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 -ENOMEM; 655 656 if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) { 657 kfree(rd); 658 return -ENOMEM; 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 struct netlink_ext_ack *extack) 1134 { 1135 struct net *net = dev_net(vxlan->dev); 1136 int err; 1137 1138 if (tb[NDA_NH_ID] && (tb[NDA_DST] || tb[NDA_VNI] || tb[NDA_IFINDEX] || 1139 tb[NDA_PORT])) { 1140 NL_SET_ERR_MSG(extack, 1141 "DST, VNI, ifindex and port are mutually exclusive with NH_ID"); 1142 return -EINVAL; 1143 } 1144 1145 if (tb[NDA_DST]) { 1146 err = vxlan_nla_get_addr(ip, tb[NDA_DST]); 1147 if (err) { 1148 NL_SET_ERR_MSG(extack, "Unsupported address family"); 1149 return err; 1150 } 1151 } else { 1152 union vxlan_addr *remote = &vxlan->default_dst.remote_ip; 1153 1154 if (remote->sa.sa_family == AF_INET) { 1155 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY); 1156 ip->sa.sa_family = AF_INET; 1157 #if IS_ENABLED(CONFIG_IPV6) 1158 } else { 1159 ip->sin6.sin6_addr = in6addr_any; 1160 ip->sa.sa_family = AF_INET6; 1161 #endif 1162 } 1163 } 1164 1165 if (tb[NDA_PORT]) { 1166 if (nla_len(tb[NDA_PORT]) != sizeof(__be16)) { 1167 NL_SET_ERR_MSG(extack, "Invalid vxlan port"); 1168 return -EINVAL; 1169 } 1170 *port = nla_get_be16(tb[NDA_PORT]); 1171 } else { 1172 *port = vxlan->cfg.dst_port; 1173 } 1174 1175 if (tb[NDA_VNI]) { 1176 if (nla_len(tb[NDA_VNI]) != sizeof(u32)) { 1177 NL_SET_ERR_MSG(extack, "Invalid vni"); 1178 return -EINVAL; 1179 } 1180 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI])); 1181 } else { 1182 *vni = vxlan->default_dst.remote_vni; 1183 } 1184 1185 if (tb[NDA_SRC_VNI]) { 1186 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32)) { 1187 NL_SET_ERR_MSG(extack, "Invalid src vni"); 1188 return -EINVAL; 1189 } 1190 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI])); 1191 } else { 1192 *src_vni = vxlan->default_dst.remote_vni; 1193 } 1194 1195 if (tb[NDA_IFINDEX]) { 1196 struct net_device *tdev; 1197 1198 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32)) { 1199 NL_SET_ERR_MSG(extack, "Invalid ifindex"); 1200 return -EINVAL; 1201 } 1202 *ifindex = nla_get_u32(tb[NDA_IFINDEX]); 1203 tdev = __dev_get_by_index(net, *ifindex); 1204 if (!tdev) { 1205 NL_SET_ERR_MSG(extack, "Device not found"); 1206 return -EADDRNOTAVAIL; 1207 } 1208 } else { 1209 *ifindex = 0; 1210 } 1211 1212 if (tb[NDA_NH_ID]) 1213 *nhid = nla_get_u32(tb[NDA_NH_ID]); 1214 else 1215 *nhid = 0; 1216 1217 return 0; 1218 } 1219 1220 /* Add static entry (via netlink) */ 1221 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[], 1222 struct net_device *dev, 1223 const unsigned char *addr, u16 vid, u16 flags, 1224 struct netlink_ext_ack *extack) 1225 { 1226 struct vxlan_dev *vxlan = netdev_priv(dev); 1227 /* struct net *net = dev_net(vxlan->dev); */ 1228 union vxlan_addr ip; 1229 __be16 port; 1230 __be32 src_vni, vni; 1231 u32 ifindex, nhid; 1232 u32 hash_index; 1233 int err; 1234 1235 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) { 1236 pr_info("RTM_NEWNEIGH with invalid state %#x\n", 1237 ndm->ndm_state); 1238 return -EINVAL; 1239 } 1240 1241 if (!tb || (!tb[NDA_DST] && !tb[NDA_NH_ID])) 1242 return -EINVAL; 1243 1244 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex, 1245 &nhid, extack); 1246 if (err) 1247 return err; 1248 1249 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family) 1250 return -EAFNOSUPPORT; 1251 1252 hash_index = fdb_head_index(vxlan, addr, src_vni); 1253 spin_lock_bh(&vxlan->hash_lock[hash_index]); 1254 err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags, 1255 port, src_vni, vni, ifindex, 1256 ndm->ndm_flags | NTF_VXLAN_ADDED_BY_USER, 1257 nhid, true, extack); 1258 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 1259 1260 return err; 1261 } 1262 1263 int __vxlan_fdb_delete(struct vxlan_dev *vxlan, 1264 const unsigned char *addr, union vxlan_addr ip, 1265 __be16 port, __be32 src_vni, __be32 vni, 1266 u32 ifindex, bool swdev_notify) 1267 { 1268 struct vxlan_rdst *rd = NULL; 1269 struct vxlan_fdb *f; 1270 int err = -ENOENT; 1271 1272 f = vxlan_find_mac(vxlan, addr, src_vni); 1273 if (!f) 1274 return err; 1275 1276 if (!vxlan_addr_any(&ip)) { 1277 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex); 1278 if (!rd) 1279 goto out; 1280 } 1281 1282 /* remove a destination if it's not the only one on the list, 1283 * otherwise destroy the fdb entry 1284 */ 1285 if (rd && !list_is_singular(&f->remotes)) { 1286 vxlan_fdb_dst_destroy(vxlan, f, rd, swdev_notify); 1287 goto out; 1288 } 1289 1290 vxlan_fdb_destroy(vxlan, f, true, swdev_notify); 1291 1292 out: 1293 return 0; 1294 } 1295 1296 /* Delete entry (via netlink) */ 1297 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[], 1298 struct net_device *dev, 1299 const unsigned char *addr, u16 vid, 1300 struct netlink_ext_ack *extack) 1301 { 1302 struct vxlan_dev *vxlan = netdev_priv(dev); 1303 union vxlan_addr ip; 1304 __be32 src_vni, vni; 1305 u32 ifindex, nhid; 1306 u32 hash_index; 1307 __be16 port; 1308 int err; 1309 1310 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex, 1311 &nhid, extack); 1312 if (err) 1313 return err; 1314 1315 hash_index = fdb_head_index(vxlan, addr, src_vni); 1316 spin_lock_bh(&vxlan->hash_lock[hash_index]); 1317 err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex, 1318 true); 1319 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 1320 1321 return err; 1322 } 1323 1324 /* Dump forwarding table */ 1325 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb, 1326 struct net_device *dev, 1327 struct net_device *filter_dev, int *idx) 1328 { 1329 struct vxlan_dev *vxlan = netdev_priv(dev); 1330 unsigned int h; 1331 int err = 0; 1332 1333 for (h = 0; h < FDB_HASH_SIZE; ++h) { 1334 struct vxlan_fdb *f; 1335 1336 rcu_read_lock(); 1337 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) { 1338 struct vxlan_rdst *rd; 1339 1340 if (rcu_access_pointer(f->nh)) { 1341 if (*idx < cb->args[2]) 1342 goto skip_nh; 1343 err = vxlan_fdb_info(skb, vxlan, f, 1344 NETLINK_CB(cb->skb).portid, 1345 cb->nlh->nlmsg_seq, 1346 RTM_NEWNEIGH, 1347 NLM_F_MULTI, NULL); 1348 if (err < 0) { 1349 rcu_read_unlock(); 1350 goto out; 1351 } 1352 skip_nh: 1353 *idx += 1; 1354 continue; 1355 } 1356 1357 list_for_each_entry_rcu(rd, &f->remotes, list) { 1358 if (*idx < cb->args[2]) 1359 goto skip; 1360 1361 err = vxlan_fdb_info(skb, vxlan, f, 1362 NETLINK_CB(cb->skb).portid, 1363 cb->nlh->nlmsg_seq, 1364 RTM_NEWNEIGH, 1365 NLM_F_MULTI, rd); 1366 if (err < 0) { 1367 rcu_read_unlock(); 1368 goto out; 1369 } 1370 skip: 1371 *idx += 1; 1372 } 1373 } 1374 rcu_read_unlock(); 1375 } 1376 out: 1377 return err; 1378 } 1379 1380 static int vxlan_fdb_get(struct sk_buff *skb, 1381 struct nlattr *tb[], 1382 struct net_device *dev, 1383 const unsigned char *addr, 1384 u16 vid, u32 portid, u32 seq, 1385 struct netlink_ext_ack *extack) 1386 { 1387 struct vxlan_dev *vxlan = netdev_priv(dev); 1388 struct vxlan_fdb *f; 1389 __be32 vni; 1390 int err; 1391 1392 if (tb[NDA_VNI]) 1393 vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI])); 1394 else 1395 vni = vxlan->default_dst.remote_vni; 1396 1397 rcu_read_lock(); 1398 1399 f = __vxlan_find_mac(vxlan, addr, vni); 1400 if (!f) { 1401 NL_SET_ERR_MSG(extack, "Fdb entry not found"); 1402 err = -ENOENT; 1403 goto errout; 1404 } 1405 1406 err = vxlan_fdb_info(skb, vxlan, f, portid, seq, 1407 RTM_NEWNEIGH, 0, first_remote_rcu(f)); 1408 errout: 1409 rcu_read_unlock(); 1410 return err; 1411 } 1412 1413 /* Watch incoming packets to learn mapping between Ethernet address 1414 * and Tunnel endpoint. 1415 * Return true if packet is bogus and should be dropped. 1416 */ 1417 static bool vxlan_snoop(struct net_device *dev, 1418 union vxlan_addr *src_ip, const u8 *src_mac, 1419 u32 src_ifindex, __be32 vni) 1420 { 1421 struct vxlan_dev *vxlan = netdev_priv(dev); 1422 struct vxlan_fdb *f; 1423 u32 ifindex = 0; 1424 1425 #if IS_ENABLED(CONFIG_IPV6) 1426 if (src_ip->sa.sa_family == AF_INET6 && 1427 (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)) 1428 ifindex = src_ifindex; 1429 #endif 1430 1431 f = vxlan_find_mac(vxlan, src_mac, vni); 1432 if (likely(f)) { 1433 struct vxlan_rdst *rdst = first_remote_rcu(f); 1434 1435 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) && 1436 rdst->remote_ifindex == ifindex)) 1437 return false; 1438 1439 /* Don't migrate static entries, drop packets */ 1440 if (f->state & (NUD_PERMANENT | NUD_NOARP)) 1441 return true; 1442 1443 /* Don't override an fdb with nexthop with a learnt entry */ 1444 if (rcu_access_pointer(f->nh)) 1445 return true; 1446 1447 if (net_ratelimit()) 1448 netdev_info(dev, 1449 "%pM migrated from %pIS to %pIS\n", 1450 src_mac, &rdst->remote_ip.sa, &src_ip->sa); 1451 1452 rdst->remote_ip = *src_ip; 1453 f->updated = jiffies; 1454 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH, true, NULL); 1455 } else { 1456 u32 hash_index = fdb_head_index(vxlan, src_mac, vni); 1457 1458 /* learned new entry */ 1459 spin_lock(&vxlan->hash_lock[hash_index]); 1460 1461 /* close off race between vxlan_flush and incoming packets */ 1462 if (netif_running(dev)) 1463 vxlan_fdb_update(vxlan, src_mac, src_ip, 1464 NUD_REACHABLE, 1465 NLM_F_EXCL|NLM_F_CREATE, 1466 vxlan->cfg.dst_port, 1467 vni, 1468 vxlan->default_dst.remote_vni, 1469 ifindex, NTF_SELF, 0, true, NULL); 1470 spin_unlock(&vxlan->hash_lock[hash_index]); 1471 } 1472 1473 return false; 1474 } 1475 1476 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs) 1477 { 1478 struct vxlan_net *vn; 1479 1480 if (!vs) 1481 return false; 1482 if (!refcount_dec_and_test(&vs->refcnt)) 1483 return false; 1484 1485 vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id); 1486 spin_lock(&vn->sock_lock); 1487 hlist_del_rcu(&vs->hlist); 1488 udp_tunnel_notify_del_rx_port(vs->sock, 1489 (vs->flags & VXLAN_F_GPE) ? 1490 UDP_TUNNEL_TYPE_VXLAN_GPE : 1491 UDP_TUNNEL_TYPE_VXLAN); 1492 spin_unlock(&vn->sock_lock); 1493 1494 return true; 1495 } 1496 1497 static void vxlan_sock_release(struct vxlan_dev *vxlan) 1498 { 1499 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock); 1500 #if IS_ENABLED(CONFIG_IPV6) 1501 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock); 1502 1503 RCU_INIT_POINTER(vxlan->vn6_sock, NULL); 1504 #endif 1505 1506 RCU_INIT_POINTER(vxlan->vn4_sock, NULL); 1507 synchronize_net(); 1508 1509 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER) 1510 vxlan_vs_del_vnigrp(vxlan); 1511 else 1512 vxlan_vs_del_dev(vxlan); 1513 1514 if (__vxlan_sock_release_prep(sock4)) { 1515 udp_tunnel_sock_release(sock4->sock); 1516 kfree(sock4); 1517 } 1518 1519 #if IS_ENABLED(CONFIG_IPV6) 1520 if (__vxlan_sock_release_prep(sock6)) { 1521 udp_tunnel_sock_release(sock6->sock); 1522 kfree(sock6); 1523 } 1524 #endif 1525 } 1526 1527 static bool vxlan_remcsum(struct vxlanhdr *unparsed, 1528 struct sk_buff *skb, u32 vxflags) 1529 { 1530 size_t start, offset; 1531 1532 if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload) 1533 goto out; 1534 1535 start = vxlan_rco_start(unparsed->vx_vni); 1536 offset = start + vxlan_rco_offset(unparsed->vx_vni); 1537 1538 if (!pskb_may_pull(skb, offset + sizeof(u16))) 1539 return false; 1540 1541 skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset, 1542 !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL)); 1543 out: 1544 unparsed->vx_flags &= ~VXLAN_HF_RCO; 1545 unparsed->vx_vni &= VXLAN_VNI_MASK; 1546 return true; 1547 } 1548 1549 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed, 1550 struct sk_buff *skb, u32 vxflags, 1551 struct vxlan_metadata *md) 1552 { 1553 struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed; 1554 struct metadata_dst *tun_dst; 1555 1556 if (!(unparsed->vx_flags & VXLAN_HF_GBP)) 1557 goto out; 1558 1559 md->gbp = ntohs(gbp->policy_id); 1560 1561 tun_dst = (struct metadata_dst *)skb_dst(skb); 1562 if (tun_dst) { 1563 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT; 1564 tun_dst->u.tun_info.options_len = sizeof(*md); 1565 } 1566 if (gbp->dont_learn) 1567 md->gbp |= VXLAN_GBP_DONT_LEARN; 1568 1569 if (gbp->policy_applied) 1570 md->gbp |= VXLAN_GBP_POLICY_APPLIED; 1571 1572 /* In flow-based mode, GBP is carried in dst_metadata */ 1573 if (!(vxflags & VXLAN_F_COLLECT_METADATA)) 1574 skb->mark = md->gbp; 1575 out: 1576 unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS; 1577 } 1578 1579 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed, 1580 __be16 *protocol, 1581 struct sk_buff *skb, u32 vxflags) 1582 { 1583 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed; 1584 1585 /* Need to have Next Protocol set for interfaces in GPE mode. */ 1586 if (!gpe->np_applied) 1587 return false; 1588 /* "The initial version is 0. If a receiver does not support the 1589 * version indicated it MUST drop the packet. 1590 */ 1591 if (gpe->version != 0) 1592 return false; 1593 /* "When the O bit is set to 1, the packet is an OAM packet and OAM 1594 * processing MUST occur." However, we don't implement OAM 1595 * processing, thus drop the packet. 1596 */ 1597 if (gpe->oam_flag) 1598 return false; 1599 1600 *protocol = tun_p_to_eth_p(gpe->next_protocol); 1601 if (!*protocol) 1602 return false; 1603 1604 unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS; 1605 return true; 1606 } 1607 1608 static bool vxlan_set_mac(struct vxlan_dev *vxlan, 1609 struct vxlan_sock *vs, 1610 struct sk_buff *skb, __be32 vni) 1611 { 1612 union vxlan_addr saddr; 1613 u32 ifindex = skb->dev->ifindex; 1614 1615 skb_reset_mac_header(skb); 1616 skb->protocol = eth_type_trans(skb, vxlan->dev); 1617 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 1618 1619 /* Ignore packet loops (and multicast echo) */ 1620 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr)) 1621 return false; 1622 1623 /* Get address from the outer IP header */ 1624 if (vxlan_get_sk_family(vs) == AF_INET) { 1625 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr; 1626 saddr.sa.sa_family = AF_INET; 1627 #if IS_ENABLED(CONFIG_IPV6) 1628 } else { 1629 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr; 1630 saddr.sa.sa_family = AF_INET6; 1631 #endif 1632 } 1633 1634 if ((vxlan->cfg.flags & VXLAN_F_LEARN) && 1635 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni)) 1636 return false; 1637 1638 return true; 1639 } 1640 1641 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph, 1642 struct sk_buff *skb) 1643 { 1644 int err = 0; 1645 1646 if (vxlan_get_sk_family(vs) == AF_INET) 1647 err = IP_ECN_decapsulate(oiph, skb); 1648 #if IS_ENABLED(CONFIG_IPV6) 1649 else 1650 err = IP6_ECN_decapsulate(oiph, skb); 1651 #endif 1652 1653 if (unlikely(err) && log_ecn_error) { 1654 if (vxlan_get_sk_family(vs) == AF_INET) 1655 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n", 1656 &((struct iphdr *)oiph)->saddr, 1657 ((struct iphdr *)oiph)->tos); 1658 else 1659 net_info_ratelimited("non-ECT from %pI6\n", 1660 &((struct ipv6hdr *)oiph)->saddr); 1661 } 1662 return err <= 1; 1663 } 1664 1665 /* Callback from net/ipv4/udp.c to receive packets */ 1666 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb) 1667 { 1668 struct vxlan_vni_node *vninode = NULL; 1669 struct vxlan_dev *vxlan; 1670 struct vxlan_sock *vs; 1671 struct vxlanhdr unparsed; 1672 struct vxlan_metadata _md; 1673 struct vxlan_metadata *md = &_md; 1674 __be16 protocol = htons(ETH_P_TEB); 1675 bool raw_proto = false; 1676 void *oiph; 1677 __be32 vni = 0; 1678 1679 /* Need UDP and VXLAN header to be present */ 1680 if (!pskb_may_pull(skb, VXLAN_HLEN)) 1681 goto drop; 1682 1683 unparsed = *vxlan_hdr(skb); 1684 /* VNI flag always required to be set */ 1685 if (!(unparsed.vx_flags & VXLAN_HF_VNI)) { 1686 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n", 1687 ntohl(vxlan_hdr(skb)->vx_flags), 1688 ntohl(vxlan_hdr(skb)->vx_vni)); 1689 /* Return non vxlan pkt */ 1690 goto drop; 1691 } 1692 unparsed.vx_flags &= ~VXLAN_HF_VNI; 1693 unparsed.vx_vni &= ~VXLAN_VNI_MASK; 1694 1695 vs = rcu_dereference_sk_user_data(sk); 1696 if (!vs) 1697 goto drop; 1698 1699 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni); 1700 1701 vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni, &vninode); 1702 if (!vxlan) 1703 goto drop; 1704 1705 /* For backwards compatibility, only allow reserved fields to be 1706 * used by VXLAN extensions if explicitly requested. 1707 */ 1708 if (vs->flags & VXLAN_F_GPE) { 1709 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags)) 1710 goto drop; 1711 raw_proto = true; 1712 } 1713 1714 if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto, 1715 !net_eq(vxlan->net, dev_net(vxlan->dev)))) 1716 goto drop; 1717 1718 if (vs->flags & VXLAN_F_REMCSUM_RX) 1719 if (unlikely(!vxlan_remcsum(&unparsed, skb, vs->flags))) 1720 goto drop; 1721 1722 if (vxlan_collect_metadata(vs)) { 1723 struct metadata_dst *tun_dst; 1724 1725 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY, 1726 key32_to_tunnel_id(vni), sizeof(*md)); 1727 1728 if (!tun_dst) 1729 goto drop; 1730 1731 md = ip_tunnel_info_opts(&tun_dst->u.tun_info); 1732 1733 skb_dst_set(skb, (struct dst_entry *)tun_dst); 1734 } else { 1735 memset(md, 0, sizeof(*md)); 1736 } 1737 1738 if (vs->flags & VXLAN_F_GBP) 1739 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md); 1740 /* Note that GBP and GPE can never be active together. This is 1741 * ensured in vxlan_dev_configure. 1742 */ 1743 1744 if (unparsed.vx_flags || unparsed.vx_vni) { 1745 /* If there are any unprocessed flags remaining treat 1746 * this as a malformed packet. This behavior diverges from 1747 * VXLAN RFC (RFC7348) which stipulates that bits in reserved 1748 * in reserved fields are to be ignored. The approach here 1749 * maintains compatibility with previous stack code, and also 1750 * is more robust and provides a little more security in 1751 * adding extensions to VXLAN. 1752 */ 1753 goto drop; 1754 } 1755 1756 if (!raw_proto) { 1757 if (!vxlan_set_mac(vxlan, vs, skb, vni)) 1758 goto drop; 1759 } else { 1760 skb_reset_mac_header(skb); 1761 skb->dev = vxlan->dev; 1762 skb->pkt_type = PACKET_HOST; 1763 } 1764 1765 oiph = skb_network_header(skb); 1766 skb_reset_network_header(skb); 1767 1768 if (!vxlan_ecn_decapsulate(vs, oiph, skb)) { 1769 ++vxlan->dev->stats.rx_frame_errors; 1770 ++vxlan->dev->stats.rx_errors; 1771 vxlan_vnifilter_count(vxlan, vni, vninode, 1772 VXLAN_VNI_STATS_RX_ERRORS, 0); 1773 goto drop; 1774 } 1775 1776 rcu_read_lock(); 1777 1778 if (unlikely(!(vxlan->dev->flags & IFF_UP))) { 1779 rcu_read_unlock(); 1780 dev_core_stats_rx_dropped_inc(vxlan->dev); 1781 vxlan_vnifilter_count(vxlan, vni, vninode, 1782 VXLAN_VNI_STATS_RX_DROPS, 0); 1783 goto drop; 1784 } 1785 1786 dev_sw_netstats_rx_add(vxlan->dev, skb->len); 1787 vxlan_vnifilter_count(vxlan, vni, vninode, VXLAN_VNI_STATS_RX, skb->len); 1788 gro_cells_receive(&vxlan->gro_cells, skb); 1789 1790 rcu_read_unlock(); 1791 1792 return 0; 1793 1794 drop: 1795 /* Consume bad packet */ 1796 kfree_skb(skb); 1797 return 0; 1798 } 1799 1800 /* Callback from net/ipv{4,6}/udp.c to check that we have a VNI for errors */ 1801 static int vxlan_err_lookup(struct sock *sk, struct sk_buff *skb) 1802 { 1803 struct vxlan_dev *vxlan; 1804 struct vxlan_sock *vs; 1805 struct vxlanhdr *hdr; 1806 __be32 vni; 1807 1808 if (!pskb_may_pull(skb, skb_transport_offset(skb) + VXLAN_HLEN)) 1809 return -EINVAL; 1810 1811 hdr = vxlan_hdr(skb); 1812 1813 if (!(hdr->vx_flags & VXLAN_HF_VNI)) 1814 return -EINVAL; 1815 1816 vs = rcu_dereference_sk_user_data(sk); 1817 if (!vs) 1818 return -ENOENT; 1819 1820 vni = vxlan_vni(hdr->vx_vni); 1821 vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni, NULL); 1822 if (!vxlan) 1823 return -ENOENT; 1824 1825 return 0; 1826 } 1827 1828 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni) 1829 { 1830 struct vxlan_dev *vxlan = netdev_priv(dev); 1831 struct arphdr *parp; 1832 u8 *arpptr, *sha; 1833 __be32 sip, tip; 1834 struct neighbour *n; 1835 1836 if (dev->flags & IFF_NOARP) 1837 goto out; 1838 1839 if (!pskb_may_pull(skb, arp_hdr_len(dev))) { 1840 dev->stats.tx_dropped++; 1841 goto out; 1842 } 1843 parp = arp_hdr(skb); 1844 1845 if ((parp->ar_hrd != htons(ARPHRD_ETHER) && 1846 parp->ar_hrd != htons(ARPHRD_IEEE802)) || 1847 parp->ar_pro != htons(ETH_P_IP) || 1848 parp->ar_op != htons(ARPOP_REQUEST) || 1849 parp->ar_hln != dev->addr_len || 1850 parp->ar_pln != 4) 1851 goto out; 1852 arpptr = (u8 *)parp + sizeof(struct arphdr); 1853 sha = arpptr; 1854 arpptr += dev->addr_len; /* sha */ 1855 memcpy(&sip, arpptr, sizeof(sip)); 1856 arpptr += sizeof(sip); 1857 arpptr += dev->addr_len; /* tha */ 1858 memcpy(&tip, arpptr, sizeof(tip)); 1859 1860 if (ipv4_is_loopback(tip) || 1861 ipv4_is_multicast(tip)) 1862 goto out; 1863 1864 n = neigh_lookup(&arp_tbl, &tip, dev); 1865 1866 if (n) { 1867 struct vxlan_fdb *f; 1868 struct sk_buff *reply; 1869 1870 if (!(n->nud_state & NUD_CONNECTED)) { 1871 neigh_release(n); 1872 goto out; 1873 } 1874 1875 f = vxlan_find_mac(vxlan, n->ha, vni); 1876 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) { 1877 /* bridge-local neighbor */ 1878 neigh_release(n); 1879 goto out; 1880 } 1881 1882 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha, 1883 n->ha, sha); 1884 1885 neigh_release(n); 1886 1887 if (reply == NULL) 1888 goto out; 1889 1890 skb_reset_mac_header(reply); 1891 __skb_pull(reply, skb_network_offset(reply)); 1892 reply->ip_summed = CHECKSUM_UNNECESSARY; 1893 reply->pkt_type = PACKET_HOST; 1894 1895 if (netif_rx(reply) == NET_RX_DROP) { 1896 dev->stats.rx_dropped++; 1897 vxlan_vnifilter_count(vxlan, vni, NULL, 1898 VXLAN_VNI_STATS_RX_DROPS, 0); 1899 } 1900 1901 } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) { 1902 union vxlan_addr ipa = { 1903 .sin.sin_addr.s_addr = tip, 1904 .sin.sin_family = AF_INET, 1905 }; 1906 1907 vxlan_ip_miss(dev, &ipa); 1908 } 1909 out: 1910 consume_skb(skb); 1911 return NETDEV_TX_OK; 1912 } 1913 1914 #if IS_ENABLED(CONFIG_IPV6) 1915 static struct sk_buff *vxlan_na_create(struct sk_buff *request, 1916 struct neighbour *n, bool isrouter) 1917 { 1918 struct net_device *dev = request->dev; 1919 struct sk_buff *reply; 1920 struct nd_msg *ns, *na; 1921 struct ipv6hdr *pip6; 1922 u8 *daddr; 1923 int na_olen = 8; /* opt hdr + ETH_ALEN for target */ 1924 int ns_olen; 1925 int i, len; 1926 1927 if (dev == NULL || !pskb_may_pull(request, request->len)) 1928 return NULL; 1929 1930 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) + 1931 sizeof(*na) + na_olen + dev->needed_tailroom; 1932 reply = alloc_skb(len, GFP_ATOMIC); 1933 if (reply == NULL) 1934 return NULL; 1935 1936 reply->protocol = htons(ETH_P_IPV6); 1937 reply->dev = dev; 1938 skb_reserve(reply, LL_RESERVED_SPACE(request->dev)); 1939 skb_push(reply, sizeof(struct ethhdr)); 1940 skb_reset_mac_header(reply); 1941 1942 ns = (struct nd_msg *)(ipv6_hdr(request) + 1); 1943 1944 daddr = eth_hdr(request)->h_source; 1945 ns_olen = request->len - skb_network_offset(request) - 1946 sizeof(struct ipv6hdr) - sizeof(*ns); 1947 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) { 1948 if (!ns->opt[i + 1]) { 1949 kfree_skb(reply); 1950 return NULL; 1951 } 1952 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) { 1953 daddr = ns->opt + i + sizeof(struct nd_opt_hdr); 1954 break; 1955 } 1956 } 1957 1958 /* Ethernet header */ 1959 ether_addr_copy(eth_hdr(reply)->h_dest, daddr); 1960 ether_addr_copy(eth_hdr(reply)->h_source, n->ha); 1961 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6); 1962 reply->protocol = htons(ETH_P_IPV6); 1963 1964 skb_pull(reply, sizeof(struct ethhdr)); 1965 skb_reset_network_header(reply); 1966 skb_put(reply, sizeof(struct ipv6hdr)); 1967 1968 /* IPv6 header */ 1969 1970 pip6 = ipv6_hdr(reply); 1971 memset(pip6, 0, sizeof(struct ipv6hdr)); 1972 pip6->version = 6; 1973 pip6->priority = ipv6_hdr(request)->priority; 1974 pip6->nexthdr = IPPROTO_ICMPV6; 1975 pip6->hop_limit = 255; 1976 pip6->daddr = ipv6_hdr(request)->saddr; 1977 pip6->saddr = *(struct in6_addr *)n->primary_key; 1978 1979 skb_pull(reply, sizeof(struct ipv6hdr)); 1980 skb_reset_transport_header(reply); 1981 1982 /* Neighbor Advertisement */ 1983 na = skb_put_zero(reply, sizeof(*na) + na_olen); 1984 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT; 1985 na->icmph.icmp6_router = isrouter; 1986 na->icmph.icmp6_override = 1; 1987 na->icmph.icmp6_solicited = 1; 1988 na->target = ns->target; 1989 ether_addr_copy(&na->opt[2], n->ha); 1990 na->opt[0] = ND_OPT_TARGET_LL_ADDR; 1991 na->opt[1] = na_olen >> 3; 1992 1993 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr, 1994 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6, 1995 csum_partial(na, sizeof(*na)+na_olen, 0)); 1996 1997 pip6->payload_len = htons(sizeof(*na)+na_olen); 1998 1999 skb_push(reply, sizeof(struct ipv6hdr)); 2000 2001 reply->ip_summed = CHECKSUM_UNNECESSARY; 2002 2003 return reply; 2004 } 2005 2006 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni) 2007 { 2008 struct vxlan_dev *vxlan = netdev_priv(dev); 2009 const struct in6_addr *daddr; 2010 const struct ipv6hdr *iphdr; 2011 struct inet6_dev *in6_dev; 2012 struct neighbour *n; 2013 struct nd_msg *msg; 2014 2015 rcu_read_lock(); 2016 in6_dev = __in6_dev_get(dev); 2017 if (!in6_dev) 2018 goto out; 2019 2020 iphdr = ipv6_hdr(skb); 2021 daddr = &iphdr->daddr; 2022 msg = (struct nd_msg *)(iphdr + 1); 2023 2024 if (ipv6_addr_loopback(daddr) || 2025 ipv6_addr_is_multicast(&msg->target)) 2026 goto out; 2027 2028 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev); 2029 2030 if (n) { 2031 struct vxlan_fdb *f; 2032 struct sk_buff *reply; 2033 2034 if (!(n->nud_state & NUD_CONNECTED)) { 2035 neigh_release(n); 2036 goto out; 2037 } 2038 2039 f = vxlan_find_mac(vxlan, n->ha, vni); 2040 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) { 2041 /* bridge-local neighbor */ 2042 neigh_release(n); 2043 goto out; 2044 } 2045 2046 reply = vxlan_na_create(skb, n, 2047 !!(f ? f->flags & NTF_ROUTER : 0)); 2048 2049 neigh_release(n); 2050 2051 if (reply == NULL) 2052 goto out; 2053 2054 if (netif_rx(reply) == NET_RX_DROP) { 2055 dev->stats.rx_dropped++; 2056 vxlan_vnifilter_count(vxlan, vni, NULL, 2057 VXLAN_VNI_STATS_RX_DROPS, 0); 2058 } 2059 } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) { 2060 union vxlan_addr ipa = { 2061 .sin6.sin6_addr = msg->target, 2062 .sin6.sin6_family = AF_INET6, 2063 }; 2064 2065 vxlan_ip_miss(dev, &ipa); 2066 } 2067 2068 out: 2069 rcu_read_unlock(); 2070 consume_skb(skb); 2071 return NETDEV_TX_OK; 2072 } 2073 #endif 2074 2075 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb) 2076 { 2077 struct vxlan_dev *vxlan = netdev_priv(dev); 2078 struct neighbour *n; 2079 2080 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest)) 2081 return false; 2082 2083 n = NULL; 2084 switch (ntohs(eth_hdr(skb)->h_proto)) { 2085 case ETH_P_IP: 2086 { 2087 struct iphdr *pip; 2088 2089 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 2090 return false; 2091 pip = ip_hdr(skb); 2092 n = neigh_lookup(&arp_tbl, &pip->daddr, dev); 2093 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) { 2094 union vxlan_addr ipa = { 2095 .sin.sin_addr.s_addr = pip->daddr, 2096 .sin.sin_family = AF_INET, 2097 }; 2098 2099 vxlan_ip_miss(dev, &ipa); 2100 return false; 2101 } 2102 2103 break; 2104 } 2105 #if IS_ENABLED(CONFIG_IPV6) 2106 case ETH_P_IPV6: 2107 { 2108 struct ipv6hdr *pip6; 2109 2110 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr))) 2111 return false; 2112 pip6 = ipv6_hdr(skb); 2113 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev); 2114 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) { 2115 union vxlan_addr ipa = { 2116 .sin6.sin6_addr = pip6->daddr, 2117 .sin6.sin6_family = AF_INET6, 2118 }; 2119 2120 vxlan_ip_miss(dev, &ipa); 2121 return false; 2122 } 2123 2124 break; 2125 } 2126 #endif 2127 default: 2128 return false; 2129 } 2130 2131 if (n) { 2132 bool diff; 2133 2134 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha); 2135 if (diff) { 2136 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest, 2137 dev->addr_len); 2138 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len); 2139 } 2140 neigh_release(n); 2141 return diff; 2142 } 2143 2144 return false; 2145 } 2146 2147 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags, 2148 struct vxlan_metadata *md) 2149 { 2150 struct vxlanhdr_gbp *gbp; 2151 2152 if (!md->gbp) 2153 return; 2154 2155 gbp = (struct vxlanhdr_gbp *)vxh; 2156 vxh->vx_flags |= VXLAN_HF_GBP; 2157 2158 if (md->gbp & VXLAN_GBP_DONT_LEARN) 2159 gbp->dont_learn = 1; 2160 2161 if (md->gbp & VXLAN_GBP_POLICY_APPLIED) 2162 gbp->policy_applied = 1; 2163 2164 gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK); 2165 } 2166 2167 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags, 2168 __be16 protocol) 2169 { 2170 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh; 2171 2172 gpe->np_applied = 1; 2173 gpe->next_protocol = tun_p_from_eth_p(protocol); 2174 if (!gpe->next_protocol) 2175 return -EPFNOSUPPORT; 2176 return 0; 2177 } 2178 2179 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst, 2180 int iphdr_len, __be32 vni, 2181 struct vxlan_metadata *md, u32 vxflags, 2182 bool udp_sum) 2183 { 2184 struct vxlanhdr *vxh; 2185 int min_headroom; 2186 int err; 2187 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL; 2188 __be16 inner_protocol = htons(ETH_P_TEB); 2189 2190 if ((vxflags & VXLAN_F_REMCSUM_TX) && 2191 skb->ip_summed == CHECKSUM_PARTIAL) { 2192 int csum_start = skb_checksum_start_offset(skb); 2193 2194 if (csum_start <= VXLAN_MAX_REMCSUM_START && 2195 !(csum_start & VXLAN_RCO_SHIFT_MASK) && 2196 (skb->csum_offset == offsetof(struct udphdr, check) || 2197 skb->csum_offset == offsetof(struct tcphdr, check))) 2198 type |= SKB_GSO_TUNNEL_REMCSUM; 2199 } 2200 2201 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len 2202 + VXLAN_HLEN + iphdr_len; 2203 2204 /* Need space for new headers (invalidates iph ptr) */ 2205 err = skb_cow_head(skb, min_headroom); 2206 if (unlikely(err)) 2207 return err; 2208 2209 err = iptunnel_handle_offloads(skb, type); 2210 if (err) 2211 return err; 2212 2213 vxh = __skb_push(skb, sizeof(*vxh)); 2214 vxh->vx_flags = VXLAN_HF_VNI; 2215 vxh->vx_vni = vxlan_vni_field(vni); 2216 2217 if (type & SKB_GSO_TUNNEL_REMCSUM) { 2218 unsigned int start; 2219 2220 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr); 2221 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset); 2222 vxh->vx_flags |= VXLAN_HF_RCO; 2223 2224 if (!skb_is_gso(skb)) { 2225 skb->ip_summed = CHECKSUM_NONE; 2226 skb->encapsulation = 0; 2227 } 2228 } 2229 2230 if (vxflags & VXLAN_F_GBP) 2231 vxlan_build_gbp_hdr(vxh, vxflags, md); 2232 if (vxflags & VXLAN_F_GPE) { 2233 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol); 2234 if (err < 0) 2235 return err; 2236 inner_protocol = skb->protocol; 2237 } 2238 2239 skb_set_inner_protocol(skb, inner_protocol); 2240 return 0; 2241 } 2242 2243 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev, 2244 struct vxlan_sock *sock4, 2245 struct sk_buff *skb, int oif, u8 tos, 2246 __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport, 2247 struct dst_cache *dst_cache, 2248 const struct ip_tunnel_info *info) 2249 { 2250 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 2251 struct rtable *rt = NULL; 2252 struct flowi4 fl4; 2253 2254 if (!sock4) 2255 return ERR_PTR(-EIO); 2256 2257 if (tos && !info) 2258 use_cache = false; 2259 if (use_cache) { 2260 rt = dst_cache_get_ip4(dst_cache, saddr); 2261 if (rt) 2262 return rt; 2263 } 2264 2265 memset(&fl4, 0, sizeof(fl4)); 2266 fl4.flowi4_oif = oif; 2267 fl4.flowi4_tos = RT_TOS(tos); 2268 fl4.flowi4_mark = skb->mark; 2269 fl4.flowi4_proto = IPPROTO_UDP; 2270 fl4.daddr = daddr; 2271 fl4.saddr = *saddr; 2272 fl4.fl4_dport = dport; 2273 fl4.fl4_sport = sport; 2274 2275 rt = ip_route_output_key(vxlan->net, &fl4); 2276 if (!IS_ERR(rt)) { 2277 if (rt->dst.dev == dev) { 2278 netdev_dbg(dev, "circular route to %pI4\n", &daddr); 2279 ip_rt_put(rt); 2280 return ERR_PTR(-ELOOP); 2281 } 2282 2283 *saddr = fl4.saddr; 2284 if (use_cache) 2285 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr); 2286 } else { 2287 netdev_dbg(dev, "no route to %pI4\n", &daddr); 2288 return ERR_PTR(-ENETUNREACH); 2289 } 2290 return rt; 2291 } 2292 2293 #if IS_ENABLED(CONFIG_IPV6) 2294 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan, 2295 struct net_device *dev, 2296 struct vxlan_sock *sock6, 2297 struct sk_buff *skb, int oif, u8 tos, 2298 __be32 label, 2299 const struct in6_addr *daddr, 2300 struct in6_addr *saddr, 2301 __be16 dport, __be16 sport, 2302 struct dst_cache *dst_cache, 2303 const struct ip_tunnel_info *info) 2304 { 2305 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 2306 struct dst_entry *ndst; 2307 struct flowi6 fl6; 2308 2309 if (!sock6) 2310 return ERR_PTR(-EIO); 2311 2312 if (tos && !info) 2313 use_cache = false; 2314 if (use_cache) { 2315 ndst = dst_cache_get_ip6(dst_cache, saddr); 2316 if (ndst) 2317 return ndst; 2318 } 2319 2320 memset(&fl6, 0, sizeof(fl6)); 2321 fl6.flowi6_oif = oif; 2322 fl6.daddr = *daddr; 2323 fl6.saddr = *saddr; 2324 fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label); 2325 fl6.flowi6_mark = skb->mark; 2326 fl6.flowi6_proto = IPPROTO_UDP; 2327 fl6.fl6_dport = dport; 2328 fl6.fl6_sport = sport; 2329 2330 ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk, 2331 &fl6, NULL); 2332 if (IS_ERR(ndst)) { 2333 netdev_dbg(dev, "no route to %pI6\n", daddr); 2334 return ERR_PTR(-ENETUNREACH); 2335 } 2336 2337 if (unlikely(ndst->dev == dev)) { 2338 netdev_dbg(dev, "circular route to %pI6\n", daddr); 2339 dst_release(ndst); 2340 return ERR_PTR(-ELOOP); 2341 } 2342 2343 *saddr = fl6.saddr; 2344 if (use_cache) 2345 dst_cache_set_ip6(dst_cache, ndst, saddr); 2346 return ndst; 2347 } 2348 #endif 2349 2350 /* Bypass encapsulation if the destination is local */ 2351 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan, 2352 struct vxlan_dev *dst_vxlan, __be32 vni, 2353 bool snoop) 2354 { 2355 struct pcpu_sw_netstats *tx_stats, *rx_stats; 2356 union vxlan_addr loopback; 2357 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip; 2358 struct net_device *dev; 2359 int len = skb->len; 2360 2361 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats); 2362 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats); 2363 skb->pkt_type = PACKET_HOST; 2364 skb->encapsulation = 0; 2365 skb->dev = dst_vxlan->dev; 2366 __skb_pull(skb, skb_network_offset(skb)); 2367 2368 if (remote_ip->sa.sa_family == AF_INET) { 2369 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK); 2370 loopback.sa.sa_family = AF_INET; 2371 #if IS_ENABLED(CONFIG_IPV6) 2372 } else { 2373 loopback.sin6.sin6_addr = in6addr_loopback; 2374 loopback.sa.sa_family = AF_INET6; 2375 #endif 2376 } 2377 2378 rcu_read_lock(); 2379 dev = skb->dev; 2380 if (unlikely(!(dev->flags & IFF_UP))) { 2381 kfree_skb(skb); 2382 goto drop; 2383 } 2384 2385 if ((dst_vxlan->cfg.flags & VXLAN_F_LEARN) && snoop) 2386 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni); 2387 2388 u64_stats_update_begin(&tx_stats->syncp); 2389 tx_stats->tx_packets++; 2390 tx_stats->tx_bytes += len; 2391 u64_stats_update_end(&tx_stats->syncp); 2392 vxlan_vnifilter_count(src_vxlan, vni, NULL, VXLAN_VNI_STATS_TX, len); 2393 2394 if (__netif_rx(skb) == NET_RX_SUCCESS) { 2395 u64_stats_update_begin(&rx_stats->syncp); 2396 rx_stats->rx_packets++; 2397 rx_stats->rx_bytes += len; 2398 u64_stats_update_end(&rx_stats->syncp); 2399 vxlan_vnifilter_count(dst_vxlan, vni, NULL, VXLAN_VNI_STATS_RX, 2400 len); 2401 } else { 2402 drop: 2403 dev->stats.rx_dropped++; 2404 vxlan_vnifilter_count(dst_vxlan, vni, NULL, 2405 VXLAN_VNI_STATS_RX_DROPS, 0); 2406 } 2407 rcu_read_unlock(); 2408 } 2409 2410 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev, 2411 struct vxlan_dev *vxlan, 2412 union vxlan_addr *daddr, 2413 __be16 dst_port, int dst_ifindex, __be32 vni, 2414 struct dst_entry *dst, 2415 u32 rt_flags) 2416 { 2417 #if IS_ENABLED(CONFIG_IPV6) 2418 /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of 2419 * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple 2420 * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry. 2421 */ 2422 BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL); 2423 #endif 2424 /* Bypass encapsulation if the destination is local */ 2425 if (rt_flags & RTCF_LOCAL && 2426 !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) { 2427 struct vxlan_dev *dst_vxlan; 2428 2429 dst_release(dst); 2430 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni, 2431 daddr->sa.sa_family, dst_port, 2432 vxlan->cfg.flags); 2433 if (!dst_vxlan) { 2434 dev->stats.tx_errors++; 2435 vxlan_vnifilter_count(vxlan, vni, NULL, 2436 VXLAN_VNI_STATS_TX_ERRORS, 0); 2437 kfree_skb(skb); 2438 2439 return -ENOENT; 2440 } 2441 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni, true); 2442 return 1; 2443 } 2444 2445 return 0; 2446 } 2447 2448 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev, 2449 __be32 default_vni, struct vxlan_rdst *rdst, 2450 bool did_rsc) 2451 { 2452 struct dst_cache *dst_cache; 2453 struct ip_tunnel_info *info; 2454 struct vxlan_dev *vxlan = netdev_priv(dev); 2455 const struct iphdr *old_iph = ip_hdr(skb); 2456 union vxlan_addr *dst; 2457 union vxlan_addr remote_ip, local_ip; 2458 struct vxlan_metadata _md; 2459 struct vxlan_metadata *md = &_md; 2460 unsigned int pkt_len = skb->len; 2461 __be16 src_port = 0, dst_port; 2462 struct dst_entry *ndst = NULL; 2463 __u8 tos, ttl; 2464 int ifindex; 2465 int err; 2466 u32 flags = vxlan->cfg.flags; 2467 bool udp_sum = false; 2468 bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev)); 2469 __be32 vni = 0; 2470 #if IS_ENABLED(CONFIG_IPV6) 2471 __be32 label; 2472 #endif 2473 2474 info = skb_tunnel_info(skb); 2475 2476 if (rdst) { 2477 dst = &rdst->remote_ip; 2478 if (vxlan_addr_any(dst)) { 2479 if (did_rsc) { 2480 /* short-circuited back to local bridge */ 2481 vxlan_encap_bypass(skb, vxlan, vxlan, 2482 default_vni, true); 2483 return; 2484 } 2485 goto drop; 2486 } 2487 2488 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port; 2489 vni = (rdst->remote_vni) ? : default_vni; 2490 ifindex = rdst->remote_ifindex; 2491 local_ip = vxlan->cfg.saddr; 2492 dst_cache = &rdst->dst_cache; 2493 md->gbp = skb->mark; 2494 if (flags & VXLAN_F_TTL_INHERIT) { 2495 ttl = ip_tunnel_get_ttl(old_iph, skb); 2496 } else { 2497 ttl = vxlan->cfg.ttl; 2498 if (!ttl && vxlan_addr_multicast(dst)) 2499 ttl = 1; 2500 } 2501 2502 tos = vxlan->cfg.tos; 2503 if (tos == 1) 2504 tos = ip_tunnel_get_dsfield(old_iph, skb); 2505 2506 if (dst->sa.sa_family == AF_INET) 2507 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX); 2508 else 2509 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX); 2510 #if IS_ENABLED(CONFIG_IPV6) 2511 label = vxlan->cfg.label; 2512 #endif 2513 } else { 2514 if (!info) { 2515 WARN_ONCE(1, "%s: Missing encapsulation instructions\n", 2516 dev->name); 2517 goto drop; 2518 } 2519 remote_ip.sa.sa_family = ip_tunnel_info_af(info); 2520 if (remote_ip.sa.sa_family == AF_INET) { 2521 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst; 2522 local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src; 2523 } else { 2524 remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst; 2525 local_ip.sin6.sin6_addr = info->key.u.ipv6.src; 2526 } 2527 dst = &remote_ip; 2528 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port; 2529 vni = tunnel_id_to_key32(info->key.tun_id); 2530 ifindex = 0; 2531 dst_cache = &info->dst_cache; 2532 if (info->key.tun_flags & TUNNEL_VXLAN_OPT) { 2533 if (info->options_len < sizeof(*md)) 2534 goto drop; 2535 md = ip_tunnel_info_opts(info); 2536 } 2537 ttl = info->key.ttl; 2538 tos = info->key.tos; 2539 #if IS_ENABLED(CONFIG_IPV6) 2540 label = info->key.label; 2541 #endif 2542 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM); 2543 } 2544 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min, 2545 vxlan->cfg.port_max, true); 2546 2547 rcu_read_lock(); 2548 if (dst->sa.sa_family == AF_INET) { 2549 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock); 2550 struct rtable *rt; 2551 __be16 df = 0; 2552 2553 if (!ifindex) 2554 ifindex = sock4->sock->sk->sk_bound_dev_if; 2555 2556 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos, 2557 dst->sin.sin_addr.s_addr, 2558 &local_ip.sin.sin_addr.s_addr, 2559 dst_port, src_port, 2560 dst_cache, info); 2561 if (IS_ERR(rt)) { 2562 err = PTR_ERR(rt); 2563 goto tx_error; 2564 } 2565 2566 if (!info) { 2567 /* Bypass encapsulation if the destination is local */ 2568 err = encap_bypass_if_local(skb, dev, vxlan, dst, 2569 dst_port, ifindex, vni, 2570 &rt->dst, rt->rt_flags); 2571 if (err) 2572 goto out_unlock; 2573 2574 if (vxlan->cfg.df == VXLAN_DF_SET) { 2575 df = htons(IP_DF); 2576 } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) { 2577 struct ethhdr *eth = eth_hdr(skb); 2578 2579 if (ntohs(eth->h_proto) == ETH_P_IPV6 || 2580 (ntohs(eth->h_proto) == ETH_P_IP && 2581 old_iph->frag_off & htons(IP_DF))) 2582 df = htons(IP_DF); 2583 } 2584 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) { 2585 df = htons(IP_DF); 2586 } 2587 2588 ndst = &rt->dst; 2589 err = skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM, 2590 netif_is_any_bridge_port(dev)); 2591 if (err < 0) { 2592 goto tx_error; 2593 } else if (err) { 2594 if (info) { 2595 struct ip_tunnel_info *unclone; 2596 struct in_addr src, dst; 2597 2598 unclone = skb_tunnel_info_unclone(skb); 2599 if (unlikely(!unclone)) 2600 goto tx_error; 2601 2602 src = remote_ip.sin.sin_addr; 2603 dst = local_ip.sin.sin_addr; 2604 unclone->key.u.ipv4.src = src.s_addr; 2605 unclone->key.u.ipv4.dst = dst.s_addr; 2606 } 2607 vxlan_encap_bypass(skb, vxlan, vxlan, vni, false); 2608 dst_release(ndst); 2609 goto out_unlock; 2610 } 2611 2612 tos = ip_tunnel_ecn_encap(tos, old_iph, skb); 2613 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst); 2614 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr), 2615 vni, md, flags, udp_sum); 2616 if (err < 0) 2617 goto tx_error; 2618 2619 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr, 2620 dst->sin.sin_addr.s_addr, tos, ttl, df, 2621 src_port, dst_port, xnet, !udp_sum); 2622 #if IS_ENABLED(CONFIG_IPV6) 2623 } else { 2624 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock); 2625 2626 if (!ifindex) 2627 ifindex = sock6->sock->sk->sk_bound_dev_if; 2628 2629 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos, 2630 label, &dst->sin6.sin6_addr, 2631 &local_ip.sin6.sin6_addr, 2632 dst_port, src_port, 2633 dst_cache, info); 2634 if (IS_ERR(ndst)) { 2635 err = PTR_ERR(ndst); 2636 ndst = NULL; 2637 goto tx_error; 2638 } 2639 2640 if (!info) { 2641 u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags; 2642 2643 err = encap_bypass_if_local(skb, dev, vxlan, dst, 2644 dst_port, ifindex, vni, 2645 ndst, rt6i_flags); 2646 if (err) 2647 goto out_unlock; 2648 } 2649 2650 err = skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM, 2651 netif_is_any_bridge_port(dev)); 2652 if (err < 0) { 2653 goto tx_error; 2654 } else if (err) { 2655 if (info) { 2656 struct ip_tunnel_info *unclone; 2657 struct in6_addr src, dst; 2658 2659 unclone = skb_tunnel_info_unclone(skb); 2660 if (unlikely(!unclone)) 2661 goto tx_error; 2662 2663 src = remote_ip.sin6.sin6_addr; 2664 dst = local_ip.sin6.sin6_addr; 2665 unclone->key.u.ipv6.src = src; 2666 unclone->key.u.ipv6.dst = dst; 2667 } 2668 2669 vxlan_encap_bypass(skb, vxlan, vxlan, vni, false); 2670 dst_release(ndst); 2671 goto out_unlock; 2672 } 2673 2674 tos = ip_tunnel_ecn_encap(tos, old_iph, skb); 2675 ttl = ttl ? : ip6_dst_hoplimit(ndst); 2676 skb_scrub_packet(skb, xnet); 2677 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr), 2678 vni, md, flags, udp_sum); 2679 if (err < 0) 2680 goto tx_error; 2681 2682 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev, 2683 &local_ip.sin6.sin6_addr, 2684 &dst->sin6.sin6_addr, tos, ttl, 2685 label, src_port, dst_port, !udp_sum); 2686 #endif 2687 } 2688 vxlan_vnifilter_count(vxlan, vni, NULL, VXLAN_VNI_STATS_TX, pkt_len); 2689 out_unlock: 2690 rcu_read_unlock(); 2691 return; 2692 2693 drop: 2694 dev->stats.tx_dropped++; 2695 vxlan_vnifilter_count(vxlan, vni, NULL, VXLAN_VNI_STATS_TX_DROPS, 0); 2696 dev_kfree_skb(skb); 2697 return; 2698 2699 tx_error: 2700 rcu_read_unlock(); 2701 if (err == -ELOOP) 2702 dev->stats.collisions++; 2703 else if (err == -ENETUNREACH) 2704 dev->stats.tx_carrier_errors++; 2705 dst_release(ndst); 2706 dev->stats.tx_errors++; 2707 vxlan_vnifilter_count(vxlan, vni, NULL, VXLAN_VNI_STATS_TX_ERRORS, 0); 2708 kfree_skb(skb); 2709 } 2710 2711 static void vxlan_xmit_nh(struct sk_buff *skb, struct net_device *dev, 2712 struct vxlan_fdb *f, __be32 vni, bool did_rsc) 2713 { 2714 struct vxlan_rdst nh_rdst; 2715 struct nexthop *nh; 2716 bool do_xmit; 2717 u32 hash; 2718 2719 memset(&nh_rdst, 0, sizeof(struct vxlan_rdst)); 2720 hash = skb_get_hash(skb); 2721 2722 rcu_read_lock(); 2723 nh = rcu_dereference(f->nh); 2724 if (!nh) { 2725 rcu_read_unlock(); 2726 goto drop; 2727 } 2728 do_xmit = vxlan_fdb_nh_path_select(nh, hash, &nh_rdst); 2729 rcu_read_unlock(); 2730 2731 if (likely(do_xmit)) 2732 vxlan_xmit_one(skb, dev, vni, &nh_rdst, did_rsc); 2733 else 2734 goto drop; 2735 2736 return; 2737 2738 drop: 2739 dev->stats.tx_dropped++; 2740 vxlan_vnifilter_count(netdev_priv(dev), vni, NULL, 2741 VXLAN_VNI_STATS_TX_DROPS, 0); 2742 dev_kfree_skb(skb); 2743 } 2744 2745 /* Transmit local packets over Vxlan 2746 * 2747 * Outer IP header inherits ECN and DF from inner header. 2748 * Outer UDP destination is the VXLAN assigned port. 2749 * source port is based on hash of flow 2750 */ 2751 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev) 2752 { 2753 struct vxlan_dev *vxlan = netdev_priv(dev); 2754 struct vxlan_rdst *rdst, *fdst = NULL; 2755 const struct ip_tunnel_info *info; 2756 bool did_rsc = false; 2757 struct vxlan_fdb *f; 2758 struct ethhdr *eth; 2759 __be32 vni = 0; 2760 2761 info = skb_tunnel_info(skb); 2762 2763 skb_reset_mac_header(skb); 2764 2765 if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) { 2766 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE && 2767 info->mode & IP_TUNNEL_INFO_TX) { 2768 vni = tunnel_id_to_key32(info->key.tun_id); 2769 } else { 2770 if (info && info->mode & IP_TUNNEL_INFO_TX) 2771 vxlan_xmit_one(skb, dev, vni, NULL, false); 2772 else 2773 kfree_skb(skb); 2774 return NETDEV_TX_OK; 2775 } 2776 } 2777 2778 if (vxlan->cfg.flags & VXLAN_F_PROXY) { 2779 eth = eth_hdr(skb); 2780 if (ntohs(eth->h_proto) == ETH_P_ARP) 2781 return arp_reduce(dev, skb, vni); 2782 #if IS_ENABLED(CONFIG_IPV6) 2783 else if (ntohs(eth->h_proto) == ETH_P_IPV6 && 2784 pskb_may_pull(skb, sizeof(struct ipv6hdr) + 2785 sizeof(struct nd_msg)) && 2786 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) { 2787 struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1); 2788 2789 if (m->icmph.icmp6_code == 0 && 2790 m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) 2791 return neigh_reduce(dev, skb, vni); 2792 } 2793 #endif 2794 } 2795 2796 eth = eth_hdr(skb); 2797 f = vxlan_find_mac(vxlan, eth->h_dest, vni); 2798 did_rsc = false; 2799 2800 if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) && 2801 (ntohs(eth->h_proto) == ETH_P_IP || 2802 ntohs(eth->h_proto) == ETH_P_IPV6)) { 2803 did_rsc = route_shortcircuit(dev, skb); 2804 if (did_rsc) 2805 f = vxlan_find_mac(vxlan, eth->h_dest, vni); 2806 } 2807 2808 if (f == NULL) { 2809 f = vxlan_find_mac(vxlan, all_zeros_mac, vni); 2810 if (f == NULL) { 2811 if ((vxlan->cfg.flags & VXLAN_F_L2MISS) && 2812 !is_multicast_ether_addr(eth->h_dest)) 2813 vxlan_fdb_miss(vxlan, eth->h_dest); 2814 2815 dev->stats.tx_dropped++; 2816 vxlan_vnifilter_count(vxlan, vni, NULL, 2817 VXLAN_VNI_STATS_TX_DROPS, 0); 2818 kfree_skb(skb); 2819 return NETDEV_TX_OK; 2820 } 2821 } 2822 2823 if (rcu_access_pointer(f->nh)) { 2824 vxlan_xmit_nh(skb, dev, f, 2825 (vni ? : vxlan->default_dst.remote_vni), did_rsc); 2826 } else { 2827 list_for_each_entry_rcu(rdst, &f->remotes, list) { 2828 struct sk_buff *skb1; 2829 2830 if (!fdst) { 2831 fdst = rdst; 2832 continue; 2833 } 2834 skb1 = skb_clone(skb, GFP_ATOMIC); 2835 if (skb1) 2836 vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc); 2837 } 2838 if (fdst) 2839 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc); 2840 else 2841 kfree_skb(skb); 2842 } 2843 2844 return NETDEV_TX_OK; 2845 } 2846 2847 /* Walk the forwarding table and purge stale entries */ 2848 static void vxlan_cleanup(struct timer_list *t) 2849 { 2850 struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer); 2851 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL; 2852 unsigned int h; 2853 2854 if (!netif_running(vxlan->dev)) 2855 return; 2856 2857 for (h = 0; h < FDB_HASH_SIZE; ++h) { 2858 struct hlist_node *p, *n; 2859 2860 spin_lock(&vxlan->hash_lock[h]); 2861 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) { 2862 struct vxlan_fdb *f 2863 = container_of(p, struct vxlan_fdb, hlist); 2864 unsigned long timeout; 2865 2866 if (f->state & (NUD_PERMANENT | NUD_NOARP)) 2867 continue; 2868 2869 if (f->flags & NTF_EXT_LEARNED) 2870 continue; 2871 2872 timeout = f->used + vxlan->cfg.age_interval * HZ; 2873 if (time_before_eq(timeout, jiffies)) { 2874 netdev_dbg(vxlan->dev, 2875 "garbage collect %pM\n", 2876 f->eth_addr); 2877 f->state = NUD_STALE; 2878 vxlan_fdb_destroy(vxlan, f, true, true); 2879 } else if (time_before(timeout, next_timer)) 2880 next_timer = timeout; 2881 } 2882 spin_unlock(&vxlan->hash_lock[h]); 2883 } 2884 2885 mod_timer(&vxlan->age_timer, next_timer); 2886 } 2887 2888 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan) 2889 { 2890 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id); 2891 2892 spin_lock(&vn->sock_lock); 2893 hlist_del_init_rcu(&vxlan->hlist4.hlist); 2894 #if IS_ENABLED(CONFIG_IPV6) 2895 hlist_del_init_rcu(&vxlan->hlist6.hlist); 2896 #endif 2897 spin_unlock(&vn->sock_lock); 2898 } 2899 2900 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan, 2901 struct vxlan_dev_node *node) 2902 { 2903 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id); 2904 __be32 vni = vxlan->default_dst.remote_vni; 2905 2906 node->vxlan = vxlan; 2907 spin_lock(&vn->sock_lock); 2908 hlist_add_head_rcu(&node->hlist, vni_head(vs, vni)); 2909 spin_unlock(&vn->sock_lock); 2910 } 2911 2912 /* Setup stats when device is created */ 2913 static int vxlan_init(struct net_device *dev) 2914 { 2915 struct vxlan_dev *vxlan = netdev_priv(dev); 2916 int err; 2917 2918 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER) 2919 vxlan_vnigroup_init(vxlan); 2920 2921 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 2922 if (!dev->tstats) 2923 return -ENOMEM; 2924 2925 err = gro_cells_init(&vxlan->gro_cells, dev); 2926 if (err) { 2927 free_percpu(dev->tstats); 2928 return err; 2929 } 2930 2931 return 0; 2932 } 2933 2934 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni) 2935 { 2936 struct vxlan_fdb *f; 2937 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, vni); 2938 2939 spin_lock_bh(&vxlan->hash_lock[hash_index]); 2940 f = __vxlan_find_mac(vxlan, all_zeros_mac, vni); 2941 if (f) 2942 vxlan_fdb_destroy(vxlan, f, true, true); 2943 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 2944 } 2945 2946 static void vxlan_uninit(struct net_device *dev) 2947 { 2948 struct vxlan_dev *vxlan = netdev_priv(dev); 2949 2950 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER) 2951 vxlan_vnigroup_uninit(vxlan); 2952 2953 gro_cells_destroy(&vxlan->gro_cells); 2954 2955 vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni); 2956 2957 free_percpu(dev->tstats); 2958 } 2959 2960 /* Start ageing timer and join group when device is brought up */ 2961 static int vxlan_open(struct net_device *dev) 2962 { 2963 struct vxlan_dev *vxlan = netdev_priv(dev); 2964 int ret; 2965 2966 ret = vxlan_sock_add(vxlan); 2967 if (ret < 0) 2968 return ret; 2969 2970 ret = vxlan_multicast_join(vxlan); 2971 if (ret) { 2972 vxlan_sock_release(vxlan); 2973 return ret; 2974 } 2975 2976 if (vxlan->cfg.age_interval) 2977 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL); 2978 2979 return ret; 2980 } 2981 2982 /* Purge the forwarding table */ 2983 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all) 2984 { 2985 unsigned int h; 2986 2987 for (h = 0; h < FDB_HASH_SIZE; ++h) { 2988 struct hlist_node *p, *n; 2989 2990 spin_lock_bh(&vxlan->hash_lock[h]); 2991 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) { 2992 struct vxlan_fdb *f 2993 = container_of(p, struct vxlan_fdb, hlist); 2994 if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP))) 2995 continue; 2996 /* the all_zeros_mac entry is deleted at vxlan_uninit */ 2997 if (is_zero_ether_addr(f->eth_addr) && 2998 f->vni == vxlan->cfg.vni) 2999 continue; 3000 vxlan_fdb_destroy(vxlan, f, true, true); 3001 } 3002 spin_unlock_bh(&vxlan->hash_lock[h]); 3003 } 3004 } 3005 3006 /* Cleanup timer and forwarding table on shutdown */ 3007 static int vxlan_stop(struct net_device *dev) 3008 { 3009 struct vxlan_dev *vxlan = netdev_priv(dev); 3010 3011 vxlan_multicast_leave(vxlan); 3012 3013 del_timer_sync(&vxlan->age_timer); 3014 3015 vxlan_flush(vxlan, false); 3016 vxlan_sock_release(vxlan); 3017 3018 return 0; 3019 } 3020 3021 /* Stub, nothing needs to be done. */ 3022 static void vxlan_set_multicast_list(struct net_device *dev) 3023 { 3024 } 3025 3026 static int vxlan_change_mtu(struct net_device *dev, int new_mtu) 3027 { 3028 struct vxlan_dev *vxlan = netdev_priv(dev); 3029 struct vxlan_rdst *dst = &vxlan->default_dst; 3030 struct net_device *lowerdev = __dev_get_by_index(vxlan->net, 3031 dst->remote_ifindex); 3032 bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6); 3033 3034 /* This check is different than dev->max_mtu, because it looks at 3035 * the lowerdev->mtu, rather than the static dev->max_mtu 3036 */ 3037 if (lowerdev) { 3038 int max_mtu = lowerdev->mtu - 3039 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM); 3040 if (new_mtu > max_mtu) 3041 return -EINVAL; 3042 } 3043 3044 dev->mtu = new_mtu; 3045 return 0; 3046 } 3047 3048 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 3049 { 3050 struct vxlan_dev *vxlan = netdev_priv(dev); 3051 struct ip_tunnel_info *info = skb_tunnel_info(skb); 3052 __be16 sport, dport; 3053 3054 sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min, 3055 vxlan->cfg.port_max, true); 3056 dport = info->key.tp_dst ? : vxlan->cfg.dst_port; 3057 3058 if (ip_tunnel_info_af(info) == AF_INET) { 3059 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock); 3060 struct rtable *rt; 3061 3062 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos, 3063 info->key.u.ipv4.dst, 3064 &info->key.u.ipv4.src, dport, sport, 3065 &info->dst_cache, info); 3066 if (IS_ERR(rt)) 3067 return PTR_ERR(rt); 3068 ip_rt_put(rt); 3069 } else { 3070 #if IS_ENABLED(CONFIG_IPV6) 3071 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock); 3072 struct dst_entry *ndst; 3073 3074 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos, 3075 info->key.label, &info->key.u.ipv6.dst, 3076 &info->key.u.ipv6.src, dport, sport, 3077 &info->dst_cache, info); 3078 if (IS_ERR(ndst)) 3079 return PTR_ERR(ndst); 3080 dst_release(ndst); 3081 #else /* !CONFIG_IPV6 */ 3082 return -EPFNOSUPPORT; 3083 #endif 3084 } 3085 info->key.tp_src = sport; 3086 info->key.tp_dst = dport; 3087 return 0; 3088 } 3089 3090 static const struct net_device_ops vxlan_netdev_ether_ops = { 3091 .ndo_init = vxlan_init, 3092 .ndo_uninit = vxlan_uninit, 3093 .ndo_open = vxlan_open, 3094 .ndo_stop = vxlan_stop, 3095 .ndo_start_xmit = vxlan_xmit, 3096 .ndo_get_stats64 = dev_get_tstats64, 3097 .ndo_set_rx_mode = vxlan_set_multicast_list, 3098 .ndo_change_mtu = vxlan_change_mtu, 3099 .ndo_validate_addr = eth_validate_addr, 3100 .ndo_set_mac_address = eth_mac_addr, 3101 .ndo_fdb_add = vxlan_fdb_add, 3102 .ndo_fdb_del = vxlan_fdb_delete, 3103 .ndo_fdb_dump = vxlan_fdb_dump, 3104 .ndo_fdb_get = vxlan_fdb_get, 3105 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst, 3106 }; 3107 3108 static const struct net_device_ops vxlan_netdev_raw_ops = { 3109 .ndo_init = vxlan_init, 3110 .ndo_uninit = vxlan_uninit, 3111 .ndo_open = vxlan_open, 3112 .ndo_stop = vxlan_stop, 3113 .ndo_start_xmit = vxlan_xmit, 3114 .ndo_get_stats64 = dev_get_tstats64, 3115 .ndo_change_mtu = vxlan_change_mtu, 3116 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst, 3117 }; 3118 3119 /* Info for udev, that this is a virtual tunnel endpoint */ 3120 static struct device_type vxlan_type = { 3121 .name = "vxlan", 3122 }; 3123 3124 /* Calls the ndo_udp_tunnel_add of the caller in order to 3125 * supply the listening VXLAN udp ports. Callers are expected 3126 * to implement the ndo_udp_tunnel_add. 3127 */ 3128 static void vxlan_offload_rx_ports(struct net_device *dev, bool push) 3129 { 3130 struct vxlan_sock *vs; 3131 struct net *net = dev_net(dev); 3132 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 3133 unsigned int i; 3134 3135 spin_lock(&vn->sock_lock); 3136 for (i = 0; i < PORT_HASH_SIZE; ++i) { 3137 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) { 3138 unsigned short type; 3139 3140 if (vs->flags & VXLAN_F_GPE) 3141 type = UDP_TUNNEL_TYPE_VXLAN_GPE; 3142 else 3143 type = UDP_TUNNEL_TYPE_VXLAN; 3144 3145 if (push) 3146 udp_tunnel_push_rx_port(dev, vs->sock, type); 3147 else 3148 udp_tunnel_drop_rx_port(dev, vs->sock, type); 3149 } 3150 } 3151 spin_unlock(&vn->sock_lock); 3152 } 3153 3154 /* Initialize the device structure. */ 3155 static void vxlan_setup(struct net_device *dev) 3156 { 3157 struct vxlan_dev *vxlan = netdev_priv(dev); 3158 unsigned int h; 3159 3160 eth_hw_addr_random(dev); 3161 ether_setup(dev); 3162 3163 dev->needs_free_netdev = true; 3164 SET_NETDEV_DEVTYPE(dev, &vxlan_type); 3165 3166 dev->features |= NETIF_F_LLTX; 3167 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST; 3168 dev->features |= NETIF_F_RXCSUM; 3169 dev->features |= NETIF_F_GSO_SOFTWARE; 3170 3171 dev->vlan_features = dev->features; 3172 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST; 3173 dev->hw_features |= NETIF_F_RXCSUM; 3174 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 3175 netif_keep_dst(dev); 3176 dev->priv_flags |= IFF_NO_QUEUE | IFF_CHANGE_PROTO_DOWN; 3177 3178 /* MTU range: 68 - 65535 */ 3179 dev->min_mtu = ETH_MIN_MTU; 3180 dev->max_mtu = ETH_MAX_MTU; 3181 3182 INIT_LIST_HEAD(&vxlan->next); 3183 3184 timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE); 3185 3186 vxlan->dev = dev; 3187 3188 for (h = 0; h < FDB_HASH_SIZE; ++h) { 3189 spin_lock_init(&vxlan->hash_lock[h]); 3190 INIT_HLIST_HEAD(&vxlan->fdb_head[h]); 3191 } 3192 } 3193 3194 static void vxlan_ether_setup(struct net_device *dev) 3195 { 3196 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 3197 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 3198 dev->netdev_ops = &vxlan_netdev_ether_ops; 3199 } 3200 3201 static void vxlan_raw_setup(struct net_device *dev) 3202 { 3203 dev->header_ops = NULL; 3204 dev->type = ARPHRD_NONE; 3205 dev->hard_header_len = 0; 3206 dev->addr_len = 0; 3207 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST; 3208 dev->netdev_ops = &vxlan_netdev_raw_ops; 3209 } 3210 3211 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = { 3212 [IFLA_VXLAN_ID] = { .type = NLA_U32 }, 3213 [IFLA_VXLAN_GROUP] = { .len = sizeof_field(struct iphdr, daddr) }, 3214 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) }, 3215 [IFLA_VXLAN_LINK] = { .type = NLA_U32 }, 3216 [IFLA_VXLAN_LOCAL] = { .len = sizeof_field(struct iphdr, saddr) }, 3217 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) }, 3218 [IFLA_VXLAN_TOS] = { .type = NLA_U8 }, 3219 [IFLA_VXLAN_TTL] = { .type = NLA_U8 }, 3220 [IFLA_VXLAN_LABEL] = { .type = NLA_U32 }, 3221 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 }, 3222 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 }, 3223 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 }, 3224 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) }, 3225 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 }, 3226 [IFLA_VXLAN_RSC] = { .type = NLA_U8 }, 3227 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 }, 3228 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 }, 3229 [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 }, 3230 [IFLA_VXLAN_PORT] = { .type = NLA_U16 }, 3231 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 }, 3232 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 }, 3233 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 }, 3234 [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 }, 3235 [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 }, 3236 [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, }, 3237 [IFLA_VXLAN_GPE] = { .type = NLA_FLAG, }, 3238 [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG }, 3239 [IFLA_VXLAN_TTL_INHERIT] = { .type = NLA_FLAG }, 3240 [IFLA_VXLAN_DF] = { .type = NLA_U8 }, 3241 [IFLA_VXLAN_VNIFILTER] = { .type = NLA_U8 }, 3242 }; 3243 3244 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[], 3245 struct netlink_ext_ack *extack) 3246 { 3247 if (tb[IFLA_ADDRESS]) { 3248 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 3249 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 3250 "Provided link layer address is not Ethernet"); 3251 return -EINVAL; 3252 } 3253 3254 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 3255 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 3256 "Provided Ethernet address is not unicast"); 3257 return -EADDRNOTAVAIL; 3258 } 3259 } 3260 3261 if (tb[IFLA_MTU]) { 3262 u32 mtu = nla_get_u32(tb[IFLA_MTU]); 3263 3264 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) { 3265 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU], 3266 "MTU must be between 68 and 65535"); 3267 return -EINVAL; 3268 } 3269 } 3270 3271 if (!data) { 3272 NL_SET_ERR_MSG(extack, 3273 "Required attributes not provided to perform the operation"); 3274 return -EINVAL; 3275 } 3276 3277 if (data[IFLA_VXLAN_ID]) { 3278 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]); 3279 3280 if (id >= VXLAN_N_VID) { 3281 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_ID], 3282 "VXLAN ID must be lower than 16777216"); 3283 return -ERANGE; 3284 } 3285 } 3286 3287 if (data[IFLA_VXLAN_PORT_RANGE]) { 3288 const struct ifla_vxlan_port_range *p 3289 = nla_data(data[IFLA_VXLAN_PORT_RANGE]); 3290 3291 if (ntohs(p->high) < ntohs(p->low)) { 3292 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_PORT_RANGE], 3293 "Invalid source port range"); 3294 return -EINVAL; 3295 } 3296 } 3297 3298 if (data[IFLA_VXLAN_DF]) { 3299 enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]); 3300 3301 if (df < 0 || df > VXLAN_DF_MAX) { 3302 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_DF], 3303 "Invalid DF attribute"); 3304 return -EINVAL; 3305 } 3306 } 3307 3308 return 0; 3309 } 3310 3311 static void vxlan_get_drvinfo(struct net_device *netdev, 3312 struct ethtool_drvinfo *drvinfo) 3313 { 3314 strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version)); 3315 strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver)); 3316 } 3317 3318 static int vxlan_get_link_ksettings(struct net_device *dev, 3319 struct ethtool_link_ksettings *cmd) 3320 { 3321 struct vxlan_dev *vxlan = netdev_priv(dev); 3322 struct vxlan_rdst *dst = &vxlan->default_dst; 3323 struct net_device *lowerdev = __dev_get_by_index(vxlan->net, 3324 dst->remote_ifindex); 3325 3326 if (!lowerdev) { 3327 cmd->base.duplex = DUPLEX_UNKNOWN; 3328 cmd->base.port = PORT_OTHER; 3329 cmd->base.speed = SPEED_UNKNOWN; 3330 3331 return 0; 3332 } 3333 3334 return __ethtool_get_link_ksettings(lowerdev, cmd); 3335 } 3336 3337 static const struct ethtool_ops vxlan_ethtool_ops = { 3338 .get_drvinfo = vxlan_get_drvinfo, 3339 .get_link = ethtool_op_get_link, 3340 .get_link_ksettings = vxlan_get_link_ksettings, 3341 }; 3342 3343 static struct socket *vxlan_create_sock(struct net *net, bool ipv6, 3344 __be16 port, u32 flags, int ifindex) 3345 { 3346 struct socket *sock; 3347 struct udp_port_cfg udp_conf; 3348 int err; 3349 3350 memset(&udp_conf, 0, sizeof(udp_conf)); 3351 3352 if (ipv6) { 3353 udp_conf.family = AF_INET6; 3354 udp_conf.use_udp6_rx_checksums = 3355 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX); 3356 udp_conf.ipv6_v6only = 1; 3357 } else { 3358 udp_conf.family = AF_INET; 3359 } 3360 3361 udp_conf.local_udp_port = port; 3362 udp_conf.bind_ifindex = ifindex; 3363 3364 /* Open UDP socket */ 3365 err = udp_sock_create(net, &udp_conf, &sock); 3366 if (err < 0) 3367 return ERR_PTR(err); 3368 3369 udp_allow_gso(sock->sk); 3370 return sock; 3371 } 3372 3373 /* Create new listen socket if needed */ 3374 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6, 3375 __be16 port, u32 flags, 3376 int ifindex) 3377 { 3378 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 3379 struct vxlan_sock *vs; 3380 struct socket *sock; 3381 unsigned int h; 3382 struct udp_tunnel_sock_cfg tunnel_cfg; 3383 3384 vs = kzalloc(sizeof(*vs), GFP_KERNEL); 3385 if (!vs) 3386 return ERR_PTR(-ENOMEM); 3387 3388 for (h = 0; h < VNI_HASH_SIZE; ++h) 3389 INIT_HLIST_HEAD(&vs->vni_list[h]); 3390 3391 sock = vxlan_create_sock(net, ipv6, port, flags, ifindex); 3392 if (IS_ERR(sock)) { 3393 kfree(vs); 3394 return ERR_CAST(sock); 3395 } 3396 3397 vs->sock = sock; 3398 refcount_set(&vs->refcnt, 1); 3399 vs->flags = (flags & VXLAN_F_RCV_FLAGS); 3400 3401 spin_lock(&vn->sock_lock); 3402 hlist_add_head_rcu(&vs->hlist, vs_head(net, port)); 3403 udp_tunnel_notify_add_rx_port(sock, 3404 (vs->flags & VXLAN_F_GPE) ? 3405 UDP_TUNNEL_TYPE_VXLAN_GPE : 3406 UDP_TUNNEL_TYPE_VXLAN); 3407 spin_unlock(&vn->sock_lock); 3408 3409 /* Mark socket as an encapsulation socket. */ 3410 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg)); 3411 tunnel_cfg.sk_user_data = vs; 3412 tunnel_cfg.encap_type = 1; 3413 tunnel_cfg.encap_rcv = vxlan_rcv; 3414 tunnel_cfg.encap_err_lookup = vxlan_err_lookup; 3415 tunnel_cfg.encap_destroy = NULL; 3416 tunnel_cfg.gro_receive = vxlan_gro_receive; 3417 tunnel_cfg.gro_complete = vxlan_gro_complete; 3418 3419 setup_udp_tunnel_sock(net, sock, &tunnel_cfg); 3420 3421 return vs; 3422 } 3423 3424 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6) 3425 { 3426 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id); 3427 bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA; 3428 struct vxlan_sock *vs = NULL; 3429 struct vxlan_dev_node *node; 3430 int l3mdev_index = 0; 3431 3432 if (vxlan->cfg.remote_ifindex) 3433 l3mdev_index = l3mdev_master_upper_ifindex_by_index( 3434 vxlan->net, vxlan->cfg.remote_ifindex); 3435 3436 if (!vxlan->cfg.no_share) { 3437 spin_lock(&vn->sock_lock); 3438 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET, 3439 vxlan->cfg.dst_port, vxlan->cfg.flags, 3440 l3mdev_index); 3441 if (vs && !refcount_inc_not_zero(&vs->refcnt)) { 3442 spin_unlock(&vn->sock_lock); 3443 return -EBUSY; 3444 } 3445 spin_unlock(&vn->sock_lock); 3446 } 3447 if (!vs) 3448 vs = vxlan_socket_create(vxlan->net, ipv6, 3449 vxlan->cfg.dst_port, vxlan->cfg.flags, 3450 l3mdev_index); 3451 if (IS_ERR(vs)) 3452 return PTR_ERR(vs); 3453 #if IS_ENABLED(CONFIG_IPV6) 3454 if (ipv6) { 3455 rcu_assign_pointer(vxlan->vn6_sock, vs); 3456 node = &vxlan->hlist6; 3457 } else 3458 #endif 3459 { 3460 rcu_assign_pointer(vxlan->vn4_sock, vs); 3461 node = &vxlan->hlist4; 3462 } 3463 3464 if (metadata && (vxlan->cfg.flags & VXLAN_F_VNIFILTER)) 3465 vxlan_vs_add_vnigrp(vxlan, vs, ipv6); 3466 else 3467 vxlan_vs_add_dev(vs, vxlan, node); 3468 3469 return 0; 3470 } 3471 3472 static int vxlan_sock_add(struct vxlan_dev *vxlan) 3473 { 3474 bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA; 3475 bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata; 3476 bool ipv4 = !ipv6 || metadata; 3477 int ret = 0; 3478 3479 RCU_INIT_POINTER(vxlan->vn4_sock, NULL); 3480 #if IS_ENABLED(CONFIG_IPV6) 3481 RCU_INIT_POINTER(vxlan->vn6_sock, NULL); 3482 if (ipv6) { 3483 ret = __vxlan_sock_add(vxlan, true); 3484 if (ret < 0 && ret != -EAFNOSUPPORT) 3485 ipv4 = false; 3486 } 3487 #endif 3488 if (ipv4) 3489 ret = __vxlan_sock_add(vxlan, false); 3490 if (ret < 0) 3491 vxlan_sock_release(vxlan); 3492 return ret; 3493 } 3494 3495 int vxlan_vni_in_use(struct net *src_net, struct vxlan_dev *vxlan, 3496 struct vxlan_config *conf, __be32 vni) 3497 { 3498 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id); 3499 struct vxlan_dev *tmp; 3500 3501 list_for_each_entry(tmp, &vn->vxlan_list, next) { 3502 if (tmp == vxlan) 3503 continue; 3504 if (tmp->cfg.flags & VXLAN_F_VNIFILTER) { 3505 if (!vxlan_vnifilter_lookup(tmp, vni)) 3506 continue; 3507 } else if (tmp->cfg.vni != vni) { 3508 continue; 3509 } 3510 if (tmp->cfg.dst_port != conf->dst_port) 3511 continue; 3512 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) != 3513 (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6))) 3514 continue; 3515 3516 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) && 3517 tmp->cfg.remote_ifindex != conf->remote_ifindex) 3518 continue; 3519 3520 return -EEXIST; 3521 } 3522 3523 return 0; 3524 } 3525 3526 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf, 3527 struct net_device **lower, 3528 struct vxlan_dev *old, 3529 struct netlink_ext_ack *extack) 3530 { 3531 bool use_ipv6 = false; 3532 3533 if (conf->flags & VXLAN_F_GPE) { 3534 /* For now, allow GPE only together with 3535 * COLLECT_METADATA. This can be relaxed later; in such 3536 * case, the other side of the PtP link will have to be 3537 * provided. 3538 */ 3539 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) || 3540 !(conf->flags & VXLAN_F_COLLECT_METADATA)) { 3541 NL_SET_ERR_MSG(extack, 3542 "VXLAN GPE does not support this combination of attributes"); 3543 return -EINVAL; 3544 } 3545 } 3546 3547 if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) { 3548 /* Unless IPv6 is explicitly requested, assume IPv4 */ 3549 conf->remote_ip.sa.sa_family = AF_INET; 3550 conf->saddr.sa.sa_family = AF_INET; 3551 } else if (!conf->remote_ip.sa.sa_family) { 3552 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family; 3553 } else if (!conf->saddr.sa.sa_family) { 3554 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family; 3555 } 3556 3557 if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) { 3558 NL_SET_ERR_MSG(extack, 3559 "Local and remote address must be from the same family"); 3560 return -EINVAL; 3561 } 3562 3563 if (vxlan_addr_multicast(&conf->saddr)) { 3564 NL_SET_ERR_MSG(extack, "Local address cannot be multicast"); 3565 return -EINVAL; 3566 } 3567 3568 if (conf->saddr.sa.sa_family == AF_INET6) { 3569 if (!IS_ENABLED(CONFIG_IPV6)) { 3570 NL_SET_ERR_MSG(extack, 3571 "IPv6 support not enabled in the kernel"); 3572 return -EPFNOSUPPORT; 3573 } 3574 use_ipv6 = true; 3575 conf->flags |= VXLAN_F_IPV6; 3576 3577 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) { 3578 int local_type = 3579 ipv6_addr_type(&conf->saddr.sin6.sin6_addr); 3580 int remote_type = 3581 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr); 3582 3583 if (local_type & IPV6_ADDR_LINKLOCAL) { 3584 if (!(remote_type & IPV6_ADDR_LINKLOCAL) && 3585 (remote_type != IPV6_ADDR_ANY)) { 3586 NL_SET_ERR_MSG(extack, 3587 "Invalid combination of local and remote address scopes"); 3588 return -EINVAL; 3589 } 3590 3591 conf->flags |= VXLAN_F_IPV6_LINKLOCAL; 3592 } else { 3593 if (remote_type == 3594 (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) { 3595 NL_SET_ERR_MSG(extack, 3596 "Invalid combination of local and remote address scopes"); 3597 return -EINVAL; 3598 } 3599 3600 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL; 3601 } 3602 } 3603 } 3604 3605 if (conf->label && !use_ipv6) { 3606 NL_SET_ERR_MSG(extack, 3607 "Label attribute only applies to IPv6 VXLAN devices"); 3608 return -EINVAL; 3609 } 3610 3611 if (conf->remote_ifindex) { 3612 struct net_device *lowerdev; 3613 3614 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex); 3615 if (!lowerdev) { 3616 NL_SET_ERR_MSG(extack, 3617 "Invalid local interface, device not found"); 3618 return -ENODEV; 3619 } 3620 3621 #if IS_ENABLED(CONFIG_IPV6) 3622 if (use_ipv6) { 3623 struct inet6_dev *idev = __in6_dev_get(lowerdev); 3624 3625 if (idev && idev->cnf.disable_ipv6) { 3626 NL_SET_ERR_MSG(extack, 3627 "IPv6 support disabled by administrator"); 3628 return -EPERM; 3629 } 3630 } 3631 #endif 3632 3633 *lower = lowerdev; 3634 } else { 3635 if (vxlan_addr_multicast(&conf->remote_ip)) { 3636 NL_SET_ERR_MSG(extack, 3637 "Local interface required for multicast remote destination"); 3638 3639 return -EINVAL; 3640 } 3641 3642 #if IS_ENABLED(CONFIG_IPV6) 3643 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) { 3644 NL_SET_ERR_MSG(extack, 3645 "Local interface required for link-local local/remote addresses"); 3646 return -EINVAL; 3647 } 3648 #endif 3649 3650 *lower = NULL; 3651 } 3652 3653 if (!conf->dst_port) { 3654 if (conf->flags & VXLAN_F_GPE) 3655 conf->dst_port = htons(IANA_VXLAN_GPE_UDP_PORT); 3656 else 3657 conf->dst_port = htons(vxlan_port); 3658 } 3659 3660 if (!conf->age_interval) 3661 conf->age_interval = FDB_AGE_DEFAULT; 3662 3663 if (vxlan_vni_in_use(src_net, old, conf, conf->vni)) { 3664 NL_SET_ERR_MSG(extack, 3665 "A VXLAN device with the specified VNI already exists"); 3666 return -EEXIST; 3667 } 3668 3669 return 0; 3670 } 3671 3672 static void vxlan_config_apply(struct net_device *dev, 3673 struct vxlan_config *conf, 3674 struct net_device *lowerdev, 3675 struct net *src_net, 3676 bool changelink) 3677 { 3678 struct vxlan_dev *vxlan = netdev_priv(dev); 3679 struct vxlan_rdst *dst = &vxlan->default_dst; 3680 unsigned short needed_headroom = ETH_HLEN; 3681 bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6); 3682 int max_mtu = ETH_MAX_MTU; 3683 3684 if (!changelink) { 3685 if (conf->flags & VXLAN_F_GPE) 3686 vxlan_raw_setup(dev); 3687 else 3688 vxlan_ether_setup(dev); 3689 3690 if (conf->mtu) 3691 dev->mtu = conf->mtu; 3692 3693 vxlan->net = src_net; 3694 } 3695 3696 dst->remote_vni = conf->vni; 3697 3698 memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip)); 3699 3700 if (lowerdev) { 3701 dst->remote_ifindex = conf->remote_ifindex; 3702 3703 netif_inherit_tso_max(dev, lowerdev); 3704 3705 needed_headroom = lowerdev->hard_header_len; 3706 needed_headroom += lowerdev->needed_headroom; 3707 3708 dev->needed_tailroom = lowerdev->needed_tailroom; 3709 3710 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : 3711 VXLAN_HEADROOM); 3712 if (max_mtu < ETH_MIN_MTU) 3713 max_mtu = ETH_MIN_MTU; 3714 3715 if (!changelink && !conf->mtu) 3716 dev->mtu = max_mtu; 3717 } 3718 3719 if (dev->mtu > max_mtu) 3720 dev->mtu = max_mtu; 3721 3722 if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA) 3723 needed_headroom += VXLAN6_HEADROOM; 3724 else 3725 needed_headroom += VXLAN_HEADROOM; 3726 dev->needed_headroom = needed_headroom; 3727 3728 memcpy(&vxlan->cfg, conf, sizeof(*conf)); 3729 } 3730 3731 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev, 3732 struct vxlan_config *conf, bool changelink, 3733 struct netlink_ext_ack *extack) 3734 { 3735 struct vxlan_dev *vxlan = netdev_priv(dev); 3736 struct net_device *lowerdev; 3737 int ret; 3738 3739 ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack); 3740 if (ret) 3741 return ret; 3742 3743 vxlan_config_apply(dev, conf, lowerdev, src_net, changelink); 3744 3745 return 0; 3746 } 3747 3748 static int __vxlan_dev_create(struct net *net, struct net_device *dev, 3749 struct vxlan_config *conf, 3750 struct netlink_ext_ack *extack) 3751 { 3752 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 3753 struct vxlan_dev *vxlan = netdev_priv(dev); 3754 struct net_device *remote_dev = NULL; 3755 struct vxlan_fdb *f = NULL; 3756 bool unregister = false; 3757 struct vxlan_rdst *dst; 3758 int err; 3759 3760 dst = &vxlan->default_dst; 3761 err = vxlan_dev_configure(net, dev, conf, false, extack); 3762 if (err) 3763 return err; 3764 3765 dev->ethtool_ops = &vxlan_ethtool_ops; 3766 3767 /* create an fdb entry for a valid default destination */ 3768 if (!vxlan_addr_any(&dst->remote_ip)) { 3769 err = vxlan_fdb_create(vxlan, all_zeros_mac, 3770 &dst->remote_ip, 3771 NUD_REACHABLE | NUD_PERMANENT, 3772 vxlan->cfg.dst_port, 3773 dst->remote_vni, 3774 dst->remote_vni, 3775 dst->remote_ifindex, 3776 NTF_SELF, 0, &f, extack); 3777 if (err) 3778 return err; 3779 } 3780 3781 err = register_netdevice(dev); 3782 if (err) 3783 goto errout; 3784 unregister = true; 3785 3786 if (dst->remote_ifindex) { 3787 remote_dev = __dev_get_by_index(net, dst->remote_ifindex); 3788 if (!remote_dev) { 3789 err = -ENODEV; 3790 goto errout; 3791 } 3792 3793 err = netdev_upper_dev_link(remote_dev, dev, extack); 3794 if (err) 3795 goto errout; 3796 } 3797 3798 err = rtnl_configure_link(dev, NULL); 3799 if (err < 0) 3800 goto unlink; 3801 3802 if (f) { 3803 vxlan_fdb_insert(vxlan, all_zeros_mac, dst->remote_vni, f); 3804 3805 /* notify default fdb entry */ 3806 err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), 3807 RTM_NEWNEIGH, true, extack); 3808 if (err) { 3809 vxlan_fdb_destroy(vxlan, f, false, false); 3810 if (remote_dev) 3811 netdev_upper_dev_unlink(remote_dev, dev); 3812 goto unregister; 3813 } 3814 } 3815 3816 list_add(&vxlan->next, &vn->vxlan_list); 3817 if (remote_dev) 3818 dst->remote_dev = remote_dev; 3819 return 0; 3820 unlink: 3821 if (remote_dev) 3822 netdev_upper_dev_unlink(remote_dev, dev); 3823 errout: 3824 /* unregister_netdevice() destroys the default FDB entry with deletion 3825 * notification. But the addition notification was not sent yet, so 3826 * destroy the entry by hand here. 3827 */ 3828 if (f) 3829 __vxlan_fdb_free(f); 3830 unregister: 3831 if (unregister) 3832 unregister_netdevice(dev); 3833 return err; 3834 } 3835 3836 /* Set/clear flags based on attribute */ 3837 static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[], 3838 int attrtype, unsigned long mask, bool changelink, 3839 bool changelink_supported, 3840 struct netlink_ext_ack *extack) 3841 { 3842 unsigned long flags; 3843 3844 if (!tb[attrtype]) 3845 return 0; 3846 3847 if (changelink && !changelink_supported) { 3848 vxlan_flag_attr_error(attrtype, extack); 3849 return -EOPNOTSUPP; 3850 } 3851 3852 if (vxlan_policy[attrtype].type == NLA_FLAG) 3853 flags = conf->flags | mask; 3854 else if (nla_get_u8(tb[attrtype])) 3855 flags = conf->flags | mask; 3856 else 3857 flags = conf->flags & ~mask; 3858 3859 conf->flags = flags; 3860 3861 return 0; 3862 } 3863 3864 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[], 3865 struct net_device *dev, struct vxlan_config *conf, 3866 bool changelink, struct netlink_ext_ack *extack) 3867 { 3868 struct vxlan_dev *vxlan = netdev_priv(dev); 3869 int err = 0; 3870 3871 memset(conf, 0, sizeof(*conf)); 3872 3873 /* if changelink operation, start with old existing cfg */ 3874 if (changelink) 3875 memcpy(conf, &vxlan->cfg, sizeof(*conf)); 3876 3877 if (data[IFLA_VXLAN_ID]) { 3878 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID])); 3879 3880 if (changelink && (vni != conf->vni)) { 3881 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI"); 3882 return -EOPNOTSUPP; 3883 } 3884 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID])); 3885 } 3886 3887 if (data[IFLA_VXLAN_GROUP]) { 3888 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) { 3889 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group"); 3890 return -EOPNOTSUPP; 3891 } 3892 3893 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]); 3894 conf->remote_ip.sa.sa_family = AF_INET; 3895 } else if (data[IFLA_VXLAN_GROUP6]) { 3896 if (!IS_ENABLED(CONFIG_IPV6)) { 3897 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel"); 3898 return -EPFNOSUPPORT; 3899 } 3900 3901 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) { 3902 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group"); 3903 return -EOPNOTSUPP; 3904 } 3905 3906 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]); 3907 conf->remote_ip.sa.sa_family = AF_INET6; 3908 } 3909 3910 if (data[IFLA_VXLAN_LOCAL]) { 3911 if (changelink && (conf->saddr.sa.sa_family != AF_INET)) { 3912 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old"); 3913 return -EOPNOTSUPP; 3914 } 3915 3916 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]); 3917 conf->saddr.sa.sa_family = AF_INET; 3918 } else if (data[IFLA_VXLAN_LOCAL6]) { 3919 if (!IS_ENABLED(CONFIG_IPV6)) { 3920 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel"); 3921 return -EPFNOSUPPORT; 3922 } 3923 3924 if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) { 3925 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old"); 3926 return -EOPNOTSUPP; 3927 } 3928 3929 /* TODO: respect scope id */ 3930 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]); 3931 conf->saddr.sa.sa_family = AF_INET6; 3932 } 3933 3934 if (data[IFLA_VXLAN_LINK]) 3935 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]); 3936 3937 if (data[IFLA_VXLAN_TOS]) 3938 conf->tos = nla_get_u8(data[IFLA_VXLAN_TOS]); 3939 3940 if (data[IFLA_VXLAN_TTL]) 3941 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]); 3942 3943 if (data[IFLA_VXLAN_TTL_INHERIT]) { 3944 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT, 3945 VXLAN_F_TTL_INHERIT, changelink, false, 3946 extack); 3947 if (err) 3948 return err; 3949 3950 } 3951 3952 if (data[IFLA_VXLAN_LABEL]) 3953 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) & 3954 IPV6_FLOWLABEL_MASK; 3955 3956 if (data[IFLA_VXLAN_LEARNING]) { 3957 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING, 3958 VXLAN_F_LEARN, changelink, true, 3959 extack); 3960 if (err) 3961 return err; 3962 } else if (!changelink) { 3963 /* default to learn on a new device */ 3964 conf->flags |= VXLAN_F_LEARN; 3965 } 3966 3967 if (data[IFLA_VXLAN_AGEING]) 3968 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]); 3969 3970 if (data[IFLA_VXLAN_PROXY]) { 3971 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY, 3972 VXLAN_F_PROXY, changelink, false, 3973 extack); 3974 if (err) 3975 return err; 3976 } 3977 3978 if (data[IFLA_VXLAN_RSC]) { 3979 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC, 3980 VXLAN_F_RSC, changelink, false, 3981 extack); 3982 if (err) 3983 return err; 3984 } 3985 3986 if (data[IFLA_VXLAN_L2MISS]) { 3987 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS, 3988 VXLAN_F_L2MISS, changelink, false, 3989 extack); 3990 if (err) 3991 return err; 3992 } 3993 3994 if (data[IFLA_VXLAN_L3MISS]) { 3995 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS, 3996 VXLAN_F_L3MISS, changelink, false, 3997 extack); 3998 if (err) 3999 return err; 4000 } 4001 4002 if (data[IFLA_VXLAN_LIMIT]) { 4003 if (changelink) { 4004 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT], 4005 "Cannot change limit"); 4006 return -EOPNOTSUPP; 4007 } 4008 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]); 4009 } 4010 4011 if (data[IFLA_VXLAN_COLLECT_METADATA]) { 4012 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA, 4013 VXLAN_F_COLLECT_METADATA, changelink, false, 4014 extack); 4015 if (err) 4016 return err; 4017 } 4018 4019 if (data[IFLA_VXLAN_PORT_RANGE]) { 4020 if (!changelink) { 4021 const struct ifla_vxlan_port_range *p 4022 = nla_data(data[IFLA_VXLAN_PORT_RANGE]); 4023 conf->port_min = ntohs(p->low); 4024 conf->port_max = ntohs(p->high); 4025 } else { 4026 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE], 4027 "Cannot change port range"); 4028 return -EOPNOTSUPP; 4029 } 4030 } 4031 4032 if (data[IFLA_VXLAN_PORT]) { 4033 if (changelink) { 4034 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT], 4035 "Cannot change port"); 4036 return -EOPNOTSUPP; 4037 } 4038 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]); 4039 } 4040 4041 if (data[IFLA_VXLAN_UDP_CSUM]) { 4042 if (changelink) { 4043 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM], 4044 "Cannot change UDP_CSUM flag"); 4045 return -EOPNOTSUPP; 4046 } 4047 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM])) 4048 conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX; 4049 } 4050 4051 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) { 4052 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX, 4053 VXLAN_F_UDP_ZERO_CSUM6_TX, changelink, 4054 false, extack); 4055 if (err) 4056 return err; 4057 } 4058 4059 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) { 4060 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX, 4061 VXLAN_F_UDP_ZERO_CSUM6_RX, changelink, 4062 false, extack); 4063 if (err) 4064 return err; 4065 } 4066 4067 if (data[IFLA_VXLAN_REMCSUM_TX]) { 4068 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX, 4069 VXLAN_F_REMCSUM_TX, changelink, false, 4070 extack); 4071 if (err) 4072 return err; 4073 } 4074 4075 if (data[IFLA_VXLAN_REMCSUM_RX]) { 4076 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX, 4077 VXLAN_F_REMCSUM_RX, changelink, false, 4078 extack); 4079 if (err) 4080 return err; 4081 } 4082 4083 if (data[IFLA_VXLAN_GBP]) { 4084 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP, 4085 VXLAN_F_GBP, changelink, false, extack); 4086 if (err) 4087 return err; 4088 } 4089 4090 if (data[IFLA_VXLAN_GPE]) { 4091 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE, 4092 VXLAN_F_GPE, changelink, false, 4093 extack); 4094 if (err) 4095 return err; 4096 } 4097 4098 if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) { 4099 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL, 4100 VXLAN_F_REMCSUM_NOPARTIAL, changelink, 4101 false, extack); 4102 if (err) 4103 return err; 4104 } 4105 4106 if (tb[IFLA_MTU]) { 4107 if (changelink) { 4108 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU], 4109 "Cannot change mtu"); 4110 return -EOPNOTSUPP; 4111 } 4112 conf->mtu = nla_get_u32(tb[IFLA_MTU]); 4113 } 4114 4115 if (data[IFLA_VXLAN_DF]) 4116 conf->df = nla_get_u8(data[IFLA_VXLAN_DF]); 4117 4118 if (data[IFLA_VXLAN_VNIFILTER]) { 4119 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_VNIFILTER, 4120 VXLAN_F_VNIFILTER, changelink, false, 4121 extack); 4122 if (err) 4123 return err; 4124 4125 if ((conf->flags & VXLAN_F_VNIFILTER) && 4126 !(conf->flags & VXLAN_F_COLLECT_METADATA)) { 4127 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_VNIFILTER], 4128 "vxlan vnifilter only valid in collect metadata mode"); 4129 return -EINVAL; 4130 } 4131 } 4132 4133 return 0; 4134 } 4135 4136 static int vxlan_newlink(struct net *src_net, struct net_device *dev, 4137 struct nlattr *tb[], struct nlattr *data[], 4138 struct netlink_ext_ack *extack) 4139 { 4140 struct vxlan_config conf; 4141 int err; 4142 4143 err = vxlan_nl2conf(tb, data, dev, &conf, false, extack); 4144 if (err) 4145 return err; 4146 4147 return __vxlan_dev_create(src_net, dev, &conf, extack); 4148 } 4149 4150 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[], 4151 struct nlattr *data[], 4152 struct netlink_ext_ack *extack) 4153 { 4154 struct vxlan_dev *vxlan = netdev_priv(dev); 4155 struct net_device *lowerdev; 4156 struct vxlan_config conf; 4157 struct vxlan_rdst *dst; 4158 int err; 4159 4160 dst = &vxlan->default_dst; 4161 err = vxlan_nl2conf(tb, data, dev, &conf, true, extack); 4162 if (err) 4163 return err; 4164 4165 err = vxlan_config_validate(vxlan->net, &conf, &lowerdev, 4166 vxlan, extack); 4167 if (err) 4168 return err; 4169 4170 if (dst->remote_dev == lowerdev) 4171 lowerdev = NULL; 4172 4173 err = netdev_adjacent_change_prepare(dst->remote_dev, lowerdev, dev, 4174 extack); 4175 if (err) 4176 return err; 4177 4178 /* handle default dst entry */ 4179 if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) { 4180 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, conf.vni); 4181 4182 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4183 if (!vxlan_addr_any(&conf.remote_ip)) { 4184 err = vxlan_fdb_update(vxlan, all_zeros_mac, 4185 &conf.remote_ip, 4186 NUD_REACHABLE | NUD_PERMANENT, 4187 NLM_F_APPEND | NLM_F_CREATE, 4188 vxlan->cfg.dst_port, 4189 conf.vni, conf.vni, 4190 conf.remote_ifindex, 4191 NTF_SELF, 0, true, extack); 4192 if (err) { 4193 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4194 netdev_adjacent_change_abort(dst->remote_dev, 4195 lowerdev, dev); 4196 return err; 4197 } 4198 } 4199 if (!vxlan_addr_any(&dst->remote_ip)) 4200 __vxlan_fdb_delete(vxlan, all_zeros_mac, 4201 dst->remote_ip, 4202 vxlan->cfg.dst_port, 4203 dst->remote_vni, 4204 dst->remote_vni, 4205 dst->remote_ifindex, 4206 true); 4207 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4208 4209 /* If vni filtering device, also update fdb entries of 4210 * all vnis that were using default remote ip 4211 */ 4212 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER) { 4213 err = vxlan_vnilist_update_group(vxlan, &dst->remote_ip, 4214 &conf.remote_ip, extack); 4215 if (err) { 4216 netdev_adjacent_change_abort(dst->remote_dev, 4217 lowerdev, dev); 4218 return err; 4219 } 4220 } 4221 } 4222 4223 if (conf.age_interval != vxlan->cfg.age_interval) 4224 mod_timer(&vxlan->age_timer, jiffies); 4225 4226 netdev_adjacent_change_commit(dst->remote_dev, lowerdev, dev); 4227 if (lowerdev && lowerdev != dst->remote_dev) 4228 dst->remote_dev = lowerdev; 4229 vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true); 4230 return 0; 4231 } 4232 4233 static void vxlan_dellink(struct net_device *dev, struct list_head *head) 4234 { 4235 struct vxlan_dev *vxlan = netdev_priv(dev); 4236 4237 vxlan_flush(vxlan, true); 4238 4239 list_del(&vxlan->next); 4240 unregister_netdevice_queue(dev, head); 4241 if (vxlan->default_dst.remote_dev) 4242 netdev_upper_dev_unlink(vxlan->default_dst.remote_dev, dev); 4243 } 4244 4245 static size_t vxlan_get_size(const struct net_device *dev) 4246 { 4247 4248 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */ 4249 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */ 4250 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */ 4251 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */ 4252 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */ 4253 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL_INHERIT */ 4254 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */ 4255 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_DF */ 4256 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */ 4257 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */ 4258 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */ 4259 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */ 4260 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */ 4261 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */ 4262 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */ 4263 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */ 4264 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */ 4265 nla_total_size(sizeof(struct ifla_vxlan_port_range)) + 4266 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */ 4267 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */ 4268 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */ 4269 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */ 4270 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */ 4271 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */ 4272 0; 4273 } 4274 4275 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev) 4276 { 4277 const struct vxlan_dev *vxlan = netdev_priv(dev); 4278 const struct vxlan_rdst *dst = &vxlan->default_dst; 4279 struct ifla_vxlan_port_range ports = { 4280 .low = htons(vxlan->cfg.port_min), 4281 .high = htons(vxlan->cfg.port_max), 4282 }; 4283 4284 if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni))) 4285 goto nla_put_failure; 4286 4287 if (!vxlan_addr_any(&dst->remote_ip)) { 4288 if (dst->remote_ip.sa.sa_family == AF_INET) { 4289 if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP, 4290 dst->remote_ip.sin.sin_addr.s_addr)) 4291 goto nla_put_failure; 4292 #if IS_ENABLED(CONFIG_IPV6) 4293 } else { 4294 if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6, 4295 &dst->remote_ip.sin6.sin6_addr)) 4296 goto nla_put_failure; 4297 #endif 4298 } 4299 } 4300 4301 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex)) 4302 goto nla_put_failure; 4303 4304 if (!vxlan_addr_any(&vxlan->cfg.saddr)) { 4305 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) { 4306 if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL, 4307 vxlan->cfg.saddr.sin.sin_addr.s_addr)) 4308 goto nla_put_failure; 4309 #if IS_ENABLED(CONFIG_IPV6) 4310 } else { 4311 if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6, 4312 &vxlan->cfg.saddr.sin6.sin6_addr)) 4313 goto nla_put_failure; 4314 #endif 4315 } 4316 } 4317 4318 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) || 4319 nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT, 4320 !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) || 4321 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) || 4322 nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) || 4323 nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) || 4324 nla_put_u8(skb, IFLA_VXLAN_LEARNING, 4325 !!(vxlan->cfg.flags & VXLAN_F_LEARN)) || 4326 nla_put_u8(skb, IFLA_VXLAN_PROXY, 4327 !!(vxlan->cfg.flags & VXLAN_F_PROXY)) || 4328 nla_put_u8(skb, IFLA_VXLAN_RSC, 4329 !!(vxlan->cfg.flags & VXLAN_F_RSC)) || 4330 nla_put_u8(skb, IFLA_VXLAN_L2MISS, 4331 !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) || 4332 nla_put_u8(skb, IFLA_VXLAN_L3MISS, 4333 !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) || 4334 nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA, 4335 !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) || 4336 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) || 4337 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) || 4338 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) || 4339 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM, 4340 !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) || 4341 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX, 4342 !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) || 4343 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX, 4344 !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) || 4345 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX, 4346 !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) || 4347 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX, 4348 !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX))) 4349 goto nla_put_failure; 4350 4351 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports)) 4352 goto nla_put_failure; 4353 4354 if (vxlan->cfg.flags & VXLAN_F_GBP && 4355 nla_put_flag(skb, IFLA_VXLAN_GBP)) 4356 goto nla_put_failure; 4357 4358 if (vxlan->cfg.flags & VXLAN_F_GPE && 4359 nla_put_flag(skb, IFLA_VXLAN_GPE)) 4360 goto nla_put_failure; 4361 4362 if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL && 4363 nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL)) 4364 goto nla_put_failure; 4365 4366 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER && 4367 nla_put_u8(skb, IFLA_VXLAN_VNIFILTER, 4368 !!(vxlan->cfg.flags & VXLAN_F_VNIFILTER))) 4369 goto nla_put_failure; 4370 4371 return 0; 4372 4373 nla_put_failure: 4374 return -EMSGSIZE; 4375 } 4376 4377 static struct net *vxlan_get_link_net(const struct net_device *dev) 4378 { 4379 struct vxlan_dev *vxlan = netdev_priv(dev); 4380 4381 return vxlan->net; 4382 } 4383 4384 static struct rtnl_link_ops vxlan_link_ops __read_mostly = { 4385 .kind = "vxlan", 4386 .maxtype = IFLA_VXLAN_MAX, 4387 .policy = vxlan_policy, 4388 .priv_size = sizeof(struct vxlan_dev), 4389 .setup = vxlan_setup, 4390 .validate = vxlan_validate, 4391 .newlink = vxlan_newlink, 4392 .changelink = vxlan_changelink, 4393 .dellink = vxlan_dellink, 4394 .get_size = vxlan_get_size, 4395 .fill_info = vxlan_fill_info, 4396 .get_link_net = vxlan_get_link_net, 4397 }; 4398 4399 struct net_device *vxlan_dev_create(struct net *net, const char *name, 4400 u8 name_assign_type, 4401 struct vxlan_config *conf) 4402 { 4403 struct nlattr *tb[IFLA_MAX + 1]; 4404 struct net_device *dev; 4405 int err; 4406 4407 memset(&tb, 0, sizeof(tb)); 4408 4409 dev = rtnl_create_link(net, name, name_assign_type, 4410 &vxlan_link_ops, tb, NULL); 4411 if (IS_ERR(dev)) 4412 return dev; 4413 4414 err = __vxlan_dev_create(net, dev, conf, NULL); 4415 if (err < 0) { 4416 free_netdev(dev); 4417 return ERR_PTR(err); 4418 } 4419 4420 err = rtnl_configure_link(dev, NULL); 4421 if (err < 0) { 4422 LIST_HEAD(list_kill); 4423 4424 vxlan_dellink(dev, &list_kill); 4425 unregister_netdevice_many(&list_kill); 4426 return ERR_PTR(err); 4427 } 4428 4429 return dev; 4430 } 4431 EXPORT_SYMBOL_GPL(vxlan_dev_create); 4432 4433 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn, 4434 struct net_device *dev) 4435 { 4436 struct vxlan_dev *vxlan, *next; 4437 LIST_HEAD(list_kill); 4438 4439 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) { 4440 struct vxlan_rdst *dst = &vxlan->default_dst; 4441 4442 /* In case we created vxlan device with carrier 4443 * and we loose the carrier due to module unload 4444 * we also need to remove vxlan device. In other 4445 * cases, it's not necessary and remote_ifindex 4446 * is 0 here, so no matches. 4447 */ 4448 if (dst->remote_ifindex == dev->ifindex) 4449 vxlan_dellink(vxlan->dev, &list_kill); 4450 } 4451 4452 unregister_netdevice_many(&list_kill); 4453 } 4454 4455 static int vxlan_netdevice_event(struct notifier_block *unused, 4456 unsigned long event, void *ptr) 4457 { 4458 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 4459 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id); 4460 4461 if (event == NETDEV_UNREGISTER) 4462 vxlan_handle_lowerdev_unregister(vn, dev); 4463 else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO) 4464 vxlan_offload_rx_ports(dev, true); 4465 else if (event == NETDEV_UDP_TUNNEL_DROP_INFO) 4466 vxlan_offload_rx_ports(dev, false); 4467 4468 return NOTIFY_DONE; 4469 } 4470 4471 static struct notifier_block vxlan_notifier_block __read_mostly = { 4472 .notifier_call = vxlan_netdevice_event, 4473 }; 4474 4475 static void 4476 vxlan_fdb_offloaded_set(struct net_device *dev, 4477 struct switchdev_notifier_vxlan_fdb_info *fdb_info) 4478 { 4479 struct vxlan_dev *vxlan = netdev_priv(dev); 4480 struct vxlan_rdst *rdst; 4481 struct vxlan_fdb *f; 4482 u32 hash_index; 4483 4484 hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni); 4485 4486 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4487 4488 f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni); 4489 if (!f) 4490 goto out; 4491 4492 rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip, 4493 fdb_info->remote_port, 4494 fdb_info->remote_vni, 4495 fdb_info->remote_ifindex); 4496 if (!rdst) 4497 goto out; 4498 4499 rdst->offloaded = fdb_info->offloaded; 4500 4501 out: 4502 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4503 } 4504 4505 static int 4506 vxlan_fdb_external_learn_add(struct net_device *dev, 4507 struct switchdev_notifier_vxlan_fdb_info *fdb_info) 4508 { 4509 struct vxlan_dev *vxlan = netdev_priv(dev); 4510 struct netlink_ext_ack *extack; 4511 u32 hash_index; 4512 int err; 4513 4514 hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni); 4515 extack = switchdev_notifier_info_to_extack(&fdb_info->info); 4516 4517 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4518 err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip, 4519 NUD_REACHABLE, 4520 NLM_F_CREATE | NLM_F_REPLACE, 4521 fdb_info->remote_port, 4522 fdb_info->vni, 4523 fdb_info->remote_vni, 4524 fdb_info->remote_ifindex, 4525 NTF_USE | NTF_SELF | NTF_EXT_LEARNED, 4526 0, false, extack); 4527 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4528 4529 return err; 4530 } 4531 4532 static int 4533 vxlan_fdb_external_learn_del(struct net_device *dev, 4534 struct switchdev_notifier_vxlan_fdb_info *fdb_info) 4535 { 4536 struct vxlan_dev *vxlan = netdev_priv(dev); 4537 struct vxlan_fdb *f; 4538 u32 hash_index; 4539 int err = 0; 4540 4541 hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni); 4542 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4543 4544 f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni); 4545 if (!f) 4546 err = -ENOENT; 4547 else if (f->flags & NTF_EXT_LEARNED) 4548 err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr, 4549 fdb_info->remote_ip, 4550 fdb_info->remote_port, 4551 fdb_info->vni, 4552 fdb_info->remote_vni, 4553 fdb_info->remote_ifindex, 4554 false); 4555 4556 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4557 4558 return err; 4559 } 4560 4561 static int vxlan_switchdev_event(struct notifier_block *unused, 4562 unsigned long event, void *ptr) 4563 { 4564 struct net_device *dev = switchdev_notifier_info_to_dev(ptr); 4565 struct switchdev_notifier_vxlan_fdb_info *fdb_info; 4566 int err = 0; 4567 4568 switch (event) { 4569 case SWITCHDEV_VXLAN_FDB_OFFLOADED: 4570 vxlan_fdb_offloaded_set(dev, ptr); 4571 break; 4572 case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE: 4573 fdb_info = ptr; 4574 err = vxlan_fdb_external_learn_add(dev, fdb_info); 4575 if (err) { 4576 err = notifier_from_errno(err); 4577 break; 4578 } 4579 fdb_info->offloaded = true; 4580 vxlan_fdb_offloaded_set(dev, fdb_info); 4581 break; 4582 case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE: 4583 fdb_info = ptr; 4584 err = vxlan_fdb_external_learn_del(dev, fdb_info); 4585 if (err) { 4586 err = notifier_from_errno(err); 4587 break; 4588 } 4589 fdb_info->offloaded = false; 4590 vxlan_fdb_offloaded_set(dev, fdb_info); 4591 break; 4592 } 4593 4594 return err; 4595 } 4596 4597 static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = { 4598 .notifier_call = vxlan_switchdev_event, 4599 }; 4600 4601 static void vxlan_fdb_nh_flush(struct nexthop *nh) 4602 { 4603 struct vxlan_fdb *fdb; 4604 struct vxlan_dev *vxlan; 4605 u32 hash_index; 4606 4607 rcu_read_lock(); 4608 list_for_each_entry_rcu(fdb, &nh->fdb_list, nh_list) { 4609 vxlan = rcu_dereference(fdb->vdev); 4610 WARN_ON(!vxlan); 4611 hash_index = fdb_head_index(vxlan, fdb->eth_addr, 4612 vxlan->default_dst.remote_vni); 4613 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4614 if (!hlist_unhashed(&fdb->hlist)) 4615 vxlan_fdb_destroy(vxlan, fdb, false, false); 4616 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4617 } 4618 rcu_read_unlock(); 4619 } 4620 4621 static int vxlan_nexthop_event(struct notifier_block *nb, 4622 unsigned long event, void *ptr) 4623 { 4624 struct nh_notifier_info *info = ptr; 4625 struct nexthop *nh; 4626 4627 if (event != NEXTHOP_EVENT_DEL) 4628 return NOTIFY_DONE; 4629 4630 nh = nexthop_find_by_id(info->net, info->id); 4631 if (!nh) 4632 return NOTIFY_DONE; 4633 4634 vxlan_fdb_nh_flush(nh); 4635 4636 return NOTIFY_DONE; 4637 } 4638 4639 static __net_init int vxlan_init_net(struct net *net) 4640 { 4641 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 4642 unsigned int h; 4643 4644 INIT_LIST_HEAD(&vn->vxlan_list); 4645 spin_lock_init(&vn->sock_lock); 4646 vn->nexthop_notifier_block.notifier_call = vxlan_nexthop_event; 4647 4648 for (h = 0; h < PORT_HASH_SIZE; ++h) 4649 INIT_HLIST_HEAD(&vn->sock_list[h]); 4650 4651 return register_nexthop_notifier(net, &vn->nexthop_notifier_block, 4652 NULL); 4653 } 4654 4655 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head) 4656 { 4657 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 4658 struct vxlan_dev *vxlan, *next; 4659 struct net_device *dev, *aux; 4660 4661 for_each_netdev_safe(net, dev, aux) 4662 if (dev->rtnl_link_ops == &vxlan_link_ops) 4663 unregister_netdevice_queue(dev, head); 4664 4665 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) { 4666 /* If vxlan->dev is in the same netns, it has already been added 4667 * to the list by the previous loop. 4668 */ 4669 if (!net_eq(dev_net(vxlan->dev), net)) 4670 unregister_netdevice_queue(vxlan->dev, head); 4671 } 4672 4673 } 4674 4675 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list) 4676 { 4677 struct net *net; 4678 LIST_HEAD(list); 4679 unsigned int h; 4680 4681 list_for_each_entry(net, net_list, exit_list) { 4682 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 4683 4684 unregister_nexthop_notifier(net, &vn->nexthop_notifier_block); 4685 } 4686 rtnl_lock(); 4687 list_for_each_entry(net, net_list, exit_list) 4688 vxlan_destroy_tunnels(net, &list); 4689 4690 unregister_netdevice_many(&list); 4691 rtnl_unlock(); 4692 4693 list_for_each_entry(net, net_list, exit_list) { 4694 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 4695 4696 for (h = 0; h < PORT_HASH_SIZE; ++h) 4697 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h])); 4698 } 4699 } 4700 4701 static struct pernet_operations vxlan_net_ops = { 4702 .init = vxlan_init_net, 4703 .exit_batch = vxlan_exit_batch_net, 4704 .id = &vxlan_net_id, 4705 .size = sizeof(struct vxlan_net), 4706 }; 4707 4708 static int __init vxlan_init_module(void) 4709 { 4710 int rc; 4711 4712 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt)); 4713 4714 rc = register_pernet_subsys(&vxlan_net_ops); 4715 if (rc) 4716 goto out1; 4717 4718 rc = register_netdevice_notifier(&vxlan_notifier_block); 4719 if (rc) 4720 goto out2; 4721 4722 rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block); 4723 if (rc) 4724 goto out3; 4725 4726 rc = rtnl_link_register(&vxlan_link_ops); 4727 if (rc) 4728 goto out4; 4729 4730 vxlan_vnifilter_init(); 4731 4732 return 0; 4733 out4: 4734 unregister_switchdev_notifier(&vxlan_switchdev_notifier_block); 4735 out3: 4736 unregister_netdevice_notifier(&vxlan_notifier_block); 4737 out2: 4738 unregister_pernet_subsys(&vxlan_net_ops); 4739 out1: 4740 return rc; 4741 } 4742 late_initcall(vxlan_init_module); 4743 4744 static void __exit vxlan_cleanup_module(void) 4745 { 4746 vxlan_vnifilter_uninit(); 4747 rtnl_link_unregister(&vxlan_link_ops); 4748 unregister_switchdev_notifier(&vxlan_switchdev_notifier_block); 4749 unregister_netdevice_notifier(&vxlan_notifier_block); 4750 unregister_pernet_subsys(&vxlan_net_ops); 4751 /* rcu_barrier() is called by netns */ 4752 } 4753 module_exit(vxlan_cleanup_module); 4754 4755 MODULE_LICENSE("GPL"); 4756 MODULE_VERSION(VXLAN_VERSION); 4757 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>"); 4758 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic"); 4759 MODULE_ALIAS_RTNL_LINK("vxlan"); 4760