drivers/net/vxlan.c

   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/ipv6_stubs.h>
  21 #include <net/ip.h>
  22 #include <net/icmp.h>
  23 #include <net/rtnetlink.h>
  24 #include <net/inet_ecn.h>
  25 #include <net/net_namespace.h>
  26 #include <net/netns/generic.h>
  27 #include <net/tun_proto.h>
  28 #include <net/vxlan.h>
  29
  30 #if IS_ENABLED(CONFIG_IPV6)
  31 #include <net/ip6_tunnel.h>
  32 #include <net/ip6_checksum.h>
  33 #endif
  34
  35 #define VXLAN_VERSION   "0.1"
  36
  37 #define PORT_HASH_BITS  8
  38 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  39 #define FDB_AGE_DEFAULT 300 /* 5 min */
  40 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  41
  42 /* UDP port for VXLAN traffic.
  43  * The IANA assigned port is 4789, but the Linux default is 8472
  44  * for compatibility with early adopters.
  45  */
  46 static unsigned short vxlan_port __read_mostly = 8472;
  47 module_param_named(udp_port, vxlan_port, ushort, 0444);
  48 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  49
  50 static bool log_ecn_error = true;
  51 module_param(log_ecn_error, bool, 0644);
  52 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  53
  54 static unsigned int vxlan_net_id;
  55 static struct rtnl_link_ops vxlan_link_ops;
  56
  57 static const u8 all_zeros_mac[ETH_ALEN + 2];
  58
  59 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  60
  61 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan);
  62
  63 /* per-network namespace private data for this module */
  64 struct vxlan_net {
  65         struct list_head  vxlan_list;
  66         struct hlist_head sock_list[PORT_HASH_SIZE];
  67         spinlock_t        sock_lock;
  68 };
  69
  70 /* Forwarding table entry */
  71 struct vxlan_fdb {
  72         struct hlist_node hlist;        /* linked list of entries */
  73         struct rcu_head   rcu;
  74         unsigned long     updated;      /* jiffies */
  75         unsigned long     used;
  76         struct list_head  remotes;
  77         u8                eth_addr[ETH_ALEN];
  78         u16               state;        /* see ndm_state */
  79         __be32            vni;
  80         u16               flags;        /* see ndm_flags and below */
  81 };
  82
  83 #define NTF_VXLAN_ADDED_BY_USER 0x100
  84
  85 /* salt for hash table */
  86 static u32 vxlan_salt __read_mostly;
  87
  88 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  89 {
  90         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  91                ip_tunnel_collect_metadata();
  92 }
  93
  94 #if IS_ENABLED(CONFIG_IPV6)
  95 static inline
  96 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
  97 {
  98         if (a->sa.sa_family != b->sa.sa_family)
  99                 return false;
 100         if (a->sa.sa_family == AF_INET6)
 101                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 102         else
 103                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 104 }
 105
 106 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 107 {
 108         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 109                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 110                 ip->sa.sa_family = AF_INET6;
 111                 return 0;
 112         } else if (nla_len(nla) >= sizeof(__be32)) {
 113                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 114                 ip->sa.sa_family = AF_INET;
 115                 return 0;
 116         } else {
 117                 return -EAFNOSUPPORT;
 118         }
 119 }
 120
 121 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 122                               const union vxlan_addr *ip)
 123 {
 124         if (ip->sa.sa_family == AF_INET6)
 125                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 126         else
 127                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 128 }
 129
 130 #else /* !CONFIG_IPV6 */
 131
 132 static inline
 133 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 134 {
 135         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 136 }
 137
 138 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 139 {
 140         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 141                 return -EAFNOSUPPORT;
 142         } else if (nla_len(nla) >= sizeof(__be32)) {
 143                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 144                 ip->sa.sa_family = AF_INET;
 145                 return 0;
 146         } else {
 147                 return -EAFNOSUPPORT;
 148         }
 149 }
 150
 151 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 152                               const union vxlan_addr *ip)
 153 {
 154         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 155 }
 156 #endif
 157
 158 /* Virtual Network hash table head */
 159 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 160 {
 161         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 162 }
 163
 164 /* Socket hash table head */
 165 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 166 {
 167         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 168
 169         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 170 }
 171
 172 /* First remote destination for a forwarding entry.
 173  * Guaranteed to be non-NULL because remotes are never deleted.
 174  */
 175 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 176 {
 177         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 178 }
 179
 180 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 181 {
 182         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 183 }
 184
 185 /* Find VXLAN socket based on network namespace, address family and UDP port
 186  * and enabled unshareable flags.
 187  */
 188 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 189                                           __be16 port, u32 flags, int ifindex)
 190 {
 191         struct vxlan_sock *vs;
 192
 193         flags &= VXLAN_F_RCV_FLAGS;
 194
 195         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 196                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 197                     vxlan_get_sk_family(vs) == family &&
 198                     vs->flags == flags &&
 199                     vs->sock->sk->sk_bound_dev_if == ifindex)
 200                         return vs;
 201         }
 202         return NULL;
 203 }
 204
 205 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, int ifindex,
 206                                            __be32 vni)
 207 {
 208         struct vxlan_dev_node *node;
 209
 210         /* For flow based devices, map all packets to VNI 0 */
 211         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 212                 vni = 0;
 213
 214         hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) {
 215                 if (node->vxlan->default_dst.remote_vni != vni)
 216                         continue;
 217
 218                 if (IS_ENABLED(CONFIG_IPV6)) {
 219                         const struct vxlan_config *cfg = &node->vxlan->cfg;
 220
 221                         if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) &&
 222                             cfg->remote_ifindex != ifindex)
 223                                 continue;
 224                 }
 225
 226                 return node->vxlan;
 227         }
 228
 229         return NULL;
 230 }
 231
 232 /* Look up VNI in a per net namespace table */
 233 static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex,
 234                                         __be32 vni, sa_family_t family,
 235                                         __be16 port, u32 flags)
 236 {
 237         struct vxlan_sock *vs;
 238
 239         vs = vxlan_find_sock(net, family, port, flags, ifindex);
 240         if (!vs)
 241                 return NULL;
 242
 243         return vxlan_vs_find_vni(vs, ifindex, vni);
 244 }
 245
 246 /* Fill in neighbour message in skbuff. */
 247 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 248                           const struct vxlan_fdb *fdb,
 249                           u32 portid, u32 seq, int type, unsigned int flags,
 250                           const struct vxlan_rdst *rdst)
 251 {
 252         unsigned long now = jiffies;
 253         struct nda_cacheinfo ci;
 254         struct nlmsghdr *nlh;
 255         struct ndmsg *ndm;
 256         bool send_ip, send_eth;
 257
 258         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 259         if (nlh == NULL)
 260                 return -EMSGSIZE;
 261
 262         ndm = nlmsg_data(nlh);
 263         memset(ndm, 0, sizeof(*ndm));
 264
 265         send_eth = send_ip = true;
 266
 267         if (type == RTM_GETNEIGH) {
 268                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 269                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 270                 ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
 271         } else
 272                 ndm->ndm_family = AF_BRIDGE;
 273         ndm->ndm_state = fdb->state;
 274         ndm->ndm_ifindex = vxlan->dev->ifindex;
 275         ndm->ndm_flags = fdb->flags;
 276         if (rdst->offloaded)
 277                 ndm->ndm_flags |= NTF_OFFLOADED;
 278         ndm->ndm_type = RTN_UNICAST;
 279
 280         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 281             nla_put_s32(skb, NDA_LINK_NETNSID,
 282                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 283                 goto nla_put_failure;
 284
 285         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 286                 goto nla_put_failure;
 287
 288         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 289                 goto nla_put_failure;
 290
 291         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 292             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 293                 goto nla_put_failure;
 294         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 295             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 296                 goto nla_put_failure;
 297         if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
 298             nla_put_u32(skb, NDA_SRC_VNI,
 299                         be32_to_cpu(fdb->vni)))
 300                 goto nla_put_failure;
 301         if (rdst->remote_ifindex &&
 302             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 303                 goto nla_put_failure;
 304
 305         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 306         ci.ndm_confirmed = 0;
 307         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 308         ci.ndm_refcnt    = 0;
 309
 310         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 311                 goto nla_put_failure;
 312
 313         nlmsg_end(skb, nlh);
 314         return 0;
 315
 316 nla_put_failure:
 317         nlmsg_cancel(skb, nlh);
 318         return -EMSGSIZE;
 319 }
 320
 321 static inline size_t vxlan_nlmsg_size(void)
 322 {
 323         return NLMSG_ALIGN(sizeof(struct ndmsg))
 324                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 325                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 326                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 327                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 328                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 329                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 330                 + nla_total_size(sizeof(struct nda_cacheinfo));
 331 }
 332
 333 static void __vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 334                                struct vxlan_rdst *rd, int type)
 335 {
 336         struct net *net = dev_net(vxlan->dev);
 337         struct sk_buff *skb;
 338         int err = -ENOBUFS;
 339
 340         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 341         if (skb == NULL)
 342                 goto errout;
 343
 344         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 345         if (err < 0) {
 346                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 347                 WARN_ON(err == -EMSGSIZE);
 348                 kfree_skb(skb);
 349                 goto errout;
 350         }
 351
 352         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 353         return;
 354 errout:
 355         if (err < 0)
 356                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 357 }
 358
 359 static void vxlan_fdb_switchdev_notifier_info(const struct vxlan_dev *vxlan,
 360                             const struct vxlan_fdb *fdb,
 361                             const struct vxlan_rdst *rd,
 362                             struct netlink_ext_ack *extack,
 363                             struct switchdev_notifier_vxlan_fdb_info *fdb_info)
 364 {
 365         fdb_info->info.dev = vxlan->dev;
 366         fdb_info->info.extack = extack;
 367         fdb_info->remote_ip = rd->remote_ip;
 368         fdb_info->remote_port = rd->remote_port;
 369         fdb_info->remote_vni = rd->remote_vni;
 370         fdb_info->remote_ifindex = rd->remote_ifindex;
 371         memcpy(fdb_info->eth_addr, fdb->eth_addr, ETH_ALEN);
 372         fdb_info->vni = fdb->vni;
 373         fdb_info->offloaded = rd->offloaded;
 374         fdb_info->added_by_user = fdb->flags & NTF_VXLAN_ADDED_BY_USER;
 375 }
 376
 377 static int vxlan_fdb_switchdev_call_notifiers(struct vxlan_dev *vxlan,
 378                                               struct vxlan_fdb *fdb,
 379                                               struct vxlan_rdst *rd,
 380                                               bool adding,
 381                                               struct netlink_ext_ack *extack)
 382 {
 383         struct switchdev_notifier_vxlan_fdb_info info;
 384         enum switchdev_notifier_type notifier_type;
 385         int ret;
 386
 387         if (WARN_ON(!rd))
 388                 return 0;
 389
 390         notifier_type = adding ? SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE
 391                                : SWITCHDEV_VXLAN_FDB_DEL_TO_DEVICE;
 392         vxlan_fdb_switchdev_notifier_info(vxlan, fdb, rd, NULL, &info);
 393         ret = call_switchdev_notifiers(notifier_type, vxlan->dev,
 394                                        &info.info, extack);
 395         return notifier_to_errno(ret);
 396 }
 397
 398 static int vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 399                             struct vxlan_rdst *rd, int type, bool swdev_notify,
 400                             struct netlink_ext_ack *extack)
 401 {
 402         int err;
 403
 404         if (swdev_notify) {
 405                 switch (type) {
 406                 case RTM_NEWNEIGH:
 407                         err = vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
 408                                                                  true, extack);
 409                         if (err)
 410                                 return err;
 411                         break;
 412                 case RTM_DELNEIGH:
 413                         vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
 414                                                            false, extack);
 415                         break;
 416                 }
 417         }
 418
 419         __vxlan_fdb_notify(vxlan, fdb, rd, type);
 420         return 0;
 421 }
 422
 423 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 424 {
 425         struct vxlan_dev *vxlan = netdev_priv(dev);
 426         struct vxlan_fdb f = {
 427                 .state = NUD_STALE,
 428         };
 429         struct vxlan_rdst remote = {
 430                 .remote_ip = *ipa, /* goes to NDA_DST */
 431                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 432         };
 433
 434         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
 435 }
 436
 437 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 438 {
 439         struct vxlan_fdb f = {
 440                 .state = NUD_STALE,
 441         };
 442         struct vxlan_rdst remote = { };
 443
 444         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 445
 446         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
 447 }
 448
 449 /* Hash Ethernet address */
 450 static u32 eth_hash(const unsigned char *addr)
 451 {
 452         u64 value = get_unaligned((u64 *)addr);
 453
 454         /* only want 6 bytes */
 455 #ifdef __BIG_ENDIAN
 456         value >>= 16;
 457 #else
 458         value <<= 16;
 459 #endif
 460         return hash_64(value, FDB_HASH_BITS);
 461 }
 462
 463 static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
 464 {
 465         /* use 1 byte of OUI and 3 bytes of NIC */
 466         u32 key = get_unaligned((u32 *)(addr + 2));
 467
 468         return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
 469 }
 470
 471 /* Hash chain to use given mac address */
 472 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 473                                                 const u8 *mac, __be32 vni)
 474 {
 475         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)
 476                 return &vxlan->fdb_head[eth_vni_hash(mac, vni)];
 477         else
 478                 return &vxlan->fdb_head[eth_hash(mac)];
 479 }
 480
 481 /* Look up Ethernet address in forwarding table */
 482 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 483                                           const u8 *mac, __be32 vni)
 484 {
 485         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
 486         struct vxlan_fdb *f;
 487
 488         hlist_for_each_entry_rcu(f, head, hlist) {
 489                 if (ether_addr_equal(mac, f->eth_addr)) {
 490                         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
 491                                 if (vni == f->vni)
 492                                         return f;
 493                         } else {
 494                                 return f;
 495                         }
 496                 }
 497         }
 498
 499         return NULL;
 500 }
 501
 502 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 503                                         const u8 *mac, __be32 vni)
 504 {
 505         struct vxlan_fdb *f;
 506
 507         f = __vxlan_find_mac(vxlan, mac, vni);
 508         if (f && f->used != jiffies)
 509                 f->used = jiffies;
 510
 511         return f;
 512 }
 513
 514 /* caller should hold vxlan->hash_lock */
 515 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 516                                               union vxlan_addr *ip, __be16 port,
 517                                               __be32 vni, __u32 ifindex)
 518 {
 519         struct vxlan_rdst *rd;
 520
 521         list_for_each_entry(rd, &f->remotes, list) {
 522                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 523                     rd->remote_port == port &&
 524                     rd->remote_vni == vni &&
 525                     rd->remote_ifindex == ifindex)
 526                         return rd;
 527         }
 528
 529         return NULL;
 530 }
 531
 532 int vxlan_fdb_find_uc(struct net_device *dev, const u8 *mac, __be32 vni,
 533                       struct switchdev_notifier_vxlan_fdb_info *fdb_info)
 534 {
 535         struct vxlan_dev *vxlan = netdev_priv(dev);
 536         u8 eth_addr[ETH_ALEN + 2] = { 0 };
 537         struct vxlan_rdst *rdst;
 538         struct vxlan_fdb *f;
 539         int rc = 0;
 540
 541         if (is_multicast_ether_addr(mac) ||
 542             is_zero_ether_addr(mac))
 543                 return -EINVAL;
 544
 545         ether_addr_copy(eth_addr, mac);
 546
 547         rcu_read_lock();
 548
 549         f = __vxlan_find_mac(vxlan, eth_addr, vni);
 550         if (!f) {
 551                 rc = -ENOENT;
 552                 goto out;
 553         }
 554
 555         rdst = first_remote_rcu(f);
 556         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, NULL, fdb_info);
 557
 558 out:
 559         rcu_read_unlock();
 560         return rc;
 561 }
 562 EXPORT_SYMBOL_GPL(vxlan_fdb_find_uc);
 563
 564 static int vxlan_fdb_notify_one(struct notifier_block *nb,
 565                                 const struct vxlan_dev *vxlan,
 566                                 const struct vxlan_fdb *f,
 567                                 const struct vxlan_rdst *rdst,
 568                                 struct netlink_ext_ack *extack)
 569 {
 570         struct switchdev_notifier_vxlan_fdb_info fdb_info;
 571         int rc;
 572
 573         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, extack, &fdb_info);
 574         rc = nb->notifier_call(nb, SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE,
 575                                &fdb_info);
 576         return notifier_to_errno(rc);
 577 }
 578
 579 int vxlan_fdb_replay(const struct net_device *dev, __be32 vni,
 580                      struct notifier_block *nb,
 581                      struct netlink_ext_ack *extack)
 582 {
 583         struct vxlan_dev *vxlan;
 584         struct vxlan_rdst *rdst;
 585         struct vxlan_fdb *f;
 586         unsigned int h;
 587         int rc = 0;
 588
 589         if (!netif_is_vxlan(dev))
 590                 return -EINVAL;
 591         vxlan = netdev_priv(dev);
 592
 593         spin_lock_bh(&vxlan->hash_lock);
 594         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 595                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) {
 596                         if (f->vni == vni) {
 597                                 list_for_each_entry(rdst, &f->remotes, list) {
 598                                         rc = vxlan_fdb_notify_one(nb, vxlan,
 599                                                                   f, rdst,
 600                                                                   extack);
 601                                         if (rc)
 602                                                 goto out;
 603                                 }
 604                         }
 605                 }
 606         }
 607
 608 out:
 609         spin_unlock_bh(&vxlan->hash_lock);
 610         return rc;
 611 }
 612 EXPORT_SYMBOL_GPL(vxlan_fdb_replay);
 613
 614 void vxlan_fdb_clear_offload(const struct net_device *dev, __be32 vni)
 615 {
 616         struct vxlan_dev *vxlan;
 617         struct vxlan_rdst *rdst;
 618         struct vxlan_fdb *f;
 619         unsigned int h;
 620
 621         if (!netif_is_vxlan(dev))
 622                 return;
 623         vxlan = netdev_priv(dev);
 624
 625         spin_lock_bh(&vxlan->hash_lock);
 626         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 627                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist)
 628                         if (f->vni == vni)
 629                                 list_for_each_entry(rdst, &f->remotes, list)
 630                                         rdst->offloaded = false;
 631         }
 632         spin_unlock_bh(&vxlan->hash_lock);
 633 }
 634 EXPORT_SYMBOL_GPL(vxlan_fdb_clear_offload);
 635
 636 /* Replace destination of unicast mac */
 637 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 638                              union vxlan_addr *ip, __be16 port, __be32 vni,
 639                              __u32 ifindex, struct vxlan_rdst *oldrd)
 640 {
 641         struct vxlan_rdst *rd;
 642
 643         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 644         if (rd)
 645                 return 0;
 646
 647         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 648         if (!rd)
 649                 return 0;
 650
 651         *oldrd = *rd;
 652         dst_cache_reset(&rd->dst_cache);
 653         rd->remote_ip = *ip;
 654         rd->remote_port = port;
 655         rd->remote_vni = vni;
 656         rd->remote_ifindex = ifindex;
 657         rd->offloaded = false;
 658         return 1;
 659 }
 660
 661 /* Add/update destinations for multicast */
 662 static int vxlan_fdb_append(struct vxlan_fdb *f,
 663                             union vxlan_addr *ip, __be16 port, __be32 vni,
 664                             __u32 ifindex, struct vxlan_rdst **rdp)
 665 {
 666         struct vxlan_rdst *rd;
 667
 668         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 669         if (rd)
 670                 return 0;
 671
 672         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 673         if (rd == NULL)
 674                 return -ENOBUFS;
 675
 676         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 677                 kfree(rd);
 678                 return -ENOBUFS;
 679         }
 680
 681         rd->remote_ip = *ip;
 682         rd->remote_port = port;
 683         rd->offloaded = false;
 684         rd->remote_vni = vni;
 685         rd->remote_ifindex = ifindex;
 686
 687         list_add_tail_rcu(&rd->list, &f->remotes);
 688
 689         *rdp = rd;
 690         return 1;
 691 }
 692
 693 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 694                                           unsigned int off,
 695                                           struct vxlanhdr *vh, size_t hdrlen,
 696                                           __be32 vni_field,
 697                                           struct gro_remcsum *grc,
 698                                           bool nopartial)
 699 {
 700         size_t start, offset;
 701
 702         if (skb->remcsum_offload)
 703                 return vh;
 704
 705         if (!NAPI_GRO_CB(skb)->csum_valid)
 706                 return NULL;
 707
 708         start = vxlan_rco_start(vni_field);
 709         offset = start + vxlan_rco_offset(vni_field);
 710
 711         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 712                                      start, offset, grc, nopartial);
 713
 714         skb->remcsum_offload = 1;
 715
 716         return vh;
 717 }
 718
 719 static struct sk_buff *vxlan_gro_receive(struct sock *sk,
 720                                          struct list_head *head,
 721                                          struct sk_buff *skb)
 722 {
 723         struct sk_buff *pp = NULL;
 724         struct sk_buff *p;
 725         struct vxlanhdr *vh, *vh2;
 726         unsigned int hlen, off_vx;
 727         int flush = 1;
 728         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 729         __be32 flags;
 730         struct gro_remcsum grc;
 731
 732         skb_gro_remcsum_init(&grc);
 733
 734         off_vx = skb_gro_offset(skb);
 735         hlen = off_vx + sizeof(*vh);
 736         vh   = skb_gro_header_fast(skb, off_vx);
 737         if (skb_gro_header_hard(skb, hlen)) {
 738                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 739                 if (unlikely(!vh))
 740                         goto out;
 741         }
 742
 743         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 744
 745         flags = vh->vx_flags;
 746
 747         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 748                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 749                                        vh->vx_vni, &grc,
 750                                        !!(vs->flags &
 751                                           VXLAN_F_REMCSUM_NOPARTIAL));
 752
 753                 if (!vh)
 754                         goto out;
 755         }
 756
 757         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 758
 759         list_for_each_entry(p, head, list) {
 760                 if (!NAPI_GRO_CB(p)->same_flow)
 761                         continue;
 762
 763                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 764                 if (vh->vx_flags != vh2->vx_flags ||
 765                     vh->vx_vni != vh2->vx_vni) {
 766                         NAPI_GRO_CB(p)->same_flow = 0;
 767                         continue;
 768                 }
 769         }
 770
 771         pp = call_gro_receive(eth_gro_receive, head, skb);
 772         flush = 0;
 773
 774 out:
 775         skb_gro_flush_final_remcsum(skb, pp, flush, &grc);
 776
 777         return pp;
 778 }
 779
 780 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 781 {
 782         /* Sets 'skb->inner_mac_header' since we are always called with
 783          * 'skb->encapsulation' set.
 784          */
 785         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 786 }
 787
 788 static struct vxlan_fdb *vxlan_fdb_alloc(struct vxlan_dev *vxlan,
 789                                          const u8 *mac, __u16 state,
 790                                          __be32 src_vni, __u16 ndm_flags)
 791 {
 792         struct vxlan_fdb *f;
 793
 794         f = kmalloc(sizeof(*f), GFP_ATOMIC);
 795         if (!f)
 796                 return NULL;
 797         f->state = state;
 798         f->flags = ndm_flags;
 799         f->updated = f->used = jiffies;
 800         f->vni = src_vni;
 801         INIT_LIST_HEAD(&f->remotes);
 802         memcpy(f->eth_addr, mac, ETH_ALEN);
 803
 804         return f;
 805 }
 806
 807 static void vxlan_fdb_insert(struct vxlan_dev *vxlan, const u8 *mac,
 808                              __be32 src_vni, struct vxlan_fdb *f)
 809 {
 810         ++vxlan->addrcnt;
 811         hlist_add_head_rcu(&f->hlist,
 812                            vxlan_fdb_head(vxlan, mac, src_vni));
 813 }
 814
 815 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 816                             const u8 *mac, union vxlan_addr *ip,
 817                             __u16 state, __be16 port, __be32 src_vni,
 818                             __be32 vni, __u32 ifindex, __u16 ndm_flags,
 819                             struct vxlan_fdb **fdb)
 820 {
 821         struct vxlan_rdst *rd = NULL;
 822         struct vxlan_fdb *f;
 823         int rc;
 824
 825         if (vxlan->cfg.addrmax &&
 826             vxlan->addrcnt >= vxlan->cfg.addrmax)
 827                 return -ENOSPC;
 828
 829         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 830         f = vxlan_fdb_alloc(vxlan, mac, state, src_vni, ndm_flags);
 831         if (!f)
 832                 return -ENOMEM;
 833
 834         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 835         if (rc < 0) {
 836                 kfree(f);
 837                 return rc;
 838         }
 839
 840         *fdb = f;
 841
 842         return 0;
 843 }
 844
 845 static void __vxlan_fdb_free(struct vxlan_fdb *f)
 846 {
 847         struct vxlan_rdst *rd, *nd;
 848
 849         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 850                 dst_cache_destroy(&rd->dst_cache);
 851                 kfree(rd);
 852         }
 853         kfree(f);
 854 }
 855
 856 static void vxlan_fdb_free(struct rcu_head *head)
 857 {
 858         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 859
 860         __vxlan_fdb_free(f);
 861 }
 862
 863 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 864                               bool do_notify, bool swdev_notify)
 865 {
 866         struct vxlan_rdst *rd;
 867
 868         netdev_dbg(vxlan->dev, "delete %pM\n", f->eth_addr);
 869
 870         --vxlan->addrcnt;
 871         if (do_notify)
 872                 list_for_each_entry(rd, &f->remotes, list)
 873                         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH,
 874                                          swdev_notify, NULL);
 875
 876         hlist_del_rcu(&f->hlist);
 877         call_rcu(&f->rcu, vxlan_fdb_free);
 878 }
 879
 880 static void vxlan_dst_free(struct rcu_head *head)
 881 {
 882         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 883
 884         dst_cache_destroy(&rd->dst_cache);
 885         kfree(rd);
 886 }
 887
 888 static int vxlan_fdb_update_existing(struct vxlan_dev *vxlan,
 889                                      union vxlan_addr *ip,
 890                                      __u16 state, __u16 flags,
 891                                      __be16 port, __be32 vni,
 892                                      __u32 ifindex, __u16 ndm_flags,
 893                                      struct vxlan_fdb *f,
 894                                      bool swdev_notify,
 895                                      struct netlink_ext_ack *extack)
 896 {
 897         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
 898         struct vxlan_rdst *rd = NULL;
 899         struct vxlan_rdst oldrd;
 900         int notify = 0;
 901         int rc = 0;
 902         int err;
 903
 904         /* Do not allow an externally learned entry to take over an entry added
 905          * by the user.
 906          */
 907         if (!(fdb_flags & NTF_EXT_LEARNED) ||
 908             !(f->flags & NTF_VXLAN_ADDED_BY_USER)) {
 909                 if (f->state != state) {
 910                         f->state = state;
 911                         f->updated = jiffies;
 912                         notify = 1;
 913                 }
 914                 if (f->flags != fdb_flags) {
 915                         f->flags = fdb_flags;
 916                         f->updated = jiffies;
 917                         notify = 1;
 918                 }
 919         }
 920
 921         if ((flags & NLM_F_REPLACE)) {
 922                 /* Only change unicasts */
 923                 if (!(is_multicast_ether_addr(f->eth_addr) ||
 924                       is_zero_ether_addr(f->eth_addr))) {
 925                         rc = vxlan_fdb_replace(f, ip, port, vni,
 926                                                ifindex, &oldrd);
 927                         notify |= rc;
 928                 } else {
 929                         return -EOPNOTSUPP;
 930                 }
 931         }
 932         if ((flags & NLM_F_APPEND) &&
 933             (is_multicast_ether_addr(f->eth_addr) ||
 934              is_zero_ether_addr(f->eth_addr))) {
 935                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 936
 937                 if (rc < 0)
 938                         return rc;
 939                 notify |= rc;
 940         }
 941
 942         if (ndm_flags & NTF_USE)
 943                 f->used = jiffies;
 944
 945         if (notify) {
 946                 if (rd == NULL)
 947                         rd = first_remote_rtnl(f);
 948
 949                 err = vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH,
 950                                        swdev_notify, extack);
 951                 if (err)
 952                         goto err_notify;
 953         }
 954
 955         return 0;
 956
 957 err_notify:
 958         if ((flags & NLM_F_REPLACE) && rc)
 959                 *rd = oldrd;
 960         else if ((flags & NLM_F_APPEND) && rc) {
 961                 list_del_rcu(&rd->list);
 962                 call_rcu(&rd->rcu, vxlan_dst_free);
 963         }
 964         return err;
 965 }
 966
 967 static int vxlan_fdb_update_create(struct vxlan_dev *vxlan,
 968                                    const u8 *mac, union vxlan_addr *ip,
 969                                    __u16 state, __u16 flags,
 970                                    __be16 port, __be32 src_vni, __be32 vni,
 971                                    __u32 ifindex, __u16 ndm_flags,
 972                                    bool swdev_notify,
 973                                    struct netlink_ext_ack *extack)
 974 {
 975         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
 976         struct vxlan_fdb *f;
 977         int rc;
 978
 979         /* Disallow replace to add a multicast entry */
 980         if ((flags & NLM_F_REPLACE) &&
 981             (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 982                 return -EOPNOTSUPP;
 983
 984         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 985         rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni,
 986                               vni, ifindex, fdb_flags, &f);
 987         if (rc < 0)
 988                 return rc;
 989
 990         vxlan_fdb_insert(vxlan, mac, src_vni, f);
 991         rc = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH,
 992                               swdev_notify, extack);
 993         if (rc)
 994                 goto err_notify;
 995
 996         return 0;
 997
 998 err_notify:
 999         vxlan_fdb_destroy(vxlan, f, false, false);
1000         return rc;
1001 }
1002
1003 /* Add new entry to forwarding table -- assumes lock held */
1004 static int vxlan_fdb_update(struct vxlan_dev *vxlan,
1005                             const u8 *mac, union vxlan_addr *ip,
1006                             __u16 state, __u16 flags,
1007                             __be16 port, __be32 src_vni, __be32 vni,
1008                             __u32 ifindex, __u16 ndm_flags,
1009                             bool swdev_notify,
1010                             struct netlink_ext_ack *extack)
1011 {
1012         struct vxlan_fdb *f;
1013
1014         f = __vxlan_find_mac(vxlan, mac, src_vni);
1015         if (f) {
1016                 if (flags & NLM_F_EXCL) {
1017                         netdev_dbg(vxlan->dev,
1018                                    "lost race to create %pM\n", mac);
1019                         return -EEXIST;
1020                 }
1021
1022                 return vxlan_fdb_update_existing(vxlan, ip, state, flags, port,
1023                                                  vni, ifindex, ndm_flags, f,
1024                                                  swdev_notify, extack);
1025         } else {
1026                 if (!(flags & NLM_F_CREATE))
1027                         return -ENOENT;
1028
1029                 return vxlan_fdb_update_create(vxlan, mac, ip, state, flags,
1030                                                port, src_vni, vni, ifindex,
1031                                                ndm_flags, swdev_notify, extack);
1032         }
1033 }
1034
1035 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
1036                                   struct vxlan_rdst *rd, bool swdev_notify)
1037 {
1038         list_del_rcu(&rd->list);
1039         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, swdev_notify, NULL);
1040         call_rcu(&rd->rcu, vxlan_dst_free);
1041 }
1042
1043 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
1044                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
1045                            __be32 *vni, u32 *ifindex)
1046 {
1047         struct net *net = dev_net(vxlan->dev);
1048         int err;
1049
1050         if (tb[NDA_DST]) {
1051                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
1052                 if (err)
1053                         return err;
1054         } else {
1055                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
1056                 if (remote->sa.sa_family == AF_INET) {
1057                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
1058                         ip->sa.sa_family = AF_INET;
1059 #if IS_ENABLED(CONFIG_IPV6)
1060                 } else {
1061                         ip->sin6.sin6_addr = in6addr_any;
1062                         ip->sa.sa_family = AF_INET6;
1063 #endif
1064                 }
1065         }
1066
1067         if (tb[NDA_PORT]) {
1068                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
1069                         return -EINVAL;
1070                 *port = nla_get_be16(tb[NDA_PORT]);
1071         } else {
1072                 *port = vxlan->cfg.dst_port;
1073         }
1074
1075         if (tb[NDA_VNI]) {
1076                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
1077                         return -EINVAL;
1078                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1079         } else {
1080                 *vni = vxlan->default_dst.remote_vni;
1081         }
1082
1083         if (tb[NDA_SRC_VNI]) {
1084                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
1085                         return -EINVAL;
1086                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
1087         } else {
1088                 *src_vni = vxlan->default_dst.remote_vni;
1089         }
1090
1091         if (tb[NDA_IFINDEX]) {
1092                 struct net_device *tdev;
1093
1094                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
1095                         return -EINVAL;
1096                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
1097                 tdev = __dev_get_by_index(net, *ifindex);
1098                 if (!tdev)
1099                         return -EADDRNOTAVAIL;
1100         } else {
1101                 *ifindex = 0;
1102         }
1103
1104         return 0;
1105 }
1106
1107 /* Add static entry (via netlink) */
1108 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
1109                          struct net_device *dev,
1110                          const unsigned char *addr, u16 vid, u16 flags,
1111                          struct netlink_ext_ack *extack)
1112 {
1113         struct vxlan_dev *vxlan = netdev_priv(dev);
1114         /* struct net *net = dev_net(vxlan->dev); */
1115         union vxlan_addr ip;
1116         __be16 port;
1117         __be32 src_vni, vni;
1118         u32 ifindex;
1119         int err;
1120
1121         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
1122                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
1123                         ndm->ndm_state);
1124                 return -EINVAL;
1125         }
1126
1127         if (tb[NDA_DST] == NULL)
1128                 return -EINVAL;
1129
1130         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
1131         if (err)
1132                 return err;
1133
1134         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
1135                 return -EAFNOSUPPORT;
1136
1137         spin_lock_bh(&vxlan->hash_lock);
1138         err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags,
1139                                port, src_vni, vni, ifindex,
1140                                ndm->ndm_flags | NTF_VXLAN_ADDED_BY_USER,
1141                                true, extack);
1142         spin_unlock_bh(&vxlan->hash_lock);
1143
1144         return err;
1145 }
1146
1147 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
1148                               const unsigned char *addr, union vxlan_addr ip,
1149                               __be16 port, __be32 src_vni, __be32 vni,
1150                               u32 ifindex, bool swdev_notify)
1151 {
1152         struct vxlan_fdb *f;
1153         struct vxlan_rdst *rd = NULL;
1154         int err = -ENOENT;
1155
1156         f = vxlan_find_mac(vxlan, addr, src_vni);
1157         if (!f)
1158                 return err;
1159
1160         if (!vxlan_addr_any(&ip)) {
1161                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
1162                 if (!rd)
1163                         goto out;
1164         }
1165
1166         /* remove a destination if it's not the only one on the list,
1167          * otherwise destroy the fdb entry
1168          */
1169         if (rd && !list_is_singular(&f->remotes)) {
1170                 vxlan_fdb_dst_destroy(vxlan, f, rd, swdev_notify);
1171                 goto out;
1172         }
1173
1174         vxlan_fdb_destroy(vxlan, f, true, swdev_notify);
1175
1176 out:
1177         return 0;
1178 }
1179
1180 /* Delete entry (via netlink) */
1181 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
1182                             struct net_device *dev,
1183                             const unsigned char *addr, u16 vid)
1184 {
1185         struct vxlan_dev *vxlan = netdev_priv(dev);
1186         union vxlan_addr ip;
1187         __be32 src_vni, vni;
1188         __be16 port;
1189         u32 ifindex;
1190         int err;
1191
1192         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
1193         if (err)
1194                 return err;
1195
1196         spin_lock_bh(&vxlan->hash_lock);
1197         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
1198                                  true);
1199         spin_unlock_bh(&vxlan->hash_lock);
1200
1201         return err;
1202 }
1203
1204 /* Dump forwarding table */
1205 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1206                           struct net_device *dev,
1207                           struct net_device *filter_dev, int *idx)
1208 {
1209         struct vxlan_dev *vxlan = netdev_priv(dev);
1210         unsigned int h;
1211         int err = 0;
1212
1213         for (h = 0; h < FDB_HASH_SIZE; ++h) {
1214                 struct vxlan_fdb *f;
1215
1216                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
1217                         struct vxlan_rdst *rd;
1218
1219                         list_for_each_entry_rcu(rd, &f->remotes, list) {
1220                                 if (*idx < cb->args[2])
1221                                         goto skip;
1222
1223                                 err = vxlan_fdb_info(skb, vxlan, f,
1224                                                      NETLINK_CB(cb->skb).portid,
1225                                                      cb->nlh->nlmsg_seq,
1226                                                      RTM_NEWNEIGH,
1227                                                      NLM_F_MULTI, rd);
1228                                 if (err < 0)
1229                                         goto out;
1230 skip:
1231                                 *idx += 1;
1232                         }
1233                 }
1234         }
1235 out:
1236         return err;
1237 }
1238
1239 static int vxlan_fdb_get(struct sk_buff *skb,
1240                          struct nlattr *tb[],
1241                          struct net_device *dev,
1242                          const unsigned char *addr,
1243                          u16 vid, u32 portid, u32 seq,
1244                          struct netlink_ext_ack *extack)
1245 {
1246         struct vxlan_dev *vxlan = netdev_priv(dev);
1247         struct vxlan_fdb *f;
1248         __be32 vni;
1249         int err;
1250
1251         if (tb[NDA_VNI])
1252                 vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1253         else
1254                 vni = vxlan->default_dst.remote_vni;
1255
1256         rcu_read_lock();
1257
1258         f = __vxlan_find_mac(vxlan, addr, vni);
1259         if (!f) {
1260                 NL_SET_ERR_MSG(extack, "Fdb entry not found");
1261                 err = -ENOENT;
1262                 goto errout;
1263         }
1264
1265         err = vxlan_fdb_info(skb, vxlan, f, portid, seq,
1266                              RTM_NEWNEIGH, 0, first_remote_rcu(f));
1267 errout:
1268         rcu_read_unlock();
1269         return err;
1270 }
1271
1272 /* Watch incoming packets to learn mapping between Ethernet address
1273  * and Tunnel endpoint.
1274  * Return true if packet is bogus and should be dropped.
1275  */
1276 static bool vxlan_snoop(struct net_device *dev,
1277                         union vxlan_addr *src_ip, const u8 *src_mac,
1278                         u32 src_ifindex, __be32 vni)
1279 {
1280         struct vxlan_dev *vxlan = netdev_priv(dev);
1281         struct vxlan_fdb *f;
1282         u32 ifindex = 0;
1283
1284 #if IS_ENABLED(CONFIG_IPV6)
1285         if (src_ip->sa.sa_family == AF_INET6 &&
1286             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
1287                 ifindex = src_ifindex;
1288 #endif
1289
1290         f = vxlan_find_mac(vxlan, src_mac, vni);
1291         if (likely(f)) {
1292                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1293
1294                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
1295                            rdst->remote_ifindex == ifindex))
1296                         return false;
1297
1298                 /* Don't migrate static entries, drop packets */
1299                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
1300                         return true;
1301
1302                 if (net_ratelimit())
1303                         netdev_info(dev,
1304                                     "%pM migrated from %pIS to %pIS\n",
1305                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1306
1307                 rdst->remote_ip = *src_ip;
1308                 f->updated = jiffies;
1309                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH, true, NULL);
1310         } else {
1311                 /* learned new entry */
1312                 spin_lock(&vxlan->hash_lock);
1313
1314                 /* close off race between vxlan_flush and incoming packets */
1315                 if (netif_running(dev))
1316                         vxlan_fdb_update(vxlan, src_mac, src_ip,
1317                                          NUD_REACHABLE,
1318                                          NLM_F_EXCL|NLM_F_CREATE,
1319                                          vxlan->cfg.dst_port,
1320                                          vni,
1321                                          vxlan->default_dst.remote_vni,
1322                                          ifindex, NTF_SELF, true, NULL);
1323                 spin_unlock(&vxlan->hash_lock);
1324         }
1325
1326         return false;
1327 }
1328
1329 /* See if multicast group is already in use by other ID */
1330 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1331 {
1332         struct vxlan_dev *vxlan;
1333         struct vxlan_sock *sock4;
1334 #if IS_ENABLED(CONFIG_IPV6)
1335         struct vxlan_sock *sock6;
1336 #endif
1337         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1338
1339         sock4 = rtnl_dereference(dev->vn4_sock);
1340
1341         /* The vxlan_sock is only used by dev, leaving group has
1342          * no effect on other vxlan devices.
1343          */
1344         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1345                 return false;
1346 #if IS_ENABLED(CONFIG_IPV6)
1347         sock6 = rtnl_dereference(dev->vn6_sock);
1348         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1349                 return false;
1350 #endif
1351
1352         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1353                 if (!netif_running(vxlan->dev) || vxlan == dev)
1354                         continue;
1355
1356                 if (family == AF_INET &&
1357                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1358                         continue;
1359 #if IS_ENABLED(CONFIG_IPV6)
1360                 if (family == AF_INET6 &&
1361                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1362                         continue;
1363 #endif
1364
1365                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1366                                       &dev->default_dst.remote_ip))
1367                         continue;
1368
1369                 if (vxlan->default_dst.remote_ifindex !=
1370                     dev->default_dst.remote_ifindex)
1371                         continue;
1372
1373                 return true;
1374         }
1375
1376         return false;
1377 }
1378
1379 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1380 {
1381         struct vxlan_net *vn;
1382
1383         if (!vs)
1384                 return false;
1385         if (!refcount_dec_and_test(&vs->refcnt))
1386                 return false;
1387
1388         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1389         spin_lock(&vn->sock_lock);
1390         hlist_del_rcu(&vs->hlist);
1391         udp_tunnel_notify_del_rx_port(vs->sock,
1392                                       (vs->flags & VXLAN_F_GPE) ?
1393                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1394                                       UDP_TUNNEL_TYPE_VXLAN);
1395         spin_unlock(&vn->sock_lock);
1396
1397         return true;
1398 }
1399
1400 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1401 {
1402         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1403 #if IS_ENABLED(CONFIG_IPV6)
1404         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1405
1406         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1407 #endif
1408
1409         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1410         synchronize_net();
1411
1412         vxlan_vs_del_dev(vxlan);
1413
1414         if (__vxlan_sock_release_prep(sock4)) {
1415                 udp_tunnel_sock_release(sock4->sock);
1416                 kfree(sock4);
1417         }
1418
1419 #if IS_ENABLED(CONFIG_IPV6)
1420         if (__vxlan_sock_release_prep(sock6)) {
1421                 udp_tunnel_sock_release(sock6->sock);
1422                 kfree(sock6);
1423         }
1424 #endif
1425 }
1426
1427 /* Update multicast group membership when first VNI on
1428  * multicast address is brought up
1429  */
1430 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1431 {
1432         struct sock *sk;
1433         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1434         int ifindex = vxlan->default_dst.remote_ifindex;
1435         int ret = -EINVAL;
1436
1437         if (ip->sa.sa_family == AF_INET) {
1438                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1439                 struct ip_mreqn mreq = {
1440                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1441                         .imr_ifindex            = ifindex,
1442                 };
1443
1444                 sk = sock4->sock->sk;
1445                 lock_sock(sk);
1446                 ret = ip_mc_join_group(sk, &mreq);
1447                 release_sock(sk);
1448 #if IS_ENABLED(CONFIG_IPV6)
1449         } else {
1450                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1451
1452                 sk = sock6->sock->sk;
1453                 lock_sock(sk);
1454                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1455                                                    &ip->sin6.sin6_addr);
1456                 release_sock(sk);
1457 #endif
1458         }
1459
1460         return ret;
1461 }
1462
1463 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1464 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1465 {
1466         struct sock *sk;
1467         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1468         int ifindex = vxlan->default_dst.remote_ifindex;
1469         int ret = -EINVAL;
1470
1471         if (ip->sa.sa_family == AF_INET) {
1472                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1473                 struct ip_mreqn mreq = {
1474                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1475                         .imr_ifindex            = ifindex,
1476                 };
1477
1478                 sk = sock4->sock->sk;
1479                 lock_sock(sk);
1480                 ret = ip_mc_leave_group(sk, &mreq);
1481                 release_sock(sk);
1482 #if IS_ENABLED(CONFIG_IPV6)
1483         } else {
1484                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1485
1486                 sk = sock6->sock->sk;
1487                 lock_sock(sk);
1488                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1489                                                    &ip->sin6.sin6_addr);
1490                 release_sock(sk);
1491 #endif
1492         }
1493
1494         return ret;
1495 }
1496
1497 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1498                           struct sk_buff *skb, u32 vxflags)
1499 {
1500         size_t start, offset;
1501
1502         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1503                 goto out;
1504
1505         start = vxlan_rco_start(unparsed->vx_vni);
1506         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1507
1508         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1509                 return false;
1510
1511         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1512                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1513 out:
1514         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1515         unparsed->vx_vni &= VXLAN_VNI_MASK;
1516         return true;
1517 }
1518
1519 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1520                                 struct sk_buff *skb, u32 vxflags,
1521                                 struct vxlan_metadata *md)
1522 {
1523         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1524         struct metadata_dst *tun_dst;
1525
1526         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1527                 goto out;
1528
1529         md->gbp = ntohs(gbp->policy_id);
1530
1531         tun_dst = (struct metadata_dst *)skb_dst(skb);
1532         if (tun_dst) {
1533                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1534                 tun_dst->u.tun_info.options_len = sizeof(*md);
1535         }
1536         if (gbp->dont_learn)
1537                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1538
1539         if (gbp->policy_applied)
1540                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1541
1542         /* In flow-based mode, GBP is carried in dst_metadata */
1543         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1544                 skb->mark = md->gbp;
1545 out:
1546         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1547 }
1548
1549 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1550                                 __be16 *protocol,
1551                                 struct sk_buff *skb, u32 vxflags)
1552 {
1553         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1554
1555         /* Need to have Next Protocol set for interfaces in GPE mode. */
1556         if (!gpe->np_applied)
1557                 return false;
1558         /* "The initial version is 0. If a receiver does not support the
1559          * version indicated it MUST drop the packet.
1560          */
1561         if (gpe->version != 0)
1562                 return false;
1563         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1564          * processing MUST occur." However, we don't implement OAM
1565          * processing, thus drop the packet.
1566          */
1567         if (gpe->oam_flag)
1568                 return false;
1569
1570         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1571         if (!*protocol)
1572                 return false;
1573
1574         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1575         return true;
1576 }
1577
1578 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1579                           struct vxlan_sock *vs,
1580                           struct sk_buff *skb, __be32 vni)
1581 {
1582         union vxlan_addr saddr;
1583         u32 ifindex = skb->dev->ifindex;
1584
1585         skb_reset_mac_header(skb);
1586         skb->protocol = eth_type_trans(skb, vxlan->dev);
1587         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1588
1589         /* Ignore packet loops (and multicast echo) */
1590         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1591                 return false;
1592
1593         /* Get address from the outer IP header */
1594         if (vxlan_get_sk_family(vs) == AF_INET) {
1595                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1596                 saddr.sa.sa_family = AF_INET;
1597 #if IS_ENABLED(CONFIG_IPV6)
1598         } else {
1599                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1600                 saddr.sa.sa_family = AF_INET6;
1601 #endif
1602         }
1603
1604         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1605             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1606                 return false;
1607
1608         return true;
1609 }
1610
1611 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1612                                   struct sk_buff *skb)
1613 {
1614         int err = 0;
1615
1616         if (vxlan_get_sk_family(vs) == AF_INET)
1617                 err = IP_ECN_decapsulate(oiph, skb);
1618 #if IS_ENABLED(CONFIG_IPV6)
1619         else
1620                 err = IP6_ECN_decapsulate(oiph, skb);
1621 #endif
1622
1623         if (unlikely(err) && log_ecn_error) {
1624                 if (vxlan_get_sk_family(vs) == AF_INET)
1625                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1626                                              &((struct iphdr *)oiph)->saddr,
1627                                              ((struct iphdr *)oiph)->tos);
1628                 else
1629                         net_info_ratelimited("non-ECT from %pI6\n",
1630                                              &((struct ipv6hdr *)oiph)->saddr);
1631         }
1632         return err <= 1;
1633 }
1634
1635 /* Callback from net/ipv4/udp.c to receive packets */
1636 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1637 {
1638         struct pcpu_sw_netstats *stats;
1639         struct vxlan_dev *vxlan;
1640         struct vxlan_sock *vs;
1641         struct vxlanhdr unparsed;
1642         struct vxlan_metadata _md;
1643         struct vxlan_metadata *md = &_md;
1644         __be16 protocol = htons(ETH_P_TEB);
1645         bool raw_proto = false;
1646         void *oiph;
1647         __be32 vni = 0;
1648
1649         /* Need UDP and VXLAN header to be present */
1650         if (!pskb_may_pull(skb, VXLAN_HLEN))
1651                 goto drop;
1652
1653         unparsed = *vxlan_hdr(skb);
1654         /* VNI flag always required to be set */
1655         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1656                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1657                            ntohl(vxlan_hdr(skb)->vx_flags),
1658                            ntohl(vxlan_hdr(skb)->vx_vni));
1659                 /* Return non vxlan pkt */
1660                 goto drop;
1661         }
1662         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1663         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1664
1665         vs = rcu_dereference_sk_user_data(sk);
1666         if (!vs)
1667                 goto drop;
1668
1669         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1670
1671         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1672         if (!vxlan)
1673                 goto drop;
1674
1675         /* For backwards compatibility, only allow reserved fields to be
1676          * used by VXLAN extensions if explicitly requested.
1677          */
1678         if (vs->flags & VXLAN_F_GPE) {
1679                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1680                         goto drop;
1681                 raw_proto = true;
1682         }
1683
1684         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1685                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1686                         goto drop;
1687
1688         if (vxlan_collect_metadata(vs)) {
1689                 struct metadata_dst *tun_dst;
1690
1691                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1692                                          key32_to_tunnel_id(vni), sizeof(*md));
1693
1694                 if (!tun_dst)
1695                         goto drop;
1696
1697                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1698
1699                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1700         } else {
1701                 memset(md, 0, sizeof(*md));
1702         }
1703
1704         if (vs->flags & VXLAN_F_REMCSUM_RX)
1705                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1706                         goto drop;
1707         if (vs->flags & VXLAN_F_GBP)
1708                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1709         /* Note that GBP and GPE can never be active together. This is
1710          * ensured in vxlan_dev_configure.
1711          */
1712
1713         if (unparsed.vx_flags || unparsed.vx_vni) {
1714                 /* If there are any unprocessed flags remaining treat
1715                  * this as a malformed packet. This behavior diverges from
1716                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1717                  * in reserved fields are to be ignored. The approach here
1718                  * maintains compatibility with previous stack code, and also
1719                  * is more robust and provides a little more security in
1720                  * adding extensions to VXLAN.
1721                  */
1722                 goto drop;
1723         }
1724
1725         if (!raw_proto) {
1726                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1727                         goto drop;
1728         } else {
1729                 skb_reset_mac_header(skb);
1730                 skb->dev = vxlan->dev;
1731                 skb->pkt_type = PACKET_HOST;
1732         }
1733
1734         oiph = skb_network_header(skb);
1735         skb_reset_network_header(skb);
1736
1737         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1738                 ++vxlan->dev->stats.rx_frame_errors;
1739                 ++vxlan->dev->stats.rx_errors;
1740                 goto drop;
1741         }
1742
1743         rcu_read_lock();
1744
1745         if (unlikely(!(vxlan->dev->flags & IFF_UP))) {
1746                 rcu_read_unlock();
1747                 atomic_long_inc(&vxlan->dev->rx_dropped);
1748                 goto drop;
1749         }
1750
1751         stats = this_cpu_ptr(vxlan->dev->tstats);
1752         u64_stats_update_begin(&stats->syncp);
1753         stats->rx_packets++;
1754         stats->rx_bytes += skb->len;
1755         u64_stats_update_end(&stats->syncp);
1756
1757         gro_cells_receive(&vxlan->gro_cells, skb);
1758
1759         rcu_read_unlock();
1760
1761         return 0;
1762
1763 drop:
1764         /* Consume bad packet */
1765         kfree_skb(skb);
1766         return 0;
1767 }
1768
1769 /* Callback from net/ipv{4,6}/udp.c to check that we have a VNI for errors */
1770 static int vxlan_err_lookup(struct sock *sk, struct sk_buff *skb)
1771 {
1772         struct vxlan_dev *vxlan;
1773         struct vxlan_sock *vs;
1774         struct vxlanhdr *hdr;
1775         __be32 vni;
1776
1777         if (!pskb_may_pull(skb, skb_transport_offset(skb) + VXLAN_HLEN))
1778                 return -EINVAL;
1779
1780         hdr = vxlan_hdr(skb);
1781
1782         if (!(hdr->vx_flags & VXLAN_HF_VNI))
1783                 return -EINVAL;
1784
1785         vs = rcu_dereference_sk_user_data(sk);
1786         if (!vs)
1787                 return -ENOENT;
1788
1789         vni = vxlan_vni(hdr->vx_vni);
1790         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1791         if (!vxlan)
1792                 return -ENOENT;
1793
1794         return 0;
1795 }
1796
1797 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1798 {
1799         struct vxlan_dev *vxlan = netdev_priv(dev);
1800         struct arphdr *parp;
1801         u8 *arpptr, *sha;
1802         __be32 sip, tip;
1803         struct neighbour *n;
1804
1805         if (dev->flags & IFF_NOARP)
1806                 goto out;
1807
1808         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1809                 dev->stats.tx_dropped++;
1810                 goto out;
1811         }
1812         parp = arp_hdr(skb);
1813
1814         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1815              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1816             parp->ar_pro != htons(ETH_P_IP) ||
1817             parp->ar_op != htons(ARPOP_REQUEST) ||
1818             parp->ar_hln != dev->addr_len ||
1819             parp->ar_pln != 4)
1820                 goto out;
1821         arpptr = (u8 *)parp + sizeof(struct arphdr);
1822         sha = arpptr;
1823         arpptr += dev->addr_len;        /* sha */
1824         memcpy(&sip, arpptr, sizeof(sip));
1825         arpptr += sizeof(sip);
1826         arpptr += dev->addr_len;        /* tha */
1827         memcpy(&tip, arpptr, sizeof(tip));
1828
1829         if (ipv4_is_loopback(tip) ||
1830             ipv4_is_multicast(tip))
1831                 goto out;
1832
1833         n = neigh_lookup(&arp_tbl, &tip, dev);
1834
1835         if (n) {
1836                 struct vxlan_fdb *f;
1837                 struct sk_buff  *reply;
1838
1839                 if (!(n->nud_state & NUD_CONNECTED)) {
1840                         neigh_release(n);
1841                         goto out;
1842                 }
1843
1844                 f = vxlan_find_mac(vxlan, n->ha, vni);
1845                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1846                         /* bridge-local neighbor */
1847                         neigh_release(n);
1848                         goto out;
1849                 }
1850
1851                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1852                                 n->ha, sha);
1853
1854                 neigh_release(n);
1855
1856                 if (reply == NULL)
1857                         goto out;
1858
1859                 skb_reset_mac_header(reply);
1860                 __skb_pull(reply, skb_network_offset(reply));
1861                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1862                 reply->pkt_type = PACKET_HOST;
1863
1864                 if (netif_rx_ni(reply) == NET_RX_DROP)
1865                         dev->stats.rx_dropped++;
1866         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1867                 union vxlan_addr ipa = {
1868                         .sin.sin_addr.s_addr = tip,
1869                         .sin.sin_family = AF_INET,
1870                 };
1871
1872                 vxlan_ip_miss(dev, &ipa);
1873         }
1874 out:
1875         consume_skb(skb);
1876         return NETDEV_TX_OK;
1877 }
1878
1879 #if IS_ENABLED(CONFIG_IPV6)
1880 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1881         struct neighbour *n, bool isrouter)
1882 {
1883         struct net_device *dev = request->dev;
1884         struct sk_buff *reply;
1885         struct nd_msg *ns, *na;
1886         struct ipv6hdr *pip6;
1887         u8 *daddr;
1888         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1889         int ns_olen;
1890         int i, len;
1891
1892         if (dev == NULL || !pskb_may_pull(request, request->len))
1893                 return NULL;
1894
1895         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1896                 sizeof(*na) + na_olen + dev->needed_tailroom;
1897         reply = alloc_skb(len, GFP_ATOMIC);
1898         if (reply == NULL)
1899                 return NULL;
1900
1901         reply->protocol = htons(ETH_P_IPV6);
1902         reply->dev = dev;
1903         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1904         skb_push(reply, sizeof(struct ethhdr));
1905         skb_reset_mac_header(reply);
1906
1907         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
1908
1909         daddr = eth_hdr(request)->h_source;
1910         ns_olen = request->len - skb_network_offset(request) -
1911                 sizeof(struct ipv6hdr) - sizeof(*ns);
1912         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1913                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1914                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1915                         break;
1916                 }
1917         }
1918
1919         /* Ethernet header */
1920         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1921         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1922         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1923         reply->protocol = htons(ETH_P_IPV6);
1924
1925         skb_pull(reply, sizeof(struct ethhdr));
1926         skb_reset_network_header(reply);
1927         skb_put(reply, sizeof(struct ipv6hdr));
1928
1929         /* IPv6 header */
1930
1931         pip6 = ipv6_hdr(reply);
1932         memset(pip6, 0, sizeof(struct ipv6hdr));
1933         pip6->version = 6;
1934         pip6->priority = ipv6_hdr(request)->priority;
1935         pip6->nexthdr = IPPROTO_ICMPV6;
1936         pip6->hop_limit = 255;
1937         pip6->daddr = ipv6_hdr(request)->saddr;
1938         pip6->saddr = *(struct in6_addr *)n->primary_key;
1939
1940         skb_pull(reply, sizeof(struct ipv6hdr));
1941         skb_reset_transport_header(reply);
1942
1943         /* Neighbor Advertisement */
1944         na = skb_put_zero(reply, sizeof(*na) + na_olen);
1945         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1946         na->icmph.icmp6_router = isrouter;
1947         na->icmph.icmp6_override = 1;
1948         na->icmph.icmp6_solicited = 1;
1949         na->target = ns->target;
1950         ether_addr_copy(&na->opt[2], n->ha);
1951         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1952         na->opt[1] = na_olen >> 3;
1953
1954         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1955                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1956                 csum_partial(na, sizeof(*na)+na_olen, 0));
1957
1958         pip6->payload_len = htons(sizeof(*na)+na_olen);
1959
1960         skb_push(reply, sizeof(struct ipv6hdr));
1961
1962         reply->ip_summed = CHECKSUM_UNNECESSARY;
1963
1964         return reply;
1965 }
1966
1967 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1968 {
1969         struct vxlan_dev *vxlan = netdev_priv(dev);
1970         const struct in6_addr *daddr;
1971         const struct ipv6hdr *iphdr;
1972         struct inet6_dev *in6_dev;
1973         struct neighbour *n;
1974         struct nd_msg *msg;
1975
1976         in6_dev = __in6_dev_get(dev);
1977         if (!in6_dev)
1978                 goto out;
1979
1980         iphdr = ipv6_hdr(skb);
1981         daddr = &iphdr->daddr;
1982         msg = (struct nd_msg *)(iphdr + 1);
1983
1984         if (ipv6_addr_loopback(daddr) ||
1985             ipv6_addr_is_multicast(&msg->target))
1986                 goto out;
1987
1988         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1989
1990         if (n) {
1991                 struct vxlan_fdb *f;
1992                 struct sk_buff *reply;
1993
1994                 if (!(n->nud_state & NUD_CONNECTED)) {
1995                         neigh_release(n);
1996                         goto out;
1997                 }
1998
1999                 f = vxlan_find_mac(vxlan, n->ha, vni);
2000                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2001                         /* bridge-local neighbor */
2002                         neigh_release(n);
2003                         goto out;
2004                 }
2005
2006                 reply = vxlan_na_create(skb, n,
2007                                         !!(f ? f->flags & NTF_ROUTER : 0));
2008
2009                 neigh_release(n);
2010
2011                 if (reply == NULL)
2012                         goto out;
2013
2014                 if (netif_rx_ni(reply) == NET_RX_DROP)
2015                         dev->stats.rx_dropped++;
2016
2017         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2018                 union vxlan_addr ipa = {
2019                         .sin6.sin6_addr = msg->target,
2020                         .sin6.sin6_family = AF_INET6,
2021                 };
2022
2023                 vxlan_ip_miss(dev, &ipa);
2024         }
2025
2026 out:
2027         consume_skb(skb);
2028         return NETDEV_TX_OK;
2029 }
2030 #endif
2031
2032 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
2033 {
2034         struct vxlan_dev *vxlan = netdev_priv(dev);
2035         struct neighbour *n;
2036
2037         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
2038                 return false;
2039
2040         n = NULL;
2041         switch (ntohs(eth_hdr(skb)->h_proto)) {
2042         case ETH_P_IP:
2043         {
2044                 struct iphdr *pip;
2045
2046                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
2047                         return false;
2048                 pip = ip_hdr(skb);
2049                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
2050                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2051                         union vxlan_addr ipa = {
2052                                 .sin.sin_addr.s_addr = pip->daddr,
2053                                 .sin.sin_family = AF_INET,
2054                         };
2055
2056                         vxlan_ip_miss(dev, &ipa);
2057                         return false;
2058                 }
2059
2060                 break;
2061         }
2062 #if IS_ENABLED(CONFIG_IPV6)
2063         case ETH_P_IPV6:
2064         {
2065                 struct ipv6hdr *pip6;
2066
2067                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
2068                         return false;
2069                 pip6 = ipv6_hdr(skb);
2070                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
2071                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2072                         union vxlan_addr ipa = {
2073                                 .sin6.sin6_addr = pip6->daddr,
2074                                 .sin6.sin6_family = AF_INET6,
2075                         };
2076
2077                         vxlan_ip_miss(dev, &ipa);
2078                         return false;
2079                 }
2080
2081                 break;
2082         }
2083 #endif
2084         default:
2085                 return false;
2086         }
2087
2088         if (n) {
2089                 bool diff;
2090
2091                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
2092                 if (diff) {
2093                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
2094                                 dev->addr_len);
2095                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
2096                 }
2097                 neigh_release(n);
2098                 return diff;
2099         }
2100
2101         return false;
2102 }
2103
2104 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
2105                                 struct vxlan_metadata *md)
2106 {
2107         struct vxlanhdr_gbp *gbp;
2108
2109         if (!md->gbp)
2110                 return;
2111
2112         gbp = (struct vxlanhdr_gbp *)vxh;
2113         vxh->vx_flags |= VXLAN_HF_GBP;
2114
2115         if (md->gbp & VXLAN_GBP_DONT_LEARN)
2116                 gbp->dont_learn = 1;
2117
2118         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
2119                 gbp->policy_applied = 1;
2120
2121         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
2122 }
2123
2124 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
2125                                __be16 protocol)
2126 {
2127         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
2128
2129         gpe->np_applied = 1;
2130         gpe->next_protocol = tun_p_from_eth_p(protocol);
2131         if (!gpe->next_protocol)
2132                 return -EPFNOSUPPORT;
2133         return 0;
2134 }
2135
2136 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
2137                            int iphdr_len, __be32 vni,
2138                            struct vxlan_metadata *md, u32 vxflags,
2139                            bool udp_sum)
2140 {
2141         struct vxlanhdr *vxh;
2142         int min_headroom;
2143         int err;
2144         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
2145         __be16 inner_protocol = htons(ETH_P_TEB);
2146
2147         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
2148             skb->ip_summed == CHECKSUM_PARTIAL) {
2149                 int csum_start = skb_checksum_start_offset(skb);
2150
2151                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
2152                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
2153                     (skb->csum_offset == offsetof(struct udphdr, check) ||
2154                      skb->csum_offset == offsetof(struct tcphdr, check)))
2155                         type |= SKB_GSO_TUNNEL_REMCSUM;
2156         }
2157
2158         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
2159                         + VXLAN_HLEN + iphdr_len;
2160
2161         /* Need space for new headers (invalidates iph ptr) */
2162         err = skb_cow_head(skb, min_headroom);
2163         if (unlikely(err))
2164                 return err;
2165
2166         err = iptunnel_handle_offloads(skb, type);
2167         if (err)
2168                 return err;
2169
2170         vxh = __skb_push(skb, sizeof(*vxh));
2171         vxh->vx_flags = VXLAN_HF_VNI;
2172         vxh->vx_vni = vxlan_vni_field(vni);
2173
2174         if (type & SKB_GSO_TUNNEL_REMCSUM) {
2175                 unsigned int start;
2176
2177                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
2178                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
2179                 vxh->vx_flags |= VXLAN_HF_RCO;
2180
2181                 if (!skb_is_gso(skb)) {
2182                         skb->ip_summed = CHECKSUM_NONE;
2183                         skb->encapsulation = 0;
2184                 }
2185         }
2186
2187         if (vxflags & VXLAN_F_GBP)
2188                 vxlan_build_gbp_hdr(vxh, vxflags, md);
2189         if (vxflags & VXLAN_F_GPE) {
2190                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
2191                 if (err < 0)
2192                         return err;
2193                 inner_protocol = skb->protocol;
2194         }
2195
2196         skb_set_inner_protocol(skb, inner_protocol);
2197         return 0;
2198 }
2199
2200 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
2201                                       struct vxlan_sock *sock4,
2202                                       struct sk_buff *skb, int oif, u8 tos,
2203                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
2204                                       struct dst_cache *dst_cache,
2205                                       const struct ip_tunnel_info *info)
2206 {
2207         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2208         struct rtable *rt = NULL;
2209         struct flowi4 fl4;
2210
2211         if (!sock4)
2212                 return ERR_PTR(-EIO);
2213
2214         if (tos && !info)
2215                 use_cache = false;
2216         if (use_cache) {
2217                 rt = dst_cache_get_ip4(dst_cache, saddr);
2218                 if (rt)
2219                         return rt;
2220         }
2221
2222         memset(&fl4, 0, sizeof(fl4));
2223         fl4.flowi4_oif = oif;
2224         fl4.flowi4_tos = RT_TOS(tos);
2225         fl4.flowi4_mark = skb->mark;
2226         fl4.flowi4_proto = IPPROTO_UDP;
2227         fl4.daddr = daddr;
2228         fl4.saddr = *saddr;
2229         fl4.fl4_dport = dport;
2230         fl4.fl4_sport = sport;
2231
2232         rt = ip_route_output_key(vxlan->net, &fl4);
2233         if (likely(!IS_ERR(rt))) {
2234                 if (rt->dst.dev == dev) {
2235                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
2236                         ip_rt_put(rt);
2237                         return ERR_PTR(-ELOOP);
2238                 }
2239
2240                 *saddr = fl4.saddr;
2241                 if (use_cache)
2242                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2243         } else {
2244                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
2245                 return ERR_PTR(-ENETUNREACH);
2246         }
2247         return rt;
2248 }
2249
2250 #if IS_ENABLED(CONFIG_IPV6)
2251 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
2252                                           struct net_device *dev,
2253                                           struct vxlan_sock *sock6,
2254                                           struct sk_buff *skb, int oif, u8 tos,
2255                                           __be32 label,
2256                                           const struct in6_addr *daddr,
2257                                           struct in6_addr *saddr,
2258                                           __be16 dport, __be16 sport,
2259                                           struct dst_cache *dst_cache,
2260                                           const struct ip_tunnel_info *info)
2261 {
2262         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2263         struct dst_entry *ndst;
2264         struct flowi6 fl6;
2265         int err;
2266
2267         if (!sock6)
2268                 return ERR_PTR(-EIO);
2269
2270         if (tos && !info)
2271                 use_cache = false;
2272         if (use_cache) {
2273                 ndst = dst_cache_get_ip6(dst_cache, saddr);
2274                 if (ndst)
2275                         return ndst;
2276         }
2277
2278         memset(&fl6, 0, sizeof(fl6));
2279         fl6.flowi6_oif = oif;
2280         fl6.daddr = *daddr;
2281         fl6.saddr = *saddr;
2282         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
2283         fl6.flowi6_mark = skb->mark;
2284         fl6.flowi6_proto = IPPROTO_UDP;
2285         fl6.fl6_dport = dport;
2286         fl6.fl6_sport = sport;
2287
2288         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
2289                                          sock6->sock->sk,
2290                                          &ndst, &fl6);
2291         if (unlikely(err < 0)) {
2292                 netdev_dbg(dev, "no route to %pI6\n", daddr);
2293                 return ERR_PTR(-ENETUNREACH);
2294         }
2295
2296         if (unlikely(ndst->dev == dev)) {
2297                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
2298                 dst_release(ndst);
2299                 return ERR_PTR(-ELOOP);
2300         }
2301
2302         *saddr = fl6.saddr;
2303         if (use_cache)
2304                 dst_cache_set_ip6(dst_cache, ndst, saddr);
2305         return ndst;
2306 }
2307 #endif
2308
2309 /* Bypass encapsulation if the destination is local */
2310 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2311                                struct vxlan_dev *dst_vxlan, __be32 vni)
2312 {
2313         struct pcpu_sw_netstats *tx_stats, *rx_stats;
2314         union vxlan_addr loopback;
2315         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2316         struct net_device *dev;
2317         int len = skb->len;
2318
2319         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2320         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2321         skb->pkt_type = PACKET_HOST;
2322         skb->encapsulation = 0;
2323         skb->dev = dst_vxlan->dev;
2324         __skb_pull(skb, skb_network_offset(skb));
2325
2326         if (remote_ip->sa.sa_family == AF_INET) {
2327                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2328                 loopback.sa.sa_family =  AF_INET;
2329 #if IS_ENABLED(CONFIG_IPV6)
2330         } else {
2331                 loopback.sin6.sin6_addr = in6addr_loopback;
2332                 loopback.sa.sa_family =  AF_INET6;
2333 #endif
2334         }
2335
2336         rcu_read_lock();
2337         dev = skb->dev;
2338         if (unlikely(!(dev->flags & IFF_UP))) {
2339                 kfree_skb(skb);
2340                 goto drop;
2341         }
2342
2343         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
2344                 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni);
2345
2346         u64_stats_update_begin(&tx_stats->syncp);
2347         tx_stats->tx_packets++;
2348         tx_stats->tx_bytes += len;
2349         u64_stats_update_end(&tx_stats->syncp);
2350
2351         if (netif_rx(skb) == NET_RX_SUCCESS) {
2352                 u64_stats_update_begin(&rx_stats->syncp);
2353                 rx_stats->rx_packets++;
2354                 rx_stats->rx_bytes += len;
2355                 u64_stats_update_end(&rx_stats->syncp);
2356         } else {
2357 drop:
2358                 dev->stats.rx_dropped++;
2359         }
2360         rcu_read_unlock();
2361 }
2362
2363 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2364                                  struct vxlan_dev *vxlan,
2365                                  union vxlan_addr *daddr,
2366                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2367                                  struct dst_entry *dst,
2368                                  u32 rt_flags)
2369 {
2370 #if IS_ENABLED(CONFIG_IPV6)
2371         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2372          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2373          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2374          */
2375         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2376 #endif
2377         /* Bypass encapsulation if the destination is local */
2378         if (rt_flags & RTCF_LOCAL &&
2379             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2380                 struct vxlan_dev *dst_vxlan;
2381
2382                 dst_release(dst);
2383                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2384                                            daddr->sa.sa_family, dst_port,
2385                                            vxlan->cfg.flags);
2386                 if (!dst_vxlan) {
2387                         dev->stats.tx_errors++;
2388                         kfree_skb(skb);
2389
2390                         return -ENOENT;
2391                 }
2392                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2393                 return 1;
2394         }
2395
2396         return 0;
2397 }
2398
2399 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2400                            __be32 default_vni, struct vxlan_rdst *rdst,
2401                            bool did_rsc)
2402 {
2403         struct dst_cache *dst_cache;
2404         struct ip_tunnel_info *info;
2405         struct vxlan_dev *vxlan = netdev_priv(dev);
2406         const struct iphdr *old_iph = ip_hdr(skb);
2407         union vxlan_addr *dst;
2408         union vxlan_addr remote_ip, local_ip;
2409         struct vxlan_metadata _md;
2410         struct vxlan_metadata *md = &_md;
2411         __be16 src_port = 0, dst_port;
2412         struct dst_entry *ndst = NULL;
2413         __be32 vni, label;
2414         __u8 tos, ttl;
2415         int ifindex;
2416         int err;
2417         u32 flags = vxlan->cfg.flags;
2418         bool udp_sum = false;
2419         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2420
2421         info = skb_tunnel_info(skb);
2422
2423         if (rdst) {
2424                 dst = &rdst->remote_ip;
2425                 if (vxlan_addr_any(dst)) {
2426                         if (did_rsc) {
2427                                 /* short-circuited back to local bridge */
2428                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2429                                 return;
2430                         }
2431                         goto drop;
2432                 }
2433
2434                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2435                 vni = (rdst->remote_vni) ? : default_vni;
2436                 ifindex = rdst->remote_ifindex;
2437                 local_ip = vxlan->cfg.saddr;
2438                 dst_cache = &rdst->dst_cache;
2439                 md->gbp = skb->mark;
2440                 if (flags & VXLAN_F_TTL_INHERIT) {
2441                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2442                 } else {
2443                         ttl = vxlan->cfg.ttl;
2444                         if (!ttl && vxlan_addr_multicast(dst))
2445                                 ttl = 1;
2446                 }
2447
2448                 tos = vxlan->cfg.tos;
2449                 if (tos == 1)
2450                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2451
2452                 if (dst->sa.sa_family == AF_INET)
2453                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2454                 else
2455                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2456                 label = vxlan->cfg.label;
2457         } else {
2458                 if (!info) {
2459                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2460                                   dev->name);
2461                         goto drop;
2462                 }
2463                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2464                 if (remote_ip.sa.sa_family == AF_INET) {
2465                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2466                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2467                 } else {
2468                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2469                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2470                 }
2471                 dst = &remote_ip;
2472                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2473                 vni = tunnel_id_to_key32(info->key.tun_id);
2474                 ifindex = 0;
2475                 dst_cache = &info->dst_cache;
2476                 if (info->options_len &&
2477                     info->key.tun_flags & TUNNEL_VXLAN_OPT)
2478                         md = ip_tunnel_info_opts(info);
2479                 ttl = info->key.ttl;
2480                 tos = info->key.tos;
2481                 label = info->key.label;
2482                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2483         }
2484         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2485                                      vxlan->cfg.port_max, true);
2486
2487         rcu_read_lock();
2488         if (dst->sa.sa_family == AF_INET) {
2489                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2490                 struct rtable *rt;
2491                 __be16 df = 0;
2492
2493                 if (!ifindex)
2494                         ifindex = sock4->sock->sk->sk_bound_dev_if;
2495
2496                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2497                                      dst->sin.sin_addr.s_addr,
2498                                      &local_ip.sin.sin_addr.s_addr,
2499                                      dst_port, src_port,
2500                                      dst_cache, info);
2501                 if (IS_ERR(rt)) {
2502                         err = PTR_ERR(rt);
2503                         goto tx_error;
2504                 }
2505
2506                 if (!info) {
2507                         /* Bypass encapsulation if the destination is local */
2508                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2509                                                     dst_port, ifindex, vni,
2510                                                     &rt->dst, rt->rt_flags);
2511                         if (err)
2512                                 goto out_unlock;
2513
2514                         if (vxlan->cfg.df == VXLAN_DF_SET) {
2515                                 df = htons(IP_DF);
2516                         } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) {
2517                                 struct ethhdr *eth = eth_hdr(skb);
2518
2519                                 if (ntohs(eth->h_proto) == ETH_P_IPV6 ||
2520                                     (ntohs(eth->h_proto) == ETH_P_IP &&
2521                                      old_iph->frag_off & htons(IP_DF)))
2522                                         df = htons(IP_DF);
2523                         }
2524                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2525                         df = htons(IP_DF);
2526                 }
2527
2528                 ndst = &rt->dst;
2529                 skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM);
2530
2531                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2532                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2533                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2534                                       vni, md, flags, udp_sum);
2535                 if (err < 0)
2536                         goto tx_error;
2537
2538                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2539                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2540                                     src_port, dst_port, xnet, !udp_sum);
2541 #if IS_ENABLED(CONFIG_IPV6)
2542         } else {
2543                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2544
2545                 if (!ifindex)
2546                         ifindex = sock6->sock->sk->sk_bound_dev_if;
2547
2548                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2549                                         label, &dst->sin6.sin6_addr,
2550                                         &local_ip.sin6.sin6_addr,
2551                                         dst_port, src_port,
2552                                         dst_cache, info);
2553                 if (IS_ERR(ndst)) {
2554                         err = PTR_ERR(ndst);
2555                         ndst = NULL;
2556                         goto tx_error;
2557                 }
2558
2559                 if (!info) {
2560                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2561
2562                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2563                                                     dst_port, ifindex, vni,
2564                                                     ndst, rt6i_flags);
2565                         if (err)
2566                                 goto out_unlock;
2567                 }
2568
2569                 skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM);
2570
2571                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2572                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2573                 skb_scrub_packet(skb, xnet);
2574                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2575                                       vni, md, flags, udp_sum);
2576                 if (err < 0)
2577                         goto tx_error;
2578
2579                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2580                                      &local_ip.sin6.sin6_addr,
2581                                      &dst->sin6.sin6_addr, tos, ttl,
2582                                      label, src_port, dst_port, !udp_sum);
2583 #endif
2584         }
2585 out_unlock:
2586         rcu_read_unlock();
2587         return;
2588
2589 drop:
2590         dev->stats.tx_dropped++;
2591         dev_kfree_skb(skb);
2592         return;
2593
2594 tx_error:
2595         rcu_read_unlock();
2596         if (err == -ELOOP)
2597                 dev->stats.collisions++;
2598         else if (err == -ENETUNREACH)
2599                 dev->stats.tx_carrier_errors++;
2600         dst_release(ndst);
2601         dev->stats.tx_errors++;
2602         kfree_skb(skb);
2603 }
2604
2605 /* Transmit local packets over Vxlan
2606  *
2607  * Outer IP header inherits ECN and DF from inner header.
2608  * Outer UDP destination is the VXLAN assigned port.
2609  *           source port is based on hash of flow
2610  */
2611 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2612 {
2613         struct vxlan_dev *vxlan = netdev_priv(dev);
2614         struct vxlan_rdst *rdst, *fdst = NULL;
2615         const struct ip_tunnel_info *info;
2616         bool did_rsc = false;
2617         struct vxlan_fdb *f;
2618         struct ethhdr *eth;
2619         __be32 vni = 0;
2620
2621         info = skb_tunnel_info(skb);
2622
2623         skb_reset_mac_header(skb);
2624
2625         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2626                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2627                     info->mode & IP_TUNNEL_INFO_TX) {
2628                         vni = tunnel_id_to_key32(info->key.tun_id);
2629                 } else {
2630                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2631                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2632                         else
2633                                 kfree_skb(skb);
2634                         return NETDEV_TX_OK;
2635                 }
2636         }
2637
2638         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2639                 eth = eth_hdr(skb);
2640                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2641                         return arp_reduce(dev, skb, vni);
2642 #if IS_ENABLED(CONFIG_IPV6)
2643                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2644                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2645                                             sizeof(struct nd_msg)) &&
2646                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2647                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2648
2649                         if (m->icmph.icmp6_code == 0 &&
2650                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2651                                 return neigh_reduce(dev, skb, vni);
2652                 }
2653 #endif
2654         }
2655
2656         eth = eth_hdr(skb);
2657         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2658         did_rsc = false;
2659
2660         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2661             (ntohs(eth->h_proto) == ETH_P_IP ||
2662              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2663                 did_rsc = route_shortcircuit(dev, skb);
2664                 if (did_rsc)
2665                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2666         }
2667
2668         if (f == NULL) {
2669                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2670                 if (f == NULL) {
2671                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2672                             !is_multicast_ether_addr(eth->h_dest))
2673                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2674
2675                         dev->stats.tx_dropped++;
2676                         kfree_skb(skb);
2677                         return NETDEV_TX_OK;
2678                 }
2679         }
2680
2681         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2682                 struct sk_buff *skb1;
2683
2684                 if (!fdst) {
2685                         fdst = rdst;
2686                         continue;
2687                 }
2688                 skb1 = skb_clone(skb, GFP_ATOMIC);
2689                 if (skb1)
2690                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2691         }
2692
2693         if (fdst)
2694                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2695         else
2696                 kfree_skb(skb);
2697         return NETDEV_TX_OK;
2698 }
2699
2700 /* Walk the forwarding table and purge stale entries */
2701 static void vxlan_cleanup(struct timer_list *t)
2702 {
2703         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2704         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2705         unsigned int h;
2706
2707         if (!netif_running(vxlan->dev))
2708                 return;
2709
2710         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2711                 struct hlist_node *p, *n;
2712
2713                 spin_lock(&vxlan->hash_lock);
2714                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2715                         struct vxlan_fdb *f
2716                                 = container_of(p, struct vxlan_fdb, hlist);
2717                         unsigned long timeout;
2718
2719                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2720                                 continue;
2721
2722                         if (f->flags & NTF_EXT_LEARNED)
2723                                 continue;
2724
2725                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2726                         if (time_before_eq(timeout, jiffies)) {
2727                                 netdev_dbg(vxlan->dev,
2728                                            "garbage collect %pM\n",
2729                                            f->eth_addr);
2730                                 f->state = NUD_STALE;
2731                                 vxlan_fdb_destroy(vxlan, f, true, true);
2732                         } else if (time_before(timeout, next_timer))
2733                                 next_timer = timeout;
2734                 }
2735                 spin_unlock(&vxlan->hash_lock);
2736         }
2737
2738         mod_timer(&vxlan->age_timer, next_timer);
2739 }
2740
2741 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2742 {
2743         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2744
2745         spin_lock(&vn->sock_lock);
2746         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2747 #if IS_ENABLED(CONFIG_IPV6)
2748         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2749 #endif
2750         spin_unlock(&vn->sock_lock);
2751 }
2752
2753 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2754                              struct vxlan_dev_node *node)
2755 {
2756         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2757         __be32 vni = vxlan->default_dst.remote_vni;
2758
2759         node->vxlan = vxlan;
2760         spin_lock(&vn->sock_lock);
2761         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2762         spin_unlock(&vn->sock_lock);
2763 }
2764
2765 /* Setup stats when device is created */
2766 static int vxlan_init(struct net_device *dev)
2767 {
2768         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2769         if (!dev->tstats)
2770                 return -ENOMEM;
2771
2772         return 0;
2773 }
2774
2775 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2776 {
2777         struct vxlan_fdb *f;
2778
2779         spin_lock_bh(&vxlan->hash_lock);
2780         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2781         if (f)
2782                 vxlan_fdb_destroy(vxlan, f, true, true);
2783         spin_unlock_bh(&vxlan->hash_lock);
2784 }
2785
2786 static void vxlan_uninit(struct net_device *dev)
2787 {
2788         struct vxlan_dev *vxlan = netdev_priv(dev);
2789
2790         gro_cells_destroy(&vxlan->gro_cells);
2791
2792         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2793
2794         free_percpu(dev->tstats);
2795 }
2796
2797 /* Start ageing timer and join group when device is brought up */
2798 static int vxlan_open(struct net_device *dev)
2799 {
2800         struct vxlan_dev *vxlan = netdev_priv(dev);
2801         int ret;
2802
2803         ret = vxlan_sock_add(vxlan);
2804         if (ret < 0)
2805                 return ret;
2806
2807         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2808                 ret = vxlan_igmp_join(vxlan);
2809                 if (ret == -EADDRINUSE)
2810                         ret = 0;
2811                 if (ret) {
2812                         vxlan_sock_release(vxlan);
2813                         return ret;
2814                 }
2815         }
2816
2817         if (vxlan->cfg.age_interval)
2818                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2819
2820         return ret;
2821 }
2822
2823 /* Purge the forwarding table */
2824 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2825 {
2826         unsigned int h;
2827
2828         spin_lock_bh(&vxlan->hash_lock);
2829         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2830                 struct hlist_node *p, *n;
2831                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2832                         struct vxlan_fdb *f
2833                                 = container_of(p, struct vxlan_fdb, hlist);
2834                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2835                                 continue;
2836                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2837                         if (!is_zero_ether_addr(f->eth_addr))
2838                                 vxlan_fdb_destroy(vxlan, f, true, true);
2839                 }
2840         }
2841         spin_unlock_bh(&vxlan->hash_lock);
2842 }
2843
2844 /* Cleanup timer and forwarding table on shutdown */
2845 static int vxlan_stop(struct net_device *dev)
2846 {
2847         struct vxlan_dev *vxlan = netdev_priv(dev);
2848         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2849         int ret = 0;
2850
2851         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2852             !vxlan_group_used(vn, vxlan))
2853                 ret = vxlan_igmp_leave(vxlan);
2854
2855         del_timer_sync(&vxlan->age_timer);
2856
2857         vxlan_flush(vxlan, false);
2858         vxlan_sock_release(vxlan);
2859
2860         return ret;
2861 }
2862
2863 /* Stub, nothing needs to be done. */
2864 static void vxlan_set_multicast_list(struct net_device *dev)
2865 {
2866 }
2867
2868 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2869 {
2870         struct vxlan_dev *vxlan = netdev_priv(dev);
2871         struct vxlan_rdst *dst = &vxlan->default_dst;
2872         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2873                                                          dst->remote_ifindex);
2874         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
2875
2876         /* This check is different than dev->max_mtu, because it looks at
2877          * the lowerdev->mtu, rather than the static dev->max_mtu
2878          */
2879         if (lowerdev) {
2880                 int max_mtu = lowerdev->mtu -
2881                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2882                 if (new_mtu > max_mtu)
2883                         return -EINVAL;
2884         }
2885
2886         dev->mtu = new_mtu;
2887         return 0;
2888 }
2889
2890 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2891 {
2892         struct vxlan_dev *vxlan = netdev_priv(dev);
2893         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2894         __be16 sport, dport;
2895
2896         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2897                                   vxlan->cfg.port_max, true);
2898         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2899
2900         if (ip_tunnel_info_af(info) == AF_INET) {
2901                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2902                 struct rtable *rt;
2903
2904                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2905                                      info->key.u.ipv4.dst,
2906                                      &info->key.u.ipv4.src, dport, sport,
2907                                      &info->dst_cache, info);
2908                 if (IS_ERR(rt))
2909                         return PTR_ERR(rt);
2910                 ip_rt_put(rt);
2911         } else {
2912 #if IS_ENABLED(CONFIG_IPV6)
2913                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2914                 struct dst_entry *ndst;
2915
2916                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2917                                         info->key.label, &info->key.u.ipv6.dst,
2918                                         &info->key.u.ipv6.src, dport, sport,
2919                                         &info->dst_cache, info);
2920                 if (IS_ERR(ndst))
2921                         return PTR_ERR(ndst);
2922                 dst_release(ndst);
2923 #else /* !CONFIG_IPV6 */
2924                 return -EPFNOSUPPORT;
2925 #endif
2926         }
2927         info->key.tp_src = sport;
2928         info->key.tp_dst = dport;
2929         return 0;
2930 }
2931
2932 static const struct net_device_ops vxlan_netdev_ether_ops = {
2933         .ndo_init               = vxlan_init,
2934         .ndo_uninit             = vxlan_uninit,
2935         .ndo_open               = vxlan_open,
2936         .ndo_stop               = vxlan_stop,
2937         .ndo_start_xmit         = vxlan_xmit,
2938         .ndo_get_stats64        = ip_tunnel_get_stats64,
2939         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2940         .ndo_change_mtu         = vxlan_change_mtu,
2941         .ndo_validate_addr      = eth_validate_addr,
2942         .ndo_set_mac_address    = eth_mac_addr,
2943         .ndo_fdb_add            = vxlan_fdb_add,
2944         .ndo_fdb_del            = vxlan_fdb_delete,
2945         .ndo_fdb_dump           = vxlan_fdb_dump,
2946         .ndo_fdb_get            = vxlan_fdb_get,
2947         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2948         .ndo_change_proto_down  = dev_change_proto_down_generic,
2949 };
2950
2951 static const struct net_device_ops vxlan_netdev_raw_ops = {
2952         .ndo_init               = vxlan_init,
2953         .ndo_uninit             = vxlan_uninit,
2954         .ndo_open               = vxlan_open,
2955         .ndo_stop               = vxlan_stop,
2956         .ndo_start_xmit         = vxlan_xmit,
2957         .ndo_get_stats64        = ip_tunnel_get_stats64,
2958         .ndo_change_mtu         = vxlan_change_mtu,
2959         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2960 };
2961
2962 /* Info for udev, that this is a virtual tunnel endpoint */
2963 static struct device_type vxlan_type = {
2964         .name = "vxlan",
2965 };
2966
2967 /* Calls the ndo_udp_tunnel_add of the caller in order to
2968  * supply the listening VXLAN udp ports. Callers are expected
2969  * to implement the ndo_udp_tunnel_add.
2970  */
2971 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
2972 {
2973         struct vxlan_sock *vs;
2974         struct net *net = dev_net(dev);
2975         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2976         unsigned int i;
2977
2978         spin_lock(&vn->sock_lock);
2979         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2980                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2981                         unsigned short type;
2982
2983                         if (vs->flags & VXLAN_F_GPE)
2984                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
2985                         else
2986                                 type = UDP_TUNNEL_TYPE_VXLAN;
2987
2988                         if (push)
2989                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
2990                         else
2991                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
2992                 }
2993         }
2994         spin_unlock(&vn->sock_lock);
2995 }
2996
2997 /* Initialize the device structure. */
2998 static void vxlan_setup(struct net_device *dev)
2999 {
3000         struct vxlan_dev *vxlan = netdev_priv(dev);
3001         unsigned int h;
3002
3003         eth_hw_addr_random(dev);
3004         ether_setup(dev);
3005
3006         dev->needs_free_netdev = true;
3007         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
3008
3009         dev->features   |= NETIF_F_LLTX;
3010         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
3011         dev->features   |= NETIF_F_RXCSUM;
3012         dev->features   |= NETIF_F_GSO_SOFTWARE;
3013
3014         dev->vlan_features = dev->features;
3015         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
3016         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
3017         netif_keep_dst(dev);
3018         dev->priv_flags |= IFF_NO_QUEUE;
3019
3020         /* MTU range: 68 - 65535 */
3021         dev->min_mtu = ETH_MIN_MTU;
3022         dev->max_mtu = ETH_MAX_MTU;
3023
3024         INIT_LIST_HEAD(&vxlan->next);
3025         spin_lock_init(&vxlan->hash_lock);
3026
3027         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
3028
3029         vxlan->dev = dev;
3030
3031         gro_cells_init(&vxlan->gro_cells, dev);
3032
3033         for (h = 0; h < FDB_HASH_SIZE; ++h)
3034                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
3035 }
3036
3037 static void vxlan_ether_setup(struct net_device *dev)
3038 {
3039         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
3040         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3041         dev->netdev_ops = &vxlan_netdev_ether_ops;
3042 }
3043
3044 static void vxlan_raw_setup(struct net_device *dev)
3045 {
3046         dev->header_ops = NULL;
3047         dev->type = ARPHRD_NONE;
3048         dev->hard_header_len = 0;
3049         dev->addr_len = 0;
3050         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
3051         dev->netdev_ops = &vxlan_netdev_raw_ops;
3052 }
3053
3054 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
3055         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
3056         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
3057         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
3058         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
3059         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
3060         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
3061         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
3062         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
3063         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
3064         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
3065         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
3066         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
3067         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
3068         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
3069         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
3070         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
3071         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
3072         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
3073         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
3074         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
3075         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
3076         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
3077         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
3078         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
3079         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
3080         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
3081         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
3082         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
3083         [IFLA_VXLAN_DF]         = { .type = NLA_U8 },
3084 };
3085
3086 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
3087                           struct netlink_ext_ack *extack)
3088 {
3089         if (tb[IFLA_ADDRESS]) {
3090                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
3091                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3092                                             "Provided link layer address is not Ethernet");
3093                         return -EINVAL;
3094                 }
3095
3096                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
3097                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3098                                             "Provided Ethernet address is not unicast");
3099                         return -EADDRNOTAVAIL;
3100                 }
3101         }
3102
3103         if (tb[IFLA_MTU]) {
3104                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
3105
3106                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
3107                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3108                                             "MTU must be between 68 and 65535");
3109                         return -EINVAL;
3110                 }
3111         }
3112
3113         if (!data) {
3114                 NL_SET_ERR_MSG(extack,
3115                                "Required attributes not provided to perform the operation");
3116                 return -EINVAL;
3117         }
3118
3119         if (data[IFLA_VXLAN_ID]) {
3120                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
3121
3122                 if (id >= VXLAN_N_VID) {
3123                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID],
3124                                             "VXLAN ID must be lower than 16777216");
3125                         return -ERANGE;
3126                 }
3127         }
3128
3129         if (data[IFLA_VXLAN_PORT_RANGE]) {
3130                 const struct ifla_vxlan_port_range *p
3131                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3132
3133                 if (ntohs(p->high) < ntohs(p->low)) {
3134                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
3135                                             "Invalid source port range");
3136                         return -EINVAL;
3137                 }
3138         }
3139
3140         if (data[IFLA_VXLAN_DF]) {
3141                 enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]);
3142
3143                 if (df < 0 || df > VXLAN_DF_MAX) {
3144                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_DF],
3145                                             "Invalid DF attribute");
3146                         return -EINVAL;
3147                 }
3148         }
3149
3150         return 0;
3151 }
3152
3153 static void vxlan_get_drvinfo(struct net_device *netdev,
3154                               struct ethtool_drvinfo *drvinfo)
3155 {
3156         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
3157         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
3158 }
3159
3160 static const struct ethtool_ops vxlan_ethtool_ops = {
3161         .get_drvinfo    = vxlan_get_drvinfo,
3162         .get_link       = ethtool_op_get_link,
3163 };
3164
3165 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
3166                                         __be16 port, u32 flags, int ifindex)
3167 {
3168         struct socket *sock;
3169         struct udp_port_cfg udp_conf;
3170         int err;
3171
3172         memset(&udp_conf, 0, sizeof(udp_conf));
3173
3174         if (ipv6) {
3175                 udp_conf.family = AF_INET6;
3176                 udp_conf.use_udp6_rx_checksums =
3177                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
3178                 udp_conf.ipv6_v6only = 1;
3179         } else {
3180                 udp_conf.family = AF_INET;
3181         }
3182
3183         udp_conf.local_udp_port = port;
3184         udp_conf.bind_ifindex = ifindex;
3185
3186         /* Open UDP socket */
3187         err = udp_sock_create(net, &udp_conf, &sock);
3188         if (err < 0)
3189                 return ERR_PTR(err);
3190
3191         return sock;
3192 }
3193
3194 /* Create new listen socket if needed */
3195 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
3196                                               __be16 port, u32 flags,
3197                                               int ifindex)
3198 {
3199         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3200         struct vxlan_sock *vs;
3201         struct socket *sock;
3202         unsigned int h;
3203         struct udp_tunnel_sock_cfg tunnel_cfg;
3204
3205         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
3206         if (!vs)
3207                 return ERR_PTR(-ENOMEM);
3208
3209         for (h = 0; h < VNI_HASH_SIZE; ++h)
3210                 INIT_HLIST_HEAD(&vs->vni_list[h]);
3211
3212         sock = vxlan_create_sock(net, ipv6, port, flags, ifindex);
3213         if (IS_ERR(sock)) {
3214                 kfree(vs);
3215                 return ERR_CAST(sock);
3216         }
3217
3218         vs->sock = sock;
3219         refcount_set(&vs->refcnt, 1);
3220         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
3221
3222         spin_lock(&vn->sock_lock);
3223         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
3224         udp_tunnel_notify_add_rx_port(sock,
3225                                       (vs->flags & VXLAN_F_GPE) ?
3226                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
3227                                       UDP_TUNNEL_TYPE_VXLAN);
3228         spin_unlock(&vn->sock_lock);
3229
3230         /* Mark socket as an encapsulation socket. */
3231         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
3232         tunnel_cfg.sk_user_data = vs;
3233         tunnel_cfg.encap_type = 1;
3234         tunnel_cfg.encap_rcv = vxlan_rcv;
3235         tunnel_cfg.encap_err_lookup = vxlan_err_lookup;
3236         tunnel_cfg.encap_destroy = NULL;
3237         tunnel_cfg.gro_receive = vxlan_gro_receive;
3238         tunnel_cfg.gro_complete = vxlan_gro_complete;
3239
3240         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
3241
3242         return vs;
3243 }
3244
3245 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
3246 {
3247         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3248         struct vxlan_sock *vs = NULL;
3249         struct vxlan_dev_node *node;
3250         int l3mdev_index = 0;
3251
3252         if (vxlan->cfg.remote_ifindex)
3253                 l3mdev_index = l3mdev_master_upper_ifindex_by_index(
3254                         vxlan->net, vxlan->cfg.remote_ifindex);
3255
3256         if (!vxlan->cfg.no_share) {
3257                 spin_lock(&vn->sock_lock);
3258                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
3259                                      vxlan->cfg.dst_port, vxlan->cfg.flags,
3260                                      l3mdev_index);
3261                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
3262                         spin_unlock(&vn->sock_lock);
3263                         return -EBUSY;
3264                 }
3265                 spin_unlock(&vn->sock_lock);
3266         }
3267         if (!vs)
3268                 vs = vxlan_socket_create(vxlan->net, ipv6,
3269                                          vxlan->cfg.dst_port, vxlan->cfg.flags,
3270                                          l3mdev_index);
3271         if (IS_ERR(vs))
3272                 return PTR_ERR(vs);
3273 #if IS_ENABLED(CONFIG_IPV6)
3274         if (ipv6) {
3275                 rcu_assign_pointer(vxlan->vn6_sock, vs);
3276                 node = &vxlan->hlist6;
3277         } else
3278 #endif
3279         {
3280                 rcu_assign_pointer(vxlan->vn4_sock, vs);
3281                 node = &vxlan->hlist4;
3282         }
3283         vxlan_vs_add_dev(vs, vxlan, node);
3284         return 0;
3285 }
3286
3287 static int vxlan_sock_add(struct vxlan_dev *vxlan)
3288 {
3289         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
3290         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
3291         bool ipv4 = !ipv6 || metadata;
3292         int ret = 0;
3293
3294         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
3295 #if IS_ENABLED(CONFIG_IPV6)
3296         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
3297         if (ipv6) {
3298                 ret = __vxlan_sock_add(vxlan, true);
3299                 if (ret < 0 && ret != -EAFNOSUPPORT)
3300                         ipv4 = false;
3301         }
3302 #endif
3303         if (ipv4)
3304                 ret = __vxlan_sock_add(vxlan, false);
3305         if (ret < 0)
3306                 vxlan_sock_release(vxlan);
3307         return ret;
3308 }
3309
3310 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
3311                                  struct net_device **lower,
3312                                  struct vxlan_dev *old,
3313                                  struct netlink_ext_ack *extack)
3314 {
3315         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
3316         struct vxlan_dev *tmp;
3317         bool use_ipv6 = false;
3318
3319         if (conf->flags & VXLAN_F_GPE) {
3320                 /* For now, allow GPE only together with
3321                  * COLLECT_METADATA. This can be relaxed later; in such
3322                  * case, the other side of the PtP link will have to be
3323                  * provided.
3324                  */
3325                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
3326                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3327                         NL_SET_ERR_MSG(extack,
3328                                        "VXLAN GPE does not support this combination of attributes");
3329                         return -EINVAL;
3330                 }
3331         }
3332
3333         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
3334                 /* Unless IPv6 is explicitly requested, assume IPv4 */
3335                 conf->remote_ip.sa.sa_family = AF_INET;
3336                 conf->saddr.sa.sa_family = AF_INET;
3337         } else if (!conf->remote_ip.sa.sa_family) {
3338                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
3339         } else if (!conf->saddr.sa.sa_family) {
3340                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
3341         }
3342
3343         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
3344                 NL_SET_ERR_MSG(extack,
3345                                "Local and remote address must be from the same family");
3346                 return -EINVAL;
3347         }
3348
3349         if (vxlan_addr_multicast(&conf->saddr)) {
3350                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
3351                 return -EINVAL;
3352         }
3353
3354         if (conf->saddr.sa.sa_family == AF_INET6) {
3355                 if (!IS_ENABLED(CONFIG_IPV6)) {
3356                         NL_SET_ERR_MSG(extack,
3357                                        "IPv6 support not enabled in the kernel");
3358                         return -EPFNOSUPPORT;
3359                 }
3360                 use_ipv6 = true;
3361                 conf->flags |= VXLAN_F_IPV6;
3362
3363                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3364                         int local_type =
3365                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3366                         int remote_type =
3367                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3368
3369                         if (local_type & IPV6_ADDR_LINKLOCAL) {
3370                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3371                                     (remote_type != IPV6_ADDR_ANY)) {
3372                                         NL_SET_ERR_MSG(extack,
3373                                                        "Invalid combination of local and remote address scopes");
3374                                         return -EINVAL;
3375                                 }
3376
3377                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3378                         } else {
3379                                 if (remote_type ==
3380                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3381                                         NL_SET_ERR_MSG(extack,
3382                                                        "Invalid combination of local and remote address scopes");
3383                                         return -EINVAL;
3384                                 }
3385
3386                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3387                         }
3388                 }
3389         }
3390
3391         if (conf->label && !use_ipv6) {
3392                 NL_SET_ERR_MSG(extack,
3393                                "Label attribute only applies to IPv6 VXLAN devices");
3394                 return -EINVAL;
3395         }
3396
3397         if (conf->remote_ifindex) {
3398                 struct net_device *lowerdev;
3399
3400                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3401                 if (!lowerdev) {
3402                         NL_SET_ERR_MSG(extack,
3403                                        "Invalid local interface, device not found");
3404                         return -ENODEV;
3405                 }
3406
3407 #if IS_ENABLED(CONFIG_IPV6)
3408                 if (use_ipv6) {
3409                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3410                         if (idev && idev->cnf.disable_ipv6) {
3411                                 NL_SET_ERR_MSG(extack,
3412                                                "IPv6 support disabled by administrator");
3413                                 return -EPERM;
3414                         }
3415                 }
3416 #endif
3417
3418                 *lower = lowerdev;
3419         } else {
3420                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3421                         NL_SET_ERR_MSG(extack,
3422                                        "Local interface required for multicast remote destination");
3423
3424                         return -EINVAL;
3425                 }
3426
3427 #if IS_ENABLED(CONFIG_IPV6)
3428                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3429                         NL_SET_ERR_MSG(extack,
3430                                        "Local interface required for link-local local/remote addresses");
3431                         return -EINVAL;
3432                 }
3433 #endif
3434
3435                 *lower = NULL;
3436         }
3437
3438         if (!conf->dst_port) {
3439                 if (conf->flags & VXLAN_F_GPE)
3440                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3441                 else
3442                         conf->dst_port = htons(vxlan_port);
3443         }
3444
3445         if (!conf->age_interval)
3446                 conf->age_interval = FDB_AGE_DEFAULT;
3447
3448         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3449                 if (tmp == old)
3450                         continue;
3451
3452                 if (tmp->cfg.vni != conf->vni)
3453                         continue;
3454                 if (tmp->cfg.dst_port != conf->dst_port)
3455                         continue;
3456                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3457                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3458                         continue;
3459
3460                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3461                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3462                         continue;
3463
3464                 NL_SET_ERR_MSG(extack,
3465                                "A VXLAN device with the specified VNI already exists");
3466                 return -EEXIST;
3467         }
3468
3469         return 0;
3470 }
3471
3472 static void vxlan_config_apply(struct net_device *dev,
3473                                struct vxlan_config *conf,
3474                                struct net_device *lowerdev,
3475                                struct net *src_net,
3476                                bool changelink)
3477 {
3478         struct vxlan_dev *vxlan = netdev_priv(dev);
3479         struct vxlan_rdst *dst = &vxlan->default_dst;
3480         unsigned short needed_headroom = ETH_HLEN;
3481         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3482         int max_mtu = ETH_MAX_MTU;
3483
3484         if (!changelink) {
3485                 if (conf->flags & VXLAN_F_GPE)
3486                         vxlan_raw_setup(dev);
3487                 else
3488                         vxlan_ether_setup(dev);
3489
3490                 if (conf->mtu)
3491                         dev->mtu = conf->mtu;
3492
3493                 vxlan->net = src_net;
3494         }
3495
3496         dst->remote_vni = conf->vni;
3497
3498         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3499
3500         if (lowerdev) {
3501                 dst->remote_ifindex = conf->remote_ifindex;
3502
3503                 dev->gso_max_size = lowerdev->gso_max_size;
3504                 dev->gso_max_segs = lowerdev->gso_max_segs;
3505
3506                 needed_headroom = lowerdev->hard_header_len;
3507
3508                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3509                                            VXLAN_HEADROOM);
3510                 if (max_mtu < ETH_MIN_MTU)
3511                         max_mtu = ETH_MIN_MTU;
3512
3513                 if (!changelink && !conf->mtu)
3514                         dev->mtu = max_mtu;
3515         }
3516
3517         if (dev->mtu > max_mtu)
3518                 dev->mtu = max_mtu;
3519
3520         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3521                 needed_headroom += VXLAN6_HEADROOM;
3522         else
3523                 needed_headroom += VXLAN_HEADROOM;
3524         dev->needed_headroom = needed_headroom;
3525
3526         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3527 }
3528
3529 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3530                                struct vxlan_config *conf, bool changelink,
3531                                struct netlink_ext_ack *extack)
3532 {
3533         struct vxlan_dev *vxlan = netdev_priv(dev);
3534         struct net_device *lowerdev;
3535         int ret;
3536
3537         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3538         if (ret)
3539                 return ret;
3540
3541         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3542
3543         return 0;
3544 }
3545
3546 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3547                               struct vxlan_config *conf,
3548                               struct netlink_ext_ack *extack)
3549 {
3550         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3551         struct vxlan_dev *vxlan = netdev_priv(dev);
3552         struct vxlan_fdb *f = NULL;
3553         bool unregister = false;
3554         int err;
3555
3556         err = vxlan_dev_configure(net, dev, conf, false, extack);
3557         if (err)
3558                 return err;
3559
3560         dev->ethtool_ops = &vxlan_ethtool_ops;
3561
3562         /* create an fdb entry for a valid default destination */
3563         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
3564                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3565                                        &vxlan->default_dst.remote_ip,
3566                                        NUD_REACHABLE | NUD_PERMANENT,
3567                                        vxlan->cfg.dst_port,
3568                                        vxlan->default_dst.remote_vni,
3569                                        vxlan->default_dst.remote_vni,
3570                                        vxlan->default_dst.remote_ifindex,
3571                                        NTF_SELF, &f);
3572                 if (err)
3573                         return err;
3574         }
3575
3576         err = register_netdevice(dev);
3577         if (err)
3578                 goto errout;
3579         unregister = true;
3580
3581         err = rtnl_configure_link(dev, NULL);
3582         if (err)
3583                 goto errout;
3584
3585         if (f) {
3586                 vxlan_fdb_insert(vxlan, all_zeros_mac,
3587                                  vxlan->default_dst.remote_vni, f);
3588
3589                 /* notify default fdb entry */
3590                 err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f),
3591                                        RTM_NEWNEIGH, true, extack);
3592                 if (err) {
3593                         vxlan_fdb_destroy(vxlan, f, false, false);
3594                         goto unregister;
3595                 }
3596         }
3597
3598         list_add(&vxlan->next, &vn->vxlan_list);
3599         return 0;
3600
3601 errout:
3602         /* unregister_netdevice() destroys the default FDB entry with deletion
3603          * notification. But the addition notification was not sent yet, so
3604          * destroy the entry by hand here.
3605          */
3606         if (f)
3607                 __vxlan_fdb_free(f);
3608 unregister:
3609         if (unregister)
3610                 unregister_netdevice(dev);
3611         return err;
3612 }
3613
3614 /* Set/clear flags based on attribute */
3615 static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[],
3616                           int attrtype, unsigned long mask, bool changelink,
3617                           bool changelink_supported,
3618                           struct netlink_ext_ack *extack)
3619 {
3620         unsigned long flags;
3621
3622         if (!tb[attrtype])
3623                 return 0;
3624
3625         if (changelink && !changelink_supported) {
3626                 vxlan_flag_attr_error(attrtype, extack);
3627                 return -EOPNOTSUPP;
3628         }
3629
3630         if (vxlan_policy[attrtype].type == NLA_FLAG)
3631                 flags = conf->flags | mask;
3632         else if (nla_get_u8(tb[attrtype]))
3633                 flags = conf->flags | mask;
3634         else
3635                 flags = conf->flags & ~mask;
3636
3637         conf->flags = flags;
3638
3639         return 0;
3640 }
3641
3642 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3643                          struct net_device *dev, struct vxlan_config *conf,
3644                          bool changelink, struct netlink_ext_ack *extack)
3645 {
3646         struct vxlan_dev *vxlan = netdev_priv(dev);
3647         int err = 0;
3648
3649         memset(conf, 0, sizeof(*conf));
3650
3651         /* if changelink operation, start with old existing cfg */
3652         if (changelink)
3653                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3654
3655         if (data[IFLA_VXLAN_ID]) {
3656                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3657
3658                 if (changelink && (vni != conf->vni)) {
3659                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI");
3660                         return -EOPNOTSUPP;
3661                 }
3662                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3663         }
3664
3665         if (data[IFLA_VXLAN_GROUP]) {
3666                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) {
3667                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group");
3668                         return -EOPNOTSUPP;
3669                 }
3670
3671                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3672                 conf->remote_ip.sa.sa_family = AF_INET;
3673         } else if (data[IFLA_VXLAN_GROUP6]) {
3674                 if (!IS_ENABLED(CONFIG_IPV6)) {
3675                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel");
3676                         return -EPFNOSUPPORT;
3677                 }
3678
3679                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) {
3680                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group");
3681                         return -EOPNOTSUPP;
3682                 }
3683
3684                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3685                 conf->remote_ip.sa.sa_family = AF_INET6;
3686         }
3687
3688         if (data[IFLA_VXLAN_LOCAL]) {
3689                 if (changelink && (conf->saddr.sa.sa_family != AF_INET)) {
3690                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old");
3691                         return -EOPNOTSUPP;
3692                 }
3693
3694                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3695                 conf->saddr.sa.sa_family = AF_INET;
3696         } else if (data[IFLA_VXLAN_LOCAL6]) {
3697                 if (!IS_ENABLED(CONFIG_IPV6)) {
3698                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel");
3699                         return -EPFNOSUPPORT;
3700                 }
3701
3702                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) {
3703                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old");
3704                         return -EOPNOTSUPP;
3705                 }
3706
3707                 /* TODO: respect scope id */
3708                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3709                 conf->saddr.sa.sa_family = AF_INET6;
3710         }
3711
3712         if (data[IFLA_VXLAN_LINK])
3713                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3714
3715         if (data[IFLA_VXLAN_TOS])
3716                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3717
3718         if (data[IFLA_VXLAN_TTL])
3719                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3720
3721         if (data[IFLA_VXLAN_TTL_INHERIT]) {
3722                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT,
3723                                     VXLAN_F_TTL_INHERIT, changelink, false,
3724                                     extack);
3725                 if (err)
3726                         return err;
3727
3728         }
3729
3730         if (data[IFLA_VXLAN_LABEL])
3731                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3732                              IPV6_FLOWLABEL_MASK;
3733
3734         if (data[IFLA_VXLAN_LEARNING]) {
3735                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING,
3736                                     VXLAN_F_LEARN, changelink, true,
3737                                     extack);
3738                 if (err)
3739                         return err;
3740         } else if (!changelink) {
3741                 /* default to learn on a new device */
3742                 conf->flags |= VXLAN_F_LEARN;
3743         }
3744
3745         if (data[IFLA_VXLAN_AGEING])
3746                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3747
3748         if (data[IFLA_VXLAN_PROXY]) {
3749                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY,
3750                                     VXLAN_F_PROXY, changelink, false,
3751                                     extack);
3752                 if (err)
3753                         return err;
3754         }
3755
3756         if (data[IFLA_VXLAN_RSC]) {
3757                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC,
3758                                     VXLAN_F_RSC, changelink, false,
3759                                     extack);
3760                 if (err)
3761                         return err;
3762         }
3763
3764         if (data[IFLA_VXLAN_L2MISS]) {
3765                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS,
3766                                     VXLAN_F_L2MISS, changelink, false,
3767                                     extack);
3768                 if (err)
3769                         return err;
3770         }
3771
3772         if (data[IFLA_VXLAN_L3MISS]) {
3773                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS,
3774                                     VXLAN_F_L3MISS, changelink, false,
3775                                     extack);
3776                 if (err)
3777                         return err;
3778         }
3779
3780         if (data[IFLA_VXLAN_LIMIT]) {
3781                 if (changelink) {
3782                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT],
3783                                             "Cannot change limit");
3784                         return -EOPNOTSUPP;
3785                 }
3786                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3787         }
3788
3789         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3790                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA,
3791                                     VXLAN_F_COLLECT_METADATA, changelink, false,
3792                                     extack);
3793                 if (err)
3794                         return err;
3795         }
3796
3797         if (data[IFLA_VXLAN_PORT_RANGE]) {
3798                 if (!changelink) {
3799                         const struct ifla_vxlan_port_range *p
3800                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3801                         conf->port_min = ntohs(p->low);
3802                         conf->port_max = ntohs(p->high);
3803                 } else {
3804                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
3805                                             "Cannot change port range");
3806                         return -EOPNOTSUPP;
3807                 }
3808         }
3809
3810         if (data[IFLA_VXLAN_PORT]) {
3811                 if (changelink) {
3812                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT],
3813                                             "Cannot change port");
3814                         return -EOPNOTSUPP;
3815                 }
3816                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3817         }
3818
3819         if (data[IFLA_VXLAN_UDP_CSUM]) {
3820                 if (changelink) {
3821                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM],
3822                                             "Cannot change UDP_CSUM flag");
3823                         return -EOPNOTSUPP;
3824                 }
3825                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3826                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3827         }
3828
3829         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3830                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3831                                     VXLAN_F_UDP_ZERO_CSUM6_TX, changelink,
3832                                     false, extack);
3833                 if (err)
3834                         return err;
3835         }
3836
3837         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3838                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3839                                     VXLAN_F_UDP_ZERO_CSUM6_RX, changelink,
3840                                     false, extack);
3841                 if (err)
3842                         return err;
3843         }
3844
3845         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3846                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX,
3847                                     VXLAN_F_REMCSUM_TX, changelink, false,
3848                                     extack);
3849                 if (err)
3850                         return err;
3851         }
3852
3853         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3854                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX,
3855                                     VXLAN_F_REMCSUM_RX, changelink, false,
3856                                     extack);
3857                 if (err)
3858                         return err;
3859         }
3860
3861         if (data[IFLA_VXLAN_GBP]) {
3862                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP,
3863                                     VXLAN_F_GBP, changelink, false, extack);
3864                 if (err)
3865                         return err;
3866         }
3867
3868         if (data[IFLA_VXLAN_GPE]) {
3869                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE,
3870                                     VXLAN_F_GPE, changelink, false,
3871                                     extack);
3872                 if (err)
3873                         return err;
3874         }
3875
3876         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3877                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL,
3878                                     VXLAN_F_REMCSUM_NOPARTIAL, changelink,
3879                                     false, extack);
3880                 if (err)
3881                         return err;
3882         }
3883
3884         if (tb[IFLA_MTU]) {
3885                 if (changelink) {
3886                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3887                                             "Cannot change mtu");
3888                         return -EOPNOTSUPP;
3889                 }
3890                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3891         }
3892
3893         if (data[IFLA_VXLAN_DF])
3894                 conf->df = nla_get_u8(data[IFLA_VXLAN_DF]);
3895
3896         return 0;
3897 }
3898
3899 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3900                          struct nlattr *tb[], struct nlattr *data[],
3901                          struct netlink_ext_ack *extack)
3902 {
3903         struct vxlan_config conf;
3904         int err;
3905
3906         err = vxlan_nl2conf(tb, data, dev, &conf, false, extack);
3907         if (err)
3908                 return err;
3909
3910         return __vxlan_dev_create(src_net, dev, &conf, extack);
3911 }
3912
3913 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3914                             struct nlattr *data[],
3915                             struct netlink_ext_ack *extack)
3916 {
3917         struct vxlan_dev *vxlan = netdev_priv(dev);
3918         struct vxlan_rdst *dst = &vxlan->default_dst;
3919         struct net_device *lowerdev;
3920         struct vxlan_config conf;
3921         int err;
3922
3923         err = vxlan_nl2conf(tb, data, dev, &conf, true, extack);
3924         if (err)
3925                 return err;
3926
3927         err = vxlan_config_validate(vxlan->net, &conf, &lowerdev,
3928                                     vxlan, extack);
3929         if (err)
3930                 return err;
3931
3932         /* handle default dst entry */
3933         if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) {
3934                 spin_lock_bh(&vxlan->hash_lock);
3935                 if (!vxlan_addr_any(&conf.remote_ip)) {
3936                         err = vxlan_fdb_update(vxlan, all_zeros_mac,
3937                                                &conf.remote_ip,
3938                                                NUD_REACHABLE | NUD_PERMANENT,
3939                                                NLM_F_APPEND | NLM_F_CREATE,
3940                                                vxlan->cfg.dst_port,
3941                                                conf.vni, conf.vni,
3942                                                conf.remote_ifindex,
3943                                                NTF_SELF, true, extack);
3944                         if (err) {
3945                                 spin_unlock_bh(&vxlan->hash_lock);
3946                                 return err;
3947                         }
3948                 }
3949                 if (!vxlan_addr_any(&dst->remote_ip))
3950                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
3951                                            dst->remote_ip,
3952                                            vxlan->cfg.dst_port,
3953                                            dst->remote_vni,
3954                                            dst->remote_vni,
3955                                            dst->remote_ifindex,
3956                                            true);
3957                 spin_unlock_bh(&vxlan->hash_lock);
3958         }
3959
3960         if (conf.age_interval != vxlan->cfg.age_interval)
3961                 mod_timer(&vxlan->age_timer, jiffies);
3962
3963         vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true);
3964         return 0;
3965 }
3966
3967 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3968 {
3969         struct vxlan_dev *vxlan = netdev_priv(dev);
3970
3971         vxlan_flush(vxlan, true);
3972
3973         list_del(&vxlan->next);
3974         unregister_netdevice_queue(dev, head);
3975 }
3976
3977 static size_t vxlan_get_size(const struct net_device *dev)
3978 {
3979
3980         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3981                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3982                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3983                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3984                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3985                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL_INHERIT */
3986                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3987                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_DF */
3988                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3989                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3990                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3991                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3992                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3993                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3994                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3995                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3996                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3997                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3998                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3999                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
4000                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
4001                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
4002                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
4003                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
4004                 0;
4005 }
4006
4007 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
4008 {
4009         const struct vxlan_dev *vxlan = netdev_priv(dev);
4010         const struct vxlan_rdst *dst = &vxlan->default_dst;
4011         struct ifla_vxlan_port_range ports = {
4012                 .low =  htons(vxlan->cfg.port_min),
4013                 .high = htons(vxlan->cfg.port_max),
4014         };
4015
4016         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
4017                 goto nla_put_failure;
4018
4019         if (!vxlan_addr_any(&dst->remote_ip)) {
4020                 if (dst->remote_ip.sa.sa_family == AF_INET) {
4021                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
4022                                             dst->remote_ip.sin.sin_addr.s_addr))
4023                                 goto nla_put_failure;
4024 #if IS_ENABLED(CONFIG_IPV6)
4025                 } else {
4026                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
4027                                              &dst->remote_ip.sin6.sin6_addr))
4028                                 goto nla_put_failure;
4029 #endif
4030                 }
4031         }
4032
4033         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
4034                 goto nla_put_failure;
4035
4036         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
4037                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
4038                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
4039                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
4040                                 goto nla_put_failure;
4041 #if IS_ENABLED(CONFIG_IPV6)
4042                 } else {
4043                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
4044                                              &vxlan->cfg.saddr.sin6.sin6_addr))
4045                                 goto nla_put_failure;
4046 #endif
4047                 }
4048         }
4049
4050         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
4051             nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT,
4052                        !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) ||
4053             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
4054             nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) ||
4055             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
4056             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
4057                         !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
4058             nla_put_u8(skb, IFLA_VXLAN_PROXY,
4059                         !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
4060             nla_put_u8(skb, IFLA_VXLAN_RSC,
4061                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
4062             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
4063                         !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
4064             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
4065                         !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
4066             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
4067                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
4068             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
4069             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
4070             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
4071             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
4072                         !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
4073             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4074                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
4075             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4076                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
4077             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
4078                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
4079             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
4080                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
4081                 goto nla_put_failure;
4082
4083         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
4084                 goto nla_put_failure;
4085
4086         if (vxlan->cfg.flags & VXLAN_F_GBP &&
4087             nla_put_flag(skb, IFLA_VXLAN_GBP))
4088                 goto nla_put_failure;
4089
4090         if (vxlan->cfg.flags & VXLAN_F_GPE &&
4091             nla_put_flag(skb, IFLA_VXLAN_GPE))
4092                 goto nla_put_failure;
4093
4094         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
4095             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
4096                 goto nla_put_failure;
4097
4098         return 0;
4099
4100 nla_put_failure:
4101         return -EMSGSIZE;
4102 }
4103
4104 static struct net *vxlan_get_link_net(const struct net_device *dev)
4105 {
4106         struct vxlan_dev *vxlan = netdev_priv(dev);
4107
4108         return vxlan->net;
4109 }
4110
4111 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
4112         .kind           = "vxlan",
4113         .maxtype        = IFLA_VXLAN_MAX,
4114         .policy         = vxlan_policy,
4115         .priv_size      = sizeof(struct vxlan_dev),
4116         .setup          = vxlan_setup,
4117         .validate       = vxlan_validate,
4118         .newlink        = vxlan_newlink,
4119         .changelink     = vxlan_changelink,
4120         .dellink        = vxlan_dellink,
4121         .get_size       = vxlan_get_size,
4122         .fill_info      = vxlan_fill_info,
4123         .get_link_net   = vxlan_get_link_net,
4124 };
4125
4126 struct net_device *vxlan_dev_create(struct net *net, const char *name,
4127                                     u8 name_assign_type,
4128                                     struct vxlan_config *conf)
4129 {
4130         struct nlattr *tb[IFLA_MAX + 1];
4131         struct net_device *dev;
4132         int err;
4133
4134         memset(&tb, 0, sizeof(tb));
4135
4136         dev = rtnl_create_link(net, name, name_assign_type,
4137                                &vxlan_link_ops, tb, NULL);
4138         if (IS_ERR(dev))
4139                 return dev;
4140
4141         err = __vxlan_dev_create(net, dev, conf, NULL);
4142         if (err < 0) {
4143                 free_netdev(dev);
4144                 return ERR_PTR(err);
4145         }
4146
4147         err = rtnl_configure_link(dev, NULL);
4148         if (err < 0) {
4149                 LIST_HEAD(list_kill);
4150
4151                 vxlan_dellink(dev, &list_kill);
4152                 unregister_netdevice_many(&list_kill);
4153                 return ERR_PTR(err);
4154         }
4155
4156         return dev;
4157 }
4158 EXPORT_SYMBOL_GPL(vxlan_dev_create);
4159
4160 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
4161                                              struct net_device *dev)
4162 {
4163         struct vxlan_dev *vxlan, *next;
4164         LIST_HEAD(list_kill);
4165
4166         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4167                 struct vxlan_rdst *dst = &vxlan->default_dst;
4168
4169                 /* In case we created vxlan device with carrier
4170                  * and we loose the carrier due to module unload
4171                  * we also need to remove vxlan device. In other
4172                  * cases, it's not necessary and remote_ifindex
4173                  * is 0 here, so no matches.
4174                  */
4175                 if (dst->remote_ifindex == dev->ifindex)
4176                         vxlan_dellink(vxlan->dev, &list_kill);
4177         }
4178
4179         unregister_netdevice_many(&list_kill);
4180 }
4181
4182 static int vxlan_netdevice_event(struct notifier_block *unused,
4183                                  unsigned long event, void *ptr)
4184 {
4185         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4186         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
4187
4188         if (event == NETDEV_UNREGISTER) {
4189                 vxlan_offload_rx_ports(dev, false);
4190                 vxlan_handle_lowerdev_unregister(vn, dev);
4191         } else if (event == NETDEV_REGISTER) {
4192                 vxlan_offload_rx_ports(dev, true);
4193         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
4194                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
4195                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
4196         }
4197
4198         return NOTIFY_DONE;
4199 }
4200
4201 static struct notifier_block vxlan_notifier_block __read_mostly = {
4202         .notifier_call = vxlan_netdevice_event,
4203 };
4204
4205 static void
4206 vxlan_fdb_offloaded_set(struct net_device *dev,
4207                         struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4208 {
4209         struct vxlan_dev *vxlan = netdev_priv(dev);
4210         struct vxlan_rdst *rdst;
4211         struct vxlan_fdb *f;
4212
4213         spin_lock_bh(&vxlan->hash_lock);
4214
4215         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4216         if (!f)
4217                 goto out;
4218
4219         rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip,
4220                                    fdb_info->remote_port,
4221                                    fdb_info->remote_vni,
4222                                    fdb_info->remote_ifindex);
4223         if (!rdst)
4224                 goto out;
4225
4226         rdst->offloaded = fdb_info->offloaded;
4227
4228 out:
4229         spin_unlock_bh(&vxlan->hash_lock);
4230 }
4231
4232 static int
4233 vxlan_fdb_external_learn_add(struct net_device *dev,
4234                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4235 {
4236         struct vxlan_dev *vxlan = netdev_priv(dev);
4237         struct netlink_ext_ack *extack;
4238         int err;
4239
4240         extack = switchdev_notifier_info_to_extack(&fdb_info->info);
4241
4242         spin_lock_bh(&vxlan->hash_lock);
4243         err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip,
4244                                NUD_REACHABLE,
4245                                NLM_F_CREATE | NLM_F_REPLACE,
4246                                fdb_info->remote_port,
4247                                fdb_info->vni,
4248                                fdb_info->remote_vni,
4249                                fdb_info->remote_ifindex,
4250                                NTF_USE | NTF_SELF | NTF_EXT_LEARNED,
4251                                false, extack);
4252         spin_unlock_bh(&vxlan->hash_lock);
4253
4254         return err;
4255 }
4256
4257 static int
4258 vxlan_fdb_external_learn_del(struct net_device *dev,
4259                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4260 {
4261         struct vxlan_dev *vxlan = netdev_priv(dev);
4262         struct vxlan_fdb *f;
4263         int err = 0;
4264
4265         spin_lock_bh(&vxlan->hash_lock);
4266
4267         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4268         if (!f)
4269                 err = -ENOENT;
4270         else if (f->flags & NTF_EXT_LEARNED)
4271                 err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr,
4272                                          fdb_info->remote_ip,
4273                                          fdb_info->remote_port,
4274                                          fdb_info->vni,
4275                                          fdb_info->remote_vni,
4276                                          fdb_info->remote_ifindex,
4277                                          false);
4278
4279         spin_unlock_bh(&vxlan->hash_lock);
4280
4281         return err;
4282 }
4283
4284 static int vxlan_switchdev_event(struct notifier_block *unused,
4285                                  unsigned long event, void *ptr)
4286 {
4287         struct net_device *dev = switchdev_notifier_info_to_dev(ptr);
4288         struct switchdev_notifier_vxlan_fdb_info *fdb_info;
4289         int err = 0;
4290
4291         switch (event) {
4292         case SWITCHDEV_VXLAN_FDB_OFFLOADED:
4293                 vxlan_fdb_offloaded_set(dev, ptr);
4294                 break;
4295         case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE:
4296                 fdb_info = ptr;
4297                 err = vxlan_fdb_external_learn_add(dev, fdb_info);
4298                 if (err) {
4299                         err = notifier_from_errno(err);
4300                         break;
4301                 }
4302                 fdb_info->offloaded = true;
4303                 vxlan_fdb_offloaded_set(dev, fdb_info);
4304                 break;
4305         case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE:
4306                 fdb_info = ptr;
4307                 err = vxlan_fdb_external_learn_del(dev, fdb_info);
4308                 if (err) {
4309                         err = notifier_from_errno(err);
4310                         break;
4311                 }
4312                 fdb_info->offloaded = false;
4313                 vxlan_fdb_offloaded_set(dev, fdb_info);
4314                 break;
4315         }
4316
4317         return err;
4318 }
4319
4320 static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = {
4321         .notifier_call = vxlan_switchdev_event,
4322 };
4323
4324 static __net_init int vxlan_init_net(struct net *net)
4325 {
4326         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4327         unsigned int h;
4328
4329         INIT_LIST_HEAD(&vn->vxlan_list);
4330         spin_lock_init(&vn->sock_lock);
4331
4332         for (h = 0; h < PORT_HASH_SIZE; ++h)
4333                 INIT_HLIST_HEAD(&vn->sock_list[h]);
4334
4335         return 0;
4336 }
4337
4338 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
4339 {
4340         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4341         struct vxlan_dev *vxlan, *next;
4342         struct net_device *dev, *aux;
4343         unsigned int h;
4344
4345         for_each_netdev_safe(net, dev, aux)
4346                 if (dev->rtnl_link_ops == &vxlan_link_ops)
4347                         unregister_netdevice_queue(dev, head);
4348
4349         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4350                 /* If vxlan->dev is in the same netns, it has already been added
4351                  * to the list by the previous loop.
4352                  */
4353                 if (!net_eq(dev_net(vxlan->dev), net))
4354                         unregister_netdevice_queue(vxlan->dev, head);
4355         }
4356
4357         for (h = 0; h < PORT_HASH_SIZE; ++h)
4358                 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
4359 }
4360
4361 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
4362 {
4363         struct net *net;
4364         LIST_HEAD(list);
4365
4366         rtnl_lock();
4367         list_for_each_entry(net, net_list, exit_list)
4368                 vxlan_destroy_tunnels(net, &list);
4369
4370         unregister_netdevice_many(&list);
4371         rtnl_unlock();
4372 }
4373
4374 static struct pernet_operations vxlan_net_ops = {
4375         .init = vxlan_init_net,
4376         .exit_batch = vxlan_exit_batch_net,
4377         .id   = &vxlan_net_id,
4378         .size = sizeof(struct vxlan_net),
4379 };
4380
4381 static int __init vxlan_init_module(void)
4382 {
4383         int rc;
4384
4385         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
4386
4387         rc = register_pernet_subsys(&vxlan_net_ops);
4388         if (rc)
4389                 goto out1;
4390
4391         rc = register_netdevice_notifier(&vxlan_notifier_block);
4392         if (rc)
4393                 goto out2;
4394
4395         rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block);
4396         if (rc)
4397                 goto out3;
4398
4399         rc = rtnl_link_register(&vxlan_link_ops);
4400         if (rc)
4401                 goto out4;
4402
4403         return 0;
4404 out4:
4405         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4406 out3:
4407         unregister_netdevice_notifier(&vxlan_notifier_block);
4408 out2:
4409         unregister_pernet_subsys(&vxlan_net_ops);
4410 out1:
4411         return rc;
4412 }
4413 late_initcall(vxlan_init_module);
4414
4415 static void __exit vxlan_cleanup_module(void)
4416 {
4417         rtnl_link_unregister(&vxlan_link_ops);
4418         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4419         unregister_netdevice_notifier(&vxlan_notifier_block);
4420         unregister_pernet_subsys(&vxlan_net_ops);
4421         /* rcu_barrier() is called by netns */
4422 }
4423 module_exit(vxlan_cleanup_module);
4424
4425 MODULE_LICENSE("GPL");
4426 MODULE_VERSION(VXLAN_VERSION);
4427 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
4428 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
4429 MODULE_ALIAS_RTNL_LINK("vxlan");