287 4288 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex)) 4289 goto nla_put_failure; 4290 4291 if (!vxlan_addr_any(&vxlan->cfg.saddr)) { 4292 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) { 4293 if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL, 4294 vxlan->cfg.saddr.sin.sin_addr.s_addr)) 4295 goto nla_put_failure; 4296 #if IS_ENABLED(CONFIG_IPV6) 4297 } else { 4298 if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6, 4299 &vxlan->cfg.saddr.sin6.sin6_addr)) 4300 goto nla_put_failure; 4301 #endif 4302 } 4303 } 4304 4305 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) || 4306 nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT, 4307 !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) || 4308 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) || 4309 nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) || 4310 nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) || 4311 nla_put_u8(skb, IFLA_VXLAN_LEARNING, 4312 !!(vxlan->cfg.flags & VXLAN_F_LEARN)) || 4313 nla_put_u8(skb, IFLA_VXLAN_PROXY, 4314 !!(vxlan->cfg.flags & VXLAN_F_PROXY)) || 4315 nla_put_u8(skb, IFLA_VXLAN_RSC, 4316 !!(vxlan->cfg.flags & VXLAN_F_RSC)) || 4317 nla_put_u8(skb, IFLA_VXLAN_L2MISS, 4318 !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) || 4319 nla_put_u8(skb, IFLA_VXLAN_L3MISS, 4320 !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) || 4321 nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA, 4322 !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) || 4323 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) || 4324 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) || 4325 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) || 4326 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM, 4327 !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) || 4328 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX, 4329 !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) || 4330 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX, 4331 !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) || 4332 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX, 4333 !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) || 4334 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX, 4335 !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX))) 4336 goto nla_put_failure; 4337 4338 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports)) 4339 goto nla_put_failure; 4340 4341 if (vxlan->cfg.flags & VXLAN_F_GBP && 4342 nla_put_flag(skb, IFLA_VXLAN_GBP)) 4343 goto nla_put_failure; 4344 4345 if (vxlan->cfg.flags & VXLAN_F_GPE && 4346 nla_put_flag(skb, IFLA_VXLAN_GPE)) 4347 goto nla_put_failure; 4348 4349 if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL && 4350 nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL)) 4351 goto nla_put_failure; 4352 4353 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER && 4354 nla_put_u8(skb, IFLA_VXLAN_VNIFILTER, 4355 !!(vxlan->cfg.flags & VXLAN_F_VNIFILTER))) 4356 goto nla_put_failure; 4357 4358 return 0; 4359 4360 nla_put_failure: 4361 return -EMSGSIZE; 4362 } 4363 4364 static struct net *vxlan_get_link_net(const struct net_device *dev) 4365 { 4366 struct vxlan_dev *vxlan = netdev_priv(dev); 4367 4368 return vxlan->net; 4369 } 4370 4371 static struct rtnl_link_ops vxlan_link_ops __read_mostly = { 4372 .kind = "vxlan", 4373 .maxtype = IFLA_VXLAN_MAX, 4374 .policy = vxlan_policy, 4375 .priv_size = sizeof(struct vxlan_dev), 4376 .setup = vxlan_setup, 4377 .validate = vxlan_validate, 4378 .newlink = vxlan_newlink, 4379 .changelink = vxlan_changelink, 4380 .dellink = vxlan_dellink, 4381 .get_size = vxlan_get_size, 4382 .fill_info = vxlan_fill_info, 4383 .get_link_net = vxlan_get_link_net, 4384 }; 4385 4386 struct net_device *vxlan_dev_create(struct net *net, const char *name, 4387 u8 name_assign_type, 4388 struct vxlan_config *conf) 4389 { 4390 struct nlattr *tb[IFLA_MAX + 1]; 4391 struct net_device *dev; 4392 int err; 4393 4394 memset(&tb, 0, sizeof(tb)); 4395 4396 dev = rtnl_create_link(net, name, name_assign_type, 4397 &vxlan_link_ops, tb, NULL); 4398 if (IS_ERR(dev)) 4399 return dev; 4400 4401 err = __vxlan_dev_create(net, dev, conf, NULL); 4402 if (err < 0) { 4403 free_netdev(dev); 4404 return ERR_PTR(err); 4405 } 4406 4407 err = rtnl_configure_link(dev, NULL); 4408 if (err < 0) { 4409 LIST_HEAD(list_kill); 4410 4411 vxlan_dellink(dev, &list_kill); 4412 unregister_netdevice_many(&list_kill); 4413 return ERR_PTR(err); 4414 } 4415 4416 return dev; 4417 } 4418 EXPORT_SYMBOL_GPL(vxlan_dev_create); 4419 4420 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn, 4421 struct net_device *dev) 4422 { 4423 struct vxlan_dev *vxlan, *next; 4424 LIST_HEAD(list_kill); 4425 4426 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) { 4427 struct vxlan_rdst *dst = &vxlan->default_dst; 4428 4429 /* In case we created vxlan device with carrier 4430 * and we loose the carrier due to module unload 4431 * we also need to remove vxlan device. In other 4432 * cases, it's not necessary and remote_ifindex 4433 * is 0 here, so no matches. 4434 */ 4435 if (dst->remote_ifindex == dev->ifindex) 4436 vxlan_dellink(vxlan->dev, &list_kill); 4437 } 4438 4439 unregister_netdevice_many(&list_kill); 4440 } 4441 4442 static int vxlan_netdevice_event(struct notifier_block *unused, 4443 unsigned long event, void *ptr) 4444 { 4445 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 4446 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id); 4447 4448 if (event == NETDEV_UNREGISTER) 4449 vxlan_handle_lowerdev_unregister(vn, dev); 4450 else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO) 4451 vxlan_offload_rx_ports(dev, true); 4452 else if (event == NETDEV_UDP_TUNNEL_DROP_INFO) 4453 vxlan_offload_rx_ports(dev, false); 4454 4455 return NOTIFY_DONE; 4456 } 4457 4458 static struct notifier_block vxlan_notifier_block __read_mostly = { 4459 .notifier_call = vxlan_netdevice_event, 4460 }; 4461 4462 static void 4463 vxlan_fdb_offloaded_set(struct net_device *dev, 4464 struct switchdev_notifier_vxlan_fdb_info *fdb_info) 4465 { 4466 struct vxlan_dev *vxlan = netdev_priv(dev); 4467 struct vxlan_rdst *rdst; 4468 struct vxlan_fdb *f; 4469 u32 hash_index; 4470 4471 hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni); 4472 4473 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4474 4475 f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni); 4476 if (!f) 4477 goto out; 4478 4479 rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip, 4480 fdb_info->remote_port, 4481 fdb_info->remote_vni, 4482 fdb_info->remote_ifindex); 4483 if (!rdst) 4484 goto out; 4485 4486 rdst->offloaded = fdb_info->offloaded; 4487 4488 out: 4489 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4490 } 4491 4492 static int 4493 vxlan_fdb_external_learn_add(struct net_device *dev, 4494 struct switchdev_notifier_vxlan_fdb_info *fdb_info) 4495 { 4496 struct vxlan_dev *vxlan = netdev_priv(dev); 4497 struct netlink_ext_ack *extack; 4498 u32 hash_index; 4499 int err; 4500 4501 hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni); 4502 extack = switchdev_notifier_info_to_extack(&fdb_info->info); 4503 4504 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4505 err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip, 4506 NUD_REACHABLE, 4507 NLM_F_CREATE | NLM_F_REPLACE, 4508 fdb_info->remote_port, 4509 fdb_info->vni, 4510 fdb_info->remote_vni, 4511 fdb_info->remote_ifindex, 4512 NTF_USE | NTF_SELF | NTF_EXT_LEARNED, 4513 0, false, extack); 4514 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4515 4516 return err; 4517 } 4518 4519 static int 4520 vxlan_fdb_external_learn_del(struct net_device *dev, 4521 struct switchdev_notifier_vxlan_fdb_info *fdb_info) 4522 { 4523 struct vxlan_dev *vxlan = netdev_priv(dev); 4524 struct vxlan_fdb *f; 4525 u32 hash_index; 4526 int err = 0; 4527 4528 hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni); 4529 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4530 4531 f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni); 4532 if (!f) 4533 err = -ENOENT; 4534 else if (f->flags & NTF_EXT_LEARNED) 4535 err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr, 4536 fdb_info->remote_ip, 4537 fdb_info->remote_port, 4538 fdb_info->vni, 4539 fdb_info->remote_vni, 4540 fdb_info->remote_ifindex, 4541 false); 4542 4543 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4544 4545 return err; 4546 } 4547 4548 static int vxlan_switchdev_event(struct notifier_block *unused, 4549 unsigned long event, void *ptr) 4550 { 4551 struct net_device *dev = switchdev_notifier_info_to_dev(ptr); 4552 struct switchdev_notifier_vxlan_fdb_info *fdb_info; 4553 int err = 0; 4554 4555 switch (event) { 4556 case SWITCHDEV_VXLAN_FDB_OFFLOADED: 4557 vxlan_fdb_offloaded_set(dev, ptr); 4558 break; 4559 case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE: 4560 fdb_info = ptr; 4561 err = vxlan_fdb_external_learn_add(dev, fdb_info); 4562 if (err) { 4563 err = notifier_from_errno(err); 4564 break; 4565 } 4566 fdb_info->offloaded = true; 4567 vxlan_fdb_offloaded_set(dev, fdb_info); 4568 break; 4569 case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE: 4570 fdb_info = ptr; 4571 err = vxlan_fdb_external_learn_del(dev, fdb_info); 4572 if (err) { 4573 err = notifier_from_errno(err); 4574 break; 4575 } 4576 fdb_info->offloaded = false; 4577 vxlan_fdb_offloaded_set(dev, fdb_info); 4578 break; 4579 } 4580 4581 return err; 4582 } 4583 4584 static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = { 4585 .notifier_call = vxlan_switchdev_event, 4586 }; 4587 4588 static void vxlan_fdb_nh_flush(struct nexthop *nh) 4589 { 4590 struct vxlan_fdb *fdb; 4591 struct vxlan_dev *vxlan; 4592 u32 hash_index; 4593 4594 rcu_read_lock(); 4595 list_for_each_entry_rcu(fdb, &nh->fdb_list, nh_list) { 4596 vxlan = rcu_dereference(fdb->vdev); 4597 WARN_ON(!vxlan); 4598 hash_index = fdb_head_index(vxlan, fdb->eth_addr, 4599 vxlan->default_dst.remote_vni); 4600 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4601 if (!hlist_unhashed(&fdb->hlist)) 4602 vxlan_fdb_destroy(vxlan, fdb, false, false); 4603 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4604 } 4605 rcu_read_unlock(); 4606 } 4607 4608 static int vxlan_nexthop_event(struct notifier_block *nb, 4609 unsigned long event, void *ptr) 4610 { 4611 struct nh_notifier_info *info = ptr; 4612 struct nexthop *nh; 4613 4614 if (event != NEXTHOP_EVENT_DEL) 4615 return NOTIFY_DONE; 4616 4617 nh = nexthop_find_by_id(info->net, info->id); 4618 if (!nh) 4619 return NOTIFY_DONE; 4620 4621 vxlan_fdb_nh_flush(nh); 4622 4623 return NOTIFY_DONE; 4624 } 4625 4626 static __net_init int vxlan_init_net(struct net *net) 4627 { 4628 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 4629 unsigned int h; 4630 4631 INIT_LIST_HEAD(&vn->vxlan_list); 4632 spin_lock_init(&vn->sock_lock); 4633 vn->nexthop_notifier_block.notifier_call = vxlan_nexthop_event; 4634 4635 for (h = 0; h < PORT_HASH_SIZE; ++h) 4636 INIT_HLIST_HEAD(&vn->sock_list[h]); 4637 4638 return register_nexthop_notifier(net, &vn->nexthop_notifier_block, 4639 NULL); 4640 } 4641 4642 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head) 4643 { 4644 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 4645 struct vxlan_dev *vxlan, *next; 4646 struct net_device *dev, *aux; 4647 4648 for_each_netdev_safe(net, dev, aux) 4649 if (dev->rtnl_link_ops == &vxlan_link_ops) 4650 unregister_netdevice_queue(dev, head); 4651 4652 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) { 4653 /* If vxlan->dev is in the same netns, it has already been added 4654 * to the list by the previous loop. 4655 */ 4656 if (!net_eq(dev_net(vxlan->dev), net)) 4657 unregister_netdevice_queue(vxlan->dev, head); 4658 } 4659 4660 } 4661 4662 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list) 4663 { 4664 struct net *net; 4665 LIST_HEAD(list); 4666 unsigned int h; 4667 4668 list_for_each_entry(net, net_list, exit_list) { 4669 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 4670 4671 unregister_nexthop_notifier(net, &vn->nexthop_notifier_block); 4672 } 4673 rtnl_lock(); 4674 list_for_each_entry(net, net_list, exit_list) 4675 vxlan_destroy_tunnels(net, &list); 4676 4677 unregister_netdevice_many(&list); 4678 rtnl_unlock(); 4679 4680 list_for_each_entry(net, net_list, exit_list) { 4681 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 4682 4683 for (h = 0; h < PORT_HASH_SIZE; ++h) 4684 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h])); 4685 } 4686 } 4687 4688 static struct pernet_operations vxlan_net_ops = { 4689 .init = vxlan_init_net, 4690 .exit_batch = vxlan_exit_batch_net, 4691 .id = &vxlan_net_id, 4692 .size = sizeof(struct vxlan_net), 4693 }; 4694 4695 static int __init vxlan_init_module(void) 4696 { 4697 int rc; 4698 4699 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt)); 4700 4701 rc = register_pernet_subsys(&vxlan_net_ops); 4702 if (rc) 4703 goto out1; 4704 4705 rc = register_netdevice_notifier(&vxlan_notifier_block); 4706 if (rc) 4707 goto out2; 4708 4709 rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block); 4710 if (rc) 4711 goto out3; 4712 4713 rc = rtnl_link_register(&vxlan_link_ops); 4714 if (rc) 4715 goto out4; 4716 4717 vxlan_vnifilter_init(); 4718 4719 return 0; 4720 out4: 4721 unregister_switchdev_notifier(&vxlan_switchdev_notifier_block); 4722 out3: 4723 unregister_netdevice_notifier(&vxlan_notifier_block); 4724 out2: 4725 unregister_pernet_subsys(&vxlan_net_ops); 4726 out1: 4727 return rc; 4728 } 4729 late_initcall(vxlan_init_module); 4730 4731 static void __exit vxlan_cleanup_module(void) 4732 { 4733 vxlan_vnifilter_uninit(); 4734 rtnl_link_unregister(&vxlan_link_ops); 4735 unregister_switchdev_notifier(&vxlan_switchdev_notifier_block); 4736 unregister_netdevice_notifier(&vxlan_notifier_block); 4737 unregister_pernet_subsys(&vxlan_net_ops); 4738 /* rcu_barrier() is called by netns */ 4739 } 4740 module_exit(vxlan_cleanup_module); 4741 4742 MODULE_LICENSE("GPL"); 4743 MODULE_VERSION(VXLAN_VERSION); 4744 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>"); 4745 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic"); 4746 MODULE_ALIAS_RTNL_LINK("vxlan"); 4747