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 /* Find VXLAN socket based on network namespace, address family, UDP port, 75 * enabled unshareable flags and socket device binding (see l3mdev with 76 * non-default VRF). 77 */ 78 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family, 79 __be16 port, u32 flags, int ifindex) 80 { 81 struct vxlan_sock *vs; 82 83 flags &= VXLAN_F_RCV_FLAGS; 84 85 hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) { 86 if (inet_sk(vs->sock->sk)->inet_sport == port && 87 vxlan_get_sk_family(vs) == family && 88 vs->flags == flags && 89 vs->sock->sk->sk_bound_dev_if == ifindex) 90 return vs; 91 } 92 return NULL; 93 } 94 95 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, 96 int ifindex, __be32 vni, 97 struct vxlan_vni_node **vninode) 98 { 99 struct vxlan_vni_node *vnode; 100 struct vxlan_dev_node *node; 101 102 /* For flow based devices, map all packets to VNI 0 */ 103 if (vs->flags & VXLAN_F_COLLECT_METADATA && 104 !(vs->flags & VXLAN_F_VNIFILTER)) 105 vni = 0; 106 107 hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) { 108 if (!node->vxlan) 109 continue; 110 vnode = NULL; 111 if (node->vxlan->cfg.flags & VXLAN_F_VNIFILTER) { 112 vnode = vxlan_vnifilter_lookup(node->vxlan, vni); 113 if (!vnode) 114 continue; 115 } else if (node->vxlan->default_dst.remote_vni != vni) { 116 continue; 117 } 118 119 if (IS_ENABLED(CONFIG_IPV6)) { 120 const struct vxlan_config *cfg = &node->vxlan->cfg; 121 122 if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) && 123 cfg->remote_ifindex != ifindex) 124 continue; 125 } 126 127 if (vninode) 128 *vninode = vnode; 129 return node->vxlan; 130 } 131 132 return NULL; 133 } 134 135 /* Look up VNI in a per net namespace table */ 136 static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex, 137 __be32 vni, sa_family_t family, 138 __be16 port, u32 flags) 139 { 140 struct vxlan_sock *vs; 141 142 vs = vxlan_find_sock(net, family, port, flags, ifindex); 143 if (!vs) 144 return NULL; 145 146 return vxlan_vs_find_vni(vs, ifindex, vni, NULL); 147 } 148 149 /* Fill in neighbour message in skbuff. */ 150 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan, 151 const struct vxlan_fdb *fdb, 152 u32 portid, u32 seq, int type, unsigned int flags, 153 const struct vxlan_rdst *rdst) 154 { 155 unsigned long now = jiffies; 156 struct nda_cacheinfo ci; 157 bool send_ip, send_eth; 158 struct nlmsghdr *nlh; 159 struct nexthop *nh; 160 struct ndmsg *ndm; 161 int nh_family; 162 u32 nh_id; 163 164 nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags); 165 if (nlh == NULL) 166 return -EMSGSIZE; 167 168 ndm = nlmsg_data(nlh); 169 memset(ndm, 0, sizeof(*ndm)); 170 171 send_eth = send_ip = true; 172 173 rcu_read_lock(); 174 nh = rcu_dereference(fdb->nh); 175 if (nh) { 176 nh_family = nexthop_get_family(nh); 177 nh_id = nh->id; 178 } 179 rcu_read_unlock(); 180 181 if (type == RTM_GETNEIGH) { 182 if (rdst) { 183 send_ip = !vxlan_addr_any(&rdst->remote_ip); 184 ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET; 185 } else if (nh) { 186 ndm->ndm_family = nh_family; 187 } 188 send_eth = !is_zero_ether_addr(fdb->eth_addr); 189 } else 190 ndm->ndm_family = AF_BRIDGE; 191 ndm->ndm_state = fdb->state; 192 ndm->ndm_ifindex = vxlan->dev->ifindex; 193 ndm->ndm_flags = fdb->flags; 194 if (rdst && rdst->offloaded) 195 ndm->ndm_flags |= NTF_OFFLOADED; 196 ndm->ndm_type = RTN_UNICAST; 197 198 if (!net_eq(dev_net(vxlan->dev), vxlan->net) && 199 nla_put_s32(skb, NDA_LINK_NETNSID, 200 peernet2id(dev_net(vxlan->dev), vxlan->net))) 201 goto nla_put_failure; 202 203 if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr)) 204 goto nla_put_failure; 205 if (nh) { 206 if (nla_put_u32(skb, NDA_NH_ID, nh_id)) 207 goto nla_put_failure; 208 } else if (rdst) { 209 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, 210 &rdst->remote_ip)) 211 goto nla_put_failure; 212 213 if (rdst->remote_port && 214 rdst->remote_port != vxlan->cfg.dst_port && 215 nla_put_be16(skb, NDA_PORT, rdst->remote_port)) 216 goto nla_put_failure; 217 if (rdst->remote_vni != vxlan->default_dst.remote_vni && 218 nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni))) 219 goto nla_put_failure; 220 if (rdst->remote_ifindex && 221 nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex)) 222 goto nla_put_failure; 223 } 224 225 if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni && 226 nla_put_u32(skb, NDA_SRC_VNI, 227 be32_to_cpu(fdb->vni))) 228 goto nla_put_failure; 229 230 ci.ndm_used = jiffies_to_clock_t(now - fdb->used); 231 ci.ndm_confirmed = 0; 232 ci.ndm_updated = jiffies_to_clock_t(now - fdb->updated); 233 ci.ndm_refcnt = 0; 234 235 if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci)) 236 goto nla_put_failure; 237 238 nlmsg_end(skb, nlh); 239 return 0; 240 241 nla_put_failure: 242 nlmsg_cancel(skb, nlh); 243 return -EMSGSIZE; 244 } 245 246 static inline size_t vxlan_nlmsg_size(void) 247 { 248 return NLMSG_ALIGN(sizeof(struct ndmsg)) 249 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */ 250 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */ 251 + nla_total_size(sizeof(__be16)) /* NDA_PORT */ 252 + nla_total_size(sizeof(__be32)) /* NDA_VNI */ 253 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */ 254 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */ 255 + nla_total_size(sizeof(struct nda_cacheinfo)); 256 } 257 258 static void __vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb, 259 struct vxlan_rdst *rd, int type) 260 { 261 struct net *net = dev_net(vxlan->dev); 262 struct sk_buff *skb; 263 int err = -ENOBUFS; 264 265 skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC); 266 if (skb == NULL) 267 goto errout; 268 269 err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd); 270 if (err < 0) { 271 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */ 272 WARN_ON(err == -EMSGSIZE); 273 kfree_skb(skb); 274 goto errout; 275 } 276 277 rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC); 278 return; 279 errout: 280 if (err < 0) 281 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err); 282 } 283 284 static void vxlan_fdb_switchdev_notifier_info(const struct vxlan_dev *vxlan, 285 const struct vxlan_fdb *fdb, 286 const struct vxlan_rdst *rd, 287 struct netlink_ext_ack *extack, 288 struct switchdev_notifier_vxlan_fdb_info *fdb_info) 289 { 290 fdb_info->info.dev = vxlan->dev; 291 fdb_info->info.extack = extack; 292 fdb_info->remote_ip = rd->remote_ip; 293 fdb_info->remote_port = rd->remote_port; 294 fdb_info->remote_vni = rd->remote_vni; 295 fdb_info->remote_ifindex = rd->remote_ifindex; 296 memcpy(fdb_info->eth_addr, fdb->eth_addr, ETH_ALEN); 297 fdb_info->vni = fdb->vni; 298 fdb_info->offloaded = rd->offloaded; 299 fdb_info->added_by_user = fdb->flags & NTF_VXLAN_ADDED_BY_USER; 300 } 301 302 static int vxlan_fdb_switchdev_call_notifiers(struct vxlan_dev *vxlan, 303 struct vxlan_fdb *fdb, 304 struct vxlan_rdst *rd, 305 bool adding, 306 struct netlink_ext_ack *extack) 307 { 308 struct switchdev_notifier_vxlan_fdb_info info; 309 enum switchdev_notifier_type notifier_type; 310 int ret; 311 312 if (WARN_ON(!rd)) 313 return 0; 314 315 notifier_type = adding ? SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE 316 : SWITCHDEV_VXLAN_FDB_DEL_TO_DEVICE; 317 vxlan_fdb_switchdev_notifier_info(vxlan, fdb, rd, NULL, &info); 318 ret = call_switchdev_notifiers(notifier_type, vxlan->dev, 319 &info.info, extack); 320 return notifier_to_errno(ret); 321 } 322 323 static int vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb, 324 struct vxlan_rdst *rd, int type, bool swdev_notify, 325 struct netlink_ext_ack *extack) 326 { 327 int err; 328 329 if (swdev_notify && rd) { 330 switch (type) { 331 case RTM_NEWNEIGH: 332 err = vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd, 333 true, extack); 334 if (err) 335 return err; 336 break; 337 case RTM_DELNEIGH: 338 vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd, 339 false, extack); 340 break; 341 } 342 } 343 344 __vxlan_fdb_notify(vxlan, fdb, rd, type); 345 return 0; 346 } 347 348 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa) 349 { 350 struct vxlan_dev *vxlan = netdev_priv(dev); 351 struct vxlan_fdb f = { 352 .state = NUD_STALE, 353 }; 354 struct vxlan_rdst remote = { 355 .remote_ip = *ipa, /* goes to NDA_DST */ 356 .remote_vni = cpu_to_be32(VXLAN_N_VID), 357 }; 358 359 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL); 360 } 361 362 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN]) 363 { 364 struct vxlan_fdb f = { 365 .state = NUD_STALE, 366 }; 367 struct vxlan_rdst remote = { }; 368 369 memcpy(f.eth_addr, eth_addr, ETH_ALEN); 370 371 vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL); 372 } 373 374 /* Hash Ethernet address */ 375 static u32 eth_hash(const unsigned char *addr) 376 { 377 u64 value = get_unaligned((u64 *)addr); 378 379 /* only want 6 bytes */ 380 #ifdef __BIG_ENDIAN 381 value >>= 16; 382 #else 383 value <<= 16; 384 #endif 385 return hash_64(value, FDB_HASH_BITS); 386 } 387 388 u32 eth_vni_hash(const unsigned char *addr, __be32 vni) 389 { 390 /* use 1 byte of OUI and 3 bytes of NIC */ 391 u32 key = get_unaligned((u32 *)(addr + 2)); 392 393 return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1); 394 } 395 396 u32 fdb_head_index(struct vxlan_dev *vxlan, const u8 *mac, __be32 vni) 397 { 398 if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) 399 return eth_vni_hash(mac, vni); 400 else 401 return eth_hash(mac); 402 } 403 404 /* Hash chain to use given mac address */ 405 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan, 406 const u8 *mac, __be32 vni) 407 { 408 return &vxlan->fdb_head[fdb_head_index(vxlan, mac, vni)]; 409 } 410 411 /* Look up Ethernet address in forwarding table */ 412 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan, 413 const u8 *mac, __be32 vni) 414 { 415 struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni); 416 struct vxlan_fdb *f; 417 418 hlist_for_each_entry_rcu(f, head, hlist) { 419 if (ether_addr_equal(mac, f->eth_addr)) { 420 if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) { 421 if (vni == f->vni) 422 return f; 423 } else { 424 return f; 425 } 426 } 427 } 428 429 return NULL; 430 } 431 432 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan, 433 const u8 *mac, __be32 vni) 434 { 435 struct vxlan_fdb *f; 436 437 f = __vxlan_find_mac(vxlan, mac, vni); 438 if (f && f->used != jiffies) 439 f->used = jiffies; 440 441 return f; 442 } 443 444 /* caller should hold vxlan->hash_lock */ 445 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f, 446 union vxlan_addr *ip, __be16 port, 447 __be32 vni, __u32 ifindex) 448 { 449 struct vxlan_rdst *rd; 450 451 list_for_each_entry(rd, &f->remotes, list) { 452 if (vxlan_addr_equal(&rd->remote_ip, ip) && 453 rd->remote_port == port && 454 rd->remote_vni == vni && 455 rd->remote_ifindex == ifindex) 456 return rd; 457 } 458 459 return NULL; 460 } 461 462 int vxlan_fdb_find_uc(struct net_device *dev, const u8 *mac, __be32 vni, 463 struct switchdev_notifier_vxlan_fdb_info *fdb_info) 464 { 465 struct vxlan_dev *vxlan = netdev_priv(dev); 466 u8 eth_addr[ETH_ALEN + 2] = { 0 }; 467 struct vxlan_rdst *rdst; 468 struct vxlan_fdb *f; 469 int rc = 0; 470 471 if (is_multicast_ether_addr(mac) || 472 is_zero_ether_addr(mac)) 473 return -EINVAL; 474 475 ether_addr_copy(eth_addr, mac); 476 477 rcu_read_lock(); 478 479 f = __vxlan_find_mac(vxlan, eth_addr, vni); 480 if (!f) { 481 rc = -ENOENT; 482 goto out; 483 } 484 485 rdst = first_remote_rcu(f); 486 vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, NULL, fdb_info); 487 488 out: 489 rcu_read_unlock(); 490 return rc; 491 } 492 EXPORT_SYMBOL_GPL(vxlan_fdb_find_uc); 493 494 static int vxlan_fdb_notify_one(struct notifier_block *nb, 495 const struct vxlan_dev *vxlan, 496 const struct vxlan_fdb *f, 497 const struct vxlan_rdst *rdst, 498 struct netlink_ext_ack *extack) 499 { 500 struct switchdev_notifier_vxlan_fdb_info fdb_info; 501 int rc; 502 503 vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, extack, &fdb_info); 504 rc = nb->notifier_call(nb, SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE, 505 &fdb_info); 506 return notifier_to_errno(rc); 507 } 508 509 int vxlan_fdb_replay(const struct net_device *dev, __be32 vni, 510 struct notifier_block *nb, 511 struct netlink_ext_ack *extack) 512 { 513 struct vxlan_dev *vxlan; 514 struct vxlan_rdst *rdst; 515 struct vxlan_fdb *f; 516 unsigned int h; 517 int rc = 0; 518 519 if (!netif_is_vxlan(dev)) 520 return -EINVAL; 521 vxlan = netdev_priv(dev); 522 523 for (h = 0; h < FDB_HASH_SIZE; ++h) { 524 spin_lock_bh(&vxlan->hash_lock[h]); 525 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) { 526 if (f->vni == vni) { 527 list_for_each_entry(rdst, &f->remotes, list) { 528 rc = vxlan_fdb_notify_one(nb, vxlan, 529 f, rdst, 530 extack); 531 if (rc) 532 goto unlock; 533 } 534 } 535 } 536 spin_unlock_bh(&vxlan->hash_lock[h]); 537 } 538 return 0; 539 540 unlock: 541 spin_unlock_bh(&vxlan->hash_lock[h]); 542 return rc; 543 } 544 EXPORT_SYMBOL_GPL(vxlan_fdb_replay); 545 546 void vxlan_fdb_clear_offload(const struct net_device *dev, __be32 vni) 547 { 548 struct vxlan_dev *vxlan; 549 struct vxlan_rdst *rdst; 550 struct vxlan_fdb *f; 551 unsigned int h; 552 553 if (!netif_is_vxlan(dev)) 554 return; 555 vxlan = netdev_priv(dev); 556 557 for (h = 0; h < FDB_HASH_SIZE; ++h) { 558 spin_lock_bh(&vxlan->hash_lock[h]); 559 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) 560 if (f->vni == vni) 561 list_for_each_entry(rdst, &f->remotes, list) 562 rdst->offloaded = false; 563 spin_unlock_bh(&vxlan->hash_lock[h]); 564 } 565 566 } 567 EXPORT_SYMBOL_GPL(vxlan_fdb_clear_offload); 568 569 /* Replace destination of unicast mac */ 570 static int vxlan_fdb_replace(struct vxlan_fdb *f, 571 union vxlan_addr *ip, __be16 port, __be32 vni, 572 __u32 ifindex, struct vxlan_rdst *oldrd) 573 { 574 struct vxlan_rdst *rd; 575 576 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex); 577 if (rd) 578 return 0; 579 580 rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list); 581 if (!rd) 582 return 0; 583 584 *oldrd = *rd; 585 dst_cache_reset(&rd->dst_cache); 586 rd->remote_ip = *ip; 587 rd->remote_port = port; 588 rd->remote_vni = vni; 589 rd->remote_ifindex = ifindex; 590 rd->offloaded = false; 591 return 1; 592 } 593 594 /* Add/update destinations for multicast */ 595 static int vxlan_fdb_append(struct vxlan_fdb *f, 596 union vxlan_addr *ip, __be16 port, __be32 vni, 597 __u32 ifindex, struct vxlan_rdst **rdp) 598 { 599 struct vxlan_rdst *rd; 600 601 rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex); 602 if (rd) 603 return 0; 604 605 rd = kmalloc(sizeof(*rd), GFP_ATOMIC); 606 if (rd == NULL) 607 return -ENOMEM; 608 609 if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) { 610 kfree(rd); 611 return -ENOMEM; 612 } 613 614 rd->remote_ip = *ip; 615 rd->remote_port = port; 616 rd->offloaded = false; 617 rd->remote_vni = vni; 618 rd->remote_ifindex = ifindex; 619 620 list_add_tail_rcu(&rd->list, &f->remotes); 621 622 *rdp = rd; 623 return 1; 624 } 625 626 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb, 627 unsigned int off, 628 struct vxlanhdr *vh, size_t hdrlen, 629 __be32 vni_field, 630 struct gro_remcsum *grc, 631 bool nopartial) 632 { 633 size_t start, offset; 634 635 if (skb->remcsum_offload) 636 return vh; 637 638 if (!NAPI_GRO_CB(skb)->csum_valid) 639 return NULL; 640 641 start = vxlan_rco_start(vni_field); 642 offset = start + vxlan_rco_offset(vni_field); 643 644 vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen, 645 start, offset, grc, nopartial); 646 647 skb->remcsum_offload = 1; 648 649 return vh; 650 } 651 652 static struct sk_buff *vxlan_gro_receive(struct sock *sk, 653 struct list_head *head, 654 struct sk_buff *skb) 655 { 656 struct sk_buff *pp = NULL; 657 struct sk_buff *p; 658 struct vxlanhdr *vh, *vh2; 659 unsigned int hlen, off_vx; 660 int flush = 1; 661 struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk); 662 __be32 flags; 663 struct gro_remcsum grc; 664 665 skb_gro_remcsum_init(&grc); 666 667 off_vx = skb_gro_offset(skb); 668 hlen = off_vx + sizeof(*vh); 669 vh = skb_gro_header(skb, hlen, off_vx); 670 if (unlikely(!vh)) 671 goto out; 672 673 skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr)); 674 675 flags = vh->vx_flags; 676 677 if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) { 678 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr), 679 vh->vx_vni, &grc, 680 !!(vs->flags & 681 VXLAN_F_REMCSUM_NOPARTIAL)); 682 683 if (!vh) 684 goto out; 685 } 686 687 skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */ 688 689 list_for_each_entry(p, head, list) { 690 if (!NAPI_GRO_CB(p)->same_flow) 691 continue; 692 693 vh2 = (struct vxlanhdr *)(p->data + off_vx); 694 if (vh->vx_flags != vh2->vx_flags || 695 vh->vx_vni != vh2->vx_vni) { 696 NAPI_GRO_CB(p)->same_flow = 0; 697 continue; 698 } 699 } 700 701 pp = call_gro_receive(eth_gro_receive, head, skb); 702 flush = 0; 703 704 out: 705 skb_gro_flush_final_remcsum(skb, pp, flush, &grc); 706 707 return pp; 708 } 709 710 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff) 711 { 712 /* Sets 'skb->inner_mac_header' since we are always called with 713 * 'skb->encapsulation' set. 714 */ 715 return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr)); 716 } 717 718 static struct vxlan_fdb *vxlan_fdb_alloc(struct vxlan_dev *vxlan, const u8 *mac, 719 __u16 state, __be32 src_vni, 720 __u16 ndm_flags) 721 { 722 struct vxlan_fdb *f; 723 724 f = kmalloc(sizeof(*f), GFP_ATOMIC); 725 if (!f) 726 return NULL; 727 f->state = state; 728 f->flags = ndm_flags; 729 f->updated = f->used = jiffies; 730 f->vni = src_vni; 731 f->nh = NULL; 732 RCU_INIT_POINTER(f->vdev, vxlan); 733 INIT_LIST_HEAD(&f->nh_list); 734 INIT_LIST_HEAD(&f->remotes); 735 memcpy(f->eth_addr, mac, ETH_ALEN); 736 737 return f; 738 } 739 740 static void vxlan_fdb_insert(struct vxlan_dev *vxlan, const u8 *mac, 741 __be32 src_vni, struct vxlan_fdb *f) 742 { 743 ++vxlan->addrcnt; 744 hlist_add_head_rcu(&f->hlist, 745 vxlan_fdb_head(vxlan, mac, src_vni)); 746 } 747 748 static int vxlan_fdb_nh_update(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb, 749 u32 nhid, struct netlink_ext_ack *extack) 750 { 751 struct nexthop *old_nh = rtnl_dereference(fdb->nh); 752 struct nexthop *nh; 753 int err = -EINVAL; 754 755 if (old_nh && old_nh->id == nhid) 756 return 0; 757 758 nh = nexthop_find_by_id(vxlan->net, nhid); 759 if (!nh) { 760 NL_SET_ERR_MSG(extack, "Nexthop id does not exist"); 761 goto err_inval; 762 } 763 764 if (!nexthop_get(nh)) { 765 NL_SET_ERR_MSG(extack, "Nexthop has been deleted"); 766 nh = NULL; 767 goto err_inval; 768 } 769 if (!nexthop_is_fdb(nh)) { 770 NL_SET_ERR_MSG(extack, "Nexthop is not a fdb nexthop"); 771 goto err_inval; 772 } 773 774 if (!nexthop_is_multipath(nh)) { 775 NL_SET_ERR_MSG(extack, "Nexthop is not a multipath group"); 776 goto err_inval; 777 } 778 779 /* check nexthop group family */ 780 switch (vxlan->default_dst.remote_ip.sa.sa_family) { 781 case AF_INET: 782 if (!nexthop_has_v4(nh)) { 783 err = -EAFNOSUPPORT; 784 NL_SET_ERR_MSG(extack, "Nexthop group family not supported"); 785 goto err_inval; 786 } 787 break; 788 case AF_INET6: 789 if (nexthop_has_v4(nh)) { 790 err = -EAFNOSUPPORT; 791 NL_SET_ERR_MSG(extack, "Nexthop group family not supported"); 792 goto err_inval; 793 } 794 } 795 796 if (old_nh) { 797 list_del_rcu(&fdb->nh_list); 798 nexthop_put(old_nh); 799 } 800 rcu_assign_pointer(fdb->nh, nh); 801 list_add_tail_rcu(&fdb->nh_list, &nh->fdb_list); 802 return 1; 803 804 err_inval: 805 if (nh) 806 nexthop_put(nh); 807 return err; 808 } 809 810 int vxlan_fdb_create(struct vxlan_dev *vxlan, 811 const u8 *mac, union vxlan_addr *ip, 812 __u16 state, __be16 port, __be32 src_vni, 813 __be32 vni, __u32 ifindex, __u16 ndm_flags, 814 u32 nhid, struct vxlan_fdb **fdb, 815 struct netlink_ext_ack *extack) 816 { 817 struct vxlan_rdst *rd = NULL; 818 struct vxlan_fdb *f; 819 int rc; 820 821 if (vxlan->cfg.addrmax && 822 vxlan->addrcnt >= vxlan->cfg.addrmax) 823 return -ENOSPC; 824 825 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip); 826 f = vxlan_fdb_alloc(vxlan, mac, state, src_vni, ndm_flags); 827 if (!f) 828 return -ENOMEM; 829 830 if (nhid) 831 rc = vxlan_fdb_nh_update(vxlan, f, nhid, extack); 832 else 833 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd); 834 if (rc < 0) 835 goto errout; 836 837 *fdb = f; 838 839 return 0; 840 841 errout: 842 kfree(f); 843 return rc; 844 } 845 846 static void __vxlan_fdb_free(struct vxlan_fdb *f) 847 { 848 struct vxlan_rdst *rd, *nd; 849 struct nexthop *nh; 850 851 nh = rcu_dereference_raw(f->nh); 852 if (nh) { 853 rcu_assign_pointer(f->nh, NULL); 854 rcu_assign_pointer(f->vdev, NULL); 855 nexthop_put(nh); 856 } 857 858 list_for_each_entry_safe(rd, nd, &f->remotes, list) { 859 dst_cache_destroy(&rd->dst_cache); 860 kfree(rd); 861 } 862 kfree(f); 863 } 864 865 static void vxlan_fdb_free(struct rcu_head *head) 866 { 867 struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu); 868 869 __vxlan_fdb_free(f); 870 } 871 872 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f, 873 bool do_notify, bool swdev_notify) 874 { 875 struct vxlan_rdst *rd; 876 877 netdev_dbg(vxlan->dev, "delete %pM\n", f->eth_addr); 878 879 --vxlan->addrcnt; 880 if (do_notify) { 881 if (rcu_access_pointer(f->nh)) 882 vxlan_fdb_notify(vxlan, f, NULL, RTM_DELNEIGH, 883 swdev_notify, NULL); 884 else 885 list_for_each_entry(rd, &f->remotes, list) 886 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, 887 swdev_notify, NULL); 888 } 889 890 hlist_del_rcu(&f->hlist); 891 list_del_rcu(&f->nh_list); 892 call_rcu(&f->rcu, vxlan_fdb_free); 893 } 894 895 static void vxlan_dst_free(struct rcu_head *head) 896 { 897 struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu); 898 899 dst_cache_destroy(&rd->dst_cache); 900 kfree(rd); 901 } 902 903 static int vxlan_fdb_update_existing(struct vxlan_dev *vxlan, 904 union vxlan_addr *ip, 905 __u16 state, __u16 flags, 906 __be16 port, __be32 vni, 907 __u32 ifindex, __u16 ndm_flags, 908 struct vxlan_fdb *f, u32 nhid, 909 bool swdev_notify, 910 struct netlink_ext_ack *extack) 911 { 912 __u16 fdb_flags = (ndm_flags & ~NTF_USE); 913 struct vxlan_rdst *rd = NULL; 914 struct vxlan_rdst oldrd; 915 int notify = 0; 916 int rc = 0; 917 int err; 918 919 if (nhid && !rcu_access_pointer(f->nh)) { 920 NL_SET_ERR_MSG(extack, 921 "Cannot replace an existing non nexthop fdb with a nexthop"); 922 return -EOPNOTSUPP; 923 } 924 925 if (nhid && (flags & NLM_F_APPEND)) { 926 NL_SET_ERR_MSG(extack, 927 "Cannot append to a nexthop fdb"); 928 return -EOPNOTSUPP; 929 } 930 931 /* Do not allow an externally learned entry to take over an entry added 932 * by the user. 933 */ 934 if (!(fdb_flags & NTF_EXT_LEARNED) || 935 !(f->flags & NTF_VXLAN_ADDED_BY_USER)) { 936 if (f->state != state) { 937 f->state = state; 938 f->updated = jiffies; 939 notify = 1; 940 } 941 if (f->flags != fdb_flags) { 942 f->flags = fdb_flags; 943 f->updated = jiffies; 944 notify = 1; 945 } 946 } 947 948 if ((flags & NLM_F_REPLACE)) { 949 /* Only change unicasts */ 950 if (!(is_multicast_ether_addr(f->eth_addr) || 951 is_zero_ether_addr(f->eth_addr))) { 952 if (nhid) { 953 rc = vxlan_fdb_nh_update(vxlan, f, nhid, extack); 954 if (rc < 0) 955 return rc; 956 } else { 957 rc = vxlan_fdb_replace(f, ip, port, vni, 958 ifindex, &oldrd); 959 } 960 notify |= rc; 961 } else { 962 NL_SET_ERR_MSG(extack, "Cannot replace non-unicast fdb entries"); 963 return -EOPNOTSUPP; 964 } 965 } 966 if ((flags & NLM_F_APPEND) && 967 (is_multicast_ether_addr(f->eth_addr) || 968 is_zero_ether_addr(f->eth_addr))) { 969 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd); 970 971 if (rc < 0) 972 return rc; 973 notify |= rc; 974 } 975 976 if (ndm_flags & NTF_USE) 977 f->used = jiffies; 978 979 if (notify) { 980 if (rd == NULL) 981 rd = first_remote_rtnl(f); 982 983 err = vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH, 984 swdev_notify, extack); 985 if (err) 986 goto err_notify; 987 } 988 989 return 0; 990 991 err_notify: 992 if (nhid) 993 return err; 994 if ((flags & NLM_F_REPLACE) && rc) 995 *rd = oldrd; 996 else if ((flags & NLM_F_APPEND) && rc) { 997 list_del_rcu(&rd->list); 998 call_rcu(&rd->rcu, vxlan_dst_free); 999 } 1000 return err; 1001 } 1002 1003 static int vxlan_fdb_update_create(struct vxlan_dev *vxlan, 1004 const u8 *mac, union vxlan_addr *ip, 1005 __u16 state, __u16 flags, 1006 __be16 port, __be32 src_vni, __be32 vni, 1007 __u32 ifindex, __u16 ndm_flags, u32 nhid, 1008 bool swdev_notify, 1009 struct netlink_ext_ack *extack) 1010 { 1011 __u16 fdb_flags = (ndm_flags & ~NTF_USE); 1012 struct vxlan_fdb *f; 1013 int rc; 1014 1015 /* Disallow replace to add a multicast entry */ 1016 if ((flags & NLM_F_REPLACE) && 1017 (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac))) 1018 return -EOPNOTSUPP; 1019 1020 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip); 1021 rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni, 1022 vni, ifindex, fdb_flags, nhid, &f, extack); 1023 if (rc < 0) 1024 return rc; 1025 1026 vxlan_fdb_insert(vxlan, mac, src_vni, f); 1027 rc = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH, 1028 swdev_notify, extack); 1029 if (rc) 1030 goto err_notify; 1031 1032 return 0; 1033 1034 err_notify: 1035 vxlan_fdb_destroy(vxlan, f, false, false); 1036 return rc; 1037 } 1038 1039 /* Add new entry to forwarding table -- assumes lock held */ 1040 int vxlan_fdb_update(struct vxlan_dev *vxlan, 1041 const u8 *mac, union vxlan_addr *ip, 1042 __u16 state, __u16 flags, 1043 __be16 port, __be32 src_vni, __be32 vni, 1044 __u32 ifindex, __u16 ndm_flags, u32 nhid, 1045 bool swdev_notify, 1046 struct netlink_ext_ack *extack) 1047 { 1048 struct vxlan_fdb *f; 1049 1050 f = __vxlan_find_mac(vxlan, mac, src_vni); 1051 if (f) { 1052 if (flags & NLM_F_EXCL) { 1053 netdev_dbg(vxlan->dev, 1054 "lost race to create %pM\n", mac); 1055 return -EEXIST; 1056 } 1057 1058 return vxlan_fdb_update_existing(vxlan, ip, state, flags, port, 1059 vni, ifindex, ndm_flags, f, 1060 nhid, swdev_notify, extack); 1061 } else { 1062 if (!(flags & NLM_F_CREATE)) 1063 return -ENOENT; 1064 1065 return vxlan_fdb_update_create(vxlan, mac, ip, state, flags, 1066 port, src_vni, vni, ifindex, 1067 ndm_flags, nhid, swdev_notify, 1068 extack); 1069 } 1070 } 1071 1072 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f, 1073 struct vxlan_rdst *rd, bool swdev_notify) 1074 { 1075 list_del_rcu(&rd->list); 1076 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, swdev_notify, NULL); 1077 call_rcu(&rd->rcu, vxlan_dst_free); 1078 } 1079 1080 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan, 1081 union vxlan_addr *ip, __be16 *port, __be32 *src_vni, 1082 __be32 *vni, u32 *ifindex, u32 *nhid, 1083 struct netlink_ext_ack *extack) 1084 { 1085 struct net *net = dev_net(vxlan->dev); 1086 int err; 1087 1088 if (tb[NDA_NH_ID] && 1089 (tb[NDA_DST] || tb[NDA_VNI] || tb[NDA_IFINDEX] || tb[NDA_PORT])) { 1090 NL_SET_ERR_MSG(extack, "DST, VNI, ifindex and port are mutually exclusive with NH_ID"); 1091 return -EINVAL; 1092 } 1093 1094 if (tb[NDA_DST]) { 1095 err = vxlan_nla_get_addr(ip, tb[NDA_DST]); 1096 if (err) { 1097 NL_SET_ERR_MSG(extack, "Unsupported address family"); 1098 return err; 1099 } 1100 } else { 1101 union vxlan_addr *remote = &vxlan->default_dst.remote_ip; 1102 1103 if (remote->sa.sa_family == AF_INET) { 1104 ip->sin.sin_addr.s_addr = htonl(INADDR_ANY); 1105 ip->sa.sa_family = AF_INET; 1106 #if IS_ENABLED(CONFIG_IPV6) 1107 } else { 1108 ip->sin6.sin6_addr = in6addr_any; 1109 ip->sa.sa_family = AF_INET6; 1110 #endif 1111 } 1112 } 1113 1114 if (tb[NDA_PORT]) { 1115 if (nla_len(tb[NDA_PORT]) != sizeof(__be16)) { 1116 NL_SET_ERR_MSG(extack, "Invalid vxlan port"); 1117 return -EINVAL; 1118 } 1119 *port = nla_get_be16(tb[NDA_PORT]); 1120 } else { 1121 *port = vxlan->cfg.dst_port; 1122 } 1123 1124 if (tb[NDA_VNI]) { 1125 if (nla_len(tb[NDA_VNI]) != sizeof(u32)) { 1126 NL_SET_ERR_MSG(extack, "Invalid vni"); 1127 return -EINVAL; 1128 } 1129 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI])); 1130 } else { 1131 *vni = vxlan->default_dst.remote_vni; 1132 } 1133 1134 if (tb[NDA_SRC_VNI]) { 1135 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32)) { 1136 NL_SET_ERR_MSG(extack, "Invalid src vni"); 1137 return -EINVAL; 1138 } 1139 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI])); 1140 } else { 1141 *src_vni = vxlan->default_dst.remote_vni; 1142 } 1143 1144 if (tb[NDA_IFINDEX]) { 1145 struct net_device *tdev; 1146 1147 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32)) { 1148 NL_SET_ERR_MSG(extack, "Invalid ifindex"); 1149 return -EINVAL; 1150 } 1151 *ifindex = nla_get_u32(tb[NDA_IFINDEX]); 1152 tdev = __dev_get_by_index(net, *ifindex); 1153 if (!tdev) { 1154 NL_SET_ERR_MSG(extack, "Device not found"); 1155 return -EADDRNOTAVAIL; 1156 } 1157 } else { 1158 *ifindex = 0; 1159 } 1160 1161 if (tb[NDA_NH_ID]) 1162 *nhid = nla_get_u32(tb[NDA_NH_ID]); 1163 else 1164 *nhid = 0; 1165 1166 return 0; 1167 } 1168 1169 /* Add static entry (via netlink) */ 1170 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[], 1171 struct net_device *dev, 1172 const unsigned char *addr, u16 vid, u16 flags, 1173 struct netlink_ext_ack *extack) 1174 { 1175 struct vxlan_dev *vxlan = netdev_priv(dev); 1176 /* struct net *net = dev_net(vxlan->dev); */ 1177 union vxlan_addr ip; 1178 __be16 port; 1179 __be32 src_vni, vni; 1180 u32 ifindex, nhid; 1181 u32 hash_index; 1182 int err; 1183 1184 if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) { 1185 pr_info("RTM_NEWNEIGH with invalid state %#x\n", 1186 ndm->ndm_state); 1187 return -EINVAL; 1188 } 1189 1190 if (!tb || (!tb[NDA_DST] && !tb[NDA_NH_ID])) 1191 return -EINVAL; 1192 1193 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex, 1194 &nhid, extack); 1195 if (err) 1196 return err; 1197 1198 if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family) 1199 return -EAFNOSUPPORT; 1200 1201 hash_index = fdb_head_index(vxlan, addr, src_vni); 1202 spin_lock_bh(&vxlan->hash_lock[hash_index]); 1203 err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags, 1204 port, src_vni, vni, ifindex, 1205 ndm->ndm_flags | NTF_VXLAN_ADDED_BY_USER, 1206 nhid, true, extack); 1207 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 1208 1209 return err; 1210 } 1211 1212 int __vxlan_fdb_delete(struct vxlan_dev *vxlan, 1213 const unsigned char *addr, union vxlan_addr ip, 1214 __be16 port, __be32 src_vni, __be32 vni, 1215 u32 ifindex, bool swdev_notify) 1216 { 1217 struct vxlan_rdst *rd = NULL; 1218 struct vxlan_fdb *f; 1219 int err = -ENOENT; 1220 1221 f = vxlan_find_mac(vxlan, addr, src_vni); 1222 if (!f) 1223 return err; 1224 1225 if (!vxlan_addr_any(&ip)) { 1226 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex); 1227 if (!rd) 1228 goto out; 1229 } 1230 1231 /* remove a destination if it's not the only one on the list, 1232 * otherwise destroy the fdb entry 1233 */ 1234 if (rd && !list_is_singular(&f->remotes)) { 1235 vxlan_fdb_dst_destroy(vxlan, f, rd, swdev_notify); 1236 goto out; 1237 } 1238 1239 vxlan_fdb_destroy(vxlan, f, true, swdev_notify); 1240 1241 out: 1242 return 0; 1243 } 1244 1245 /* Delete entry (via netlink) */ 1246 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[], 1247 struct net_device *dev, 1248 const unsigned char *addr, u16 vid, 1249 struct netlink_ext_ack *extack) 1250 { 1251 struct vxlan_dev *vxlan = netdev_priv(dev); 1252 union vxlan_addr ip; 1253 __be32 src_vni, vni; 1254 u32 ifindex, nhid; 1255 u32 hash_index; 1256 __be16 port; 1257 int err; 1258 1259 err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex, 1260 &nhid, extack); 1261 if (err) 1262 return err; 1263 1264 hash_index = fdb_head_index(vxlan, addr, src_vni); 1265 spin_lock_bh(&vxlan->hash_lock[hash_index]); 1266 err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex, 1267 true); 1268 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 1269 1270 return err; 1271 } 1272 1273 /* Dump forwarding table */ 1274 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb, 1275 struct net_device *dev, 1276 struct net_device *filter_dev, int *idx) 1277 { 1278 struct vxlan_dev *vxlan = netdev_priv(dev); 1279 unsigned int h; 1280 int err = 0; 1281 1282 for (h = 0; h < FDB_HASH_SIZE; ++h) { 1283 struct vxlan_fdb *f; 1284 1285 rcu_read_lock(); 1286 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) { 1287 struct vxlan_rdst *rd; 1288 1289 if (rcu_access_pointer(f->nh)) { 1290 if (*idx < cb->args[2]) 1291 goto skip_nh; 1292 err = vxlan_fdb_info(skb, vxlan, f, 1293 NETLINK_CB(cb->skb).portid, 1294 cb->nlh->nlmsg_seq, 1295 RTM_NEWNEIGH, 1296 NLM_F_MULTI, NULL); 1297 if (err < 0) { 1298 rcu_read_unlock(); 1299 goto out; 1300 } 1301 skip_nh: 1302 *idx += 1; 1303 continue; 1304 } 1305 1306 list_for_each_entry_rcu(rd, &f->remotes, list) { 1307 if (*idx < cb->args[2]) 1308 goto skip; 1309 1310 err = vxlan_fdb_info(skb, vxlan, f, 1311 NETLINK_CB(cb->skb).portid, 1312 cb->nlh->nlmsg_seq, 1313 RTM_NEWNEIGH, 1314 NLM_F_MULTI, rd); 1315 if (err < 0) { 1316 rcu_read_unlock(); 1317 goto out; 1318 } 1319 skip: 1320 *idx += 1; 1321 } 1322 } 1323 rcu_read_unlock(); 1324 } 1325 out: 1326 return err; 1327 } 1328 1329 static int vxlan_fdb_get(struct sk_buff *skb, 1330 struct nlattr *tb[], 1331 struct net_device *dev, 1332 const unsigned char *addr, 1333 u16 vid, u32 portid, u32 seq, 1334 struct netlink_ext_ack *extack) 1335 { 1336 struct vxlan_dev *vxlan = netdev_priv(dev); 1337 struct vxlan_fdb *f; 1338 __be32 vni; 1339 int err; 1340 1341 if (tb[NDA_VNI]) 1342 vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI])); 1343 else 1344 vni = vxlan->default_dst.remote_vni; 1345 1346 rcu_read_lock(); 1347 1348 f = __vxlan_find_mac(vxlan, addr, vni); 1349 if (!f) { 1350 NL_SET_ERR_MSG(extack, "Fdb entry not found"); 1351 err = -ENOENT; 1352 goto errout; 1353 } 1354 1355 err = vxlan_fdb_info(skb, vxlan, f, portid, seq, 1356 RTM_NEWNEIGH, 0, first_remote_rcu(f)); 1357 errout: 1358 rcu_read_unlock(); 1359 return err; 1360 } 1361 1362 /* Watch incoming packets to learn mapping between Ethernet address 1363 * and Tunnel endpoint. 1364 * Return true if packet is bogus and should be dropped. 1365 */ 1366 static bool vxlan_snoop(struct net_device *dev, 1367 union vxlan_addr *src_ip, const u8 *src_mac, 1368 u32 src_ifindex, __be32 vni) 1369 { 1370 struct vxlan_dev *vxlan = netdev_priv(dev); 1371 struct vxlan_fdb *f; 1372 u32 ifindex = 0; 1373 1374 #if IS_ENABLED(CONFIG_IPV6) 1375 if (src_ip->sa.sa_family == AF_INET6 && 1376 (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL)) 1377 ifindex = src_ifindex; 1378 #endif 1379 1380 f = vxlan_find_mac(vxlan, src_mac, vni); 1381 if (likely(f)) { 1382 struct vxlan_rdst *rdst = first_remote_rcu(f); 1383 1384 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) && 1385 rdst->remote_ifindex == ifindex)) 1386 return false; 1387 1388 /* Don't migrate static entries, drop packets */ 1389 if (f->state & (NUD_PERMANENT | NUD_NOARP)) 1390 return true; 1391 1392 /* Don't override an fdb with nexthop with a learnt entry */ 1393 if (rcu_access_pointer(f->nh)) 1394 return true; 1395 1396 if (net_ratelimit()) 1397 netdev_info(dev, 1398 "%pM migrated from %pIS to %pIS\n", 1399 src_mac, &rdst->remote_ip.sa, &src_ip->sa); 1400 1401 rdst->remote_ip = *src_ip; 1402 f->updated = jiffies; 1403 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH, true, NULL); 1404 } else { 1405 u32 hash_index = fdb_head_index(vxlan, src_mac, vni); 1406 1407 /* learned new entry */ 1408 spin_lock(&vxlan->hash_lock[hash_index]); 1409 1410 /* close off race between vxlan_flush and incoming packets */ 1411 if (netif_running(dev)) 1412 vxlan_fdb_update(vxlan, src_mac, src_ip, 1413 NUD_REACHABLE, 1414 NLM_F_EXCL|NLM_F_CREATE, 1415 vxlan->cfg.dst_port, 1416 vni, 1417 vxlan->default_dst.remote_vni, 1418 ifindex, NTF_SELF, 0, true, NULL); 1419 spin_unlock(&vxlan->hash_lock[hash_index]); 1420 } 1421 1422 return false; 1423 } 1424 1425 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs) 1426 { 1427 struct vxlan_net *vn; 1428 1429 if (!vs) 1430 return false; 1431 if (!refcount_dec_and_test(&vs->refcnt)) 1432 return false; 1433 1434 vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id); 1435 spin_lock(&vn->sock_lock); 1436 hlist_del_rcu(&vs->hlist); 1437 udp_tunnel_notify_del_rx_port(vs->sock, 1438 (vs->flags & VXLAN_F_GPE) ? 1439 UDP_TUNNEL_TYPE_VXLAN_GPE : 1440 UDP_TUNNEL_TYPE_VXLAN); 1441 spin_unlock(&vn->sock_lock); 1442 1443 return true; 1444 } 1445 1446 static void vxlan_sock_release(struct vxlan_dev *vxlan) 1447 { 1448 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock); 1449 #if IS_ENABLED(CONFIG_IPV6) 1450 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock); 1451 1452 RCU_INIT_POINTER(vxlan->vn6_sock, NULL); 1453 #endif 1454 1455 RCU_INIT_POINTER(vxlan->vn4_sock, NULL); 1456 synchronize_net(); 1457 1458 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER) 1459 vxlan_vs_del_vnigrp(vxlan); 1460 else 1461 vxlan_vs_del_dev(vxlan); 1462 1463 if (__vxlan_sock_release_prep(sock4)) { 1464 udp_tunnel_sock_release(sock4->sock); 1465 kfree(sock4); 1466 } 1467 1468 #if IS_ENABLED(CONFIG_IPV6) 1469 if (__vxlan_sock_release_prep(sock6)) { 1470 udp_tunnel_sock_release(sock6->sock); 1471 kfree(sock6); 1472 } 1473 #endif 1474 } 1475 1476 static bool vxlan_remcsum(struct vxlanhdr *unparsed, 1477 struct sk_buff *skb, u32 vxflags) 1478 { 1479 size_t start, offset; 1480 1481 if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload) 1482 goto out; 1483 1484 start = vxlan_rco_start(unparsed->vx_vni); 1485 offset = start + vxlan_rco_offset(unparsed->vx_vni); 1486 1487 if (!pskb_may_pull(skb, offset + sizeof(u16))) 1488 return false; 1489 1490 skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset, 1491 !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL)); 1492 out: 1493 unparsed->vx_flags &= ~VXLAN_HF_RCO; 1494 unparsed->vx_vni &= VXLAN_VNI_MASK; 1495 return true; 1496 } 1497 1498 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed, 1499 struct sk_buff *skb, u32 vxflags, 1500 struct vxlan_metadata *md) 1501 { 1502 struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed; 1503 struct metadata_dst *tun_dst; 1504 1505 if (!(unparsed->vx_flags & VXLAN_HF_GBP)) 1506 goto out; 1507 1508 md->gbp = ntohs(gbp->policy_id); 1509 1510 tun_dst = (struct metadata_dst *)skb_dst(skb); 1511 if (tun_dst) { 1512 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT; 1513 tun_dst->u.tun_info.options_len = sizeof(*md); 1514 } 1515 if (gbp->dont_learn) 1516 md->gbp |= VXLAN_GBP_DONT_LEARN; 1517 1518 if (gbp->policy_applied) 1519 md->gbp |= VXLAN_GBP_POLICY_APPLIED; 1520 1521 /* In flow-based mode, GBP is carried in dst_metadata */ 1522 if (!(vxflags & VXLAN_F_COLLECT_METADATA)) 1523 skb->mark = md->gbp; 1524 out: 1525 unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS; 1526 } 1527 1528 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed, 1529 __be16 *protocol, 1530 struct sk_buff *skb, u32 vxflags) 1531 { 1532 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed; 1533 1534 /* Need to have Next Protocol set for interfaces in GPE mode. */ 1535 if (!gpe->np_applied) 1536 return false; 1537 /* "The initial version is 0. If a receiver does not support the 1538 * version indicated it MUST drop the packet. 1539 */ 1540 if (gpe->version != 0) 1541 return false; 1542 /* "When the O bit is set to 1, the packet is an OAM packet and OAM 1543 * processing MUST occur." However, we don't implement OAM 1544 * processing, thus drop the packet. 1545 */ 1546 if (gpe->oam_flag) 1547 return false; 1548 1549 *protocol = tun_p_to_eth_p(gpe->next_protocol); 1550 if (!*protocol) 1551 return false; 1552 1553 unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS; 1554 return true; 1555 } 1556 1557 static bool vxlan_set_mac(struct vxlan_dev *vxlan, 1558 struct vxlan_sock *vs, 1559 struct sk_buff *skb, __be32 vni) 1560 { 1561 union vxlan_addr saddr; 1562 u32 ifindex = skb->dev->ifindex; 1563 1564 skb_reset_mac_header(skb); 1565 skb->protocol = eth_type_trans(skb, vxlan->dev); 1566 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); 1567 1568 /* Ignore packet loops (and multicast echo) */ 1569 if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr)) 1570 return false; 1571 1572 /* Get address from the outer IP header */ 1573 if (vxlan_get_sk_family(vs) == AF_INET) { 1574 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr; 1575 saddr.sa.sa_family = AF_INET; 1576 #if IS_ENABLED(CONFIG_IPV6) 1577 } else { 1578 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr; 1579 saddr.sa.sa_family = AF_INET6; 1580 #endif 1581 } 1582 1583 if ((vxlan->cfg.flags & VXLAN_F_LEARN) && 1584 vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni)) 1585 return false; 1586 1587 return true; 1588 } 1589 1590 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph, 1591 struct sk_buff *skb) 1592 { 1593 int err = 0; 1594 1595 if (vxlan_get_sk_family(vs) == AF_INET) 1596 err = IP_ECN_decapsulate(oiph, skb); 1597 #if IS_ENABLED(CONFIG_IPV6) 1598 else 1599 err = IP6_ECN_decapsulate(oiph, skb); 1600 #endif 1601 1602 if (unlikely(err) && log_ecn_error) { 1603 if (vxlan_get_sk_family(vs) == AF_INET) 1604 net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n", 1605 &((struct iphdr *)oiph)->saddr, 1606 ((struct iphdr *)oiph)->tos); 1607 else 1608 net_info_ratelimited("non-ECT from %pI6\n", 1609 &((struct ipv6hdr *)oiph)->saddr); 1610 } 1611 return err <= 1; 1612 } 1613 1614 /* Callback from net/ipv4/udp.c to receive packets */ 1615 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb) 1616 { 1617 struct vxlan_vni_node *vninode = NULL; 1618 struct vxlan_dev *vxlan; 1619 struct vxlan_sock *vs; 1620 struct vxlanhdr unparsed; 1621 struct vxlan_metadata _md; 1622 struct vxlan_metadata *md = &_md; 1623 __be16 protocol = htons(ETH_P_TEB); 1624 bool raw_proto = false; 1625 void *oiph; 1626 __be32 vni = 0; 1627 1628 /* Need UDP and VXLAN header to be present */ 1629 if (!pskb_may_pull(skb, VXLAN_HLEN)) 1630 goto drop; 1631 1632 unparsed = *vxlan_hdr(skb); 1633 /* VNI flag always required to be set */ 1634 if (!(unparsed.vx_flags & VXLAN_HF_VNI)) { 1635 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n", 1636 ntohl(vxlan_hdr(skb)->vx_flags), 1637 ntohl(vxlan_hdr(skb)->vx_vni)); 1638 /* Return non vxlan pkt */ 1639 goto drop; 1640 } 1641 unparsed.vx_flags &= ~VXLAN_HF_VNI; 1642 unparsed.vx_vni &= ~VXLAN_VNI_MASK; 1643 1644 vs = rcu_dereference_sk_user_data(sk); 1645 if (!vs) 1646 goto drop; 1647 1648 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni); 1649 1650 vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni, &vninode); 1651 if (!vxlan) 1652 goto drop; 1653 1654 /* For backwards compatibility, only allow reserved fields to be 1655 * used by VXLAN extensions if explicitly requested. 1656 */ 1657 if (vs->flags & VXLAN_F_GPE) { 1658 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags)) 1659 goto drop; 1660 raw_proto = true; 1661 } 1662 1663 if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto, 1664 !net_eq(vxlan->net, dev_net(vxlan->dev)))) 1665 goto drop; 1666 1667 if (vs->flags & VXLAN_F_REMCSUM_RX) 1668 if (unlikely(!vxlan_remcsum(&unparsed, skb, vs->flags))) 1669 goto drop; 1670 1671 if (vxlan_collect_metadata(vs)) { 1672 struct metadata_dst *tun_dst; 1673 1674 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY, 1675 key32_to_tunnel_id(vni), sizeof(*md)); 1676 1677 if (!tun_dst) 1678 goto drop; 1679 1680 md = ip_tunnel_info_opts(&tun_dst->u.tun_info); 1681 1682 skb_dst_set(skb, (struct dst_entry *)tun_dst); 1683 } else { 1684 memset(md, 0, sizeof(*md)); 1685 } 1686 1687 if (vs->flags & VXLAN_F_GBP) 1688 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md); 1689 /* Note that GBP and GPE can never be active together. This is 1690 * ensured in vxlan_dev_configure. 1691 */ 1692 1693 if (unparsed.vx_flags || unparsed.vx_vni) { 1694 /* If there are any unprocessed flags remaining treat 1695 * this as a malformed packet. This behavior diverges from 1696 * VXLAN RFC (RFC7348) which stipulates that bits in reserved 1697 * in reserved fields are to be ignored. The approach here 1698 * maintains compatibility with previous stack code, and also 1699 * is more robust and provides a little more security in 1700 * adding extensions to VXLAN. 1701 */ 1702 goto drop; 1703 } 1704 1705 if (!raw_proto) { 1706 if (!vxlan_set_mac(vxlan, vs, skb, vni)) 1707 goto drop; 1708 } else { 1709 skb_reset_mac_header(skb); 1710 skb->dev = vxlan->dev; 1711 skb->pkt_type = PACKET_HOST; 1712 } 1713 1714 oiph = skb_network_header(skb); 1715 skb_reset_network_header(skb); 1716 1717 if (!vxlan_ecn_decapsulate(vs, oiph, skb)) { 1718 ++vxlan->dev->stats.rx_frame_errors; 1719 ++vxlan->dev->stats.rx_errors; 1720 vxlan_vnifilter_count(vxlan, vni, vninode, 1721 VXLAN_VNI_STATS_RX_ERRORS, 0); 1722 goto drop; 1723 } 1724 1725 rcu_read_lock(); 1726 1727 if (unlikely(!(vxlan->dev->flags & IFF_UP))) { 1728 rcu_read_unlock(); 1729 dev_core_stats_rx_dropped_inc(vxlan->dev); 1730 vxlan_vnifilter_count(vxlan, vni, vninode, 1731 VXLAN_VNI_STATS_RX_DROPS, 0); 1732 goto drop; 1733 } 1734 1735 dev_sw_netstats_rx_add(vxlan->dev, skb->len); 1736 vxlan_vnifilter_count(vxlan, vni, vninode, VXLAN_VNI_STATS_RX, skb->len); 1737 gro_cells_receive(&vxlan->gro_cells, skb); 1738 1739 rcu_read_unlock(); 1740 1741 return 0; 1742 1743 drop: 1744 /* Consume bad packet */ 1745 kfree_skb(skb); 1746 return 0; 1747 } 1748 1749 /* Callback from net/ipv{4,6}/udp.c to check that we have a VNI for errors */ 1750 static int vxlan_err_lookup(struct sock *sk, struct sk_buff *skb) 1751 { 1752 struct vxlan_dev *vxlan; 1753 struct vxlan_sock *vs; 1754 struct vxlanhdr *hdr; 1755 __be32 vni; 1756 1757 if (!pskb_may_pull(skb, skb_transport_offset(skb) + VXLAN_HLEN)) 1758 return -EINVAL; 1759 1760 hdr = vxlan_hdr(skb); 1761 1762 if (!(hdr->vx_flags & VXLAN_HF_VNI)) 1763 return -EINVAL; 1764 1765 vs = rcu_dereference_sk_user_data(sk); 1766 if (!vs) 1767 return -ENOENT; 1768 1769 vni = vxlan_vni(hdr->vx_vni); 1770 vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni, NULL); 1771 if (!vxlan) 1772 return -ENOENT; 1773 1774 return 0; 1775 } 1776 1777 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni) 1778 { 1779 struct vxlan_dev *vxlan = netdev_priv(dev); 1780 struct arphdr *parp; 1781 u8 *arpptr, *sha; 1782 __be32 sip, tip; 1783 struct neighbour *n; 1784 1785 if (dev->flags & IFF_NOARP) 1786 goto out; 1787 1788 if (!pskb_may_pull(skb, arp_hdr_len(dev))) { 1789 dev->stats.tx_dropped++; 1790 goto out; 1791 } 1792 parp = arp_hdr(skb); 1793 1794 if ((parp->ar_hrd != htons(ARPHRD_ETHER) && 1795 parp->ar_hrd != htons(ARPHRD_IEEE802)) || 1796 parp->ar_pro != htons(ETH_P_IP) || 1797 parp->ar_op != htons(ARPOP_REQUEST) || 1798 parp->ar_hln != dev->addr_len || 1799 parp->ar_pln != 4) 1800 goto out; 1801 arpptr = (u8 *)parp + sizeof(struct arphdr); 1802 sha = arpptr; 1803 arpptr += dev->addr_len; /* sha */ 1804 memcpy(&sip, arpptr, sizeof(sip)); 1805 arpptr += sizeof(sip); 1806 arpptr += dev->addr_len; /* tha */ 1807 memcpy(&tip, arpptr, sizeof(tip)); 1808 1809 if (ipv4_is_loopback(tip) || 1810 ipv4_is_multicast(tip)) 1811 goto out; 1812 1813 n = neigh_lookup(&arp_tbl, &tip, dev); 1814 1815 if (n) { 1816 struct vxlan_fdb *f; 1817 struct sk_buff *reply; 1818 1819 if (!(READ_ONCE(n->nud_state) & NUD_CONNECTED)) { 1820 neigh_release(n); 1821 goto out; 1822 } 1823 1824 f = vxlan_find_mac(vxlan, n->ha, vni); 1825 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) { 1826 /* bridge-local neighbor */ 1827 neigh_release(n); 1828 goto out; 1829 } 1830 1831 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha, 1832 n->ha, sha); 1833 1834 neigh_release(n); 1835 1836 if (reply == NULL) 1837 goto out; 1838 1839 skb_reset_mac_header(reply); 1840 __skb_pull(reply, skb_network_offset(reply)); 1841 reply->ip_summed = CHECKSUM_UNNECESSARY; 1842 reply->pkt_type = PACKET_HOST; 1843 1844 if (netif_rx(reply) == NET_RX_DROP) { 1845 dev->stats.rx_dropped++; 1846 vxlan_vnifilter_count(vxlan, vni, NULL, 1847 VXLAN_VNI_STATS_RX_DROPS, 0); 1848 } 1849 1850 } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) { 1851 union vxlan_addr ipa = { 1852 .sin.sin_addr.s_addr = tip, 1853 .sin.sin_family = AF_INET, 1854 }; 1855 1856 vxlan_ip_miss(dev, &ipa); 1857 } 1858 out: 1859 consume_skb(skb); 1860 return NETDEV_TX_OK; 1861 } 1862 1863 #if IS_ENABLED(CONFIG_IPV6) 1864 static struct sk_buff *vxlan_na_create(struct sk_buff *request, 1865 struct neighbour *n, bool isrouter) 1866 { 1867 struct net_device *dev = request->dev; 1868 struct sk_buff *reply; 1869 struct nd_msg *ns, *na; 1870 struct ipv6hdr *pip6; 1871 u8 *daddr; 1872 int na_olen = 8; /* opt hdr + ETH_ALEN for target */ 1873 int ns_olen; 1874 int i, len; 1875 1876 if (dev == NULL || !pskb_may_pull(request, request->len)) 1877 return NULL; 1878 1879 len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) + 1880 sizeof(*na) + na_olen + dev->needed_tailroom; 1881 reply = alloc_skb(len, GFP_ATOMIC); 1882 if (reply == NULL) 1883 return NULL; 1884 1885 reply->protocol = htons(ETH_P_IPV6); 1886 reply->dev = dev; 1887 skb_reserve(reply, LL_RESERVED_SPACE(request->dev)); 1888 skb_push(reply, sizeof(struct ethhdr)); 1889 skb_reset_mac_header(reply); 1890 1891 ns = (struct nd_msg *)(ipv6_hdr(request) + 1); 1892 1893 daddr = eth_hdr(request)->h_source; 1894 ns_olen = request->len - skb_network_offset(request) - 1895 sizeof(struct ipv6hdr) - sizeof(*ns); 1896 for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) { 1897 if (!ns->opt[i + 1]) { 1898 kfree_skb(reply); 1899 return NULL; 1900 } 1901 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) { 1902 daddr = ns->opt + i + sizeof(struct nd_opt_hdr); 1903 break; 1904 } 1905 } 1906 1907 /* Ethernet header */ 1908 ether_addr_copy(eth_hdr(reply)->h_dest, daddr); 1909 ether_addr_copy(eth_hdr(reply)->h_source, n->ha); 1910 eth_hdr(reply)->h_proto = htons(ETH_P_IPV6); 1911 reply->protocol = htons(ETH_P_IPV6); 1912 1913 skb_pull(reply, sizeof(struct ethhdr)); 1914 skb_reset_network_header(reply); 1915 skb_put(reply, sizeof(struct ipv6hdr)); 1916 1917 /* IPv6 header */ 1918 1919 pip6 = ipv6_hdr(reply); 1920 memset(pip6, 0, sizeof(struct ipv6hdr)); 1921 pip6->version = 6; 1922 pip6->priority = ipv6_hdr(request)->priority; 1923 pip6->nexthdr = IPPROTO_ICMPV6; 1924 pip6->hop_limit = 255; 1925 pip6->daddr = ipv6_hdr(request)->saddr; 1926 pip6->saddr = *(struct in6_addr *)n->primary_key; 1927 1928 skb_pull(reply, sizeof(struct ipv6hdr)); 1929 skb_reset_transport_header(reply); 1930 1931 /* Neighbor Advertisement */ 1932 na = skb_put_zero(reply, sizeof(*na) + na_olen); 1933 na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT; 1934 na->icmph.icmp6_router = isrouter; 1935 na->icmph.icmp6_override = 1; 1936 na->icmph.icmp6_solicited = 1; 1937 na->target = ns->target; 1938 ether_addr_copy(&na->opt[2], n->ha); 1939 na->opt[0] = ND_OPT_TARGET_LL_ADDR; 1940 na->opt[1] = na_olen >> 3; 1941 1942 na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr, 1943 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6, 1944 csum_partial(na, sizeof(*na)+na_olen, 0)); 1945 1946 pip6->payload_len = htons(sizeof(*na)+na_olen); 1947 1948 skb_push(reply, sizeof(struct ipv6hdr)); 1949 1950 reply->ip_summed = CHECKSUM_UNNECESSARY; 1951 1952 return reply; 1953 } 1954 1955 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni) 1956 { 1957 struct vxlan_dev *vxlan = netdev_priv(dev); 1958 const struct in6_addr *daddr; 1959 const struct ipv6hdr *iphdr; 1960 struct inet6_dev *in6_dev; 1961 struct neighbour *n; 1962 struct nd_msg *msg; 1963 1964 rcu_read_lock(); 1965 in6_dev = __in6_dev_get(dev); 1966 if (!in6_dev) 1967 goto out; 1968 1969 iphdr = ipv6_hdr(skb); 1970 daddr = &iphdr->daddr; 1971 msg = (struct nd_msg *)(iphdr + 1); 1972 1973 if (ipv6_addr_loopback(daddr) || 1974 ipv6_addr_is_multicast(&msg->target)) 1975 goto out; 1976 1977 n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev); 1978 1979 if (n) { 1980 struct vxlan_fdb *f; 1981 struct sk_buff *reply; 1982 1983 if (!(READ_ONCE(n->nud_state) & NUD_CONNECTED)) { 1984 neigh_release(n); 1985 goto out; 1986 } 1987 1988 f = vxlan_find_mac(vxlan, n->ha, vni); 1989 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) { 1990 /* bridge-local neighbor */ 1991 neigh_release(n); 1992 goto out; 1993 } 1994 1995 reply = vxlan_na_create(skb, n, 1996 !!(f ? f->flags & NTF_ROUTER : 0)); 1997 1998 neigh_release(n); 1999 2000 if (reply == NULL) 2001 goto out; 2002 2003 if (netif_rx(reply) == NET_RX_DROP) { 2004 dev->stats.rx_dropped++; 2005 vxlan_vnifilter_count(vxlan, vni, NULL, 2006 VXLAN_VNI_STATS_RX_DROPS, 0); 2007 } 2008 } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) { 2009 union vxlan_addr ipa = { 2010 .sin6.sin6_addr = msg->target, 2011 .sin6.sin6_family = AF_INET6, 2012 }; 2013 2014 vxlan_ip_miss(dev, &ipa); 2015 } 2016 2017 out: 2018 rcu_read_unlock(); 2019 consume_skb(skb); 2020 return NETDEV_TX_OK; 2021 } 2022 #endif 2023 2024 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb) 2025 { 2026 struct vxlan_dev *vxlan = netdev_priv(dev); 2027 struct neighbour *n; 2028 2029 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest)) 2030 return false; 2031 2032 n = NULL; 2033 switch (ntohs(eth_hdr(skb)->h_proto)) { 2034 case ETH_P_IP: 2035 { 2036 struct iphdr *pip; 2037 2038 if (!pskb_may_pull(skb, sizeof(struct iphdr))) 2039 return false; 2040 pip = ip_hdr(skb); 2041 n = neigh_lookup(&arp_tbl, &pip->daddr, dev); 2042 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) { 2043 union vxlan_addr ipa = { 2044 .sin.sin_addr.s_addr = pip->daddr, 2045 .sin.sin_family = AF_INET, 2046 }; 2047 2048 vxlan_ip_miss(dev, &ipa); 2049 return false; 2050 } 2051 2052 break; 2053 } 2054 #if IS_ENABLED(CONFIG_IPV6) 2055 case ETH_P_IPV6: 2056 { 2057 struct ipv6hdr *pip6; 2058 2059 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr))) 2060 return false; 2061 pip6 = ipv6_hdr(skb); 2062 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev); 2063 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) { 2064 union vxlan_addr ipa = { 2065 .sin6.sin6_addr = pip6->daddr, 2066 .sin6.sin6_family = AF_INET6, 2067 }; 2068 2069 vxlan_ip_miss(dev, &ipa); 2070 return false; 2071 } 2072 2073 break; 2074 } 2075 #endif 2076 default: 2077 return false; 2078 } 2079 2080 if (n) { 2081 bool diff; 2082 2083 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha); 2084 if (diff) { 2085 memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest, 2086 dev->addr_len); 2087 memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len); 2088 } 2089 neigh_release(n); 2090 return diff; 2091 } 2092 2093 return false; 2094 } 2095 2096 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, __be16 protocol) 2097 { 2098 struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh; 2099 2100 gpe->np_applied = 1; 2101 gpe->next_protocol = tun_p_from_eth_p(protocol); 2102 if (!gpe->next_protocol) 2103 return -EPFNOSUPPORT; 2104 return 0; 2105 } 2106 2107 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst, 2108 int iphdr_len, __be32 vni, 2109 struct vxlan_metadata *md, u32 vxflags, 2110 bool udp_sum) 2111 { 2112 struct vxlanhdr *vxh; 2113 int min_headroom; 2114 int err; 2115 int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL; 2116 __be16 inner_protocol = htons(ETH_P_TEB); 2117 2118 if ((vxflags & VXLAN_F_REMCSUM_TX) && 2119 skb->ip_summed == CHECKSUM_PARTIAL) { 2120 int csum_start = skb_checksum_start_offset(skb); 2121 2122 if (csum_start <= VXLAN_MAX_REMCSUM_START && 2123 !(csum_start & VXLAN_RCO_SHIFT_MASK) && 2124 (skb->csum_offset == offsetof(struct udphdr, check) || 2125 skb->csum_offset == offsetof(struct tcphdr, check))) 2126 type |= SKB_GSO_TUNNEL_REMCSUM; 2127 } 2128 2129 min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len 2130 + VXLAN_HLEN + iphdr_len; 2131 2132 /* Need space for new headers (invalidates iph ptr) */ 2133 err = skb_cow_head(skb, min_headroom); 2134 if (unlikely(err)) 2135 return err; 2136 2137 err = iptunnel_handle_offloads(skb, type); 2138 if (err) 2139 return err; 2140 2141 vxh = __skb_push(skb, sizeof(*vxh)); 2142 vxh->vx_flags = VXLAN_HF_VNI; 2143 vxh->vx_vni = vxlan_vni_field(vni); 2144 2145 if (type & SKB_GSO_TUNNEL_REMCSUM) { 2146 unsigned int start; 2147 2148 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr); 2149 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset); 2150 vxh->vx_flags |= VXLAN_HF_RCO; 2151 2152 if (!skb_is_gso(skb)) { 2153 skb->ip_summed = CHECKSUM_NONE; 2154 skb->encapsulation = 0; 2155 } 2156 } 2157 2158 if (vxflags & VXLAN_F_GBP) 2159 vxlan_build_gbp_hdr(vxh, md); 2160 if (vxflags & VXLAN_F_GPE) { 2161 err = vxlan_build_gpe_hdr(vxh, skb->protocol); 2162 if (err < 0) 2163 return err; 2164 inner_protocol = skb->protocol; 2165 } 2166 2167 skb_set_inner_protocol(skb, inner_protocol); 2168 return 0; 2169 } 2170 2171 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev, 2172 struct vxlan_sock *sock4, 2173 struct sk_buff *skb, int oif, u8 tos, 2174 __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport, 2175 __u8 flow_flags, struct dst_cache *dst_cache, 2176 const struct ip_tunnel_info *info) 2177 { 2178 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 2179 struct rtable *rt = NULL; 2180 struct flowi4 fl4; 2181 2182 if (!sock4) 2183 return ERR_PTR(-EIO); 2184 2185 if (tos && !info) 2186 use_cache = false; 2187 if (use_cache) { 2188 rt = dst_cache_get_ip4(dst_cache, saddr); 2189 if (rt) 2190 return rt; 2191 } 2192 2193 memset(&fl4, 0, sizeof(fl4)); 2194 fl4.flowi4_oif = oif; 2195 fl4.flowi4_tos = RT_TOS(tos); 2196 fl4.flowi4_mark = skb->mark; 2197 fl4.flowi4_proto = IPPROTO_UDP; 2198 fl4.daddr = daddr; 2199 fl4.saddr = *saddr; 2200 fl4.fl4_dport = dport; 2201 fl4.fl4_sport = sport; 2202 fl4.flowi4_flags = flow_flags; 2203 2204 rt = ip_route_output_key(vxlan->net, &fl4); 2205 if (!IS_ERR(rt)) { 2206 if (rt->dst.dev == dev) { 2207 netdev_dbg(dev, "circular route to %pI4\n", &daddr); 2208 ip_rt_put(rt); 2209 return ERR_PTR(-ELOOP); 2210 } 2211 2212 *saddr = fl4.saddr; 2213 if (use_cache) 2214 dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr); 2215 } else { 2216 netdev_dbg(dev, "no route to %pI4\n", &daddr); 2217 return ERR_PTR(-ENETUNREACH); 2218 } 2219 return rt; 2220 } 2221 2222 #if IS_ENABLED(CONFIG_IPV6) 2223 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan, 2224 struct net_device *dev, 2225 struct vxlan_sock *sock6, 2226 struct sk_buff *skb, int oif, u8 tos, 2227 __be32 label, 2228 const struct in6_addr *daddr, 2229 struct in6_addr *saddr, 2230 __be16 dport, __be16 sport, 2231 struct dst_cache *dst_cache, 2232 const struct ip_tunnel_info *info) 2233 { 2234 bool use_cache = ip_tunnel_dst_cache_usable(skb, info); 2235 struct dst_entry *ndst; 2236 struct flowi6 fl6; 2237 2238 if (!sock6) 2239 return ERR_PTR(-EIO); 2240 2241 if (tos && !info) 2242 use_cache = false; 2243 if (use_cache) { 2244 ndst = dst_cache_get_ip6(dst_cache, saddr); 2245 if (ndst) 2246 return ndst; 2247 } 2248 2249 memset(&fl6, 0, sizeof(fl6)); 2250 fl6.flowi6_oif = oif; 2251 fl6.daddr = *daddr; 2252 fl6.saddr = *saddr; 2253 fl6.flowlabel = ip6_make_flowinfo(tos, label); 2254 fl6.flowi6_mark = skb->mark; 2255 fl6.flowi6_proto = IPPROTO_UDP; 2256 fl6.fl6_dport = dport; 2257 fl6.fl6_sport = sport; 2258 2259 ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk, 2260 &fl6, NULL); 2261 if (IS_ERR(ndst)) { 2262 netdev_dbg(dev, "no route to %pI6\n", daddr); 2263 return ERR_PTR(-ENETUNREACH); 2264 } 2265 2266 if (unlikely(ndst->dev == dev)) { 2267 netdev_dbg(dev, "circular route to %pI6\n", daddr); 2268 dst_release(ndst); 2269 return ERR_PTR(-ELOOP); 2270 } 2271 2272 *saddr = fl6.saddr; 2273 if (use_cache) 2274 dst_cache_set_ip6(dst_cache, ndst, saddr); 2275 return ndst; 2276 } 2277 #endif 2278 2279 /* Bypass encapsulation if the destination is local */ 2280 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan, 2281 struct vxlan_dev *dst_vxlan, __be32 vni, 2282 bool snoop) 2283 { 2284 struct pcpu_sw_netstats *tx_stats, *rx_stats; 2285 union vxlan_addr loopback; 2286 union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip; 2287 struct net_device *dev; 2288 int len = skb->len; 2289 2290 tx_stats = this_cpu_ptr(src_vxlan->dev->tstats); 2291 rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats); 2292 skb->pkt_type = PACKET_HOST; 2293 skb->encapsulation = 0; 2294 skb->dev = dst_vxlan->dev; 2295 __skb_pull(skb, skb_network_offset(skb)); 2296 2297 if (remote_ip->sa.sa_family == AF_INET) { 2298 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK); 2299 loopback.sa.sa_family = AF_INET; 2300 #if IS_ENABLED(CONFIG_IPV6) 2301 } else { 2302 loopback.sin6.sin6_addr = in6addr_loopback; 2303 loopback.sa.sa_family = AF_INET6; 2304 #endif 2305 } 2306 2307 rcu_read_lock(); 2308 dev = skb->dev; 2309 if (unlikely(!(dev->flags & IFF_UP))) { 2310 kfree_skb(skb); 2311 goto drop; 2312 } 2313 2314 if ((dst_vxlan->cfg.flags & VXLAN_F_LEARN) && snoop) 2315 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni); 2316 2317 u64_stats_update_begin(&tx_stats->syncp); 2318 u64_stats_inc(&tx_stats->tx_packets); 2319 u64_stats_add(&tx_stats->tx_bytes, len); 2320 u64_stats_update_end(&tx_stats->syncp); 2321 vxlan_vnifilter_count(src_vxlan, vni, NULL, VXLAN_VNI_STATS_TX, len); 2322 2323 if (__netif_rx(skb) == NET_RX_SUCCESS) { 2324 u64_stats_update_begin(&rx_stats->syncp); 2325 u64_stats_inc(&rx_stats->rx_packets); 2326 u64_stats_add(&rx_stats->rx_bytes, len); 2327 u64_stats_update_end(&rx_stats->syncp); 2328 vxlan_vnifilter_count(dst_vxlan, vni, NULL, VXLAN_VNI_STATS_RX, 2329 len); 2330 } else { 2331 drop: 2332 dev->stats.rx_dropped++; 2333 vxlan_vnifilter_count(dst_vxlan, vni, NULL, 2334 VXLAN_VNI_STATS_RX_DROPS, 0); 2335 } 2336 rcu_read_unlock(); 2337 } 2338 2339 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev, 2340 struct vxlan_dev *vxlan, 2341 union vxlan_addr *daddr, 2342 __be16 dst_port, int dst_ifindex, __be32 vni, 2343 struct dst_entry *dst, 2344 u32 rt_flags) 2345 { 2346 #if IS_ENABLED(CONFIG_IPV6) 2347 /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of 2348 * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple 2349 * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry. 2350 */ 2351 BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL); 2352 #endif 2353 /* Bypass encapsulation if the destination is local */ 2354 if (rt_flags & RTCF_LOCAL && 2355 !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) && 2356 vxlan->cfg.flags & VXLAN_F_LOCALBYPASS) { 2357 struct vxlan_dev *dst_vxlan; 2358 2359 dst_release(dst); 2360 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni, 2361 daddr->sa.sa_family, dst_port, 2362 vxlan->cfg.flags); 2363 if (!dst_vxlan) { 2364 dev->stats.tx_errors++; 2365 vxlan_vnifilter_count(vxlan, vni, NULL, 2366 VXLAN_VNI_STATS_TX_ERRORS, 0); 2367 kfree_skb(skb); 2368 2369 return -ENOENT; 2370 } 2371 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni, true); 2372 return 1; 2373 } 2374 2375 return 0; 2376 } 2377 2378 void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev, 2379 __be32 default_vni, struct vxlan_rdst *rdst, bool did_rsc) 2380 { 2381 struct dst_cache *dst_cache; 2382 struct ip_tunnel_info *info; 2383 struct vxlan_dev *vxlan = netdev_priv(dev); 2384 const struct iphdr *old_iph = ip_hdr(skb); 2385 union vxlan_addr *dst; 2386 union vxlan_addr remote_ip, local_ip; 2387 struct vxlan_metadata _md; 2388 struct vxlan_metadata *md = &_md; 2389 unsigned int pkt_len = skb->len; 2390 __be16 src_port = 0, dst_port; 2391 struct dst_entry *ndst = NULL; 2392 __u8 tos, ttl, flow_flags = 0; 2393 int ifindex; 2394 int err; 2395 u32 flags = vxlan->cfg.flags; 2396 bool udp_sum = false; 2397 bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev)); 2398 __be32 vni = 0; 2399 #if IS_ENABLED(CONFIG_IPV6) 2400 __be32 label; 2401 #endif 2402 2403 info = skb_tunnel_info(skb); 2404 2405 if (rdst) { 2406 dst = &rdst->remote_ip; 2407 if (vxlan_addr_any(dst)) { 2408 if (did_rsc) { 2409 /* short-circuited back to local bridge */ 2410 vxlan_encap_bypass(skb, vxlan, vxlan, 2411 default_vni, true); 2412 return; 2413 } 2414 goto drop; 2415 } 2416 2417 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port; 2418 vni = (rdst->remote_vni) ? : default_vni; 2419 ifindex = rdst->remote_ifindex; 2420 local_ip = vxlan->cfg.saddr; 2421 dst_cache = &rdst->dst_cache; 2422 md->gbp = skb->mark; 2423 if (flags & VXLAN_F_TTL_INHERIT) { 2424 ttl = ip_tunnel_get_ttl(old_iph, skb); 2425 } else { 2426 ttl = vxlan->cfg.ttl; 2427 if (!ttl && vxlan_addr_multicast(dst)) 2428 ttl = 1; 2429 } 2430 2431 tos = vxlan->cfg.tos; 2432 if (tos == 1) 2433 tos = ip_tunnel_get_dsfield(old_iph, skb); 2434 2435 if (dst->sa.sa_family == AF_INET) 2436 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX); 2437 else 2438 udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX); 2439 #if IS_ENABLED(CONFIG_IPV6) 2440 label = vxlan->cfg.label; 2441 #endif 2442 } else { 2443 if (!info) { 2444 WARN_ONCE(1, "%s: Missing encapsulation instructions\n", 2445 dev->name); 2446 goto drop; 2447 } 2448 remote_ip.sa.sa_family = ip_tunnel_info_af(info); 2449 if (remote_ip.sa.sa_family == AF_INET) { 2450 remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst; 2451 local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src; 2452 } else { 2453 remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst; 2454 local_ip.sin6.sin6_addr = info->key.u.ipv6.src; 2455 } 2456 dst = &remote_ip; 2457 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port; 2458 flow_flags = info->key.flow_flags; 2459 vni = tunnel_id_to_key32(info->key.tun_id); 2460 ifindex = 0; 2461 dst_cache = &info->dst_cache; 2462 if (info->key.tun_flags & TUNNEL_VXLAN_OPT) { 2463 if (info->options_len < sizeof(*md)) 2464 goto drop; 2465 md = ip_tunnel_info_opts(info); 2466 } 2467 ttl = info->key.ttl; 2468 tos = info->key.tos; 2469 #if IS_ENABLED(CONFIG_IPV6) 2470 label = info->key.label; 2471 #endif 2472 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM); 2473 } 2474 src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min, 2475 vxlan->cfg.port_max, true); 2476 2477 rcu_read_lock(); 2478 if (dst->sa.sa_family == AF_INET) { 2479 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock); 2480 struct rtable *rt; 2481 __be16 df = 0; 2482 2483 if (!ifindex) 2484 ifindex = sock4->sock->sk->sk_bound_dev_if; 2485 2486 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos, 2487 dst->sin.sin_addr.s_addr, 2488 &local_ip.sin.sin_addr.s_addr, 2489 dst_port, src_port, flow_flags, 2490 dst_cache, info); 2491 if (IS_ERR(rt)) { 2492 err = PTR_ERR(rt); 2493 goto tx_error; 2494 } 2495 2496 if (!info) { 2497 /* Bypass encapsulation if the destination is local */ 2498 err = encap_bypass_if_local(skb, dev, vxlan, dst, 2499 dst_port, ifindex, vni, 2500 &rt->dst, rt->rt_flags); 2501 if (err) 2502 goto out_unlock; 2503 2504 if (vxlan->cfg.df == VXLAN_DF_SET) { 2505 df = htons(IP_DF); 2506 } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) { 2507 struct ethhdr *eth = eth_hdr(skb); 2508 2509 if (ntohs(eth->h_proto) == ETH_P_IPV6 || 2510 (ntohs(eth->h_proto) == ETH_P_IP && 2511 old_iph->frag_off & htons(IP_DF))) 2512 df = htons(IP_DF); 2513 } 2514 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) { 2515 df = htons(IP_DF); 2516 } 2517 2518 ndst = &rt->dst; 2519 err = skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM, 2520 netif_is_any_bridge_port(dev)); 2521 if (err < 0) { 2522 goto tx_error; 2523 } else if (err) { 2524 if (info) { 2525 struct ip_tunnel_info *unclone; 2526 struct in_addr src, dst; 2527 2528 unclone = skb_tunnel_info_unclone(skb); 2529 if (unlikely(!unclone)) 2530 goto tx_error; 2531 2532 src = remote_ip.sin.sin_addr; 2533 dst = local_ip.sin.sin_addr; 2534 unclone->key.u.ipv4.src = src.s_addr; 2535 unclone->key.u.ipv4.dst = dst.s_addr; 2536 } 2537 vxlan_encap_bypass(skb, vxlan, vxlan, vni, false); 2538 dst_release(ndst); 2539 goto out_unlock; 2540 } 2541 2542 tos = ip_tunnel_ecn_encap(tos, old_iph, skb); 2543 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst); 2544 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr), 2545 vni, md, flags, udp_sum); 2546 if (err < 0) 2547 goto tx_error; 2548 2549 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr, 2550 dst->sin.sin_addr.s_addr, tos, ttl, df, 2551 src_port, dst_port, xnet, !udp_sum); 2552 #if IS_ENABLED(CONFIG_IPV6) 2553 } else { 2554 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock); 2555 2556 if (!ifindex) 2557 ifindex = sock6->sock->sk->sk_bound_dev_if; 2558 2559 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos, 2560 label, &dst->sin6.sin6_addr, 2561 &local_ip.sin6.sin6_addr, 2562 dst_port, src_port, 2563 dst_cache, info); 2564 if (IS_ERR(ndst)) { 2565 err = PTR_ERR(ndst); 2566 ndst = NULL; 2567 goto tx_error; 2568 } 2569 2570 if (!info) { 2571 u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags; 2572 2573 err = encap_bypass_if_local(skb, dev, vxlan, dst, 2574 dst_port, ifindex, vni, 2575 ndst, rt6i_flags); 2576 if (err) 2577 goto out_unlock; 2578 } 2579 2580 err = skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM, 2581 netif_is_any_bridge_port(dev)); 2582 if (err < 0) { 2583 goto tx_error; 2584 } else if (err) { 2585 if (info) { 2586 struct ip_tunnel_info *unclone; 2587 struct in6_addr src, dst; 2588 2589 unclone = skb_tunnel_info_unclone(skb); 2590 if (unlikely(!unclone)) 2591 goto tx_error; 2592 2593 src = remote_ip.sin6.sin6_addr; 2594 dst = local_ip.sin6.sin6_addr; 2595 unclone->key.u.ipv6.src = src; 2596 unclone->key.u.ipv6.dst = dst; 2597 } 2598 2599 vxlan_encap_bypass(skb, vxlan, vxlan, vni, false); 2600 dst_release(ndst); 2601 goto out_unlock; 2602 } 2603 2604 tos = ip_tunnel_ecn_encap(tos, old_iph, skb); 2605 ttl = ttl ? : ip6_dst_hoplimit(ndst); 2606 skb_scrub_packet(skb, xnet); 2607 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr), 2608 vni, md, flags, udp_sum); 2609 if (err < 0) 2610 goto tx_error; 2611 2612 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev, 2613 &local_ip.sin6.sin6_addr, 2614 &dst->sin6.sin6_addr, tos, ttl, 2615 label, src_port, dst_port, !udp_sum); 2616 #endif 2617 } 2618 vxlan_vnifilter_count(vxlan, vni, NULL, VXLAN_VNI_STATS_TX, pkt_len); 2619 out_unlock: 2620 rcu_read_unlock(); 2621 return; 2622 2623 drop: 2624 dev->stats.tx_dropped++; 2625 vxlan_vnifilter_count(vxlan, vni, NULL, VXLAN_VNI_STATS_TX_DROPS, 0); 2626 dev_kfree_skb(skb); 2627 return; 2628 2629 tx_error: 2630 rcu_read_unlock(); 2631 if (err == -ELOOP) 2632 dev->stats.collisions++; 2633 else if (err == -ENETUNREACH) 2634 dev->stats.tx_carrier_errors++; 2635 dst_release(ndst); 2636 dev->stats.tx_errors++; 2637 vxlan_vnifilter_count(vxlan, vni, NULL, VXLAN_VNI_STATS_TX_ERRORS, 0); 2638 kfree_skb(skb); 2639 } 2640 2641 static void vxlan_xmit_nh(struct sk_buff *skb, struct net_device *dev, 2642 struct vxlan_fdb *f, __be32 vni, bool did_rsc) 2643 { 2644 struct vxlan_rdst nh_rdst; 2645 struct nexthop *nh; 2646 bool do_xmit; 2647 u32 hash; 2648 2649 memset(&nh_rdst, 0, sizeof(struct vxlan_rdst)); 2650 hash = skb_get_hash(skb); 2651 2652 rcu_read_lock(); 2653 nh = rcu_dereference(f->nh); 2654 if (!nh) { 2655 rcu_read_unlock(); 2656 goto drop; 2657 } 2658 do_xmit = vxlan_fdb_nh_path_select(nh, hash, &nh_rdst); 2659 rcu_read_unlock(); 2660 2661 if (likely(do_xmit)) 2662 vxlan_xmit_one(skb, dev, vni, &nh_rdst, did_rsc); 2663 else 2664 goto drop; 2665 2666 return; 2667 2668 drop: 2669 dev->stats.tx_dropped++; 2670 vxlan_vnifilter_count(netdev_priv(dev), vni, NULL, 2671 VXLAN_VNI_STATS_TX_DROPS, 0); 2672 dev_kfree_skb(skb); 2673 } 2674 2675 /* Transmit local packets over Vxlan 2676 * 2677 * Outer IP header inherits ECN and DF from inner header. 2678 * Outer UDP destination is the VXLAN assigned port. 2679 * source port is based on hash of flow 2680 */ 2681 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev) 2682 { 2683 struct vxlan_dev *vxlan = netdev_priv(dev); 2684 struct vxlan_rdst *rdst, *fdst = NULL; 2685 const struct ip_tunnel_info *info; 2686 bool did_rsc = false; 2687 struct vxlan_fdb *f; 2688 struct ethhdr *eth; 2689 __be32 vni = 0; 2690 2691 info = skb_tunnel_info(skb); 2692 2693 skb_reset_mac_header(skb); 2694 2695 if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) { 2696 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE && 2697 info->mode & IP_TUNNEL_INFO_TX) { 2698 vni = tunnel_id_to_key32(info->key.tun_id); 2699 } else { 2700 if (info && info->mode & IP_TUNNEL_INFO_TX) 2701 vxlan_xmit_one(skb, dev, vni, NULL, false); 2702 else 2703 kfree_skb(skb); 2704 return NETDEV_TX_OK; 2705 } 2706 } 2707 2708 if (vxlan->cfg.flags & VXLAN_F_PROXY) { 2709 eth = eth_hdr(skb); 2710 if (ntohs(eth->h_proto) == ETH_P_ARP) 2711 return arp_reduce(dev, skb, vni); 2712 #if IS_ENABLED(CONFIG_IPV6) 2713 else if (ntohs(eth->h_proto) == ETH_P_IPV6 && 2714 pskb_may_pull(skb, sizeof(struct ipv6hdr) + 2715 sizeof(struct nd_msg)) && 2716 ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) { 2717 struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1); 2718 2719 if (m->icmph.icmp6_code == 0 && 2720 m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION) 2721 return neigh_reduce(dev, skb, vni); 2722 } 2723 #endif 2724 } 2725 2726 if (vxlan->cfg.flags & VXLAN_F_MDB) { 2727 struct vxlan_mdb_entry *mdb_entry; 2728 2729 rcu_read_lock(); 2730 mdb_entry = vxlan_mdb_entry_skb_get(vxlan, skb, vni); 2731 if (mdb_entry) { 2732 netdev_tx_t ret; 2733 2734 ret = vxlan_mdb_xmit(vxlan, mdb_entry, skb); 2735 rcu_read_unlock(); 2736 return ret; 2737 } 2738 rcu_read_unlock(); 2739 } 2740 2741 eth = eth_hdr(skb); 2742 f = vxlan_find_mac(vxlan, eth->h_dest, vni); 2743 did_rsc = false; 2744 2745 if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) && 2746 (ntohs(eth->h_proto) == ETH_P_IP || 2747 ntohs(eth->h_proto) == ETH_P_IPV6)) { 2748 did_rsc = route_shortcircuit(dev, skb); 2749 if (did_rsc) 2750 f = vxlan_find_mac(vxlan, eth->h_dest, vni); 2751 } 2752 2753 if (f == NULL) { 2754 f = vxlan_find_mac(vxlan, all_zeros_mac, vni); 2755 if (f == NULL) { 2756 if ((vxlan->cfg.flags & VXLAN_F_L2MISS) && 2757 !is_multicast_ether_addr(eth->h_dest)) 2758 vxlan_fdb_miss(vxlan, eth->h_dest); 2759 2760 dev->stats.tx_dropped++; 2761 vxlan_vnifilter_count(vxlan, vni, NULL, 2762 VXLAN_VNI_STATS_TX_DROPS, 0); 2763 kfree_skb(skb); 2764 return NETDEV_TX_OK; 2765 } 2766 } 2767 2768 if (rcu_access_pointer(f->nh)) { 2769 vxlan_xmit_nh(skb, dev, f, 2770 (vni ? : vxlan->default_dst.remote_vni), did_rsc); 2771 } else { 2772 list_for_each_entry_rcu(rdst, &f->remotes, list) { 2773 struct sk_buff *skb1; 2774 2775 if (!fdst) { 2776 fdst = rdst; 2777 continue; 2778 } 2779 skb1 = skb_clone(skb, GFP_ATOMIC); 2780 if (skb1) 2781 vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc); 2782 } 2783 if (fdst) 2784 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc); 2785 else 2786 kfree_skb(skb); 2787 } 2788 2789 return NETDEV_TX_OK; 2790 } 2791 2792 /* Walk the forwarding table and purge stale entries */ 2793 static void vxlan_cleanup(struct timer_list *t) 2794 { 2795 struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer); 2796 unsigned long next_timer = jiffies + FDB_AGE_INTERVAL; 2797 unsigned int h; 2798 2799 if (!netif_running(vxlan->dev)) 2800 return; 2801 2802 for (h = 0; h < FDB_HASH_SIZE; ++h) { 2803 struct hlist_node *p, *n; 2804 2805 spin_lock(&vxlan->hash_lock[h]); 2806 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) { 2807 struct vxlan_fdb *f 2808 = container_of(p, struct vxlan_fdb, hlist); 2809 unsigned long timeout; 2810 2811 if (f->state & (NUD_PERMANENT | NUD_NOARP)) 2812 continue; 2813 2814 if (f->flags & NTF_EXT_LEARNED) 2815 continue; 2816 2817 timeout = f->used + vxlan->cfg.age_interval * HZ; 2818 if (time_before_eq(timeout, jiffies)) { 2819 netdev_dbg(vxlan->dev, 2820 "garbage collect %pM\n", 2821 f->eth_addr); 2822 f->state = NUD_STALE; 2823 vxlan_fdb_destroy(vxlan, f, true, true); 2824 } else if (time_before(timeout, next_timer)) 2825 next_timer = timeout; 2826 } 2827 spin_unlock(&vxlan->hash_lock[h]); 2828 } 2829 2830 mod_timer(&vxlan->age_timer, next_timer); 2831 } 2832 2833 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan) 2834 { 2835 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id); 2836 2837 spin_lock(&vn->sock_lock); 2838 hlist_del_init_rcu(&vxlan->hlist4.hlist); 2839 #if IS_ENABLED(CONFIG_IPV6) 2840 hlist_del_init_rcu(&vxlan->hlist6.hlist); 2841 #endif 2842 spin_unlock(&vn->sock_lock); 2843 } 2844 2845 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan, 2846 struct vxlan_dev_node *node) 2847 { 2848 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id); 2849 __be32 vni = vxlan->default_dst.remote_vni; 2850 2851 node->vxlan = vxlan; 2852 spin_lock(&vn->sock_lock); 2853 hlist_add_head_rcu(&node->hlist, vni_head(vs, vni)); 2854 spin_unlock(&vn->sock_lock); 2855 } 2856 2857 /* Setup stats when device is created */ 2858 static int vxlan_init(struct net_device *dev) 2859 { 2860 struct vxlan_dev *vxlan = netdev_priv(dev); 2861 int err; 2862 2863 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER) 2864 vxlan_vnigroup_init(vxlan); 2865 2866 dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats); 2867 if (!dev->tstats) { 2868 err = -ENOMEM; 2869 goto err_vnigroup_uninit; 2870 } 2871 2872 err = gro_cells_init(&vxlan->gro_cells, dev); 2873 if (err) 2874 goto err_free_percpu; 2875 2876 err = vxlan_mdb_init(vxlan); 2877 if (err) 2878 goto err_gro_cells_destroy; 2879 2880 return 0; 2881 2882 err_gro_cells_destroy: 2883 gro_cells_destroy(&vxlan->gro_cells); 2884 err_free_percpu: 2885 free_percpu(dev->tstats); 2886 err_vnigroup_uninit: 2887 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER) 2888 vxlan_vnigroup_uninit(vxlan); 2889 return err; 2890 } 2891 2892 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni) 2893 { 2894 struct vxlan_fdb *f; 2895 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, vni); 2896 2897 spin_lock_bh(&vxlan->hash_lock[hash_index]); 2898 f = __vxlan_find_mac(vxlan, all_zeros_mac, vni); 2899 if (f) 2900 vxlan_fdb_destroy(vxlan, f, true, true); 2901 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 2902 } 2903 2904 static void vxlan_uninit(struct net_device *dev) 2905 { 2906 struct vxlan_dev *vxlan = netdev_priv(dev); 2907 2908 vxlan_mdb_fini(vxlan); 2909 2910 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER) 2911 vxlan_vnigroup_uninit(vxlan); 2912 2913 gro_cells_destroy(&vxlan->gro_cells); 2914 2915 vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni); 2916 2917 free_percpu(dev->tstats); 2918 } 2919 2920 /* Start ageing timer and join group when device is brought up */ 2921 static int vxlan_open(struct net_device *dev) 2922 { 2923 struct vxlan_dev *vxlan = netdev_priv(dev); 2924 int ret; 2925 2926 ret = vxlan_sock_add(vxlan); 2927 if (ret < 0) 2928 return ret; 2929 2930 ret = vxlan_multicast_join(vxlan); 2931 if (ret) { 2932 vxlan_sock_release(vxlan); 2933 return ret; 2934 } 2935 2936 if (vxlan->cfg.age_interval) 2937 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL); 2938 2939 return ret; 2940 } 2941 2942 /* Purge the forwarding table */ 2943 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all) 2944 { 2945 unsigned int h; 2946 2947 for (h = 0; h < FDB_HASH_SIZE; ++h) { 2948 struct hlist_node *p, *n; 2949 2950 spin_lock_bh(&vxlan->hash_lock[h]); 2951 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) { 2952 struct vxlan_fdb *f 2953 = container_of(p, struct vxlan_fdb, hlist); 2954 if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP))) 2955 continue; 2956 /* the all_zeros_mac entry is deleted at vxlan_uninit */ 2957 if (is_zero_ether_addr(f->eth_addr) && 2958 f->vni == vxlan->cfg.vni) 2959 continue; 2960 vxlan_fdb_destroy(vxlan, f, true, true); 2961 } 2962 spin_unlock_bh(&vxlan->hash_lock[h]); 2963 } 2964 } 2965 2966 /* Cleanup timer and forwarding table on shutdown */ 2967 static int vxlan_stop(struct net_device *dev) 2968 { 2969 struct vxlan_dev *vxlan = netdev_priv(dev); 2970 2971 vxlan_multicast_leave(vxlan); 2972 2973 del_timer_sync(&vxlan->age_timer); 2974 2975 vxlan_flush(vxlan, false); 2976 vxlan_sock_release(vxlan); 2977 2978 return 0; 2979 } 2980 2981 /* Stub, nothing needs to be done. */ 2982 static void vxlan_set_multicast_list(struct net_device *dev) 2983 { 2984 } 2985 2986 static int vxlan_change_mtu(struct net_device *dev, int new_mtu) 2987 { 2988 struct vxlan_dev *vxlan = netdev_priv(dev); 2989 struct vxlan_rdst *dst = &vxlan->default_dst; 2990 struct net_device *lowerdev = __dev_get_by_index(vxlan->net, 2991 dst->remote_ifindex); 2992 bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6); 2993 2994 /* This check is different than dev->max_mtu, because it looks at 2995 * the lowerdev->mtu, rather than the static dev->max_mtu 2996 */ 2997 if (lowerdev) { 2998 int max_mtu = lowerdev->mtu - 2999 (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM); 3000 if (new_mtu > max_mtu) 3001 return -EINVAL; 3002 } 3003 3004 dev->mtu = new_mtu; 3005 return 0; 3006 } 3007 3008 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) 3009 { 3010 struct vxlan_dev *vxlan = netdev_priv(dev); 3011 struct ip_tunnel_info *info = skb_tunnel_info(skb); 3012 __be16 sport, dport; 3013 3014 sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min, 3015 vxlan->cfg.port_max, true); 3016 dport = info->key.tp_dst ? : vxlan->cfg.dst_port; 3017 3018 if (ip_tunnel_info_af(info) == AF_INET) { 3019 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock); 3020 struct rtable *rt; 3021 3022 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos, 3023 info->key.u.ipv4.dst, 3024 &info->key.u.ipv4.src, dport, sport, 3025 info->key.flow_flags, &info->dst_cache, 3026 info); 3027 if (IS_ERR(rt)) 3028 return PTR_ERR(rt); 3029 ip_rt_put(rt); 3030 } else { 3031 #if IS_ENABLED(CONFIG_IPV6) 3032 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock); 3033 struct dst_entry *ndst; 3034 3035 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos, 3036 info->key.label, &info->key.u.ipv6.dst, 3037 &info->key.u.ipv6.src, dport, sport, 3038 &info->dst_cache, info); 3039 if (IS_ERR(ndst)) 3040 return PTR_ERR(ndst); 3041 dst_release(ndst); 3042 #else /* !CONFIG_IPV6 */ 3043 return -EPFNOSUPPORT; 3044 #endif 3045 } 3046 info->key.tp_src = sport; 3047 info->key.tp_dst = dport; 3048 return 0; 3049 } 3050 3051 static const struct net_device_ops vxlan_netdev_ether_ops = { 3052 .ndo_init = vxlan_init, 3053 .ndo_uninit = vxlan_uninit, 3054 .ndo_open = vxlan_open, 3055 .ndo_stop = vxlan_stop, 3056 .ndo_start_xmit = vxlan_xmit, 3057 .ndo_get_stats64 = dev_get_tstats64, 3058 .ndo_set_rx_mode = vxlan_set_multicast_list, 3059 .ndo_change_mtu = vxlan_change_mtu, 3060 .ndo_validate_addr = eth_validate_addr, 3061 .ndo_set_mac_address = eth_mac_addr, 3062 .ndo_fdb_add = vxlan_fdb_add, 3063 .ndo_fdb_del = vxlan_fdb_delete, 3064 .ndo_fdb_dump = vxlan_fdb_dump, 3065 .ndo_fdb_get = vxlan_fdb_get, 3066 .ndo_mdb_add = vxlan_mdb_add, 3067 .ndo_mdb_del = vxlan_mdb_del, 3068 .ndo_mdb_dump = vxlan_mdb_dump, 3069 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst, 3070 }; 3071 3072 static const struct net_device_ops vxlan_netdev_raw_ops = { 3073 .ndo_init = vxlan_init, 3074 .ndo_uninit = vxlan_uninit, 3075 .ndo_open = vxlan_open, 3076 .ndo_stop = vxlan_stop, 3077 .ndo_start_xmit = vxlan_xmit, 3078 .ndo_get_stats64 = dev_get_tstats64, 3079 .ndo_change_mtu = vxlan_change_mtu, 3080 .ndo_fill_metadata_dst = vxlan_fill_metadata_dst, 3081 }; 3082 3083 /* Info for udev, that this is a virtual tunnel endpoint */ 3084 static struct device_type vxlan_type = { 3085 .name = "vxlan", 3086 }; 3087 3088 /* Calls the ndo_udp_tunnel_add of the caller in order to 3089 * supply the listening VXLAN udp ports. Callers are expected 3090 * to implement the ndo_udp_tunnel_add. 3091 */ 3092 static void vxlan_offload_rx_ports(struct net_device *dev, bool push) 3093 { 3094 struct vxlan_sock *vs; 3095 struct net *net = dev_net(dev); 3096 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 3097 unsigned int i; 3098 3099 spin_lock(&vn->sock_lock); 3100 for (i = 0; i < PORT_HASH_SIZE; ++i) { 3101 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) { 3102 unsigned short type; 3103 3104 if (vs->flags & VXLAN_F_GPE) 3105 type = UDP_TUNNEL_TYPE_VXLAN_GPE; 3106 else 3107 type = UDP_TUNNEL_TYPE_VXLAN; 3108 3109 if (push) 3110 udp_tunnel_push_rx_port(dev, vs->sock, type); 3111 else 3112 udp_tunnel_drop_rx_port(dev, vs->sock, type); 3113 } 3114 } 3115 spin_unlock(&vn->sock_lock); 3116 } 3117 3118 /* Initialize the device structure. */ 3119 static void vxlan_setup(struct net_device *dev) 3120 { 3121 struct vxlan_dev *vxlan = netdev_priv(dev); 3122 unsigned int h; 3123 3124 eth_hw_addr_random(dev); 3125 ether_setup(dev); 3126 3127 dev->needs_free_netdev = true; 3128 SET_NETDEV_DEVTYPE(dev, &vxlan_type); 3129 3130 dev->features |= NETIF_F_LLTX; 3131 dev->features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST; 3132 dev->features |= NETIF_F_RXCSUM; 3133 dev->features |= NETIF_F_GSO_SOFTWARE; 3134 3135 dev->vlan_features = dev->features; 3136 dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST; 3137 dev->hw_features |= NETIF_F_RXCSUM; 3138 dev->hw_features |= NETIF_F_GSO_SOFTWARE; 3139 netif_keep_dst(dev); 3140 dev->priv_flags |= IFF_NO_QUEUE | IFF_CHANGE_PROTO_DOWN; 3141 3142 /* MTU range: 68 - 65535 */ 3143 dev->min_mtu = ETH_MIN_MTU; 3144 dev->max_mtu = ETH_MAX_MTU; 3145 3146 INIT_LIST_HEAD(&vxlan->next); 3147 3148 timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE); 3149 3150 vxlan->dev = dev; 3151 3152 for (h = 0; h < FDB_HASH_SIZE; ++h) { 3153 spin_lock_init(&vxlan->hash_lock[h]); 3154 INIT_HLIST_HEAD(&vxlan->fdb_head[h]); 3155 } 3156 } 3157 3158 static void vxlan_ether_setup(struct net_device *dev) 3159 { 3160 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 3161 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 3162 dev->netdev_ops = &vxlan_netdev_ether_ops; 3163 } 3164 3165 static void vxlan_raw_setup(struct net_device *dev) 3166 { 3167 dev->header_ops = NULL; 3168 dev->type = ARPHRD_NONE; 3169 dev->hard_header_len = 0; 3170 dev->addr_len = 0; 3171 dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST; 3172 dev->netdev_ops = &vxlan_netdev_raw_ops; 3173 } 3174 3175 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = { 3176 [IFLA_VXLAN_UNSPEC] = { .strict_start_type = IFLA_VXLAN_LOCALBYPASS }, 3177 [IFLA_VXLAN_ID] = { .type = NLA_U32 }, 3178 [IFLA_VXLAN_GROUP] = { .len = sizeof_field(struct iphdr, daddr) }, 3179 [IFLA_VXLAN_GROUP6] = { .len = sizeof(struct in6_addr) }, 3180 [IFLA_VXLAN_LINK] = { .type = NLA_U32 }, 3181 [IFLA_VXLAN_LOCAL] = { .len = sizeof_field(struct iphdr, saddr) }, 3182 [IFLA_VXLAN_LOCAL6] = { .len = sizeof(struct in6_addr) }, 3183 [IFLA_VXLAN_TOS] = { .type = NLA_U8 }, 3184 [IFLA_VXLAN_TTL] = { .type = NLA_U8 }, 3185 [IFLA_VXLAN_LABEL] = { .type = NLA_U32 }, 3186 [IFLA_VXLAN_LEARNING] = { .type = NLA_U8 }, 3187 [IFLA_VXLAN_AGEING] = { .type = NLA_U32 }, 3188 [IFLA_VXLAN_LIMIT] = { .type = NLA_U32 }, 3189 [IFLA_VXLAN_PORT_RANGE] = { .len = sizeof(struct ifla_vxlan_port_range) }, 3190 [IFLA_VXLAN_PROXY] = { .type = NLA_U8 }, 3191 [IFLA_VXLAN_RSC] = { .type = NLA_U8 }, 3192 [IFLA_VXLAN_L2MISS] = { .type = NLA_U8 }, 3193 [IFLA_VXLAN_L3MISS] = { .type = NLA_U8 }, 3194 [IFLA_VXLAN_COLLECT_METADATA] = { .type = NLA_U8 }, 3195 [IFLA_VXLAN_PORT] = { .type = NLA_U16 }, 3196 [IFLA_VXLAN_UDP_CSUM] = { .type = NLA_U8 }, 3197 [IFLA_VXLAN_UDP_ZERO_CSUM6_TX] = { .type = NLA_U8 }, 3198 [IFLA_VXLAN_UDP_ZERO_CSUM6_RX] = { .type = NLA_U8 }, 3199 [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 }, 3200 [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 }, 3201 [IFLA_VXLAN_GBP] = { .type = NLA_FLAG, }, 3202 [IFLA_VXLAN_GPE] = { .type = NLA_FLAG, }, 3203 [IFLA_VXLAN_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG }, 3204 [IFLA_VXLAN_TTL_INHERIT] = { .type = NLA_FLAG }, 3205 [IFLA_VXLAN_DF] = { .type = NLA_U8 }, 3206 [IFLA_VXLAN_VNIFILTER] = { .type = NLA_U8 }, 3207 [IFLA_VXLAN_LOCALBYPASS] = NLA_POLICY_MAX(NLA_U8, 1), 3208 }; 3209 3210 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[], 3211 struct netlink_ext_ack *extack) 3212 { 3213 if (tb[IFLA_ADDRESS]) { 3214 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) { 3215 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 3216 "Provided link layer address is not Ethernet"); 3217 return -EINVAL; 3218 } 3219 3220 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) { 3221 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS], 3222 "Provided Ethernet address is not unicast"); 3223 return -EADDRNOTAVAIL; 3224 } 3225 } 3226 3227 if (tb[IFLA_MTU]) { 3228 u32 mtu = nla_get_u32(tb[IFLA_MTU]); 3229 3230 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) { 3231 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU], 3232 "MTU must be between 68 and 65535"); 3233 return -EINVAL; 3234 } 3235 } 3236 3237 if (!data) { 3238 NL_SET_ERR_MSG(extack, 3239 "Required attributes not provided to perform the operation"); 3240 return -EINVAL; 3241 } 3242 3243 if (data[IFLA_VXLAN_ID]) { 3244 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]); 3245 3246 if (id >= VXLAN_N_VID) { 3247 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_ID], 3248 "VXLAN ID must be lower than 16777216"); 3249 return -ERANGE; 3250 } 3251 } 3252 3253 if (data[IFLA_VXLAN_PORT_RANGE]) { 3254 const struct ifla_vxlan_port_range *p 3255 = nla_data(data[IFLA_VXLAN_PORT_RANGE]); 3256 3257 if (ntohs(p->high) < ntohs(p->low)) { 3258 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_PORT_RANGE], 3259 "Invalid source port range"); 3260 return -EINVAL; 3261 } 3262 } 3263 3264 if (data[IFLA_VXLAN_DF]) { 3265 enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]); 3266 3267 if (df < 0 || df > VXLAN_DF_MAX) { 3268 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_DF], 3269 "Invalid DF attribute"); 3270 return -EINVAL; 3271 } 3272 } 3273 3274 return 0; 3275 } 3276 3277 static void vxlan_get_drvinfo(struct net_device *netdev, 3278 struct ethtool_drvinfo *drvinfo) 3279 { 3280 strscpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version)); 3281 strscpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver)); 3282 } 3283 3284 static int vxlan_get_link_ksettings(struct net_device *dev, 3285 struct ethtool_link_ksettings *cmd) 3286 { 3287 struct vxlan_dev *vxlan = netdev_priv(dev); 3288 struct vxlan_rdst *dst = &vxlan->default_dst; 3289 struct net_device *lowerdev = __dev_get_by_index(vxlan->net, 3290 dst->remote_ifindex); 3291 3292 if (!lowerdev) { 3293 cmd->base.duplex = DUPLEX_UNKNOWN; 3294 cmd->base.port = PORT_OTHER; 3295 cmd->base.speed = SPEED_UNKNOWN; 3296 3297 return 0; 3298 } 3299 3300 return __ethtool_get_link_ksettings(lowerdev, cmd); 3301 } 3302 3303 static const struct ethtool_ops vxlan_ethtool_ops = { 3304 .get_drvinfo = vxlan_get_drvinfo, 3305 .get_link = ethtool_op_get_link, 3306 .get_link_ksettings = vxlan_get_link_ksettings, 3307 }; 3308 3309 static struct socket *vxlan_create_sock(struct net *net, bool ipv6, 3310 __be16 port, u32 flags, int ifindex) 3311 { 3312 struct socket *sock; 3313 struct udp_port_cfg udp_conf; 3314 int err; 3315 3316 memset(&udp_conf, 0, sizeof(udp_conf)); 3317 3318 if (ipv6) { 3319 udp_conf.family = AF_INET6; 3320 udp_conf.use_udp6_rx_checksums = 3321 !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX); 3322 udp_conf.ipv6_v6only = 1; 3323 } else { 3324 udp_conf.family = AF_INET; 3325 } 3326 3327 udp_conf.local_udp_port = port; 3328 udp_conf.bind_ifindex = ifindex; 3329 3330 /* Open UDP socket */ 3331 err = udp_sock_create(net, &udp_conf, &sock); 3332 if (err < 0) 3333 return ERR_PTR(err); 3334 3335 udp_allow_gso(sock->sk); 3336 return sock; 3337 } 3338 3339 /* Create new listen socket if needed */ 3340 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6, 3341 __be16 port, u32 flags, 3342 int ifindex) 3343 { 3344 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 3345 struct vxlan_sock *vs; 3346 struct socket *sock; 3347 unsigned int h; 3348 struct udp_tunnel_sock_cfg tunnel_cfg; 3349 3350 vs = kzalloc(sizeof(*vs), GFP_KERNEL); 3351 if (!vs) 3352 return ERR_PTR(-ENOMEM); 3353 3354 for (h = 0; h < VNI_HASH_SIZE; ++h) 3355 INIT_HLIST_HEAD(&vs->vni_list[h]); 3356 3357 sock = vxlan_create_sock(net, ipv6, port, flags, ifindex); 3358 if (IS_ERR(sock)) { 3359 kfree(vs); 3360 return ERR_CAST(sock); 3361 } 3362 3363 vs->sock = sock; 3364 refcount_set(&vs->refcnt, 1); 3365 vs->flags = (flags & VXLAN_F_RCV_FLAGS); 3366 3367 spin_lock(&vn->sock_lock); 3368 hlist_add_head_rcu(&vs->hlist, vs_head(net, port)); 3369 udp_tunnel_notify_add_rx_port(sock, 3370 (vs->flags & VXLAN_F_GPE) ? 3371 UDP_TUNNEL_TYPE_VXLAN_GPE : 3372 UDP_TUNNEL_TYPE_VXLAN); 3373 spin_unlock(&vn->sock_lock); 3374 3375 /* Mark socket as an encapsulation socket. */ 3376 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg)); 3377 tunnel_cfg.sk_user_data = vs; 3378 tunnel_cfg.encap_type = 1; 3379 tunnel_cfg.encap_rcv = vxlan_rcv; 3380 tunnel_cfg.encap_err_lookup = vxlan_err_lookup; 3381 tunnel_cfg.encap_destroy = NULL; 3382 tunnel_cfg.gro_receive = vxlan_gro_receive; 3383 tunnel_cfg.gro_complete = vxlan_gro_complete; 3384 3385 setup_udp_tunnel_sock(net, sock, &tunnel_cfg); 3386 3387 return vs; 3388 } 3389 3390 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6) 3391 { 3392 struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id); 3393 bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA; 3394 struct vxlan_sock *vs = NULL; 3395 struct vxlan_dev_node *node; 3396 int l3mdev_index = 0; 3397 3398 if (vxlan->cfg.remote_ifindex) 3399 l3mdev_index = l3mdev_master_upper_ifindex_by_index( 3400 vxlan->net, vxlan->cfg.remote_ifindex); 3401 3402 if (!vxlan->cfg.no_share) { 3403 spin_lock(&vn->sock_lock); 3404 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET, 3405 vxlan->cfg.dst_port, vxlan->cfg.flags, 3406 l3mdev_index); 3407 if (vs && !refcount_inc_not_zero(&vs->refcnt)) { 3408 spin_unlock(&vn->sock_lock); 3409 return -EBUSY; 3410 } 3411 spin_unlock(&vn->sock_lock); 3412 } 3413 if (!vs) 3414 vs = vxlan_socket_create(vxlan->net, ipv6, 3415 vxlan->cfg.dst_port, vxlan->cfg.flags, 3416 l3mdev_index); 3417 if (IS_ERR(vs)) 3418 return PTR_ERR(vs); 3419 #if IS_ENABLED(CONFIG_IPV6) 3420 if (ipv6) { 3421 rcu_assign_pointer(vxlan->vn6_sock, vs); 3422 node = &vxlan->hlist6; 3423 } else 3424 #endif 3425 { 3426 rcu_assign_pointer(vxlan->vn4_sock, vs); 3427 node = &vxlan->hlist4; 3428 } 3429 3430 if (metadata && (vxlan->cfg.flags & VXLAN_F_VNIFILTER)) 3431 vxlan_vs_add_vnigrp(vxlan, vs, ipv6); 3432 else 3433 vxlan_vs_add_dev(vs, vxlan, node); 3434 3435 return 0; 3436 } 3437 3438 static int vxlan_sock_add(struct vxlan_dev *vxlan) 3439 { 3440 bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA; 3441 bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata; 3442 bool ipv4 = !ipv6 || metadata; 3443 int ret = 0; 3444 3445 RCU_INIT_POINTER(vxlan->vn4_sock, NULL); 3446 #if IS_ENABLED(CONFIG_IPV6) 3447 RCU_INIT_POINTER(vxlan->vn6_sock, NULL); 3448 if (ipv6) { 3449 ret = __vxlan_sock_add(vxlan, true); 3450 if (ret < 0 && ret != -EAFNOSUPPORT) 3451 ipv4 = false; 3452 } 3453 #endif 3454 if (ipv4) 3455 ret = __vxlan_sock_add(vxlan, false); 3456 if (ret < 0) 3457 vxlan_sock_release(vxlan); 3458 return ret; 3459 } 3460 3461 int vxlan_vni_in_use(struct net *src_net, struct vxlan_dev *vxlan, 3462 struct vxlan_config *conf, __be32 vni) 3463 { 3464 struct vxlan_net *vn = net_generic(src_net, vxlan_net_id); 3465 struct vxlan_dev *tmp; 3466 3467 list_for_each_entry(tmp, &vn->vxlan_list, next) { 3468 if (tmp == vxlan) 3469 continue; 3470 if (tmp->cfg.flags & VXLAN_F_VNIFILTER) { 3471 if (!vxlan_vnifilter_lookup(tmp, vni)) 3472 continue; 3473 } else if (tmp->cfg.vni != vni) { 3474 continue; 3475 } 3476 if (tmp->cfg.dst_port != conf->dst_port) 3477 continue; 3478 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) != 3479 (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6))) 3480 continue; 3481 3482 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) && 3483 tmp->cfg.remote_ifindex != conf->remote_ifindex) 3484 continue; 3485 3486 return -EEXIST; 3487 } 3488 3489 return 0; 3490 } 3491 3492 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf, 3493 struct net_device **lower, 3494 struct vxlan_dev *old, 3495 struct netlink_ext_ack *extack) 3496 { 3497 bool use_ipv6 = false; 3498 3499 if (conf->flags & VXLAN_F_GPE) { 3500 /* For now, allow GPE only together with 3501 * COLLECT_METADATA. This can be relaxed later; in such 3502 * case, the other side of the PtP link will have to be 3503 * provided. 3504 */ 3505 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) || 3506 !(conf->flags & VXLAN_F_COLLECT_METADATA)) { 3507 NL_SET_ERR_MSG(extack, 3508 "VXLAN GPE does not support this combination of attributes"); 3509 return -EINVAL; 3510 } 3511 } 3512 3513 if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) { 3514 /* Unless IPv6 is explicitly requested, assume IPv4 */ 3515 conf->remote_ip.sa.sa_family = AF_INET; 3516 conf->saddr.sa.sa_family = AF_INET; 3517 } else if (!conf->remote_ip.sa.sa_family) { 3518 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family; 3519 } else if (!conf->saddr.sa.sa_family) { 3520 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family; 3521 } 3522 3523 if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) { 3524 NL_SET_ERR_MSG(extack, 3525 "Local and remote address must be from the same family"); 3526 return -EINVAL; 3527 } 3528 3529 if (vxlan_addr_multicast(&conf->saddr)) { 3530 NL_SET_ERR_MSG(extack, "Local address cannot be multicast"); 3531 return -EINVAL; 3532 } 3533 3534 if (conf->saddr.sa.sa_family == AF_INET6) { 3535 if (!IS_ENABLED(CONFIG_IPV6)) { 3536 NL_SET_ERR_MSG(extack, 3537 "IPv6 support not enabled in the kernel"); 3538 return -EPFNOSUPPORT; 3539 } 3540 use_ipv6 = true; 3541 conf->flags |= VXLAN_F_IPV6; 3542 3543 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) { 3544 int local_type = 3545 ipv6_addr_type(&conf->saddr.sin6.sin6_addr); 3546 int remote_type = 3547 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr); 3548 3549 if (local_type & IPV6_ADDR_LINKLOCAL) { 3550 if (!(remote_type & IPV6_ADDR_LINKLOCAL) && 3551 (remote_type != IPV6_ADDR_ANY)) { 3552 NL_SET_ERR_MSG(extack, 3553 "Invalid combination of local and remote address scopes"); 3554 return -EINVAL; 3555 } 3556 3557 conf->flags |= VXLAN_F_IPV6_LINKLOCAL; 3558 } else { 3559 if (remote_type == 3560 (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) { 3561 NL_SET_ERR_MSG(extack, 3562 "Invalid combination of local and remote address scopes"); 3563 return -EINVAL; 3564 } 3565 3566 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL; 3567 } 3568 } 3569 } 3570 3571 if (conf->label && !use_ipv6) { 3572 NL_SET_ERR_MSG(extack, 3573 "Label attribute only applies to IPv6 VXLAN devices"); 3574 return -EINVAL; 3575 } 3576 3577 if (conf->remote_ifindex) { 3578 struct net_device *lowerdev; 3579 3580 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex); 3581 if (!lowerdev) { 3582 NL_SET_ERR_MSG(extack, 3583 "Invalid local interface, device not found"); 3584 return -ENODEV; 3585 } 3586 3587 #if IS_ENABLED(CONFIG_IPV6) 3588 if (use_ipv6) { 3589 struct inet6_dev *idev = __in6_dev_get(lowerdev); 3590 3591 if (idev && idev->cnf.disable_ipv6) { 3592 NL_SET_ERR_MSG(extack, 3593 "IPv6 support disabled by administrator"); 3594 return -EPERM; 3595 } 3596 } 3597 #endif 3598 3599 *lower = lowerdev; 3600 } else { 3601 if (vxlan_addr_multicast(&conf->remote_ip)) { 3602 NL_SET_ERR_MSG(extack, 3603 "Local interface required for multicast remote destination"); 3604 3605 return -EINVAL; 3606 } 3607 3608 #if IS_ENABLED(CONFIG_IPV6) 3609 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) { 3610 NL_SET_ERR_MSG(extack, 3611 "Local interface required for link-local local/remote addresses"); 3612 return -EINVAL; 3613 } 3614 #endif 3615 3616 *lower = NULL; 3617 } 3618 3619 if (!conf->dst_port) { 3620 if (conf->flags & VXLAN_F_GPE) 3621 conf->dst_port = htons(IANA_VXLAN_GPE_UDP_PORT); 3622 else 3623 conf->dst_port = htons(vxlan_port); 3624 } 3625 3626 if (!conf->age_interval) 3627 conf->age_interval = FDB_AGE_DEFAULT; 3628 3629 if (vxlan_vni_in_use(src_net, old, conf, conf->vni)) { 3630 NL_SET_ERR_MSG(extack, 3631 "A VXLAN device with the specified VNI already exists"); 3632 return -EEXIST; 3633 } 3634 3635 return 0; 3636 } 3637 3638 static void vxlan_config_apply(struct net_device *dev, 3639 struct vxlan_config *conf, 3640 struct net_device *lowerdev, 3641 struct net *src_net, 3642 bool changelink) 3643 { 3644 struct vxlan_dev *vxlan = netdev_priv(dev); 3645 struct vxlan_rdst *dst = &vxlan->default_dst; 3646 unsigned short needed_headroom = ETH_HLEN; 3647 bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6); 3648 int max_mtu = ETH_MAX_MTU; 3649 3650 if (!changelink) { 3651 if (conf->flags & VXLAN_F_GPE) 3652 vxlan_raw_setup(dev); 3653 else 3654 vxlan_ether_setup(dev); 3655 3656 if (conf->mtu) 3657 dev->mtu = conf->mtu; 3658 3659 vxlan->net = src_net; 3660 } 3661 3662 dst->remote_vni = conf->vni; 3663 3664 memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip)); 3665 3666 if (lowerdev) { 3667 dst->remote_ifindex = conf->remote_ifindex; 3668 3669 netif_inherit_tso_max(dev, lowerdev); 3670 3671 needed_headroom = lowerdev->hard_header_len; 3672 needed_headroom += lowerdev->needed_headroom; 3673 3674 dev->needed_tailroom = lowerdev->needed_tailroom; 3675 3676 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : 3677 VXLAN_HEADROOM); 3678 if (max_mtu < ETH_MIN_MTU) 3679 max_mtu = ETH_MIN_MTU; 3680 3681 if (!changelink && !conf->mtu) 3682 dev->mtu = max_mtu; 3683 } 3684 3685 if (dev->mtu > max_mtu) 3686 dev->mtu = max_mtu; 3687 3688 if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA) 3689 needed_headroom += VXLAN6_HEADROOM; 3690 else 3691 needed_headroom += VXLAN_HEADROOM; 3692 dev->needed_headroom = needed_headroom; 3693 3694 memcpy(&vxlan->cfg, conf, sizeof(*conf)); 3695 } 3696 3697 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev, 3698 struct vxlan_config *conf, bool changelink, 3699 struct netlink_ext_ack *extack) 3700 { 3701 struct vxlan_dev *vxlan = netdev_priv(dev); 3702 struct net_device *lowerdev; 3703 int ret; 3704 3705 ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack); 3706 if (ret) 3707 return ret; 3708 3709 vxlan_config_apply(dev, conf, lowerdev, src_net, changelink); 3710 3711 return 0; 3712 } 3713 3714 static int __vxlan_dev_create(struct net *net, struct net_device *dev, 3715 struct vxlan_config *conf, 3716 struct netlink_ext_ack *extack) 3717 { 3718 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 3719 struct vxlan_dev *vxlan = netdev_priv(dev); 3720 struct net_device *remote_dev = NULL; 3721 struct vxlan_fdb *f = NULL; 3722 bool unregister = false; 3723 struct vxlan_rdst *dst; 3724 int err; 3725 3726 dst = &vxlan->default_dst; 3727 err = vxlan_dev_configure(net, dev, conf, false, extack); 3728 if (err) 3729 return err; 3730 3731 dev->ethtool_ops = &vxlan_ethtool_ops; 3732 3733 /* create an fdb entry for a valid default destination */ 3734 if (!vxlan_addr_any(&dst->remote_ip)) { 3735 err = vxlan_fdb_create(vxlan, all_zeros_mac, 3736 &dst->remote_ip, 3737 NUD_REACHABLE | NUD_PERMANENT, 3738 vxlan->cfg.dst_port, 3739 dst->remote_vni, 3740 dst->remote_vni, 3741 dst->remote_ifindex, 3742 NTF_SELF, 0, &f, extack); 3743 if (err) 3744 return err; 3745 } 3746 3747 err = register_netdevice(dev); 3748 if (err) 3749 goto errout; 3750 unregister = true; 3751 3752 if (dst->remote_ifindex) { 3753 remote_dev = __dev_get_by_index(net, dst->remote_ifindex); 3754 if (!remote_dev) { 3755 err = -ENODEV; 3756 goto errout; 3757 } 3758 3759 err = netdev_upper_dev_link(remote_dev, dev, extack); 3760 if (err) 3761 goto errout; 3762 } 3763 3764 err = rtnl_configure_link(dev, NULL, 0, NULL); 3765 if (err < 0) 3766 goto unlink; 3767 3768 if (f) { 3769 vxlan_fdb_insert(vxlan, all_zeros_mac, dst->remote_vni, f); 3770 3771 /* notify default fdb entry */ 3772 err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), 3773 RTM_NEWNEIGH, true, extack); 3774 if (err) { 3775 vxlan_fdb_destroy(vxlan, f, false, false); 3776 if (remote_dev) 3777 netdev_upper_dev_unlink(remote_dev, dev); 3778 goto unregister; 3779 } 3780 } 3781 3782 list_add(&vxlan->next, &vn->vxlan_list); 3783 if (remote_dev) 3784 dst->remote_dev = remote_dev; 3785 return 0; 3786 unlink: 3787 if (remote_dev) 3788 netdev_upper_dev_unlink(remote_dev, dev); 3789 errout: 3790 /* unregister_netdevice() destroys the default FDB entry with deletion 3791 * notification. But the addition notification was not sent yet, so 3792 * destroy the entry by hand here. 3793 */ 3794 if (f) 3795 __vxlan_fdb_free(f); 3796 unregister: 3797 if (unregister) 3798 unregister_netdevice(dev); 3799 return err; 3800 } 3801 3802 /* Set/clear flags based on attribute */ 3803 static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[], 3804 int attrtype, unsigned long mask, bool changelink, 3805 bool changelink_supported, 3806 struct netlink_ext_ack *extack) 3807 { 3808 unsigned long flags; 3809 3810 if (!tb[attrtype]) 3811 return 0; 3812 3813 if (changelink && !changelink_supported) { 3814 vxlan_flag_attr_error(attrtype, extack); 3815 return -EOPNOTSUPP; 3816 } 3817 3818 if (vxlan_policy[attrtype].type == NLA_FLAG) 3819 flags = conf->flags | mask; 3820 else if (nla_get_u8(tb[attrtype])) 3821 flags = conf->flags | mask; 3822 else 3823 flags = conf->flags & ~mask; 3824 3825 conf->flags = flags; 3826 3827 return 0; 3828 } 3829 3830 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[], 3831 struct net_device *dev, struct vxlan_config *conf, 3832 bool changelink, struct netlink_ext_ack *extack) 3833 { 3834 struct vxlan_dev *vxlan = netdev_priv(dev); 3835 int err = 0; 3836 3837 memset(conf, 0, sizeof(*conf)); 3838 3839 /* if changelink operation, start with old existing cfg */ 3840 if (changelink) 3841 memcpy(conf, &vxlan->cfg, sizeof(*conf)); 3842 3843 if (data[IFLA_VXLAN_ID]) { 3844 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID])); 3845 3846 if (changelink && (vni != conf->vni)) { 3847 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI"); 3848 return -EOPNOTSUPP; 3849 } 3850 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID])); 3851 } 3852 3853 if (data[IFLA_VXLAN_GROUP]) { 3854 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) { 3855 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group"); 3856 return -EOPNOTSUPP; 3857 } 3858 3859 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]); 3860 conf->remote_ip.sa.sa_family = AF_INET; 3861 } else if (data[IFLA_VXLAN_GROUP6]) { 3862 if (!IS_ENABLED(CONFIG_IPV6)) { 3863 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel"); 3864 return -EPFNOSUPPORT; 3865 } 3866 3867 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) { 3868 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group"); 3869 return -EOPNOTSUPP; 3870 } 3871 3872 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]); 3873 conf->remote_ip.sa.sa_family = AF_INET6; 3874 } 3875 3876 if (data[IFLA_VXLAN_LOCAL]) { 3877 if (changelink && (conf->saddr.sa.sa_family != AF_INET)) { 3878 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old"); 3879 return -EOPNOTSUPP; 3880 } 3881 3882 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]); 3883 conf->saddr.sa.sa_family = AF_INET; 3884 } else if (data[IFLA_VXLAN_LOCAL6]) { 3885 if (!IS_ENABLED(CONFIG_IPV6)) { 3886 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel"); 3887 return -EPFNOSUPPORT; 3888 } 3889 3890 if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) { 3891 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old"); 3892 return -EOPNOTSUPP; 3893 } 3894 3895 /* TODO: respect scope id */ 3896 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]); 3897 conf->saddr.sa.sa_family = AF_INET6; 3898 } 3899 3900 if (data[IFLA_VXLAN_LINK]) 3901 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]); 3902 3903 if (data[IFLA_VXLAN_TOS]) 3904 conf->tos = nla_get_u8(data[IFLA_VXLAN_TOS]); 3905 3906 if (data[IFLA_VXLAN_TTL]) 3907 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]); 3908 3909 if (data[IFLA_VXLAN_TTL_INHERIT]) { 3910 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT, 3911 VXLAN_F_TTL_INHERIT, changelink, false, 3912 extack); 3913 if (err) 3914 return err; 3915 3916 } 3917 3918 if (data[IFLA_VXLAN_LABEL]) 3919 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) & 3920 IPV6_FLOWLABEL_MASK; 3921 3922 if (data[IFLA_VXLAN_LEARNING]) { 3923 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING, 3924 VXLAN_F_LEARN, changelink, true, 3925 extack); 3926 if (err) 3927 return err; 3928 } else if (!changelink) { 3929 /* default to learn on a new device */ 3930 conf->flags |= VXLAN_F_LEARN; 3931 } 3932 3933 if (data[IFLA_VXLAN_AGEING]) 3934 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]); 3935 3936 if (data[IFLA_VXLAN_PROXY]) { 3937 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY, 3938 VXLAN_F_PROXY, changelink, false, 3939 extack); 3940 if (err) 3941 return err; 3942 } 3943 3944 if (data[IFLA_VXLAN_RSC]) { 3945 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC, 3946 VXLAN_F_RSC, changelink, false, 3947 extack); 3948 if (err) 3949 return err; 3950 } 3951 3952 if (data[IFLA_VXLAN_L2MISS]) { 3953 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS, 3954 VXLAN_F_L2MISS, changelink, false, 3955 extack); 3956 if (err) 3957 return err; 3958 } 3959 3960 if (data[IFLA_VXLAN_L3MISS]) { 3961 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS, 3962 VXLAN_F_L3MISS, changelink, false, 3963 extack); 3964 if (err) 3965 return err; 3966 } 3967 3968 if (data[IFLA_VXLAN_LIMIT]) { 3969 if (changelink) { 3970 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT], 3971 "Cannot change limit"); 3972 return -EOPNOTSUPP; 3973 } 3974 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]); 3975 } 3976 3977 if (data[IFLA_VXLAN_COLLECT_METADATA]) { 3978 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA, 3979 VXLAN_F_COLLECT_METADATA, changelink, false, 3980 extack); 3981 if (err) 3982 return err; 3983 } 3984 3985 if (data[IFLA_VXLAN_PORT_RANGE]) { 3986 if (!changelink) { 3987 const struct ifla_vxlan_port_range *p 3988 = nla_data(data[IFLA_VXLAN_PORT_RANGE]); 3989 conf->port_min = ntohs(p->low); 3990 conf->port_max = ntohs(p->high); 3991 } else { 3992 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE], 3993 "Cannot change port range"); 3994 return -EOPNOTSUPP; 3995 } 3996 } 3997 3998 if (data[IFLA_VXLAN_PORT]) { 3999 if (changelink) { 4000 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT], 4001 "Cannot change port"); 4002 return -EOPNOTSUPP; 4003 } 4004 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]); 4005 } 4006 4007 if (data[IFLA_VXLAN_UDP_CSUM]) { 4008 if (changelink) { 4009 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM], 4010 "Cannot change UDP_CSUM flag"); 4011 return -EOPNOTSUPP; 4012 } 4013 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM])) 4014 conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX; 4015 } 4016 4017 if (data[IFLA_VXLAN_LOCALBYPASS]) { 4018 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LOCALBYPASS, 4019 VXLAN_F_LOCALBYPASS, changelink, 4020 true, extack); 4021 if (err) 4022 return err; 4023 } else if (!changelink) { 4024 /* default to local bypass on a new device */ 4025 conf->flags |= VXLAN_F_LOCALBYPASS; 4026 } 4027 4028 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) { 4029 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX, 4030 VXLAN_F_UDP_ZERO_CSUM6_TX, changelink, 4031 false, extack); 4032 if (err) 4033 return err; 4034 } 4035 4036 if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) { 4037 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX, 4038 VXLAN_F_UDP_ZERO_CSUM6_RX, changelink, 4039 false, extack); 4040 if (err) 4041 return err; 4042 } 4043 4044 if (data[IFLA_VXLAN_REMCSUM_TX]) { 4045 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX, 4046 VXLAN_F_REMCSUM_TX, changelink, false, 4047 extack); 4048 if (err) 4049 return err; 4050 } 4051 4052 if (data[IFLA_VXLAN_REMCSUM_RX]) { 4053 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX, 4054 VXLAN_F_REMCSUM_RX, changelink, false, 4055 extack); 4056 if (err) 4057 return err; 4058 } 4059 4060 if (data[IFLA_VXLAN_GBP]) { 4061 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP, 4062 VXLAN_F_GBP, changelink, false, extack); 4063 if (err) 4064 return err; 4065 } 4066 4067 if (data[IFLA_VXLAN_GPE]) { 4068 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE, 4069 VXLAN_F_GPE, changelink, false, 4070 extack); 4071 if (err) 4072 return err; 4073 } 4074 4075 if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) { 4076 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL, 4077 VXLAN_F_REMCSUM_NOPARTIAL, changelink, 4078 false, extack); 4079 if (err) 4080 return err; 4081 } 4082 4083 if (tb[IFLA_MTU]) { 4084 if (changelink) { 4085 NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU], 4086 "Cannot change mtu"); 4087 return -EOPNOTSUPP; 4088 } 4089 conf->mtu = nla_get_u32(tb[IFLA_MTU]); 4090 } 4091 4092 if (data[IFLA_VXLAN_DF]) 4093 conf->df = nla_get_u8(data[IFLA_VXLAN_DF]); 4094 4095 if (data[IFLA_VXLAN_VNIFILTER]) { 4096 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_VNIFILTER, 4097 VXLAN_F_VNIFILTER, changelink, false, 4098 extack); 4099 if (err) 4100 return err; 4101 4102 if ((conf->flags & VXLAN_F_VNIFILTER) && 4103 !(conf->flags & VXLAN_F_COLLECT_METADATA)) { 4104 NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_VNIFILTER], 4105 "vxlan vnifilter only valid in collect metadata mode"); 4106 return -EINVAL; 4107 } 4108 } 4109 4110 return 0; 4111 } 4112 4113 static int vxlan_newlink(struct net *src_net, struct net_device *dev, 4114 struct nlattr *tb[], struct nlattr *data[], 4115 struct netlink_ext_ack *extack) 4116 { 4117 struct vxlan_config conf; 4118 int err; 4119 4120 err = vxlan_nl2conf(tb, data, dev, &conf, false, extack); 4121 if (err) 4122 return err; 4123 4124 return __vxlan_dev_create(src_net, dev, &conf, extack); 4125 } 4126 4127 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[], 4128 struct nlattr *data[], 4129 struct netlink_ext_ack *extack) 4130 { 4131 struct vxlan_dev *vxlan = netdev_priv(dev); 4132 struct net_device *lowerdev; 4133 struct vxlan_config conf; 4134 struct vxlan_rdst *dst; 4135 int err; 4136 4137 dst = &vxlan->default_dst; 4138 err = vxlan_nl2conf(tb, data, dev, &conf, true, extack); 4139 if (err) 4140 return err; 4141 4142 err = vxlan_config_validate(vxlan->net, &conf, &lowerdev, 4143 vxlan, extack); 4144 if (err) 4145 return err; 4146 4147 if (dst->remote_dev == lowerdev) 4148 lowerdev = NULL; 4149 4150 err = netdev_adjacent_change_prepare(dst->remote_dev, lowerdev, dev, 4151 extack); 4152 if (err) 4153 return err; 4154 4155 /* handle default dst entry */ 4156 if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) { 4157 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, conf.vni); 4158 4159 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4160 if (!vxlan_addr_any(&conf.remote_ip)) { 4161 err = vxlan_fdb_update(vxlan, all_zeros_mac, 4162 &conf.remote_ip, 4163 NUD_REACHABLE | NUD_PERMANENT, 4164 NLM_F_APPEND | NLM_F_CREATE, 4165 vxlan->cfg.dst_port, 4166 conf.vni, conf.vni, 4167 conf.remote_ifindex, 4168 NTF_SELF, 0, true, extack); 4169 if (err) { 4170 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4171 netdev_adjacent_change_abort(dst->remote_dev, 4172 lowerdev, dev); 4173 return err; 4174 } 4175 } 4176 if (!vxlan_addr_any(&dst->remote_ip)) 4177 __vxlan_fdb_delete(vxlan, all_zeros_mac, 4178 dst->remote_ip, 4179 vxlan->cfg.dst_port, 4180 dst->remote_vni, 4181 dst->remote_vni, 4182 dst->remote_ifindex, 4183 true); 4184 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4185 4186 /* If vni filtering device, also update fdb entries of 4187 * all vnis that were using default remote ip 4188 */ 4189 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER) { 4190 err = vxlan_vnilist_update_group(vxlan, &dst->remote_ip, 4191 &conf.remote_ip, extack); 4192 if (err) { 4193 netdev_adjacent_change_abort(dst->remote_dev, 4194 lowerdev, dev); 4195 return err; 4196 } 4197 } 4198 } 4199 4200 if (conf.age_interval != vxlan->cfg.age_interval) 4201 mod_timer(&vxlan->age_timer, jiffies); 4202 4203 netdev_adjacent_change_commit(dst->remote_dev, lowerdev, dev); 4204 if (lowerdev && lowerdev != dst->remote_dev) 4205 dst->remote_dev = lowerdev; 4206 vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true); 4207 return 0; 4208 } 4209 4210 static void vxlan_dellink(struct net_device *dev, struct list_head *head) 4211 { 4212 struct vxlan_dev *vxlan = netdev_priv(dev); 4213 4214 vxlan_flush(vxlan, true); 4215 4216 list_del(&vxlan->next); 4217 unregister_netdevice_queue(dev, head); 4218 if (vxlan->default_dst.remote_dev) 4219 netdev_upper_dev_unlink(vxlan->default_dst.remote_dev, dev); 4220 } 4221 4222 static size_t vxlan_get_size(const struct net_device *dev) 4223 { 4224 4225 return nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_ID */ 4226 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */ 4227 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */ 4228 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */ 4229 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL */ 4230 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TTL_INHERIT */ 4231 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_TOS */ 4232 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_DF */ 4233 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */ 4234 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LEARNING */ 4235 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_PROXY */ 4236 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_RSC */ 4237 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L2MISS */ 4238 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_L3MISS */ 4239 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_COLLECT_METADATA */ 4240 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */ 4241 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */ 4242 nla_total_size(sizeof(struct ifla_vxlan_port_range)) + 4243 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */ 4244 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */ 4245 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */ 4246 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */ 4247 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */ 4248 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */ 4249 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LOCALBYPASS */ 4250 0; 4251 } 4252 4253 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev) 4254 { 4255 const struct vxlan_dev *vxlan = netdev_priv(dev); 4256 const struct vxlan_rdst *dst = &vxlan->default_dst; 4257 struct ifla_vxlan_port_range ports = { 4258 .low = htons(vxlan->cfg.port_min), 4259 .high = htons(vxlan->cfg.port_max), 4260 }; 4261 4262 if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni))) 4263 goto nla_put_failure; 4264 4265 if (!vxlan_addr_any(&dst->remote_ip)) { 4266 if (dst->remote_ip.sa.sa_family == AF_INET) { 4267 if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP, 4268 dst->remote_ip.sin.sin_addr.s_addr)) 4269 goto nla_put_failure; 4270 #if IS_ENABLED(CONFIG_IPV6) 4271 } else { 4272 if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6, 4273 &dst->remote_ip.sin6.sin6_addr)) 4274 goto nla_put_failure; 4275 #endif 4276 } 4277 } 4278 4279 if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex)) 4280 goto nla_put_failure; 4281 4282 if (!vxlan_addr_any(&vxlan->cfg.saddr)) { 4283 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) { 4284 if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL, 4285 vxlan->cfg.saddr.sin.sin_addr.s_addr)) 4286 goto nla_put_failure; 4287 #if IS_ENABLED(CONFIG_IPV6) 4288 } else { 4289 if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6, 4290 &vxlan->cfg.saddr.sin6.sin6_addr)) 4291 goto nla_put_failure; 4292 #endif 4293 } 4294 } 4295 4296 if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) || 4297 nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT, 4298 !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) || 4299 nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) || 4300 nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) || 4301 nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) || 4302 nla_put_u8(skb, IFLA_VXLAN_LEARNING, 4303 !!(vxlan->cfg.flags & VXLAN_F_LEARN)) || 4304 nla_put_u8(skb, IFLA_VXLAN_PROXY, 4305 !!(vxlan->cfg.flags & VXLAN_F_PROXY)) || 4306 nla_put_u8(skb, IFLA_VXLAN_RSC, 4307 !!(vxlan->cfg.flags & VXLAN_F_RSC)) || 4308 nla_put_u8(skb, IFLA_VXLAN_L2MISS, 4309 !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) || 4310 nla_put_u8(skb, IFLA_VXLAN_L3MISS, 4311 !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) || 4312 nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA, 4313 !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) || 4314 nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) || 4315 nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) || 4316 nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) || 4317 nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM, 4318 !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) || 4319 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX, 4320 !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) || 4321 nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX, 4322 !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) || 4323 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX, 4324 !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) || 4325 nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX, 4326 !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)) || 4327 nla_put_u8(skb, IFLA_VXLAN_LOCALBYPASS, 4328 !!(vxlan->cfg.flags & VXLAN_F_LOCALBYPASS))) 4329 goto nla_put_failure; 4330 4331 if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports)) 4332 goto nla_put_failure; 4333 4334 if (vxlan->cfg.flags & VXLAN_F_GBP && 4335 nla_put_flag(skb, IFLA_VXLAN_GBP)) 4336 goto nla_put_failure; 4337 4338 if (vxlan->cfg.flags & VXLAN_F_GPE && 4339 nla_put_flag(skb, IFLA_VXLAN_GPE)) 4340 goto nla_put_failure; 4341 4342 if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL && 4343 nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL)) 4344 goto nla_put_failure; 4345 4346 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER && 4347 nla_put_u8(skb, IFLA_VXLAN_VNIFILTER, 4348 !!(vxlan->cfg.flags & VXLAN_F_VNIFILTER))) 4349 goto nla_put_failure; 4350 4351 return 0; 4352 4353 nla_put_failure: 4354 return -EMSGSIZE; 4355 } 4356 4357 static struct net *vxlan_get_link_net(const struct net_device *dev) 4358 { 4359 struct vxlan_dev *vxlan = netdev_priv(dev); 4360 4361 return vxlan->net; 4362 } 4363 4364 static struct rtnl_link_ops vxlan_link_ops __read_mostly = { 4365 .kind = "vxlan", 4366 .maxtype = IFLA_VXLAN_MAX, 4367 .policy = vxlan_policy, 4368 .priv_size = sizeof(struct vxlan_dev), 4369 .setup = vxlan_setup, 4370 .validate = vxlan_validate, 4371 .newlink = vxlan_newlink, 4372 .changelink = vxlan_changelink, 4373 .dellink = vxlan_dellink, 4374 .get_size = vxlan_get_size, 4375 .fill_info = vxlan_fill_info, 4376 .get_link_net = vxlan_get_link_net, 4377 }; 4378 4379 struct net_device *vxlan_dev_create(struct net *net, const char *name, 4380 u8 name_assign_type, 4381 struct vxlan_config *conf) 4382 { 4383 struct nlattr *tb[IFLA_MAX + 1]; 4384 struct net_device *dev; 4385 int err; 4386 4387 memset(&tb, 0, sizeof(tb)); 4388 4389 dev = rtnl_create_link(net, name, name_assign_type, 4390 &vxlan_link_ops, tb, NULL); 4391 if (IS_ERR(dev)) 4392 return dev; 4393 4394 err = __vxlan_dev_create(net, dev, conf, NULL); 4395 if (err < 0) { 4396 free_netdev(dev); 4397 return ERR_PTR(err); 4398 } 4399 4400 err = rtnl_configure_link(dev, NULL, 0, NULL); 4401 if (err < 0) { 4402 LIST_HEAD(list_kill); 4403 4404 vxlan_dellink(dev, &list_kill); 4405 unregister_netdevice_many(&list_kill); 4406 return ERR_PTR(err); 4407 } 4408 4409 return dev; 4410 } 4411 EXPORT_SYMBOL_GPL(vxlan_dev_create); 4412 4413 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn, 4414 struct net_device *dev) 4415 { 4416 struct vxlan_dev *vxlan, *next; 4417 LIST_HEAD(list_kill); 4418 4419 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) { 4420 struct vxlan_rdst *dst = &vxlan->default_dst; 4421 4422 /* In case we created vxlan device with carrier 4423 * and we loose the carrier due to module unload 4424 * we also need to remove vxlan device. In other 4425 * cases, it's not necessary and remote_ifindex 4426 * is 0 here, so no matches. 4427 */ 4428 if (dst->remote_ifindex == dev->ifindex) 4429 vxlan_dellink(vxlan->dev, &list_kill); 4430 } 4431 4432 unregister_netdevice_many(&list_kill); 4433 } 4434 4435 static int vxlan_netdevice_event(struct notifier_block *unused, 4436 unsigned long event, void *ptr) 4437 { 4438 struct net_device *dev = netdev_notifier_info_to_dev(ptr); 4439 struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id); 4440 4441 if (event == NETDEV_UNREGISTER) 4442 vxlan_handle_lowerdev_unregister(vn, dev); 4443 else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO) 4444 vxlan_offload_rx_ports(dev, true); 4445 else if (event == NETDEV_UDP_TUNNEL_DROP_INFO) 4446 vxlan_offload_rx_ports(dev, false); 4447 4448 return NOTIFY_DONE; 4449 } 4450 4451 static struct notifier_block vxlan_notifier_block __read_mostly = { 4452 .notifier_call = vxlan_netdevice_event, 4453 }; 4454 4455 static void 4456 vxlan_fdb_offloaded_set(struct net_device *dev, 4457 struct switchdev_notifier_vxlan_fdb_info *fdb_info) 4458 { 4459 struct vxlan_dev *vxlan = netdev_priv(dev); 4460 struct vxlan_rdst *rdst; 4461 struct vxlan_fdb *f; 4462 u32 hash_index; 4463 4464 hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni); 4465 4466 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4467 4468 f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni); 4469 if (!f) 4470 goto out; 4471 4472 rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip, 4473 fdb_info->remote_port, 4474 fdb_info->remote_vni, 4475 fdb_info->remote_ifindex); 4476 if (!rdst) 4477 goto out; 4478 4479 rdst->offloaded = fdb_info->offloaded; 4480 4481 out: 4482 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4483 } 4484 4485 static int 4486 vxlan_fdb_external_learn_add(struct net_device *dev, 4487 struct switchdev_notifier_vxlan_fdb_info *fdb_info) 4488 { 4489 struct vxlan_dev *vxlan = netdev_priv(dev); 4490 struct netlink_ext_ack *extack; 4491 u32 hash_index; 4492 int err; 4493 4494 hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni); 4495 extack = switchdev_notifier_info_to_extack(&fdb_info->info); 4496 4497 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4498 err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip, 4499 NUD_REACHABLE, 4500 NLM_F_CREATE | NLM_F_REPLACE, 4501 fdb_info->remote_port, 4502 fdb_info->vni, 4503 fdb_info->remote_vni, 4504 fdb_info->remote_ifindex, 4505 NTF_USE | NTF_SELF | NTF_EXT_LEARNED, 4506 0, false, extack); 4507 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4508 4509 return err; 4510 } 4511 4512 static int 4513 vxlan_fdb_external_learn_del(struct net_device *dev, 4514 struct switchdev_notifier_vxlan_fdb_info *fdb_info) 4515 { 4516 struct vxlan_dev *vxlan = netdev_priv(dev); 4517 struct vxlan_fdb *f; 4518 u32 hash_index; 4519 int err = 0; 4520 4521 hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni); 4522 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4523 4524 f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni); 4525 if (!f) 4526 err = -ENOENT; 4527 else if (f->flags & NTF_EXT_LEARNED) 4528 err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr, 4529 fdb_info->remote_ip, 4530 fdb_info->remote_port, 4531 fdb_info->vni, 4532 fdb_info->remote_vni, 4533 fdb_info->remote_ifindex, 4534 false); 4535 4536 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4537 4538 return err; 4539 } 4540 4541 static int vxlan_switchdev_event(struct notifier_block *unused, 4542 unsigned long event, void *ptr) 4543 { 4544 struct net_device *dev = switchdev_notifier_info_to_dev(ptr); 4545 struct switchdev_notifier_vxlan_fdb_info *fdb_info; 4546 int err = 0; 4547 4548 switch (event) { 4549 case SWITCHDEV_VXLAN_FDB_OFFLOADED: 4550 vxlan_fdb_offloaded_set(dev, ptr); 4551 break; 4552 case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE: 4553 fdb_info = ptr; 4554 err = vxlan_fdb_external_learn_add(dev, fdb_info); 4555 if (err) { 4556 err = notifier_from_errno(err); 4557 break; 4558 } 4559 fdb_info->offloaded = true; 4560 vxlan_fdb_offloaded_set(dev, fdb_info); 4561 break; 4562 case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE: 4563 fdb_info = ptr; 4564 err = vxlan_fdb_external_learn_del(dev, fdb_info); 4565 if (err) { 4566 err = notifier_from_errno(err); 4567 break; 4568 } 4569 fdb_info->offloaded = false; 4570 vxlan_fdb_offloaded_set(dev, fdb_info); 4571 break; 4572 } 4573 4574 return err; 4575 } 4576 4577 static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = { 4578 .notifier_call = vxlan_switchdev_event, 4579 }; 4580 4581 static void vxlan_fdb_nh_flush(struct nexthop *nh) 4582 { 4583 struct vxlan_fdb *fdb; 4584 struct vxlan_dev *vxlan; 4585 u32 hash_index; 4586 4587 rcu_read_lock(); 4588 list_for_each_entry_rcu(fdb, &nh->fdb_list, nh_list) { 4589 vxlan = rcu_dereference(fdb->vdev); 4590 WARN_ON(!vxlan); 4591 hash_index = fdb_head_index(vxlan, fdb->eth_addr, 4592 vxlan->default_dst.remote_vni); 4593 spin_lock_bh(&vxlan->hash_lock[hash_index]); 4594 if (!hlist_unhashed(&fdb->hlist)) 4595 vxlan_fdb_destroy(vxlan, fdb, false, false); 4596 spin_unlock_bh(&vxlan->hash_lock[hash_index]); 4597 } 4598 rcu_read_unlock(); 4599 } 4600 4601 static int vxlan_nexthop_event(struct notifier_block *nb, 4602 unsigned long event, void *ptr) 4603 { 4604 struct nh_notifier_info *info = ptr; 4605 struct nexthop *nh; 4606 4607 if (event != NEXTHOP_EVENT_DEL) 4608 return NOTIFY_DONE; 4609 4610 nh = nexthop_find_by_id(info->net, info->id); 4611 if (!nh) 4612 return NOTIFY_DONE; 4613 4614 vxlan_fdb_nh_flush(nh); 4615 4616 return NOTIFY_DONE; 4617 } 4618 4619 static __net_init int vxlan_init_net(struct net *net) 4620 { 4621 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 4622 unsigned int h; 4623 4624 INIT_LIST_HEAD(&vn->vxlan_list); 4625 spin_lock_init(&vn->sock_lock); 4626 vn->nexthop_notifier_block.notifier_call = vxlan_nexthop_event; 4627 4628 for (h = 0; h < PORT_HASH_SIZE; ++h) 4629 INIT_HLIST_HEAD(&vn->sock_list[h]); 4630 4631 return register_nexthop_notifier(net, &vn->nexthop_notifier_block, 4632 NULL); 4633 } 4634 4635 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head) 4636 { 4637 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 4638 struct vxlan_dev *vxlan, *next; 4639 struct net_device *dev, *aux; 4640 4641 for_each_netdev_safe(net, dev, aux) 4642 if (dev->rtnl_link_ops == &vxlan_link_ops) 4643 unregister_netdevice_queue(dev, head); 4644 4645 list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) { 4646 /* If vxlan->dev is in the same netns, it has already been added 4647 * to the list by the previous loop. 4648 */ 4649 if (!net_eq(dev_net(vxlan->dev), net)) 4650 unregister_netdevice_queue(vxlan->dev, head); 4651 } 4652 4653 } 4654 4655 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list) 4656 { 4657 struct net *net; 4658 LIST_HEAD(list); 4659 unsigned int h; 4660 4661 list_for_each_entry(net, net_list, exit_list) { 4662 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 4663 4664 unregister_nexthop_notifier(net, &vn->nexthop_notifier_block); 4665 } 4666 rtnl_lock(); 4667 list_for_each_entry(net, net_list, exit_list) 4668 vxlan_destroy_tunnels(net, &list); 4669 4670 unregister_netdevice_many(&list); 4671 rtnl_unlock(); 4672 4673 list_for_each_entry(net, net_list, exit_list) { 4674 struct vxlan_net *vn = net_generic(net, vxlan_net_id); 4675 4676 for (h = 0; h < PORT_HASH_SIZE; ++h) 4677 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h])); 4678 } 4679 } 4680 4681 static struct pernet_operations vxlan_net_ops = { 4682 .init = vxlan_init_net, 4683 .exit_batch = vxlan_exit_batch_net, 4684 .id = &vxlan_net_id, 4685 .size = sizeof(struct vxlan_net), 4686 }; 4687 4688 static int __init vxlan_init_module(void) 4689 { 4690 int rc; 4691 4692 get_random_bytes(&vxlan_salt, sizeof(vxlan_salt)); 4693 4694 rc = register_pernet_subsys(&vxlan_net_ops); 4695 if (rc) 4696 goto out1; 4697 4698 rc = register_netdevice_notifier(&vxlan_notifier_block); 4699 if (rc) 4700 goto out2; 4701 4702 rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block); 4703 if (rc) 4704 goto out3; 4705 4706 rc = rtnl_link_register(&vxlan_link_ops); 4707 if (rc) 4708 goto out4; 4709 4710 vxlan_vnifilter_init(); 4711 4712 return 0; 4713 out4: 4714 unregister_switchdev_notifier(&vxlan_switchdev_notifier_block); 4715 out3: 4716 unregister_netdevice_notifier(&vxlan_notifier_block); 4717 out2: 4718 unregister_pernet_subsys(&vxlan_net_ops); 4719 out1: 4720 return rc; 4721 } 4722 late_initcall(vxlan_init_module); 4723 4724 static void __exit vxlan_cleanup_module(void) 4725 { 4726 vxlan_vnifilter_uninit(); 4727 rtnl_link_unregister(&vxlan_link_ops); 4728 unregister_switchdev_notifier(&vxlan_switchdev_notifier_block); 4729 unregister_netdevice_notifier(&vxlan_notifier_block); 4730 unregister_pernet_subsys(&vxlan_net_ops); 4731 /* rcu_barrier() is called by netns */ 4732 } 4733 module_exit(vxlan_cleanup_module); 4734 4735 MODULE_LICENSE("GPL"); 4736 MODULE_VERSION(VXLAN_VERSION); 4737 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>"); 4738 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic"); 4739 MODULE_ALIAS_RTNL_LINK("vxlan"); 4740