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