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 static u32 fdb_head_index(struct vxlan_dev *vxlan, const u8 *mac, __be32 vni)
 472 {
 473         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)
 474                 return eth_vni_hash(mac, vni);
 475         else
 476                 return eth_hash(mac);
 477 }
 478
 479 /* Hash chain to use given mac address */
 480 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 481                                                 const u8 *mac, __be32 vni)
 482 {
 483         return &vxlan->fdb_head[fdb_head_index(vxlan, mac, vni)];
 484 }
 485
 486 /* Look up Ethernet address in forwarding table */
 487 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 488                                           const u8 *mac, __be32 vni)
 489 {
 490         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
 491         struct vxlan_fdb *f;
 492
 493         hlist_for_each_entry_rcu(f, head, hlist) {
 494                 if (ether_addr_equal(mac, f->eth_addr)) {
 495                         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
 496                                 if (vni == f->vni)
 497                                         return f;
 498                         } else {
 499                                 return f;
 500                         }
 501                 }
 502         }
 503
 504         return NULL;
 505 }
 506
 507 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 508                                         const u8 *mac, __be32 vni)
 509 {
 510         struct vxlan_fdb *f;
 511
 512         f = __vxlan_find_mac(vxlan, mac, vni);
 513         if (f && f->used != jiffies)
 514                 f->used = jiffies;
 515
 516         return f;
 517 }
 518
 519 /* caller should hold vxlan->hash_lock */
 520 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 521                                               union vxlan_addr *ip, __be16 port,
 522                                               __be32 vni, __u32 ifindex)
 523 {
 524         struct vxlan_rdst *rd;
 525
 526         list_for_each_entry(rd, &f->remotes, list) {
 527                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 528                     rd->remote_port == port &&
 529                     rd->remote_vni == vni &&
 530                     rd->remote_ifindex == ifindex)
 531                         return rd;
 532         }
 533
 534         return NULL;
 535 }
 536
 537 int vxlan_fdb_find_uc(struct net_device *dev, const u8 *mac, __be32 vni,
 538                       struct switchdev_notifier_vxlan_fdb_info *fdb_info)
 539 {
 540         struct vxlan_dev *vxlan = netdev_priv(dev);
 541         u8 eth_addr[ETH_ALEN + 2] = { 0 };
 542         struct vxlan_rdst *rdst;
 543         struct vxlan_fdb *f;
 544         int rc = 0;
 545
 546         if (is_multicast_ether_addr(mac) ||
 547             is_zero_ether_addr(mac))
 548                 return -EINVAL;
 549
 550         ether_addr_copy(eth_addr, mac);
 551
 552         rcu_read_lock();
 553
 554         f = __vxlan_find_mac(vxlan, eth_addr, vni);
 555         if (!f) {
 556                 rc = -ENOENT;
 557                 goto out;
 558         }
 559
 560         rdst = first_remote_rcu(f);
 561         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, NULL, fdb_info);
 562
 563 out:
 564         rcu_read_unlock();
 565         return rc;
 566 }
 567 EXPORT_SYMBOL_GPL(vxlan_fdb_find_uc);
 568
 569 static int vxlan_fdb_notify_one(struct notifier_block *nb,
 570                                 const struct vxlan_dev *vxlan,
 571                                 const struct vxlan_fdb *f,
 572                                 const struct vxlan_rdst *rdst,
 573                                 struct netlink_ext_ack *extack)
 574 {
 575         struct switchdev_notifier_vxlan_fdb_info fdb_info;
 576         int rc;
 577
 578         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, extack, &fdb_info);
 579         rc = nb->notifier_call(nb, SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE,
 580                                &fdb_info);
 581         return notifier_to_errno(rc);
 582 }
 583
 584 int vxlan_fdb_replay(const struct net_device *dev, __be32 vni,
 585                      struct notifier_block *nb,
 586                      struct netlink_ext_ack *extack)
 587 {
 588         struct vxlan_dev *vxlan;
 589         struct vxlan_rdst *rdst;
 590         struct vxlan_fdb *f;
 591         unsigned int h;
 592         int rc = 0;
 593
 594         if (!netif_is_vxlan(dev))
 595                 return -EINVAL;
 596         vxlan = netdev_priv(dev);
 597
 598         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 599                 spin_lock_bh(&vxlan->hash_lock[h]);
 600                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) {
 601                         if (f->vni == vni) {
 602                                 list_for_each_entry(rdst, &f->remotes, list) {
 603                                         rc = vxlan_fdb_notify_one(nb, vxlan,
 604                                                                   f, rdst,
 605                                                                   extack);
 606                                         if (rc)
 607                                                 goto unlock;
 608                                 }
 609                         }
 610                 }
 611                 spin_unlock_bh(&vxlan->hash_lock[h]);
 612         }
 613         return 0;
 614
 615 unlock:
 616         spin_unlock_bh(&vxlan->hash_lock[h]);
 617         return rc;
 618 }
 619 EXPORT_SYMBOL_GPL(vxlan_fdb_replay);
 620
 621 void vxlan_fdb_clear_offload(const struct net_device *dev, __be32 vni)
 622 {
 623         struct vxlan_dev *vxlan;
 624         struct vxlan_rdst *rdst;
 625         struct vxlan_fdb *f;
 626         unsigned int h;
 627
 628         if (!netif_is_vxlan(dev))
 629                 return;
 630         vxlan = netdev_priv(dev);
 631
 632         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 633                 spin_lock_bh(&vxlan->hash_lock[h]);
 634                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist)
 635                         if (f->vni == vni)
 636                                 list_for_each_entry(rdst, &f->remotes, list)
 637                                         rdst->offloaded = false;
 638                 spin_unlock_bh(&vxlan->hash_lock[h]);
 639         }
 640
 641 }
 642 EXPORT_SYMBOL_GPL(vxlan_fdb_clear_offload);
 643
 644 /* Replace destination of unicast mac */
 645 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 646                              union vxlan_addr *ip, __be16 port, __be32 vni,
 647                              __u32 ifindex, struct vxlan_rdst *oldrd)
 648 {
 649         struct vxlan_rdst *rd;
 650
 651         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 652         if (rd)
 653                 return 0;
 654
 655         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 656         if (!rd)
 657                 return 0;
 658
 659         *oldrd = *rd;
 660         dst_cache_reset(&rd->dst_cache);
 661         rd->remote_ip = *ip;
 662         rd->remote_port = port;
 663         rd->remote_vni = vni;
 664         rd->remote_ifindex = ifindex;
 665         rd->offloaded = false;
 666         return 1;
 667 }
 668
 669 /* Add/update destinations for multicast */
 670 static int vxlan_fdb_append(struct vxlan_fdb *f,
 671                             union vxlan_addr *ip, __be16 port, __be32 vni,
 672                             __u32 ifindex, struct vxlan_rdst **rdp)
 673 {
 674         struct vxlan_rdst *rd;
 675
 676         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 677         if (rd)
 678                 return 0;
 679
 680         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 681         if (rd == NULL)
 682                 return -ENOMEM;
 683
 684         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 685                 kfree(rd);
 686                 return -ENOMEM;
 687         }
 688
 689         rd->remote_ip = *ip;
 690         rd->remote_port = port;
 691         rd->offloaded = false;
 692         rd->remote_vni = vni;
 693         rd->remote_ifindex = ifindex;
 694
 695         list_add_tail_rcu(&rd->list, &f->remotes);
 696
 697         *rdp = rd;
 698         return 1;
 699 }
 700
 701 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 702                                           unsigned int off,
 703                                           struct vxlanhdr *vh, size_t hdrlen,
 704                                           __be32 vni_field,
 705                                           struct gro_remcsum *grc,
 706                                           bool nopartial)
 707 {
 708         size_t start, offset;
 709
 710         if (skb->remcsum_offload)
 711                 return vh;
 712
 713         if (!NAPI_GRO_CB(skb)->csum_valid)
 714                 return NULL;
 715
 716         start = vxlan_rco_start(vni_field);
 717         offset = start + vxlan_rco_offset(vni_field);
 718
 719         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 720                                      start, offset, grc, nopartial);
 721
 722         skb->remcsum_offload = 1;
 723
 724         return vh;
 725 }
 726
 727 static struct sk_buff *vxlan_gro_receive(struct sock *sk,
 728                                          struct list_head *head,
 729                                          struct sk_buff *skb)
 730 {
 731         struct sk_buff *pp = NULL;
 732         struct sk_buff *p;
 733         struct vxlanhdr *vh, *vh2;
 734         unsigned int hlen, off_vx;
 735         int flush = 1;
 736         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 737         __be32 flags;
 738         struct gro_remcsum grc;
 739
 740         skb_gro_remcsum_init(&grc);
 741
 742         off_vx = skb_gro_offset(skb);
 743         hlen = off_vx + sizeof(*vh);
 744         vh   = skb_gro_header_fast(skb, off_vx);
 745         if (skb_gro_header_hard(skb, hlen)) {
 746                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 747                 if (unlikely(!vh))
 748                         goto out;
 749         }
 750
 751         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 752
 753         flags = vh->vx_flags;
 754
 755         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 756                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 757                                        vh->vx_vni, &grc,
 758                                        !!(vs->flags &
 759                                           VXLAN_F_REMCSUM_NOPARTIAL));
 760
 761                 if (!vh)
 762                         goto out;
 763         }
 764
 765         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 766
 767         list_for_each_entry(p, head, list) {
 768                 if (!NAPI_GRO_CB(p)->same_flow)
 769                         continue;
 770
 771                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 772                 if (vh->vx_flags != vh2->vx_flags ||
 773                     vh->vx_vni != vh2->vx_vni) {
 774                         NAPI_GRO_CB(p)->same_flow = 0;
 775                         continue;
 776                 }
 777         }
 778
 779         pp = call_gro_receive(eth_gro_receive, head, skb);
 780         flush = 0;
 781
 782 out:
 783         skb_gro_flush_final_remcsum(skb, pp, flush, &grc);
 784
 785         return pp;
 786 }
 787
 788 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 789 {
 790         /* Sets 'skb->inner_mac_header' since we are always called with
 791          * 'skb->encapsulation' set.
 792          */
 793         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 794 }
 795
 796 static struct vxlan_fdb *vxlan_fdb_alloc(struct vxlan_dev *vxlan,
 797                                          const u8 *mac, __u16 state,
 798                                          __be32 src_vni, __u16 ndm_flags)
 799 {
 800         struct vxlan_fdb *f;
 801
 802         f = kmalloc(sizeof(*f), GFP_ATOMIC);
 803         if (!f)
 804                 return NULL;
 805         f->state = state;
 806         f->flags = ndm_flags;
 807         f->updated = f->used = jiffies;
 808         f->vni = src_vni;
 809         INIT_LIST_HEAD(&f->remotes);
 810         memcpy(f->eth_addr, mac, ETH_ALEN);
 811
 812         return f;
 813 }
 814
 815 static void vxlan_fdb_insert(struct vxlan_dev *vxlan, const u8 *mac,
 816                              __be32 src_vni, struct vxlan_fdb *f)
 817 {
 818         ++vxlan->addrcnt;
 819         hlist_add_head_rcu(&f->hlist,
 820                            vxlan_fdb_head(vxlan, mac, src_vni));
 821 }
 822
 823 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 824                             const u8 *mac, union vxlan_addr *ip,
 825                             __u16 state, __be16 port, __be32 src_vni,
 826                             __be32 vni, __u32 ifindex, __u16 ndm_flags,
 827                             struct vxlan_fdb **fdb)
 828 {
 829         struct vxlan_rdst *rd = NULL;
 830         struct vxlan_fdb *f;
 831         int rc;
 832
 833         if (vxlan->cfg.addrmax &&
 834             vxlan->addrcnt >= vxlan->cfg.addrmax)
 835                 return -ENOSPC;
 836
 837         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 838         f = vxlan_fdb_alloc(vxlan, mac, state, src_vni, ndm_flags);
 839         if (!f)
 840                 return -ENOMEM;
 841
 842         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 843         if (rc < 0) {
 844                 kfree(f);
 845                 return rc;
 846         }
 847
 848         *fdb = f;
 849
 850         return 0;
 851 }
 852
 853 static void __vxlan_fdb_free(struct vxlan_fdb *f)
 854 {
 855         struct vxlan_rdst *rd, *nd;
 856
 857         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 858                 dst_cache_destroy(&rd->dst_cache);
 859                 kfree(rd);
 860         }
 861         kfree(f);
 862 }
 863
 864 static void vxlan_fdb_free(struct rcu_head *head)
 865 {
 866         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 867
 868         __vxlan_fdb_free(f);
 869 }
 870
 871 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 872                               bool do_notify, bool swdev_notify)
 873 {
 874         struct vxlan_rdst *rd;
 875
 876         netdev_dbg(vxlan->dev, "delete %pM\n", f->eth_addr);
 877
 878         --vxlan->addrcnt;
 879         if (do_notify)
 880                 list_for_each_entry(rd, &f->remotes, list)
 881                         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH,
 882                                          swdev_notify, NULL);
 883
 884         hlist_del_rcu(&f->hlist);
 885         call_rcu(&f->rcu, vxlan_fdb_free);
 886 }
 887
 888 static void vxlan_dst_free(struct rcu_head *head)
 889 {
 890         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 891
 892         dst_cache_destroy(&rd->dst_cache);
 893         kfree(rd);
 894 }
 895
 896 static int vxlan_fdb_update_existing(struct vxlan_dev *vxlan,
 897                                      union vxlan_addr *ip,
 898                                      __u16 state, __u16 flags,
 899                                      __be16 port, __be32 vni,
 900                                      __u32 ifindex, __u16 ndm_flags,
 901                                      struct vxlan_fdb *f,
 902                                      bool swdev_notify,
 903                                      struct netlink_ext_ack *extack)
 904 {
 905         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
 906         struct vxlan_rdst *rd = NULL;
 907         struct vxlan_rdst oldrd;
 908         int notify = 0;
 909         int rc = 0;
 910         int err;
 911
 912         /* Do not allow an externally learned entry to take over an entry added
 913          * by the user.
 914          */
 915         if (!(fdb_flags & NTF_EXT_LEARNED) ||
 916             !(f->flags & NTF_VXLAN_ADDED_BY_USER)) {
 917                 if (f->state != state) {
 918                         f->state = state;
 919                         f->updated = jiffies;
 920                         notify = 1;
 921                 }
 922                 if (f->flags != fdb_flags) {
 923                         f->flags = fdb_flags;
 924                         f->updated = jiffies;
 925                         notify = 1;
 926                 }
 927         }
 928
 929         if ((flags & NLM_F_REPLACE)) {
 930                 /* Only change unicasts */
 931                 if (!(is_multicast_ether_addr(f->eth_addr) ||
 932                       is_zero_ether_addr(f->eth_addr))) {
 933                         rc = vxlan_fdb_replace(f, ip, port, vni,
 934                                                ifindex, &oldrd);
 935                         notify |= rc;
 936                 } else {
 937                         return -EOPNOTSUPP;
 938                 }
 939         }
 940         if ((flags & NLM_F_APPEND) &&
 941             (is_multicast_ether_addr(f->eth_addr) ||
 942              is_zero_ether_addr(f->eth_addr))) {
 943                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 944
 945                 if (rc < 0)
 946                         return rc;
 947                 notify |= rc;
 948         }
 949
 950         if (ndm_flags & NTF_USE)
 951                 f->used = jiffies;
 952
 953         if (notify) {
 954                 if (rd == NULL)
 955                         rd = first_remote_rtnl(f);
 956
 957                 err = vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH,
 958                                        swdev_notify, extack);
 959                 if (err)
 960                         goto err_notify;
 961         }
 962
 963         return 0;
 964
 965 err_notify:
 966         if ((flags & NLM_F_REPLACE) && rc)
 967                 *rd = oldrd;
 968         else if ((flags & NLM_F_APPEND) && rc) {
 969                 list_del_rcu(&rd->list);
 970                 call_rcu(&rd->rcu, vxlan_dst_free);
 971         }
 972         return err;
 973 }
 974
 975 static int vxlan_fdb_update_create(struct vxlan_dev *vxlan,
 976                                    const u8 *mac, union vxlan_addr *ip,
 977                                    __u16 state, __u16 flags,
 978                                    __be16 port, __be32 src_vni, __be32 vni,
 979                                    __u32 ifindex, __u16 ndm_flags,
 980                                    bool swdev_notify,
 981                                    struct netlink_ext_ack *extack)
 982 {
 983         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
 984         struct vxlan_fdb *f;
 985         int rc;
 986
 987         /* Disallow replace to add a multicast entry */
 988         if ((flags & NLM_F_REPLACE) &&
 989             (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 990                 return -EOPNOTSUPP;
 991
 992         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 993         rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni,
 994                               vni, ifindex, fdb_flags, &f);
 995         if (rc < 0)
 996                 return rc;
 997
 998         vxlan_fdb_insert(vxlan, mac, src_vni, f);
 999         rc = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH,
1000                               swdev_notify, extack);
1001         if (rc)
1002                 goto err_notify;
1003
1004         return 0;
1005
1006 err_notify:
1007         vxlan_fdb_destroy(vxlan, f, false, false);
1008         return rc;
1009 }
1010
1011 /* Add new entry to forwarding table -- assumes lock held */
1012 static int vxlan_fdb_update(struct vxlan_dev *vxlan,
1013                             const u8 *mac, union vxlan_addr *ip,
1014                             __u16 state, __u16 flags,
1015                             __be16 port, __be32 src_vni, __be32 vni,
1016                             __u32 ifindex, __u16 ndm_flags,
1017                             bool swdev_notify,
1018                             struct netlink_ext_ack *extack)
1019 {
1020         struct vxlan_fdb *f;
1021
1022         f = __vxlan_find_mac(vxlan, mac, src_vni);
1023         if (f) {
1024                 if (flags & NLM_F_EXCL) {
1025                         netdev_dbg(vxlan->dev,
1026                                    "lost race to create %pM\n", mac);
1027                         return -EEXIST;
1028                 }
1029
1030                 return vxlan_fdb_update_existing(vxlan, ip, state, flags, port,
1031                                                  vni, ifindex, ndm_flags, f,
1032                                                  swdev_notify, extack);
1033         } else {
1034                 if (!(flags & NLM_F_CREATE))
1035                         return -ENOENT;
1036
1037                 return vxlan_fdb_update_create(vxlan, mac, ip, state, flags,
1038                                                port, src_vni, vni, ifindex,
1039                                                ndm_flags, swdev_notify, extack);
1040         }
1041 }
1042
1043 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
1044                                   struct vxlan_rdst *rd, bool swdev_notify)
1045 {
1046         list_del_rcu(&rd->list);
1047         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, swdev_notify, NULL);
1048         call_rcu(&rd->rcu, vxlan_dst_free);
1049 }
1050
1051 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
1052                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
1053                            __be32 *vni, u32 *ifindex)
1054 {
1055         struct net *net = dev_net(vxlan->dev);
1056         int err;
1057
1058         if (tb[NDA_DST]) {
1059                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
1060                 if (err)
1061                         return err;
1062         } else {
1063                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
1064                 if (remote->sa.sa_family == AF_INET) {
1065                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
1066                         ip->sa.sa_family = AF_INET;
1067 #if IS_ENABLED(CONFIG_IPV6)
1068                 } else {
1069                         ip->sin6.sin6_addr = in6addr_any;
1070                         ip->sa.sa_family = AF_INET6;
1071 #endif
1072                 }
1073         }
1074
1075         if (tb[NDA_PORT]) {
1076                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
1077                         return -EINVAL;
1078                 *port = nla_get_be16(tb[NDA_PORT]);
1079         } else {
1080                 *port = vxlan->cfg.dst_port;
1081         }
1082
1083         if (tb[NDA_VNI]) {
1084                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
1085                         return -EINVAL;
1086                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1087         } else {
1088                 *vni = vxlan->default_dst.remote_vni;
1089         }
1090
1091         if (tb[NDA_SRC_VNI]) {
1092                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
1093                         return -EINVAL;
1094                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
1095         } else {
1096                 *src_vni = vxlan->default_dst.remote_vni;
1097         }
1098
1099         if (tb[NDA_IFINDEX]) {
1100                 struct net_device *tdev;
1101
1102                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
1103                         return -EINVAL;
1104                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
1105                 tdev = __dev_get_by_index(net, *ifindex);
1106                 if (!tdev)
1107                         return -EADDRNOTAVAIL;
1108         } else {
1109                 *ifindex = 0;
1110         }
1111
1112         return 0;
1113 }
1114
1115 /* Add static entry (via netlink) */
1116 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
1117                          struct net_device *dev,
1118                          const unsigned char *addr, u16 vid, u16 flags,
1119                          struct netlink_ext_ack *extack)
1120 {
1121         struct vxlan_dev *vxlan = netdev_priv(dev);
1122         /* struct net *net = dev_net(vxlan->dev); */
1123         union vxlan_addr ip;
1124         __be16 port;
1125         __be32 src_vni, vni;
1126         u32 ifindex;
1127         u32 hash_index;
1128         int err;
1129
1130         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
1131                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
1132                         ndm->ndm_state);
1133                 return -EINVAL;
1134         }
1135
1136         if (tb[NDA_DST] == NULL)
1137                 return -EINVAL;
1138
1139         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
1140         if (err)
1141                 return err;
1142
1143         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
1144                 return -EAFNOSUPPORT;
1145
1146         hash_index = fdb_head_index(vxlan, addr, src_vni);
1147         spin_lock_bh(&vxlan->hash_lock[hash_index]);
1148         err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags,
1149                                port, src_vni, vni, ifindex,
1150                                ndm->ndm_flags | NTF_VXLAN_ADDED_BY_USER,
1151                                true, extack);
1152         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1153
1154         return err;
1155 }
1156
1157 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
1158                               const unsigned char *addr, union vxlan_addr ip,
1159                               __be16 port, __be32 src_vni, __be32 vni,
1160                               u32 ifindex, bool swdev_notify)
1161 {
1162         struct vxlan_fdb *f;
1163         struct vxlan_rdst *rd = NULL;
1164         int err = -ENOENT;
1165
1166         f = vxlan_find_mac(vxlan, addr, src_vni);
1167         if (!f)
1168                 return err;
1169
1170         if (!vxlan_addr_any(&ip)) {
1171                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
1172                 if (!rd)
1173                         goto out;
1174         }
1175
1176         /* remove a destination if it's not the only one on the list,
1177          * otherwise destroy the fdb entry
1178          */
1179         if (rd && !list_is_singular(&f->remotes)) {
1180                 vxlan_fdb_dst_destroy(vxlan, f, rd, swdev_notify);
1181                 goto out;
1182         }
1183
1184         vxlan_fdb_destroy(vxlan, f, true, swdev_notify);
1185
1186 out:
1187         return 0;
1188 }
1189
1190 /* Delete entry (via netlink) */
1191 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
1192                             struct net_device *dev,
1193                             const unsigned char *addr, u16 vid)
1194 {
1195         struct vxlan_dev *vxlan = netdev_priv(dev);
1196         union vxlan_addr ip;
1197         __be32 src_vni, vni;
1198         __be16 port;
1199         u32 ifindex;
1200         u32 hash_index;
1201         int err;
1202
1203         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
1204         if (err)
1205                 return err;
1206
1207         hash_index = fdb_head_index(vxlan, addr, src_vni);
1208         spin_lock_bh(&vxlan->hash_lock[hash_index]);
1209         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
1210                                  true);
1211         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1212
1213         return err;
1214 }
1215
1216 /* Dump forwarding table */
1217 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1218                           struct net_device *dev,
1219                           struct net_device *filter_dev, int *idx)
1220 {
1221         struct vxlan_dev *vxlan = netdev_priv(dev);
1222         unsigned int h;
1223         int err = 0;
1224
1225         for (h = 0; h < FDB_HASH_SIZE; ++h) {
1226                 struct vxlan_fdb *f;
1227
1228                 rcu_read_lock();
1229                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
1230                         struct vxlan_rdst *rd;
1231
1232                         list_for_each_entry_rcu(rd, &f->remotes, list) {
1233                                 if (*idx < cb->args[2])
1234                                         goto skip;
1235
1236                                 err = vxlan_fdb_info(skb, vxlan, f,
1237                                                      NETLINK_CB(cb->skb).portid,
1238                                                      cb->nlh->nlmsg_seq,
1239                                                      RTM_NEWNEIGH,
1240                                                      NLM_F_MULTI, rd);
1241                                 if (err < 0) {
1242                                         rcu_read_unlock();
1243                                         goto out;
1244                                 }
1245 skip:
1246                                 *idx += 1;
1247                         }
1248                 }
1249                 rcu_read_unlock();
1250         }
1251 out:
1252         return err;
1253 }
1254
1255 static int vxlan_fdb_get(struct sk_buff *skb,
1256                          struct nlattr *tb[],
1257                          struct net_device *dev,
1258                          const unsigned char *addr,
1259                          u16 vid, u32 portid, u32 seq,
1260                          struct netlink_ext_ack *extack)
1261 {
1262         struct vxlan_dev *vxlan = netdev_priv(dev);
1263         struct vxlan_fdb *f;
1264         __be32 vni;
1265         int err;
1266
1267         if (tb[NDA_VNI])
1268                 vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1269         else
1270                 vni = vxlan->default_dst.remote_vni;
1271
1272         rcu_read_lock();
1273
1274         f = __vxlan_find_mac(vxlan, addr, vni);
1275         if (!f) {
1276                 NL_SET_ERR_MSG(extack, "Fdb entry not found");
1277                 err = -ENOENT;
1278                 goto errout;
1279         }
1280
1281         err = vxlan_fdb_info(skb, vxlan, f, portid, seq,
1282                              RTM_NEWNEIGH, 0, first_remote_rcu(f));
1283 errout:
1284         rcu_read_unlock();
1285         return err;
1286 }
1287
1288 /* Watch incoming packets to learn mapping between Ethernet address
1289  * and Tunnel endpoint.
1290  * Return true if packet is bogus and should be dropped.
1291  */
1292 static bool vxlan_snoop(struct net_device *dev,
1293                         union vxlan_addr *src_ip, const u8 *src_mac,
1294                         u32 src_ifindex, __be32 vni)
1295 {
1296         struct vxlan_dev *vxlan = netdev_priv(dev);
1297         struct vxlan_fdb *f;
1298         u32 ifindex = 0;
1299
1300 #if IS_ENABLED(CONFIG_IPV6)
1301         if (src_ip->sa.sa_family == AF_INET6 &&
1302             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
1303                 ifindex = src_ifindex;
1304 #endif
1305
1306         f = vxlan_find_mac(vxlan, src_mac, vni);
1307         if (likely(f)) {
1308                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1309
1310                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
1311                            rdst->remote_ifindex == ifindex))
1312                         return false;
1313
1314                 /* Don't migrate static entries, drop packets */
1315                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
1316                         return true;
1317
1318                 if (net_ratelimit())
1319                         netdev_info(dev,
1320                                     "%pM migrated from %pIS to %pIS\n",
1321                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1322
1323                 rdst->remote_ip = *src_ip;
1324                 f->updated = jiffies;
1325                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH, true, NULL);
1326         } else {
1327                 u32 hash_index = fdb_head_index(vxlan, src_mac, vni);
1328
1329                 /* learned new entry */
1330                 spin_lock(&vxlan->hash_lock[hash_index]);
1331
1332                 /* close off race between vxlan_flush and incoming packets */
1333                 if (netif_running(dev))
1334                         vxlan_fdb_update(vxlan, src_mac, src_ip,
1335                                          NUD_REACHABLE,
1336                                          NLM_F_EXCL|NLM_F_CREATE,
1337                                          vxlan->cfg.dst_port,
1338                                          vni,
1339                                          vxlan->default_dst.remote_vni,
1340                                          ifindex, NTF_SELF, true, NULL);
1341                 spin_unlock(&vxlan->hash_lock[hash_index]);
1342         }
1343
1344         return false;
1345 }
1346
1347 /* See if multicast group is already in use by other ID */
1348 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1349 {
1350         struct vxlan_dev *vxlan;
1351         struct vxlan_sock *sock4;
1352 #if IS_ENABLED(CONFIG_IPV6)
1353         struct vxlan_sock *sock6;
1354 #endif
1355         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1356
1357         sock4 = rtnl_dereference(dev->vn4_sock);
1358
1359         /* The vxlan_sock is only used by dev, leaving group has
1360          * no effect on other vxlan devices.
1361          */
1362         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1363                 return false;
1364 #if IS_ENABLED(CONFIG_IPV6)
1365         sock6 = rtnl_dereference(dev->vn6_sock);
1366         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1367                 return false;
1368 #endif
1369
1370         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1371                 if (!netif_running(vxlan->dev) || vxlan == dev)
1372                         continue;
1373
1374                 if (family == AF_INET &&
1375                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1376                         continue;
1377 #if IS_ENABLED(CONFIG_IPV6)
1378                 if (family == AF_INET6 &&
1379                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1380                         continue;
1381 #endif
1382
1383                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1384                                       &dev->default_dst.remote_ip))
1385                         continue;
1386
1387                 if (vxlan->default_dst.remote_ifindex !=
1388                     dev->default_dst.remote_ifindex)
1389                         continue;
1390
1391                 return true;
1392         }
1393
1394         return false;
1395 }
1396
1397 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1398 {
1399         struct vxlan_net *vn;
1400
1401         if (!vs)
1402                 return false;
1403         if (!refcount_dec_and_test(&vs->refcnt))
1404                 return false;
1405
1406         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1407         spin_lock(&vn->sock_lock);
1408         hlist_del_rcu(&vs->hlist);
1409         udp_tunnel_notify_del_rx_port(vs->sock,
1410                                       (vs->flags & VXLAN_F_GPE) ?
1411                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1412                                       UDP_TUNNEL_TYPE_VXLAN);
1413         spin_unlock(&vn->sock_lock);
1414
1415         return true;
1416 }
1417
1418 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1419 {
1420         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1421 #if IS_ENABLED(CONFIG_IPV6)
1422         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1423
1424         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1425 #endif
1426
1427         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1428         synchronize_net();
1429
1430         vxlan_vs_del_dev(vxlan);
1431
1432         if (__vxlan_sock_release_prep(sock4)) {
1433                 udp_tunnel_sock_release(sock4->sock);
1434                 kfree(sock4);
1435         }
1436
1437 #if IS_ENABLED(CONFIG_IPV6)
1438         if (__vxlan_sock_release_prep(sock6)) {
1439                 udp_tunnel_sock_release(sock6->sock);
1440                 kfree(sock6);
1441         }
1442 #endif
1443 }
1444
1445 /* Update multicast group membership when first VNI on
1446  * multicast address is brought up
1447  */
1448 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1449 {
1450         struct sock *sk;
1451         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1452         int ifindex = vxlan->default_dst.remote_ifindex;
1453         int ret = -EINVAL;
1454
1455         if (ip->sa.sa_family == AF_INET) {
1456                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1457                 struct ip_mreqn mreq = {
1458                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1459                         .imr_ifindex            = ifindex,
1460                 };
1461
1462                 sk = sock4->sock->sk;
1463                 lock_sock(sk);
1464                 ret = ip_mc_join_group(sk, &mreq);
1465                 release_sock(sk);
1466 #if IS_ENABLED(CONFIG_IPV6)
1467         } else {
1468                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1469
1470                 sk = sock6->sock->sk;
1471                 lock_sock(sk);
1472                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1473                                                    &ip->sin6.sin6_addr);
1474                 release_sock(sk);
1475 #endif
1476         }
1477
1478         return ret;
1479 }
1480
1481 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1482 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1483 {
1484         struct sock *sk;
1485         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1486         int ifindex = vxlan->default_dst.remote_ifindex;
1487         int ret = -EINVAL;
1488
1489         if (ip->sa.sa_family == AF_INET) {
1490                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1491                 struct ip_mreqn mreq = {
1492                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1493                         .imr_ifindex            = ifindex,
1494                 };
1495
1496                 sk = sock4->sock->sk;
1497                 lock_sock(sk);
1498                 ret = ip_mc_leave_group(sk, &mreq);
1499                 release_sock(sk);
1500 #if IS_ENABLED(CONFIG_IPV6)
1501         } else {
1502                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1503
1504                 sk = sock6->sock->sk;
1505                 lock_sock(sk);
1506                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1507                                                    &ip->sin6.sin6_addr);
1508                 release_sock(sk);
1509 #endif
1510         }
1511
1512         return ret;
1513 }
1514
1515 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1516                           struct sk_buff *skb, u32 vxflags)
1517 {
1518         size_t start, offset;
1519
1520         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1521                 goto out;
1522
1523         start = vxlan_rco_start(unparsed->vx_vni);
1524         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1525
1526         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1527                 return false;
1528
1529         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1530                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1531 out:
1532         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1533         unparsed->vx_vni &= VXLAN_VNI_MASK;
1534         return true;
1535 }
1536
1537 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1538                                 struct sk_buff *skb, u32 vxflags,
1539                                 struct vxlan_metadata *md)
1540 {
1541         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1542         struct metadata_dst *tun_dst;
1543
1544         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1545                 goto out;
1546
1547         md->gbp = ntohs(gbp->policy_id);
1548
1549         tun_dst = (struct metadata_dst *)skb_dst(skb);
1550         if (tun_dst) {
1551                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1552                 tun_dst->u.tun_info.options_len = sizeof(*md);
1553         }
1554         if (gbp->dont_learn)
1555                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1556
1557         if (gbp->policy_applied)
1558                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1559
1560         /* In flow-based mode, GBP is carried in dst_metadata */
1561         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1562                 skb->mark = md->gbp;
1563 out:
1564         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1565 }
1566
1567 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1568                                 __be16 *protocol,
1569                                 struct sk_buff *skb, u32 vxflags)
1570 {
1571         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1572
1573         /* Need to have Next Protocol set for interfaces in GPE mode. */
1574         if (!gpe->np_applied)
1575                 return false;
1576         /* "The initial version is 0. If a receiver does not support the
1577          * version indicated it MUST drop the packet.
1578          */
1579         if (gpe->version != 0)
1580                 return false;
1581         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1582          * processing MUST occur." However, we don't implement OAM
1583          * processing, thus drop the packet.
1584          */
1585         if (gpe->oam_flag)
1586                 return false;
1587
1588         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1589         if (!*protocol)
1590                 return false;
1591
1592         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1593         return true;
1594 }
1595
1596 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1597                           struct vxlan_sock *vs,
1598                           struct sk_buff *skb, __be32 vni)
1599 {
1600         union vxlan_addr saddr;
1601         u32 ifindex = skb->dev->ifindex;
1602
1603         skb_reset_mac_header(skb);
1604         skb->protocol = eth_type_trans(skb, vxlan->dev);
1605         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1606
1607         /* Ignore packet loops (and multicast echo) */
1608         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1609                 return false;
1610
1611         /* Get address from the outer IP header */
1612         if (vxlan_get_sk_family(vs) == AF_INET) {
1613                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1614                 saddr.sa.sa_family = AF_INET;
1615 #if IS_ENABLED(CONFIG_IPV6)
1616         } else {
1617                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1618                 saddr.sa.sa_family = AF_INET6;
1619 #endif
1620         }
1621
1622         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1623             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1624                 return false;
1625
1626         return true;
1627 }
1628
1629 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1630                                   struct sk_buff *skb)
1631 {
1632         int err = 0;
1633
1634         if (vxlan_get_sk_family(vs) == AF_INET)
1635                 err = IP_ECN_decapsulate(oiph, skb);
1636 #if IS_ENABLED(CONFIG_IPV6)
1637         else
1638                 err = IP6_ECN_decapsulate(oiph, skb);
1639 #endif
1640
1641         if (unlikely(err) && log_ecn_error) {
1642                 if (vxlan_get_sk_family(vs) == AF_INET)
1643                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1644                                              &((struct iphdr *)oiph)->saddr,
1645                                              ((struct iphdr *)oiph)->tos);
1646                 else
1647                         net_info_ratelimited("non-ECT from %pI6\n",
1648                                              &((struct ipv6hdr *)oiph)->saddr);
1649         }
1650         return err <= 1;
1651 }
1652
1653 /* Callback from net/ipv4/udp.c to receive packets */
1654 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1655 {
1656         struct pcpu_sw_netstats *stats;
1657         struct vxlan_dev *vxlan;
1658         struct vxlan_sock *vs;
1659         struct vxlanhdr unparsed;
1660         struct vxlan_metadata _md;
1661         struct vxlan_metadata *md = &_md;
1662         __be16 protocol = htons(ETH_P_TEB);
1663         bool raw_proto = false;
1664         void *oiph;
1665         __be32 vni = 0;
1666
1667         /* Need UDP and VXLAN header to be present */
1668         if (!pskb_may_pull(skb, VXLAN_HLEN))
1669                 goto drop;
1670
1671         unparsed = *vxlan_hdr(skb);
1672         /* VNI flag always required to be set */
1673         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1674                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1675                            ntohl(vxlan_hdr(skb)->vx_flags),
1676                            ntohl(vxlan_hdr(skb)->vx_vni));
1677                 /* Return non vxlan pkt */
1678                 goto drop;
1679         }
1680         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1681         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1682
1683         vs = rcu_dereference_sk_user_data(sk);
1684         if (!vs)
1685                 goto drop;
1686
1687         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1688
1689         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1690         if (!vxlan)
1691                 goto drop;
1692
1693         /* For backwards compatibility, only allow reserved fields to be
1694          * used by VXLAN extensions if explicitly requested.
1695          */
1696         if (vs->flags & VXLAN_F_GPE) {
1697                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1698                         goto drop;
1699                 raw_proto = true;
1700         }
1701
1702         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1703                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1704                         goto drop;
1705
1706         if (vxlan_collect_metadata(vs)) {
1707                 struct metadata_dst *tun_dst;
1708
1709                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1710                                          key32_to_tunnel_id(vni), sizeof(*md));
1711
1712                 if (!tun_dst)
1713                         goto drop;
1714
1715                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1716
1717                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1718         } else {
1719                 memset(md, 0, sizeof(*md));
1720         }
1721
1722         if (vs->flags & VXLAN_F_REMCSUM_RX)
1723                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1724                         goto drop;
1725         if (vs->flags & VXLAN_F_GBP)
1726                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1727         /* Note that GBP and GPE can never be active together. This is
1728          * ensured in vxlan_dev_configure.
1729          */
1730
1731         if (unparsed.vx_flags || unparsed.vx_vni) {
1732                 /* If there are any unprocessed flags remaining treat
1733                  * this as a malformed packet. This behavior diverges from
1734                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1735                  * in reserved fields are to be ignored. The approach here
1736                  * maintains compatibility with previous stack code, and also
1737                  * is more robust and provides a little more security in
1738                  * adding extensions to VXLAN.
1739                  */
1740                 goto drop;
1741         }
1742
1743         if (!raw_proto) {
1744                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1745                         goto drop;
1746         } else {
1747                 skb_reset_mac_header(skb);
1748                 skb->dev = vxlan->dev;
1749                 skb->pkt_type = PACKET_HOST;
1750         }
1751
1752         oiph = skb_network_header(skb);
1753         skb_reset_network_header(skb);
1754
1755         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1756                 ++vxlan->dev->stats.rx_frame_errors;
1757                 ++vxlan->dev->stats.rx_errors;
1758                 goto drop;
1759         }
1760
1761         rcu_read_lock();
1762
1763         if (unlikely(!(vxlan->dev->flags & IFF_UP))) {
1764                 rcu_read_unlock();
1765                 atomic_long_inc(&vxlan->dev->rx_dropped);
1766                 goto drop;
1767         }
1768
1769         stats = this_cpu_ptr(vxlan->dev->tstats);
1770         u64_stats_update_begin(&stats->syncp);
1771         stats->rx_packets++;
1772         stats->rx_bytes += skb->len;
1773         u64_stats_update_end(&stats->syncp);
1774
1775         gro_cells_receive(&vxlan->gro_cells, skb);
1776
1777         rcu_read_unlock();
1778
1779         return 0;
1780
1781 drop:
1782         /* Consume bad packet */
1783         kfree_skb(skb);
1784         return 0;
1785 }
1786
1787 /* Callback from net/ipv{4,6}/udp.c to check that we have a VNI for errors */
1788 static int vxlan_err_lookup(struct sock *sk, struct sk_buff *skb)
1789 {
1790         struct vxlan_dev *vxlan;
1791         struct vxlan_sock *vs;
1792         struct vxlanhdr *hdr;
1793         __be32 vni;
1794
1795         if (!pskb_may_pull(skb, skb_transport_offset(skb) + VXLAN_HLEN))
1796                 return -EINVAL;
1797
1798         hdr = vxlan_hdr(skb);
1799
1800         if (!(hdr->vx_flags & VXLAN_HF_VNI))
1801                 return -EINVAL;
1802
1803         vs = rcu_dereference_sk_user_data(sk);
1804         if (!vs)
1805                 return -ENOENT;
1806
1807         vni = vxlan_vni(hdr->vx_vni);
1808         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1809         if (!vxlan)
1810                 return -ENOENT;
1811
1812         return 0;
1813 }
1814
1815 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1816 {
1817         struct vxlan_dev *vxlan = netdev_priv(dev);
1818         struct arphdr *parp;
1819         u8 *arpptr, *sha;
1820         __be32 sip, tip;
1821         struct neighbour *n;
1822
1823         if (dev->flags & IFF_NOARP)
1824                 goto out;
1825
1826         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1827                 dev->stats.tx_dropped++;
1828                 goto out;
1829         }
1830         parp = arp_hdr(skb);
1831
1832         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1833              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1834             parp->ar_pro != htons(ETH_P_IP) ||
1835             parp->ar_op != htons(ARPOP_REQUEST) ||
1836             parp->ar_hln != dev->addr_len ||
1837             parp->ar_pln != 4)
1838                 goto out;
1839         arpptr = (u8 *)parp + sizeof(struct arphdr);
1840         sha = arpptr;
1841         arpptr += dev->addr_len;        /* sha */
1842         memcpy(&sip, arpptr, sizeof(sip));
1843         arpptr += sizeof(sip);
1844         arpptr += dev->addr_len;        /* tha */
1845         memcpy(&tip, arpptr, sizeof(tip));
1846
1847         if (ipv4_is_loopback(tip) ||
1848             ipv4_is_multicast(tip))
1849                 goto out;
1850
1851         n = neigh_lookup(&arp_tbl, &tip, dev);
1852
1853         if (n) {
1854                 struct vxlan_fdb *f;
1855                 struct sk_buff  *reply;
1856
1857                 if (!(n->nud_state & NUD_CONNECTED)) {
1858                         neigh_release(n);
1859                         goto out;
1860                 }
1861
1862                 f = vxlan_find_mac(vxlan, n->ha, vni);
1863                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1864                         /* bridge-local neighbor */
1865                         neigh_release(n);
1866                         goto out;
1867                 }
1868
1869                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1870                                 n->ha, sha);
1871
1872                 neigh_release(n);
1873
1874                 if (reply == NULL)
1875                         goto out;
1876
1877                 skb_reset_mac_header(reply);
1878                 __skb_pull(reply, skb_network_offset(reply));
1879                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1880                 reply->pkt_type = PACKET_HOST;
1881
1882                 if (netif_rx_ni(reply) == NET_RX_DROP)
1883                         dev->stats.rx_dropped++;
1884         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1885                 union vxlan_addr ipa = {
1886                         .sin.sin_addr.s_addr = tip,
1887                         .sin.sin_family = AF_INET,
1888                 };
1889
1890                 vxlan_ip_miss(dev, &ipa);
1891         }
1892 out:
1893         consume_skb(skb);
1894         return NETDEV_TX_OK;
1895 }
1896
1897 #if IS_ENABLED(CONFIG_IPV6)
1898 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1899         struct neighbour *n, bool isrouter)
1900 {
1901         struct net_device *dev = request->dev;
1902         struct sk_buff *reply;
1903         struct nd_msg *ns, *na;
1904         struct ipv6hdr *pip6;
1905         u8 *daddr;
1906         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1907         int ns_olen;
1908         int i, len;
1909
1910         if (dev == NULL || !pskb_may_pull(request, request->len))
1911                 return NULL;
1912
1913         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1914                 sizeof(*na) + na_olen + dev->needed_tailroom;
1915         reply = alloc_skb(len, GFP_ATOMIC);
1916         if (reply == NULL)
1917                 return NULL;
1918
1919         reply->protocol = htons(ETH_P_IPV6);
1920         reply->dev = dev;
1921         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1922         skb_push(reply, sizeof(struct ethhdr));
1923         skb_reset_mac_header(reply);
1924
1925         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
1926
1927         daddr = eth_hdr(request)->h_source;
1928         ns_olen = request->len - skb_network_offset(request) -
1929                 sizeof(struct ipv6hdr) - sizeof(*ns);
1930         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1931                 if (!ns->opt[i + 1]) {
1932                         kfree_skb(reply);
1933                         return NULL;
1934                 }
1935                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1936                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1937                         break;
1938                 }
1939         }
1940
1941         /* Ethernet header */
1942         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1943         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1944         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1945         reply->protocol = htons(ETH_P_IPV6);
1946
1947         skb_pull(reply, sizeof(struct ethhdr));
1948         skb_reset_network_header(reply);
1949         skb_put(reply, sizeof(struct ipv6hdr));
1950
1951         /* IPv6 header */
1952
1953         pip6 = ipv6_hdr(reply);
1954         memset(pip6, 0, sizeof(struct ipv6hdr));
1955         pip6->version = 6;
1956         pip6->priority = ipv6_hdr(request)->priority;
1957         pip6->nexthdr = IPPROTO_ICMPV6;
1958         pip6->hop_limit = 255;
1959         pip6->daddr = ipv6_hdr(request)->saddr;
1960         pip6->saddr = *(struct in6_addr *)n->primary_key;
1961
1962         skb_pull(reply, sizeof(struct ipv6hdr));
1963         skb_reset_transport_header(reply);
1964
1965         /* Neighbor Advertisement */
1966         na = skb_put_zero(reply, sizeof(*na) + na_olen);
1967         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1968         na->icmph.icmp6_router = isrouter;
1969         na->icmph.icmp6_override = 1;
1970         na->icmph.icmp6_solicited = 1;
1971         na->target = ns->target;
1972         ether_addr_copy(&na->opt[2], n->ha);
1973         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1974         na->opt[1] = na_olen >> 3;
1975
1976         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1977                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1978                 csum_partial(na, sizeof(*na)+na_olen, 0));
1979
1980         pip6->payload_len = htons(sizeof(*na)+na_olen);
1981
1982         skb_push(reply, sizeof(struct ipv6hdr));
1983
1984         reply->ip_summed = CHECKSUM_UNNECESSARY;
1985
1986         return reply;
1987 }
1988
1989 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1990 {
1991         struct vxlan_dev *vxlan = netdev_priv(dev);
1992         const struct in6_addr *daddr;
1993         const struct ipv6hdr *iphdr;
1994         struct inet6_dev *in6_dev;
1995         struct neighbour *n;
1996         struct nd_msg *msg;
1997
1998         rcu_read_lock();
1999         in6_dev = __in6_dev_get(dev);
2000         if (!in6_dev)
2001                 goto out;
2002
2003         iphdr = ipv6_hdr(skb);
2004         daddr = &iphdr->daddr;
2005         msg = (struct nd_msg *)(iphdr + 1);
2006
2007         if (ipv6_addr_loopback(daddr) ||
2008             ipv6_addr_is_multicast(&msg->target))
2009                 goto out;
2010
2011         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
2012
2013         if (n) {
2014                 struct vxlan_fdb *f;
2015                 struct sk_buff *reply;
2016
2017                 if (!(n->nud_state & NUD_CONNECTED)) {
2018                         neigh_release(n);
2019                         goto out;
2020                 }
2021
2022                 f = vxlan_find_mac(vxlan, n->ha, vni);
2023                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2024                         /* bridge-local neighbor */
2025                         neigh_release(n);
2026                         goto out;
2027                 }
2028
2029                 reply = vxlan_na_create(skb, n,
2030                                         !!(f ? f->flags & NTF_ROUTER : 0));
2031
2032                 neigh_release(n);
2033
2034                 if (reply == NULL)
2035                         goto out;
2036
2037                 if (netif_rx_ni(reply) == NET_RX_DROP)
2038                         dev->stats.rx_dropped++;
2039
2040         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2041                 union vxlan_addr ipa = {
2042                         .sin6.sin6_addr = msg->target,
2043                         .sin6.sin6_family = AF_INET6,
2044                 };
2045
2046                 vxlan_ip_miss(dev, &ipa);
2047         }
2048
2049 out:
2050         rcu_read_unlock();
2051         consume_skb(skb);
2052         return NETDEV_TX_OK;
2053 }
2054 #endif
2055
2056 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
2057 {
2058         struct vxlan_dev *vxlan = netdev_priv(dev);
2059         struct neighbour *n;
2060
2061         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
2062                 return false;
2063
2064         n = NULL;
2065         switch (ntohs(eth_hdr(skb)->h_proto)) {
2066         case ETH_P_IP:
2067         {
2068                 struct iphdr *pip;
2069
2070                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
2071                         return false;
2072                 pip = ip_hdr(skb);
2073                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
2074                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2075                         union vxlan_addr ipa = {
2076                                 .sin.sin_addr.s_addr = pip->daddr,
2077                                 .sin.sin_family = AF_INET,
2078                         };
2079
2080                         vxlan_ip_miss(dev, &ipa);
2081                         return false;
2082                 }
2083
2084                 break;
2085         }
2086 #if IS_ENABLED(CONFIG_IPV6)
2087         case ETH_P_IPV6:
2088         {
2089                 struct ipv6hdr *pip6;
2090
2091                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
2092                         return false;
2093                 pip6 = ipv6_hdr(skb);
2094                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
2095                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2096                         union vxlan_addr ipa = {
2097                                 .sin6.sin6_addr = pip6->daddr,
2098                                 .sin6.sin6_family = AF_INET6,
2099                         };
2100
2101                         vxlan_ip_miss(dev, &ipa);
2102                         return false;
2103                 }
2104
2105                 break;
2106         }
2107 #endif
2108         default:
2109                 return false;
2110         }
2111
2112         if (n) {
2113                 bool diff;
2114
2115                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
2116                 if (diff) {
2117                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
2118                                 dev->addr_len);
2119                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
2120                 }
2121                 neigh_release(n);
2122                 return diff;
2123         }
2124
2125         return false;
2126 }
2127
2128 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
2129                                 struct vxlan_metadata *md)
2130 {
2131         struct vxlanhdr_gbp *gbp;
2132
2133         if (!md->gbp)
2134                 return;
2135
2136         gbp = (struct vxlanhdr_gbp *)vxh;
2137         vxh->vx_flags |= VXLAN_HF_GBP;
2138
2139         if (md->gbp & VXLAN_GBP_DONT_LEARN)
2140                 gbp->dont_learn = 1;
2141
2142         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
2143                 gbp->policy_applied = 1;
2144
2145         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
2146 }
2147
2148 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
2149                                __be16 protocol)
2150 {
2151         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
2152
2153         gpe->np_applied = 1;
2154         gpe->next_protocol = tun_p_from_eth_p(protocol);
2155         if (!gpe->next_protocol)
2156                 return -EPFNOSUPPORT;
2157         return 0;
2158 }
2159
2160 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
2161                            int iphdr_len, __be32 vni,
2162                            struct vxlan_metadata *md, u32 vxflags,
2163                            bool udp_sum)
2164 {
2165         struct vxlanhdr *vxh;
2166         int min_headroom;
2167         int err;
2168         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
2169         __be16 inner_protocol = htons(ETH_P_TEB);
2170
2171         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
2172             skb->ip_summed == CHECKSUM_PARTIAL) {
2173                 int csum_start = skb_checksum_start_offset(skb);
2174
2175                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
2176                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
2177                     (skb->csum_offset == offsetof(struct udphdr, check) ||
2178                      skb->csum_offset == offsetof(struct tcphdr, check)))
2179                         type |= SKB_GSO_TUNNEL_REMCSUM;
2180         }
2181
2182         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
2183                         + VXLAN_HLEN + iphdr_len;
2184
2185         /* Need space for new headers (invalidates iph ptr) */
2186         err = skb_cow_head(skb, min_headroom);
2187         if (unlikely(err))
2188                 return err;
2189
2190         err = iptunnel_handle_offloads(skb, type);
2191         if (err)
2192                 return err;
2193
2194         vxh = __skb_push(skb, sizeof(*vxh));
2195         vxh->vx_flags = VXLAN_HF_VNI;
2196         vxh->vx_vni = vxlan_vni_field(vni);
2197
2198         if (type & SKB_GSO_TUNNEL_REMCSUM) {
2199                 unsigned int start;
2200
2201                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
2202                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
2203                 vxh->vx_flags |= VXLAN_HF_RCO;
2204
2205                 if (!skb_is_gso(skb)) {
2206                         skb->ip_summed = CHECKSUM_NONE;
2207                         skb->encapsulation = 0;
2208                 }
2209         }
2210
2211         if (vxflags & VXLAN_F_GBP)
2212                 vxlan_build_gbp_hdr(vxh, vxflags, md);
2213         if (vxflags & VXLAN_F_GPE) {
2214                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
2215                 if (err < 0)
2216                         return err;
2217                 inner_protocol = skb->protocol;
2218         }
2219
2220         skb_set_inner_protocol(skb, inner_protocol);
2221         return 0;
2222 }
2223
2224 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
2225                                       struct vxlan_sock *sock4,
2226                                       struct sk_buff *skb, int oif, u8 tos,
2227                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
2228                                       struct dst_cache *dst_cache,
2229                                       const struct ip_tunnel_info *info)
2230 {
2231         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2232         struct rtable *rt = NULL;
2233         struct flowi4 fl4;
2234
2235         if (!sock4)
2236                 return ERR_PTR(-EIO);
2237
2238         if (tos && !info)
2239                 use_cache = false;
2240         if (use_cache) {
2241                 rt = dst_cache_get_ip4(dst_cache, saddr);
2242                 if (rt)
2243                         return rt;
2244         }
2245
2246         memset(&fl4, 0, sizeof(fl4));
2247         fl4.flowi4_oif = oif;
2248         fl4.flowi4_tos = RT_TOS(tos);
2249         fl4.flowi4_mark = skb->mark;
2250         fl4.flowi4_proto = IPPROTO_UDP;
2251         fl4.daddr = daddr;
2252         fl4.saddr = *saddr;
2253         fl4.fl4_dport = dport;
2254         fl4.fl4_sport = sport;
2255
2256         rt = ip_route_output_key(vxlan->net, &fl4);
2257         if (!IS_ERR(rt)) {
2258                 if (rt->dst.dev == dev) {
2259                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
2260                         ip_rt_put(rt);
2261                         return ERR_PTR(-ELOOP);
2262                 }
2263
2264                 *saddr = fl4.saddr;
2265                 if (use_cache)
2266                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2267         } else {
2268                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
2269                 return ERR_PTR(-ENETUNREACH);
2270         }
2271         return rt;
2272 }
2273
2274 #if IS_ENABLED(CONFIG_IPV6)
2275 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
2276                                           struct net_device *dev,
2277                                           struct vxlan_sock *sock6,
2278                                           struct sk_buff *skb, int oif, u8 tos,
2279                                           __be32 label,
2280                                           const struct in6_addr *daddr,
2281                                           struct in6_addr *saddr,
2282                                           __be16 dport, __be16 sport,
2283                                           struct dst_cache *dst_cache,
2284                                           const struct ip_tunnel_info *info)
2285 {
2286         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2287         struct dst_entry *ndst;
2288         struct flowi6 fl6;
2289
2290         if (!sock6)
2291                 return ERR_PTR(-EIO);
2292
2293         if (tos && !info)
2294                 use_cache = false;
2295         if (use_cache) {
2296                 ndst = dst_cache_get_ip6(dst_cache, saddr);
2297                 if (ndst)
2298                         return ndst;
2299         }
2300
2301         memset(&fl6, 0, sizeof(fl6));
2302         fl6.flowi6_oif = oif;
2303         fl6.daddr = *daddr;
2304         fl6.saddr = *saddr;
2305         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
2306         fl6.flowi6_mark = skb->mark;
2307         fl6.flowi6_proto = IPPROTO_UDP;
2308         fl6.fl6_dport = dport;
2309         fl6.fl6_sport = sport;
2310
2311         ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk,
2312                                                &fl6, NULL);
2313         if (unlikely(IS_ERR(ndst))) {
2314                 netdev_dbg(dev, "no route to %pI6\n", daddr);
2315                 return ERR_PTR(-ENETUNREACH);
2316         }
2317
2318         if (unlikely(ndst->dev == dev)) {
2319                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
2320                 dst_release(ndst);
2321                 return ERR_PTR(-ELOOP);
2322         }
2323
2324         *saddr = fl6.saddr;
2325         if (use_cache)
2326                 dst_cache_set_ip6(dst_cache, ndst, saddr);
2327         return ndst;
2328 }
2329 #endif
2330
2331 /* Bypass encapsulation if the destination is local */
2332 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2333                                struct vxlan_dev *dst_vxlan, __be32 vni)
2334 {
2335         struct pcpu_sw_netstats *tx_stats, *rx_stats;
2336         union vxlan_addr loopback;
2337         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2338         struct net_device *dev;
2339         int len = skb->len;
2340
2341         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2342         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2343         skb->pkt_type = PACKET_HOST;
2344         skb->encapsulation = 0;
2345         skb->dev = dst_vxlan->dev;
2346         __skb_pull(skb, skb_network_offset(skb));
2347
2348         if (remote_ip->sa.sa_family == AF_INET) {
2349                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2350                 loopback.sa.sa_family =  AF_INET;
2351 #if IS_ENABLED(CONFIG_IPV6)
2352         } else {
2353                 loopback.sin6.sin6_addr = in6addr_loopback;
2354                 loopback.sa.sa_family =  AF_INET6;
2355 #endif
2356         }
2357
2358         rcu_read_lock();
2359         dev = skb->dev;
2360         if (unlikely(!(dev->flags & IFF_UP))) {
2361                 kfree_skb(skb);
2362                 goto drop;
2363         }
2364
2365         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
2366                 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni);
2367
2368         u64_stats_update_begin(&tx_stats->syncp);
2369         tx_stats->tx_packets++;
2370         tx_stats->tx_bytes += len;
2371         u64_stats_update_end(&tx_stats->syncp);
2372
2373         if (netif_rx(skb) == NET_RX_SUCCESS) {
2374                 u64_stats_update_begin(&rx_stats->syncp);
2375                 rx_stats->rx_packets++;
2376                 rx_stats->rx_bytes += len;
2377                 u64_stats_update_end(&rx_stats->syncp);
2378         } else {
2379 drop:
2380                 dev->stats.rx_dropped++;
2381         }
2382         rcu_read_unlock();
2383 }
2384
2385 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2386                                  struct vxlan_dev *vxlan,
2387                                  union vxlan_addr *daddr,
2388                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2389                                  struct dst_entry *dst,
2390                                  u32 rt_flags)
2391 {
2392 #if IS_ENABLED(CONFIG_IPV6)
2393         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2394          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2395          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2396          */
2397         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2398 #endif
2399         /* Bypass encapsulation if the destination is local */
2400         if (rt_flags & RTCF_LOCAL &&
2401             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2402                 struct vxlan_dev *dst_vxlan;
2403
2404                 dst_release(dst);
2405                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2406                                            daddr->sa.sa_family, dst_port,
2407                                            vxlan->cfg.flags);
2408                 if (!dst_vxlan) {
2409                         dev->stats.tx_errors++;
2410                         kfree_skb(skb);
2411
2412                         return -ENOENT;
2413                 }
2414                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2415                 return 1;
2416         }
2417
2418         return 0;
2419 }
2420
2421 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2422                            __be32 default_vni, struct vxlan_rdst *rdst,
2423                            bool did_rsc)
2424 {
2425         struct dst_cache *dst_cache;
2426         struct ip_tunnel_info *info;
2427         struct vxlan_dev *vxlan = netdev_priv(dev);
2428         const struct iphdr *old_iph = ip_hdr(skb);
2429         union vxlan_addr *dst;
2430         union vxlan_addr remote_ip, local_ip;
2431         struct vxlan_metadata _md;
2432         struct vxlan_metadata *md = &_md;
2433         __be16 src_port = 0, dst_port;
2434         struct dst_entry *ndst = NULL;
2435         __be32 vni, label;
2436         __u8 tos, ttl;
2437         int ifindex;
2438         int err;
2439         u32 flags = vxlan->cfg.flags;
2440         bool udp_sum = false;
2441         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2442
2443         info = skb_tunnel_info(skb);
2444
2445         if (rdst) {
2446                 dst = &rdst->remote_ip;
2447                 if (vxlan_addr_any(dst)) {
2448                         if (did_rsc) {
2449                                 /* short-circuited back to local bridge */
2450                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2451                                 return;
2452                         }
2453                         goto drop;
2454                 }
2455
2456                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2457                 vni = (rdst->remote_vni) ? : default_vni;
2458                 ifindex = rdst->remote_ifindex;
2459                 local_ip = vxlan->cfg.saddr;
2460                 dst_cache = &rdst->dst_cache;
2461                 md->gbp = skb->mark;
2462                 if (flags & VXLAN_F_TTL_INHERIT) {
2463                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2464                 } else {
2465                         ttl = vxlan->cfg.ttl;
2466                         if (!ttl && vxlan_addr_multicast(dst))
2467                                 ttl = 1;
2468                 }
2469
2470                 tos = vxlan->cfg.tos;
2471                 if (tos == 1)
2472                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2473
2474                 if (dst->sa.sa_family == AF_INET)
2475                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2476                 else
2477                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2478                 label = vxlan->cfg.label;
2479         } else {
2480                 if (!info) {
2481                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2482                                   dev->name);
2483                         goto drop;
2484                 }
2485                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2486                 if (remote_ip.sa.sa_family == AF_INET) {
2487                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2488                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2489                 } else {
2490                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2491                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2492                 }
2493                 dst = &remote_ip;
2494                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2495                 vni = tunnel_id_to_key32(info->key.tun_id);
2496                 ifindex = 0;
2497                 dst_cache = &info->dst_cache;
2498                 if (info->key.tun_flags & TUNNEL_VXLAN_OPT) {
2499                         if (info->options_len < sizeof(*md))
2500                                 goto drop;
2501                         md = ip_tunnel_info_opts(info);
2502                 }
2503                 ttl = info->key.ttl;
2504                 tos = info->key.tos;
2505                 label = info->key.label;
2506                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2507         }
2508         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2509                                      vxlan->cfg.port_max, true);
2510
2511         rcu_read_lock();
2512         if (dst->sa.sa_family == AF_INET) {
2513                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2514                 struct rtable *rt;
2515                 __be16 df = 0;
2516
2517                 if (!ifindex)
2518                         ifindex = sock4->sock->sk->sk_bound_dev_if;
2519
2520                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2521                                      dst->sin.sin_addr.s_addr,
2522                                      &local_ip.sin.sin_addr.s_addr,
2523                                      dst_port, src_port,
2524                                      dst_cache, info);
2525                 if (IS_ERR(rt)) {
2526                         err = PTR_ERR(rt);
2527                         goto tx_error;
2528                 }
2529
2530                 if (!info) {
2531                         /* Bypass encapsulation if the destination is local */
2532                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2533                                                     dst_port, ifindex, vni,
2534                                                     &rt->dst, rt->rt_flags);
2535                         if (err)
2536                                 goto out_unlock;
2537
2538                         if (vxlan->cfg.df == VXLAN_DF_SET) {
2539                                 df = htons(IP_DF);
2540                         } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) {
2541                                 struct ethhdr *eth = eth_hdr(skb);
2542
2543                                 if (ntohs(eth->h_proto) == ETH_P_IPV6 ||
2544                                     (ntohs(eth->h_proto) == ETH_P_IP &&
2545                                      old_iph->frag_off & htons(IP_DF)))
2546                                         df = htons(IP_DF);
2547                         }
2548                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2549                         df = htons(IP_DF);
2550                 }
2551
2552                 ndst = &rt->dst;
2553                 skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM);
2554
2555                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2556                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2557                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2558                                       vni, md, flags, udp_sum);
2559                 if (err < 0)
2560                         goto tx_error;
2561
2562                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2563                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2564                                     src_port, dst_port, xnet, !udp_sum);
2565 #if IS_ENABLED(CONFIG_IPV6)
2566         } else {
2567                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2568
2569                 if (!ifindex)
2570                         ifindex = sock6->sock->sk->sk_bound_dev_if;
2571
2572                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2573                                         label, &dst->sin6.sin6_addr,
2574                                         &local_ip.sin6.sin6_addr,
2575                                         dst_port, src_port,
2576                                         dst_cache, info);
2577                 if (IS_ERR(ndst)) {
2578                         err = PTR_ERR(ndst);
2579                         ndst = NULL;
2580                         goto tx_error;
2581                 }
2582
2583                 if (!info) {
2584                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2585
2586                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2587                                                     dst_port, ifindex, vni,
2588                                                     ndst, rt6i_flags);
2589                         if (err)
2590                                 goto out_unlock;
2591                 }
2592
2593                 skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM);
2594
2595                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2596                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2597                 skb_scrub_packet(skb, xnet);
2598                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2599                                       vni, md, flags, udp_sum);
2600                 if (err < 0)
2601                         goto tx_error;
2602
2603                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2604                                      &local_ip.sin6.sin6_addr,
2605                                      &dst->sin6.sin6_addr, tos, ttl,
2606                                      label, src_port, dst_port, !udp_sum);
2607 #endif
2608         }
2609 out_unlock:
2610         rcu_read_unlock();
2611         return;
2612
2613 drop:
2614         dev->stats.tx_dropped++;
2615         dev_kfree_skb(skb);
2616         return;
2617
2618 tx_error:
2619         rcu_read_unlock();
2620         if (err == -ELOOP)
2621                 dev->stats.collisions++;
2622         else if (err == -ENETUNREACH)
2623                 dev->stats.tx_carrier_errors++;
2624         dst_release(ndst);
2625         dev->stats.tx_errors++;
2626         kfree_skb(skb);
2627 }
2628
2629 /* Transmit local packets over Vxlan
2630  *
2631  * Outer IP header inherits ECN and DF from inner header.
2632  * Outer UDP destination is the VXLAN assigned port.
2633  *           source port is based on hash of flow
2634  */
2635 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2636 {
2637         struct vxlan_dev *vxlan = netdev_priv(dev);
2638         struct vxlan_rdst *rdst, *fdst = NULL;
2639         const struct ip_tunnel_info *info;
2640         bool did_rsc = false;
2641         struct vxlan_fdb *f;
2642         struct ethhdr *eth;
2643         __be32 vni = 0;
2644
2645         info = skb_tunnel_info(skb);
2646
2647         skb_reset_mac_header(skb);
2648
2649         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2650                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2651                     info->mode & IP_TUNNEL_INFO_TX) {
2652                         vni = tunnel_id_to_key32(info->key.tun_id);
2653                 } else {
2654                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2655                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2656                         else
2657                                 kfree_skb(skb);
2658                         return NETDEV_TX_OK;
2659                 }
2660         }
2661
2662         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2663                 eth = eth_hdr(skb);
2664                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2665                         return arp_reduce(dev, skb, vni);
2666 #if IS_ENABLED(CONFIG_IPV6)
2667                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2668                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2669                                             sizeof(struct nd_msg)) &&
2670                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2671                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2672
2673                         if (m->icmph.icmp6_code == 0 &&
2674                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2675                                 return neigh_reduce(dev, skb, vni);
2676                 }
2677 #endif
2678         }
2679
2680         eth = eth_hdr(skb);
2681         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2682         did_rsc = false;
2683
2684         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2685             (ntohs(eth->h_proto) == ETH_P_IP ||
2686              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2687                 did_rsc = route_shortcircuit(dev, skb);
2688                 if (did_rsc)
2689                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2690         }
2691
2692         if (f == NULL) {
2693                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2694                 if (f == NULL) {
2695                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2696                             !is_multicast_ether_addr(eth->h_dest))
2697                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2698
2699                         dev->stats.tx_dropped++;
2700                         kfree_skb(skb);
2701                         return NETDEV_TX_OK;
2702                 }
2703         }
2704
2705         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2706                 struct sk_buff *skb1;
2707
2708                 if (!fdst) {
2709                         fdst = rdst;
2710                         continue;
2711                 }
2712                 skb1 = skb_clone(skb, GFP_ATOMIC);
2713                 if (skb1)
2714                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2715         }
2716
2717         if (fdst)
2718                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2719         else
2720                 kfree_skb(skb);
2721         return NETDEV_TX_OK;
2722 }
2723
2724 /* Walk the forwarding table and purge stale entries */
2725 static void vxlan_cleanup(struct timer_list *t)
2726 {
2727         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2728         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2729         unsigned int h;
2730
2731         if (!netif_running(vxlan->dev))
2732                 return;
2733
2734         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2735                 struct hlist_node *p, *n;
2736
2737                 spin_lock(&vxlan->hash_lock[h]);
2738                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2739                         struct vxlan_fdb *f
2740                                 = container_of(p, struct vxlan_fdb, hlist);
2741                         unsigned long timeout;
2742
2743                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2744                                 continue;
2745
2746                         if (f->flags & NTF_EXT_LEARNED)
2747                                 continue;
2748
2749                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2750                         if (time_before_eq(timeout, jiffies)) {
2751                                 netdev_dbg(vxlan->dev,
2752                                            "garbage collect %pM\n",
2753                                            f->eth_addr);
2754                                 f->state = NUD_STALE;
2755                                 vxlan_fdb_destroy(vxlan, f, true, true);
2756                         } else if (time_before(timeout, next_timer))
2757                                 next_timer = timeout;
2758                 }
2759                 spin_unlock(&vxlan->hash_lock[h]);
2760         }
2761
2762         mod_timer(&vxlan->age_timer, next_timer);
2763 }
2764
2765 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2766 {
2767         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2768
2769         spin_lock(&vn->sock_lock);
2770         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2771 #if IS_ENABLED(CONFIG_IPV6)
2772         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2773 #endif
2774         spin_unlock(&vn->sock_lock);
2775 }
2776
2777 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2778                              struct vxlan_dev_node *node)
2779 {
2780         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2781         __be32 vni = vxlan->default_dst.remote_vni;
2782
2783         node->vxlan = vxlan;
2784         spin_lock(&vn->sock_lock);
2785         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2786         spin_unlock(&vn->sock_lock);
2787 }
2788
2789 /* Setup stats when device is created */
2790 static int vxlan_init(struct net_device *dev)
2791 {
2792         struct vxlan_dev *vxlan = netdev_priv(dev);
2793         int err;
2794
2795         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2796         if (!dev->tstats)
2797                 return -ENOMEM;
2798
2799         err = gro_cells_init(&vxlan->gro_cells, dev);
2800         if (err) {
2801                 free_percpu(dev->tstats);
2802                 return err;
2803         }
2804
2805         return 0;
2806 }
2807
2808 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2809 {
2810         struct vxlan_fdb *f;
2811         u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, vni);
2812
2813         spin_lock_bh(&vxlan->hash_lock[hash_index]);
2814         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2815         if (f)
2816                 vxlan_fdb_destroy(vxlan, f, true, true);
2817         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
2818 }
2819
2820 static void vxlan_uninit(struct net_device *dev)
2821 {
2822         struct vxlan_dev *vxlan = netdev_priv(dev);
2823
2824         gro_cells_destroy(&vxlan->gro_cells);
2825
2826         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2827
2828         free_percpu(dev->tstats);
2829 }
2830
2831 /* Start ageing timer and join group when device is brought up */
2832 static int vxlan_open(struct net_device *dev)
2833 {
2834         struct vxlan_dev *vxlan = netdev_priv(dev);
2835         int ret;
2836
2837         ret = vxlan_sock_add(vxlan);
2838         if (ret < 0)
2839                 return ret;
2840
2841         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2842                 ret = vxlan_igmp_join(vxlan);
2843                 if (ret == -EADDRINUSE)
2844                         ret = 0;
2845                 if (ret) {
2846                         vxlan_sock_release(vxlan);
2847                         return ret;
2848                 }
2849         }
2850
2851         if (vxlan->cfg.age_interval)
2852                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2853
2854         return ret;
2855 }
2856
2857 /* Purge the forwarding table */
2858 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2859 {
2860         unsigned int h;
2861
2862         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2863                 struct hlist_node *p, *n;
2864
2865                 spin_lock_bh(&vxlan->hash_lock[h]);
2866                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2867                         struct vxlan_fdb *f
2868                                 = container_of(p, struct vxlan_fdb, hlist);
2869                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2870                                 continue;
2871                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2872                         if (is_zero_ether_addr(f->eth_addr) &&
2873                             f->vni == vxlan->cfg.vni)
2874                                 continue;
2875                         vxlan_fdb_destroy(vxlan, f, true, true);
2876                 }
2877                 spin_unlock_bh(&vxlan->hash_lock[h]);
2878         }
2879 }
2880
2881 /* Cleanup timer and forwarding table on shutdown */
2882 static int vxlan_stop(struct net_device *dev)
2883 {
2884         struct vxlan_dev *vxlan = netdev_priv(dev);
2885         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2886         int ret = 0;
2887
2888         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2889             !vxlan_group_used(vn, vxlan))
2890                 ret = vxlan_igmp_leave(vxlan);
2891
2892         del_timer_sync(&vxlan->age_timer);
2893
2894         vxlan_flush(vxlan, false);
2895         vxlan_sock_release(vxlan);
2896
2897         return ret;
2898 }
2899
2900 /* Stub, nothing needs to be done. */
2901 static void vxlan_set_multicast_list(struct net_device *dev)
2902 {
2903 }
2904
2905 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2906 {
2907         struct vxlan_dev *vxlan = netdev_priv(dev);
2908         struct vxlan_rdst *dst = &vxlan->default_dst;
2909         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2910                                                          dst->remote_ifindex);
2911         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
2912
2913         /* This check is different than dev->max_mtu, because it looks at
2914          * the lowerdev->mtu, rather than the static dev->max_mtu
2915          */
2916         if (lowerdev) {
2917                 int max_mtu = lowerdev->mtu -
2918                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2919                 if (new_mtu > max_mtu)
2920                         return -EINVAL;
2921         }
2922
2923         dev->mtu = new_mtu;
2924         return 0;
2925 }
2926
2927 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2928 {
2929         struct vxlan_dev *vxlan = netdev_priv(dev);
2930         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2931         __be16 sport, dport;
2932
2933         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2934                                   vxlan->cfg.port_max, true);
2935         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2936
2937         if (ip_tunnel_info_af(info) == AF_INET) {
2938                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2939                 struct rtable *rt;
2940
2941                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2942                                      info->key.u.ipv4.dst,
2943                                      &info->key.u.ipv4.src, dport, sport,
2944                                      &info->dst_cache, info);
2945                 if (IS_ERR(rt))
2946                         return PTR_ERR(rt);
2947                 ip_rt_put(rt);
2948         } else {
2949 #if IS_ENABLED(CONFIG_IPV6)
2950                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2951                 struct dst_entry *ndst;
2952
2953                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2954                                         info->key.label, &info->key.u.ipv6.dst,
2955                                         &info->key.u.ipv6.src, dport, sport,
2956                                         &info->dst_cache, info);
2957                 if (IS_ERR(ndst))
2958                         return PTR_ERR(ndst);
2959                 dst_release(ndst);
2960 #else /* !CONFIG_IPV6 */
2961                 return -EPFNOSUPPORT;
2962 #endif
2963         }
2964         info->key.tp_src = sport;
2965         info->key.tp_dst = dport;
2966         return 0;
2967 }
2968
2969 static const struct net_device_ops vxlan_netdev_ether_ops = {
2970         .ndo_init               = vxlan_init,
2971         .ndo_uninit             = vxlan_uninit,
2972         .ndo_open               = vxlan_open,
2973         .ndo_stop               = vxlan_stop,
2974         .ndo_start_xmit         = vxlan_xmit,
2975         .ndo_get_stats64        = ip_tunnel_get_stats64,
2976         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2977         .ndo_change_mtu         = vxlan_change_mtu,
2978         .ndo_validate_addr      = eth_validate_addr,
2979         .ndo_set_mac_address    = eth_mac_addr,
2980         .ndo_fdb_add            = vxlan_fdb_add,
2981         .ndo_fdb_del            = vxlan_fdb_delete,
2982         .ndo_fdb_dump           = vxlan_fdb_dump,
2983         .ndo_fdb_get            = vxlan_fdb_get,
2984         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2985         .ndo_change_proto_down  = dev_change_proto_down_generic,
2986 };
2987
2988 static const struct net_device_ops vxlan_netdev_raw_ops = {
2989         .ndo_init               = vxlan_init,
2990         .ndo_uninit             = vxlan_uninit,
2991         .ndo_open               = vxlan_open,
2992         .ndo_stop               = vxlan_stop,
2993         .ndo_start_xmit         = vxlan_xmit,
2994         .ndo_get_stats64        = ip_tunnel_get_stats64,
2995         .ndo_change_mtu         = vxlan_change_mtu,
2996         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2997 };
2998
2999 /* Info for udev, that this is a virtual tunnel endpoint */
3000 static struct device_type vxlan_type = {
3001         .name = "vxlan",
3002 };
3003
3004 /* Calls the ndo_udp_tunnel_add of the caller in order to
3005  * supply the listening VXLAN udp ports. Callers are expected
3006  * to implement the ndo_udp_tunnel_add.
3007  */
3008 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
3009 {
3010         struct vxlan_sock *vs;
3011         struct net *net = dev_net(dev);
3012         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3013         unsigned int i;
3014
3015         spin_lock(&vn->sock_lock);
3016         for (i = 0; i < PORT_HASH_SIZE; ++i) {
3017                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
3018                         unsigned short type;
3019
3020                         if (vs->flags & VXLAN_F_GPE)
3021                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
3022                         else
3023                                 type = UDP_TUNNEL_TYPE_VXLAN;
3024
3025                         if (push)
3026                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
3027                         else
3028                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
3029                 }
3030         }
3031         spin_unlock(&vn->sock_lock);
3032 }
3033
3034 /* Initialize the device structure. */
3035 static void vxlan_setup(struct net_device *dev)
3036 {
3037         struct vxlan_dev *vxlan = netdev_priv(dev);
3038         unsigned int h;
3039
3040         eth_hw_addr_random(dev);
3041         ether_setup(dev);
3042
3043         dev->needs_free_netdev = true;
3044         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
3045
3046         dev->features   |= NETIF_F_LLTX;
3047         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
3048         dev->features   |= NETIF_F_RXCSUM;
3049         dev->features   |= NETIF_F_GSO_SOFTWARE;
3050
3051         dev->vlan_features = dev->features;
3052         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
3053         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
3054         netif_keep_dst(dev);
3055         dev->priv_flags |= IFF_NO_QUEUE;
3056
3057         /* MTU range: 68 - 65535 */
3058         dev->min_mtu = ETH_MIN_MTU;
3059         dev->max_mtu = ETH_MAX_MTU;
3060
3061         INIT_LIST_HEAD(&vxlan->next);
3062
3063         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
3064
3065         vxlan->dev = dev;
3066
3067         for (h = 0; h < FDB_HASH_SIZE; ++h) {
3068                 spin_lock_init(&vxlan->hash_lock[h]);
3069                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
3070         }
3071 }
3072
3073 static void vxlan_ether_setup(struct net_device *dev)
3074 {
3075         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
3076         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3077         dev->netdev_ops = &vxlan_netdev_ether_ops;
3078 }
3079
3080 static void vxlan_raw_setup(struct net_device *dev)
3081 {
3082         dev->header_ops = NULL;
3083         dev->type = ARPHRD_NONE;
3084         dev->hard_header_len = 0;
3085         dev->addr_len = 0;
3086         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
3087         dev->netdev_ops = &vxlan_netdev_raw_ops;
3088 }
3089
3090 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
3091         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
3092         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
3093         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
3094         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
3095         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
3096         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
3097         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
3098         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
3099         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
3100         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
3101         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
3102         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
3103         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
3104         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
3105         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
3106         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
3107         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
3108         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
3109         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
3110         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
3111         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
3112         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
3113         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
3114         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
3115         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
3116         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
3117         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
3118         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
3119         [IFLA_VXLAN_DF]         = { .type = NLA_U8 },
3120 };
3121
3122 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
3123                           struct netlink_ext_ack *extack)
3124 {
3125         if (tb[IFLA_ADDRESS]) {
3126                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
3127                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3128                                             "Provided link layer address is not Ethernet");
3129                         return -EINVAL;
3130                 }
3131
3132                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
3133                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3134                                             "Provided Ethernet address is not unicast");
3135                         return -EADDRNOTAVAIL;
3136                 }
3137         }
3138
3139         if (tb[IFLA_MTU]) {
3140                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
3141
3142                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
3143                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3144                                             "MTU must be between 68 and 65535");
3145                         return -EINVAL;
3146                 }
3147         }
3148
3149         if (!data) {
3150                 NL_SET_ERR_MSG(extack,
3151                                "Required attributes not provided to perform the operation");
3152                 return -EINVAL;
3153         }
3154
3155         if (data[IFLA_VXLAN_ID]) {
3156                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
3157
3158                 if (id >= VXLAN_N_VID) {
3159                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_ID],
3160                                             "VXLAN ID must be lower than 16777216");
3161                         return -ERANGE;
3162                 }
3163         }
3164
3165         if (data[IFLA_VXLAN_PORT_RANGE]) {
3166                 const struct ifla_vxlan_port_range *p
3167                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3168
3169                 if (ntohs(p->high) < ntohs(p->low)) {
3170                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_PORT_RANGE],
3171                                             "Invalid source port range");
3172                         return -EINVAL;
3173                 }
3174         }
3175
3176         if (data[IFLA_VXLAN_DF]) {
3177                 enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]);
3178
3179                 if (df < 0 || df > VXLAN_DF_MAX) {
3180                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_DF],
3181                                             "Invalid DF attribute");
3182                         return -EINVAL;
3183                 }
3184         }
3185
3186         return 0;
3187 }
3188
3189 static void vxlan_get_drvinfo(struct net_device *netdev,
3190                               struct ethtool_drvinfo *drvinfo)
3191 {
3192         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
3193         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
3194 }
3195
3196 static const struct ethtool_ops vxlan_ethtool_ops = {
3197         .get_drvinfo    = vxlan_get_drvinfo,
3198         .get_link       = ethtool_op_get_link,
3199 };
3200
3201 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
3202                                         __be16 port, u32 flags, int ifindex)
3203 {
3204         struct socket *sock;
3205         struct udp_port_cfg udp_conf;
3206         int err;
3207
3208         memset(&udp_conf, 0, sizeof(udp_conf));
3209
3210         if (ipv6) {
3211                 udp_conf.family = AF_INET6;
3212                 udp_conf.use_udp6_rx_checksums =
3213                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
3214                 udp_conf.ipv6_v6only = 1;
3215         } else {
3216                 udp_conf.family = AF_INET;
3217         }
3218
3219         udp_conf.local_udp_port = port;
3220         udp_conf.bind_ifindex = ifindex;
3221
3222         /* Open UDP socket */
3223         err = udp_sock_create(net, &udp_conf, &sock);
3224         if (err < 0)
3225                 return ERR_PTR(err);
3226
3227         return sock;
3228 }
3229
3230 /* Create new listen socket if needed */
3231 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
3232                                               __be16 port, u32 flags,
3233                                               int ifindex)
3234 {
3235         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3236         struct vxlan_sock *vs;
3237         struct socket *sock;
3238         unsigned int h;
3239         struct udp_tunnel_sock_cfg tunnel_cfg;
3240
3241         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
3242         if (!vs)
3243                 return ERR_PTR(-ENOMEM);
3244
3245         for (h = 0; h < VNI_HASH_SIZE; ++h)
3246                 INIT_HLIST_HEAD(&vs->vni_list[h]);
3247
3248         sock = vxlan_create_sock(net, ipv6, port, flags, ifindex);
3249         if (IS_ERR(sock)) {
3250                 kfree(vs);
3251                 return ERR_CAST(sock);
3252         }
3253
3254         vs->sock = sock;
3255         refcount_set(&vs->refcnt, 1);
3256         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
3257
3258         spin_lock(&vn->sock_lock);
3259         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
3260         udp_tunnel_notify_add_rx_port(sock,
3261                                       (vs->flags & VXLAN_F_GPE) ?
3262                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
3263                                       UDP_TUNNEL_TYPE_VXLAN);
3264         spin_unlock(&vn->sock_lock);
3265
3266         /* Mark socket as an encapsulation socket. */
3267         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
3268         tunnel_cfg.sk_user_data = vs;
3269         tunnel_cfg.encap_type = 1;
3270         tunnel_cfg.encap_rcv = vxlan_rcv;
3271         tunnel_cfg.encap_err_lookup = vxlan_err_lookup;
3272         tunnel_cfg.encap_destroy = NULL;
3273         tunnel_cfg.gro_receive = vxlan_gro_receive;
3274         tunnel_cfg.gro_complete = vxlan_gro_complete;
3275
3276         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
3277
3278         return vs;
3279 }
3280
3281 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
3282 {
3283         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3284         struct vxlan_sock *vs = NULL;
3285         struct vxlan_dev_node *node;
3286         int l3mdev_index = 0;
3287
3288         if (vxlan->cfg.remote_ifindex)
3289                 l3mdev_index = l3mdev_master_upper_ifindex_by_index(
3290                         vxlan->net, vxlan->cfg.remote_ifindex);
3291
3292         if (!vxlan->cfg.no_share) {
3293                 spin_lock(&vn->sock_lock);
3294                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
3295                                      vxlan->cfg.dst_port, vxlan->cfg.flags,
3296                                      l3mdev_index);
3297                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
3298                         spin_unlock(&vn->sock_lock);
3299                         return -EBUSY;
3300                 }
3301                 spin_unlock(&vn->sock_lock);
3302         }
3303         if (!vs)
3304                 vs = vxlan_socket_create(vxlan->net, ipv6,
3305                                          vxlan->cfg.dst_port, vxlan->cfg.flags,
3306                                          l3mdev_index);
3307         if (IS_ERR(vs))
3308                 return PTR_ERR(vs);
3309 #if IS_ENABLED(CONFIG_IPV6)
3310         if (ipv6) {
3311                 rcu_assign_pointer(vxlan->vn6_sock, vs);
3312                 node = &vxlan->hlist6;
3313         } else
3314 #endif
3315         {
3316                 rcu_assign_pointer(vxlan->vn4_sock, vs);
3317                 node = &vxlan->hlist4;
3318         }
3319         vxlan_vs_add_dev(vs, vxlan, node);
3320         return 0;
3321 }
3322
3323 static int vxlan_sock_add(struct vxlan_dev *vxlan)
3324 {
3325         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
3326         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
3327         bool ipv4 = !ipv6 || metadata;
3328         int ret = 0;
3329
3330         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
3331 #if IS_ENABLED(CONFIG_IPV6)
3332         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
3333         if (ipv6) {
3334                 ret = __vxlan_sock_add(vxlan, true);
3335                 if (ret < 0 && ret != -EAFNOSUPPORT)
3336                         ipv4 = false;
3337         }
3338 #endif
3339         if (ipv4)
3340                 ret = __vxlan_sock_add(vxlan, false);
3341         if (ret < 0)
3342                 vxlan_sock_release(vxlan);
3343         return ret;
3344 }
3345
3346 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
3347                                  struct net_device **lower,
3348                                  struct vxlan_dev *old,
3349                                  struct netlink_ext_ack *extack)
3350 {
3351         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
3352         struct vxlan_dev *tmp;
3353         bool use_ipv6 = false;
3354
3355         if (conf->flags & VXLAN_F_GPE) {
3356                 /* For now, allow GPE only together with
3357                  * COLLECT_METADATA. This can be relaxed later; in such
3358                  * case, the other side of the PtP link will have to be
3359                  * provided.
3360                  */
3361                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
3362                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3363                         NL_SET_ERR_MSG(extack,
3364                                        "VXLAN GPE does not support this combination of attributes");
3365                         return -EINVAL;
3366                 }
3367         }
3368
3369         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
3370                 /* Unless IPv6 is explicitly requested, assume IPv4 */
3371                 conf->remote_ip.sa.sa_family = AF_INET;
3372                 conf->saddr.sa.sa_family = AF_INET;
3373         } else if (!conf->remote_ip.sa.sa_family) {
3374                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
3375         } else if (!conf->saddr.sa.sa_family) {
3376                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
3377         }
3378
3379         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
3380                 NL_SET_ERR_MSG(extack,
3381                                "Local and remote address must be from the same family");
3382                 return -EINVAL;
3383         }
3384
3385         if (vxlan_addr_multicast(&conf->saddr)) {
3386                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
3387                 return -EINVAL;
3388         }
3389
3390         if (conf->saddr.sa.sa_family == AF_INET6) {
3391                 if (!IS_ENABLED(CONFIG_IPV6)) {
3392                         NL_SET_ERR_MSG(extack,
3393                                        "IPv6 support not enabled in the kernel");
3394                         return -EPFNOSUPPORT;
3395                 }
3396                 use_ipv6 = true;
3397                 conf->flags |= VXLAN_F_IPV6;
3398
3399                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3400                         int local_type =
3401                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3402                         int remote_type =
3403                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3404
3405                         if (local_type & IPV6_ADDR_LINKLOCAL) {
3406                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3407                                     (remote_type != IPV6_ADDR_ANY)) {
3408                                         NL_SET_ERR_MSG(extack,
3409                                                        "Invalid combination of local and remote address scopes");
3410                                         return -EINVAL;
3411                                 }
3412
3413                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3414                         } else {
3415                                 if (remote_type ==
3416                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3417                                         NL_SET_ERR_MSG(extack,
3418                                                        "Invalid combination of local and remote address scopes");
3419                                         return -EINVAL;
3420                                 }
3421
3422                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3423                         }
3424                 }
3425         }
3426
3427         if (conf->label && !use_ipv6) {
3428                 NL_SET_ERR_MSG(extack,
3429                                "Label attribute only applies to IPv6 VXLAN devices");
3430                 return -EINVAL;
3431         }
3432
3433         if (conf->remote_ifindex) {
3434                 struct net_device *lowerdev;
3435
3436                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3437                 if (!lowerdev) {
3438                         NL_SET_ERR_MSG(extack,
3439                                        "Invalid local interface, device not found");
3440                         return -ENODEV;
3441                 }
3442
3443 #if IS_ENABLED(CONFIG_IPV6)
3444                 if (use_ipv6) {
3445                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3446                         if (idev && idev->cnf.disable_ipv6) {
3447                                 NL_SET_ERR_MSG(extack,
3448                                                "IPv6 support disabled by administrator");
3449                                 return -EPERM;
3450                         }
3451                 }
3452 #endif
3453
3454                 *lower = lowerdev;
3455         } else {
3456                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3457                         NL_SET_ERR_MSG(extack,
3458                                        "Local interface required for multicast remote destination");
3459
3460                         return -EINVAL;
3461                 }
3462
3463 #if IS_ENABLED(CONFIG_IPV6)
3464                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3465                         NL_SET_ERR_MSG(extack,
3466                                        "Local interface required for link-local local/remote addresses");
3467                         return -EINVAL;
3468                 }
3469 #endif
3470
3471                 *lower = NULL;
3472         }
3473
3474         if (!conf->dst_port) {
3475                 if (conf->flags & VXLAN_F_GPE)
3476                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3477                 else
3478                         conf->dst_port = htons(vxlan_port);
3479         }
3480
3481         if (!conf->age_interval)
3482                 conf->age_interval = FDB_AGE_DEFAULT;
3483
3484         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3485                 if (tmp == old)
3486                         continue;
3487
3488                 if (tmp->cfg.vni != conf->vni)
3489                         continue;
3490                 if (tmp->cfg.dst_port != conf->dst_port)
3491                         continue;
3492                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3493                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3494                         continue;
3495
3496                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3497                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3498                         continue;
3499
3500                 NL_SET_ERR_MSG(extack,
3501                                "A VXLAN device with the specified VNI already exists");
3502                 return -EEXIST;
3503         }
3504
3505         return 0;
3506 }
3507
3508 static void vxlan_config_apply(struct net_device *dev,
3509                                struct vxlan_config *conf,
3510                                struct net_device *lowerdev,
3511                                struct net *src_net,
3512                                bool changelink)
3513 {
3514         struct vxlan_dev *vxlan = netdev_priv(dev);
3515         struct vxlan_rdst *dst = &vxlan->default_dst;
3516         unsigned short needed_headroom = ETH_HLEN;
3517         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3518         int max_mtu = ETH_MAX_MTU;
3519
3520         if (!changelink) {
3521                 if (conf->flags & VXLAN_F_GPE)
3522                         vxlan_raw_setup(dev);
3523                 else
3524                         vxlan_ether_setup(dev);
3525
3526                 if (conf->mtu)
3527                         dev->mtu = conf->mtu;
3528
3529                 vxlan->net = src_net;
3530         }
3531
3532         dst->remote_vni = conf->vni;
3533
3534         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3535
3536         if (lowerdev) {
3537                 dst->remote_ifindex = conf->remote_ifindex;
3538
3539                 dev->gso_max_size = lowerdev->gso_max_size;
3540                 dev->gso_max_segs = lowerdev->gso_max_segs;
3541
3542                 needed_headroom = lowerdev->hard_header_len;
3543                 needed_headroom += lowerdev->needed_headroom;
3544
3545                 dev->needed_tailroom = lowerdev->needed_tailroom;
3546
3547                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3548                                            VXLAN_HEADROOM);
3549                 if (max_mtu < ETH_MIN_MTU)
3550                         max_mtu = ETH_MIN_MTU;
3551
3552                 if (!changelink && !conf->mtu)
3553                         dev->mtu = max_mtu;
3554         }
3555
3556         if (dev->mtu > max_mtu)
3557                 dev->mtu = max_mtu;
3558
3559         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3560                 needed_headroom += VXLAN6_HEADROOM;
3561         else
3562                 needed_headroom += VXLAN_HEADROOM;
3563         dev->needed_headroom = needed_headroom;
3564
3565         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3566 }
3567
3568 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3569                                struct vxlan_config *conf, bool changelink,
3570                                struct netlink_ext_ack *extack)
3571 {
3572         struct vxlan_dev *vxlan = netdev_priv(dev);
3573         struct net_device *lowerdev;
3574         int ret;
3575
3576         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3577         if (ret)
3578                 return ret;
3579
3580         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3581
3582         return 0;
3583 }
3584
3585 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3586                               struct vxlan_config *conf,
3587                               struct netlink_ext_ack *extack)
3588 {
3589         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3590         struct vxlan_dev *vxlan = netdev_priv(dev);
3591         struct net_device *remote_dev = NULL;
3592         struct vxlan_fdb *f = NULL;
3593         bool unregister = false;
3594         struct vxlan_rdst *dst;
3595         int err;
3596
3597         dst = &vxlan->default_dst;
3598         err = vxlan_dev_configure(net, dev, conf, false, extack);
3599         if (err)
3600                 return err;
3601
3602         dev->ethtool_ops = &vxlan_ethtool_ops;
3603
3604         /* create an fdb entry for a valid default destination */
3605         if (!vxlan_addr_any(&dst->remote_ip)) {
3606                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3607                                        &dst->remote_ip,
3608                                        NUD_REACHABLE | NUD_PERMANENT,
3609                                        vxlan->cfg.dst_port,
3610                                        dst->remote_vni,
3611                                        dst->remote_vni,
3612                                        dst->remote_ifindex,
3613                                        NTF_SELF, &f);
3614                 if (err)
3615                         return err;
3616         }
3617
3618         err = register_netdevice(dev);
3619         if (err)
3620                 goto errout;
3621         unregister = true;
3622
3623         if (dst->remote_ifindex) {
3624                 remote_dev = __dev_get_by_index(net, dst->remote_ifindex);
3625                 if (!remote_dev) {
3626                         err = -ENODEV;
3627                         goto errout;
3628                 }
3629
3630                 err = netdev_upper_dev_link(remote_dev, dev, extack);
3631                 if (err)
3632                         goto errout;
3633         }
3634
3635         err = rtnl_configure_link(dev, NULL);
3636         if (err)
3637                 goto unlink;
3638
3639         if (f) {
3640                 vxlan_fdb_insert(vxlan, all_zeros_mac, dst->remote_vni, f);
3641
3642                 /* notify default fdb entry */
3643                 err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f),
3644                                        RTM_NEWNEIGH, true, extack);
3645                 if (err) {
3646                         vxlan_fdb_destroy(vxlan, f, false, false);
3647                         if (remote_dev)
3648                                 netdev_upper_dev_unlink(remote_dev, dev);
3649                         goto unregister;
3650                 }
3651         }
3652
3653         list_add(&vxlan->next, &vn->vxlan_list);
3654         if (remote_dev)
3655                 dst->remote_dev = remote_dev;
3656         return 0;
3657 unlink:
3658         if (remote_dev)
3659                 netdev_upper_dev_unlink(remote_dev, dev);
3660 errout:
3661         /* unregister_netdevice() destroys the default FDB entry with deletion
3662          * notification. But the addition notification was not sent yet, so
3663          * destroy the entry by hand here.
3664          */
3665         if (f)
3666                 __vxlan_fdb_free(f);
3667 unregister:
3668         if (unregister)
3669                 unregister_netdevice(dev);
3670         return err;
3671 }
3672
3673 /* Set/clear flags based on attribute */
3674 static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[],
3675                           int attrtype, unsigned long mask, bool changelink,
3676                           bool changelink_supported,
3677                           struct netlink_ext_ack *extack)
3678 {
3679         unsigned long flags;
3680
3681         if (!tb[attrtype])
3682                 return 0;
3683
3684         if (changelink && !changelink_supported) {
3685                 vxlan_flag_attr_error(attrtype, extack);
3686                 return -EOPNOTSUPP;
3687         }
3688
3689         if (vxlan_policy[attrtype].type == NLA_FLAG)
3690                 flags = conf->flags | mask;
3691         else if (nla_get_u8(tb[attrtype]))
3692                 flags = conf->flags | mask;
3693         else
3694                 flags = conf->flags & ~mask;
3695
3696         conf->flags = flags;
3697
3698         return 0;
3699 }
3700
3701 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3702                          struct net_device *dev, struct vxlan_config *conf,
3703                          bool changelink, struct netlink_ext_ack *extack)
3704 {
3705         struct vxlan_dev *vxlan = netdev_priv(dev);
3706         int err = 0;
3707
3708         memset(conf, 0, sizeof(*conf));
3709
3710         /* if changelink operation, start with old existing cfg */
3711         if (changelink)
3712                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3713
3714         if (data[IFLA_VXLAN_ID]) {
3715                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3716
3717                 if (changelink && (vni != conf->vni)) {
3718                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI");
3719                         return -EOPNOTSUPP;
3720                 }
3721                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3722         }
3723
3724         if (data[IFLA_VXLAN_GROUP]) {
3725                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) {
3726                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group");
3727                         return -EOPNOTSUPP;
3728                 }
3729
3730                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3731                 conf->remote_ip.sa.sa_family = AF_INET;
3732         } else if (data[IFLA_VXLAN_GROUP6]) {
3733                 if (!IS_ENABLED(CONFIG_IPV6)) {
3734                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel");
3735                         return -EPFNOSUPPORT;
3736                 }
3737
3738                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) {
3739                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group");
3740                         return -EOPNOTSUPP;
3741                 }
3742
3743                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3744                 conf->remote_ip.sa.sa_family = AF_INET6;
3745         }
3746
3747         if (data[IFLA_VXLAN_LOCAL]) {
3748                 if (changelink && (conf->saddr.sa.sa_family != AF_INET)) {
3749                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old");
3750                         return -EOPNOTSUPP;
3751                 }
3752
3753                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3754                 conf->saddr.sa.sa_family = AF_INET;
3755         } else if (data[IFLA_VXLAN_LOCAL6]) {
3756                 if (!IS_ENABLED(CONFIG_IPV6)) {
3757                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel");
3758                         return -EPFNOSUPPORT;
3759                 }
3760
3761                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) {
3762                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old");
3763                         return -EOPNOTSUPP;
3764                 }
3765
3766                 /* TODO: respect scope id */
3767                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3768                 conf->saddr.sa.sa_family = AF_INET6;
3769         }
3770
3771         if (data[IFLA_VXLAN_LINK])
3772                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3773
3774         if (data[IFLA_VXLAN_TOS])
3775                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3776
3777         if (data[IFLA_VXLAN_TTL])
3778                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3779
3780         if (data[IFLA_VXLAN_TTL_INHERIT]) {
3781                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT,
3782                                     VXLAN_F_TTL_INHERIT, changelink, false,
3783                                     extack);
3784                 if (err)
3785                         return err;
3786
3787         }
3788
3789         if (data[IFLA_VXLAN_LABEL])
3790                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3791                              IPV6_FLOWLABEL_MASK;
3792
3793         if (data[IFLA_VXLAN_LEARNING]) {
3794                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING,
3795                                     VXLAN_F_LEARN, changelink, true,
3796                                     extack);
3797                 if (err)
3798                         return err;
3799         } else if (!changelink) {
3800                 /* default to learn on a new device */
3801                 conf->flags |= VXLAN_F_LEARN;
3802         }
3803
3804         if (data[IFLA_VXLAN_AGEING])
3805                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3806
3807         if (data[IFLA_VXLAN_PROXY]) {
3808                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY,
3809                                     VXLAN_F_PROXY, changelink, false,
3810                                     extack);
3811                 if (err)
3812                         return err;
3813         }
3814
3815         if (data[IFLA_VXLAN_RSC]) {
3816                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC,
3817                                     VXLAN_F_RSC, changelink, false,
3818                                     extack);
3819                 if (err)
3820                         return err;
3821         }
3822
3823         if (data[IFLA_VXLAN_L2MISS]) {
3824                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS,
3825                                     VXLAN_F_L2MISS, changelink, false,
3826                                     extack);
3827                 if (err)
3828                         return err;
3829         }
3830
3831         if (data[IFLA_VXLAN_L3MISS]) {
3832                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS,
3833                                     VXLAN_F_L3MISS, changelink, false,
3834                                     extack);
3835                 if (err)
3836                         return err;
3837         }
3838
3839         if (data[IFLA_VXLAN_LIMIT]) {
3840                 if (changelink) {
3841                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT],
3842                                             "Cannot change limit");
3843                         return -EOPNOTSUPP;
3844                 }
3845                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3846         }
3847
3848         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3849                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA,
3850                                     VXLAN_F_COLLECT_METADATA, changelink, false,
3851                                     extack);
3852                 if (err)
3853                         return err;
3854         }
3855
3856         if (data[IFLA_VXLAN_PORT_RANGE]) {
3857                 if (!changelink) {
3858                         const struct ifla_vxlan_port_range *p
3859                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3860                         conf->port_min = ntohs(p->low);
3861                         conf->port_max = ntohs(p->high);
3862                 } else {
3863                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
3864                                             "Cannot change port range");
3865                         return -EOPNOTSUPP;
3866                 }
3867         }
3868
3869         if (data[IFLA_VXLAN_PORT]) {
3870                 if (changelink) {
3871                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT],
3872                                             "Cannot change port");
3873                         return -EOPNOTSUPP;
3874                 }
3875                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3876         }
3877
3878         if (data[IFLA_VXLAN_UDP_CSUM]) {
3879                 if (changelink) {
3880                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM],
3881                                             "Cannot change UDP_CSUM flag");
3882                         return -EOPNOTSUPP;
3883                 }
3884                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3885                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3886         }
3887
3888         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3889                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3890                                     VXLAN_F_UDP_ZERO_CSUM6_TX, changelink,
3891                                     false, extack);
3892                 if (err)
3893                         return err;
3894         }
3895
3896         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3897                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3898                                     VXLAN_F_UDP_ZERO_CSUM6_RX, changelink,
3899                                     false, extack);
3900                 if (err)
3901                         return err;
3902         }
3903
3904         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3905                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX,
3906                                     VXLAN_F_REMCSUM_TX, changelink, false,
3907                                     extack);
3908                 if (err)
3909                         return err;
3910         }
3911
3912         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3913                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX,
3914                                     VXLAN_F_REMCSUM_RX, changelink, false,
3915                                     extack);
3916                 if (err)
3917                         return err;
3918         }
3919
3920         if (data[IFLA_VXLAN_GBP]) {
3921                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP,
3922                                     VXLAN_F_GBP, changelink, false, extack);
3923                 if (err)
3924                         return err;
3925         }
3926
3927         if (data[IFLA_VXLAN_GPE]) {
3928                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE,
3929                                     VXLAN_F_GPE, changelink, false,
3930                                     extack);
3931                 if (err)
3932                         return err;
3933         }
3934
3935         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3936                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL,
3937                                     VXLAN_F_REMCSUM_NOPARTIAL, changelink,
3938                                     false, extack);
3939                 if (err)
3940                         return err;
3941         }
3942
3943         if (tb[IFLA_MTU]) {
3944                 if (changelink) {
3945                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3946                                             "Cannot change mtu");
3947                         return -EOPNOTSUPP;
3948                 }
3949                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3950         }
3951
3952         if (data[IFLA_VXLAN_DF])
3953                 conf->df = nla_get_u8(data[IFLA_VXLAN_DF]);
3954
3955         return 0;
3956 }
3957
3958 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3959                          struct nlattr *tb[], struct nlattr *data[],
3960                          struct netlink_ext_ack *extack)
3961 {
3962         struct vxlan_config conf;
3963         int err;
3964
3965         err = vxlan_nl2conf(tb, data, dev, &conf, false, extack);
3966         if (err)
3967                 return err;
3968
3969         return __vxlan_dev_create(src_net, dev, &conf, extack);
3970 }
3971
3972 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3973                             struct nlattr *data[],
3974                             struct netlink_ext_ack *extack)
3975 {
3976         struct vxlan_dev *vxlan = netdev_priv(dev);
3977         struct net_device *lowerdev;
3978         struct vxlan_config conf;
3979         struct vxlan_rdst *dst;
3980         int err;
3981
3982         dst = &vxlan->default_dst;
3983         err = vxlan_nl2conf(tb, data, dev, &conf, true, extack);
3984         if (err)
3985                 return err;
3986
3987         err = vxlan_config_validate(vxlan->net, &conf, &lowerdev,
3988                                     vxlan, extack);
3989         if (err)
3990                 return err;
3991
3992         if (dst->remote_dev == lowerdev)
3993                 lowerdev = NULL;
3994
3995         err = netdev_adjacent_change_prepare(dst->remote_dev, lowerdev, dev,
3996                                              extack);
3997         if (err)
3998                 return err;
3999
4000         /* handle default dst entry */
4001         if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) {
4002                 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, conf.vni);
4003
4004                 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4005                 if (!vxlan_addr_any(&conf.remote_ip)) {
4006                         err = vxlan_fdb_update(vxlan, all_zeros_mac,
4007                                                &conf.remote_ip,
4008                                                NUD_REACHABLE | NUD_PERMANENT,
4009                                                NLM_F_APPEND | NLM_F_CREATE,
4010                                                vxlan->cfg.dst_port,
4011                                                conf.vni, conf.vni,
4012                                                conf.remote_ifindex,
4013                                                NTF_SELF, true, extack);
4014                         if (err) {
4015                                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4016                                 netdev_adjacent_change_abort(dst->remote_dev,
4017                                                              lowerdev, dev);
4018                                 return err;
4019                         }
4020                 }
4021                 if (!vxlan_addr_any(&dst->remote_ip))
4022                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
4023                                            dst->remote_ip,
4024                                            vxlan->cfg.dst_port,
4025                                            dst->remote_vni,
4026                                            dst->remote_vni,
4027                                            dst->remote_ifindex,
4028                                            true);
4029                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4030         }
4031
4032         if (conf.age_interval != vxlan->cfg.age_interval)
4033                 mod_timer(&vxlan->age_timer, jiffies);
4034
4035         netdev_adjacent_change_commit(dst->remote_dev, lowerdev, dev);
4036         if (lowerdev && lowerdev != dst->remote_dev) {
4037                 dst->remote_dev = lowerdev;
4038                 netdev_update_lockdep_key(lowerdev);
4039         }
4040         vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true);
4041         return 0;
4042 }
4043
4044 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
4045 {
4046         struct vxlan_dev *vxlan = netdev_priv(dev);
4047
4048         vxlan_flush(vxlan, true);
4049
4050         list_del(&vxlan->next);
4051         unregister_netdevice_queue(dev, head);
4052         if (vxlan->default_dst.remote_dev)
4053                 netdev_upper_dev_unlink(vxlan->default_dst.remote_dev, dev);
4054 }
4055
4056 static size_t vxlan_get_size(const struct net_device *dev)
4057 {
4058
4059         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
4060                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
4061                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
4062                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
4063                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
4064                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL_INHERIT */
4065                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
4066                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_DF */
4067                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
4068                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
4069                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
4070                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
4071                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
4072                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
4073                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
4074                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
4075                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
4076                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
4077                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
4078                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
4079                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
4080                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
4081                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
4082                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
4083                 0;
4084 }
4085
4086 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
4087 {
4088         const struct vxlan_dev *vxlan = netdev_priv(dev);
4089         const struct vxlan_rdst *dst = &vxlan->default_dst;
4090         struct ifla_vxlan_port_range ports = {
4091                 .low =  htons(vxlan->cfg.port_min),
4092                 .high = htons(vxlan->cfg.port_max),
4093         };
4094
4095         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
4096                 goto nla_put_failure;
4097
4098         if (!vxlan_addr_any(&dst->remote_ip)) {
4099                 if (dst->remote_ip.sa.sa_family == AF_INET) {
4100                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
4101                                             dst->remote_ip.sin.sin_addr.s_addr))
4102                                 goto nla_put_failure;
4103 #if IS_ENABLED(CONFIG_IPV6)
4104                 } else {
4105                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
4106                                              &dst->remote_ip.sin6.sin6_addr))
4107                                 goto nla_put_failure;
4108 #endif
4109                 }
4110         }
4111
4112         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
4113                 goto nla_put_failure;
4114
4115         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
4116                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
4117                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
4118                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
4119                                 goto nla_put_failure;
4120 #if IS_ENABLED(CONFIG_IPV6)
4121                 } else {
4122                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
4123                                              &vxlan->cfg.saddr.sin6.sin6_addr))
4124                                 goto nla_put_failure;
4125 #endif
4126                 }
4127         }
4128
4129         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
4130             nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT,
4131                        !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) ||
4132             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
4133             nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) ||
4134             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
4135             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
4136                         !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
4137             nla_put_u8(skb, IFLA_VXLAN_PROXY,
4138                         !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
4139             nla_put_u8(skb, IFLA_VXLAN_RSC,
4140                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
4141             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
4142                         !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
4143             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
4144                         !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
4145             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
4146                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
4147             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
4148             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
4149             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
4150             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
4151                         !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
4152             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4153                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
4154             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4155                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
4156             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
4157                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
4158             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
4159                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
4160                 goto nla_put_failure;
4161
4162         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
4163                 goto nla_put_failure;
4164
4165         if (vxlan->cfg.flags & VXLAN_F_GBP &&
4166             nla_put_flag(skb, IFLA_VXLAN_GBP))
4167                 goto nla_put_failure;
4168
4169         if (vxlan->cfg.flags & VXLAN_F_GPE &&
4170             nla_put_flag(skb, IFLA_VXLAN_GPE))
4171                 goto nla_put_failure;
4172
4173         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
4174             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
4175                 goto nla_put_failure;
4176
4177         return 0;
4178
4179 nla_put_failure:
4180         return -EMSGSIZE;
4181 }
4182
4183 static struct net *vxlan_get_link_net(const struct net_device *dev)
4184 {
4185         struct vxlan_dev *vxlan = netdev_priv(dev);
4186
4187         return vxlan->net;
4188 }
4189
4190 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
4191         .kind           = "vxlan",
4192         .maxtype        = IFLA_VXLAN_MAX,
4193         .policy         = vxlan_policy,
4194         .priv_size      = sizeof(struct vxlan_dev),
4195         .setup          = vxlan_setup,
4196         .validate       = vxlan_validate,
4197         .newlink        = vxlan_newlink,
4198         .changelink     = vxlan_changelink,
4199         .dellink        = vxlan_dellink,
4200         .get_size       = vxlan_get_size,
4201         .fill_info      = vxlan_fill_info,
4202         .get_link_net   = vxlan_get_link_net,
4203 };
4204
4205 struct net_device *vxlan_dev_create(struct net *net, const char *name,
4206                                     u8 name_assign_type,
4207                                     struct vxlan_config *conf)
4208 {
4209         struct nlattr *tb[IFLA_MAX + 1];
4210         struct net_device *dev;
4211         int err;
4212
4213         memset(&tb, 0, sizeof(tb));
4214
4215         dev = rtnl_create_link(net, name, name_assign_type,
4216                                &vxlan_link_ops, tb, NULL);
4217         if (IS_ERR(dev))
4218                 return dev;
4219
4220         err = __vxlan_dev_create(net, dev, conf, NULL);
4221         if (err < 0) {
4222                 free_netdev(dev);
4223                 return ERR_PTR(err);
4224         }
4225
4226         err = rtnl_configure_link(dev, NULL);
4227         if (err < 0) {
4228                 LIST_HEAD(list_kill);
4229
4230                 vxlan_dellink(dev, &list_kill);
4231                 unregister_netdevice_many(&list_kill);
4232                 return ERR_PTR(err);
4233         }
4234
4235         return dev;
4236 }
4237 EXPORT_SYMBOL_GPL(vxlan_dev_create);
4238
4239 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
4240                                              struct net_device *dev)
4241 {
4242         struct vxlan_dev *vxlan, *next;
4243         LIST_HEAD(list_kill);
4244
4245         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4246                 struct vxlan_rdst *dst = &vxlan->default_dst;
4247
4248                 /* In case we created vxlan device with carrier
4249                  * and we loose the carrier due to module unload
4250                  * we also need to remove vxlan device. In other
4251                  * cases, it's not necessary and remote_ifindex
4252                  * is 0 here, so no matches.
4253                  */
4254                 if (dst->remote_ifindex == dev->ifindex)
4255                         vxlan_dellink(vxlan->dev, &list_kill);
4256         }
4257
4258         unregister_netdevice_many(&list_kill);
4259 }
4260
4261 static int vxlan_netdevice_event(struct notifier_block *unused,
4262                                  unsigned long event, void *ptr)
4263 {
4264         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4265         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
4266
4267         if (event == NETDEV_UNREGISTER) {
4268                 vxlan_offload_rx_ports(dev, false);
4269                 vxlan_handle_lowerdev_unregister(vn, dev);
4270         } else if (event == NETDEV_REGISTER) {
4271                 vxlan_offload_rx_ports(dev, true);
4272         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
4273                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
4274                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
4275         }
4276
4277         return NOTIFY_DONE;
4278 }
4279
4280 static struct notifier_block vxlan_notifier_block __read_mostly = {
4281         .notifier_call = vxlan_netdevice_event,
4282 };
4283
4284 static void
4285 vxlan_fdb_offloaded_set(struct net_device *dev,
4286                         struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4287 {
4288         struct vxlan_dev *vxlan = netdev_priv(dev);
4289         struct vxlan_rdst *rdst;
4290         struct vxlan_fdb *f;
4291         u32 hash_index;
4292
4293         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4294
4295         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4296
4297         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4298         if (!f)
4299                 goto out;
4300
4301         rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip,
4302                                    fdb_info->remote_port,
4303                                    fdb_info->remote_vni,
4304                                    fdb_info->remote_ifindex);
4305         if (!rdst)
4306                 goto out;
4307
4308         rdst->offloaded = fdb_info->offloaded;
4309
4310 out:
4311         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4312 }
4313
4314 static int
4315 vxlan_fdb_external_learn_add(struct net_device *dev,
4316                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4317 {
4318         struct vxlan_dev *vxlan = netdev_priv(dev);
4319         struct netlink_ext_ack *extack;
4320         u32 hash_index;
4321         int err;
4322
4323         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4324         extack = switchdev_notifier_info_to_extack(&fdb_info->info);
4325
4326         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4327         err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip,
4328                                NUD_REACHABLE,
4329                                NLM_F_CREATE | NLM_F_REPLACE,
4330                                fdb_info->remote_port,
4331                                fdb_info->vni,
4332                                fdb_info->remote_vni,
4333                                fdb_info->remote_ifindex,
4334                                NTF_USE | NTF_SELF | NTF_EXT_LEARNED,
4335                                false, extack);
4336         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4337
4338         return err;
4339 }
4340
4341 static int
4342 vxlan_fdb_external_learn_del(struct net_device *dev,
4343                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4344 {
4345         struct vxlan_dev *vxlan = netdev_priv(dev);
4346         struct vxlan_fdb *f;
4347         u32 hash_index;
4348         int err = 0;
4349
4350         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4351         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4352
4353         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4354         if (!f)
4355                 err = -ENOENT;
4356         else if (f->flags & NTF_EXT_LEARNED)
4357                 err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr,
4358                                          fdb_info->remote_ip,
4359                                          fdb_info->remote_port,
4360                                          fdb_info->vni,
4361                                          fdb_info->remote_vni,
4362                                          fdb_info->remote_ifindex,
4363                                          false);
4364
4365         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4366
4367         return err;
4368 }
4369
4370 static int vxlan_switchdev_event(struct notifier_block *unused,
4371                                  unsigned long event, void *ptr)
4372 {
4373         struct net_device *dev = switchdev_notifier_info_to_dev(ptr);
4374         struct switchdev_notifier_vxlan_fdb_info *fdb_info;
4375         int err = 0;
4376
4377         switch (event) {
4378         case SWITCHDEV_VXLAN_FDB_OFFLOADED:
4379                 vxlan_fdb_offloaded_set(dev, ptr);
4380                 break;
4381         case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE:
4382                 fdb_info = ptr;
4383                 err = vxlan_fdb_external_learn_add(dev, fdb_info);
4384                 if (err) {
4385                         err = notifier_from_errno(err);
4386                         break;
4387                 }
4388                 fdb_info->offloaded = true;
4389                 vxlan_fdb_offloaded_set(dev, fdb_info);
4390                 break;
4391         case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE:
4392                 fdb_info = ptr;
4393                 err = vxlan_fdb_external_learn_del(dev, fdb_info);
4394                 if (err) {
4395                         err = notifier_from_errno(err);
4396                         break;
4397                 }
4398                 fdb_info->offloaded = false;
4399                 vxlan_fdb_offloaded_set(dev, fdb_info);
4400                 break;
4401         }
4402
4403         return err;
4404 }
4405
4406 static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = {
4407         .notifier_call = vxlan_switchdev_event,
4408 };
4409
4410 static __net_init int vxlan_init_net(struct net *net)
4411 {
4412         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4413         unsigned int h;
4414
4415         INIT_LIST_HEAD(&vn->vxlan_list);
4416         spin_lock_init(&vn->sock_lock);
4417
4418         for (h = 0; h < PORT_HASH_SIZE; ++h)
4419                 INIT_HLIST_HEAD(&vn->sock_list[h]);
4420
4421         return 0;
4422 }
4423
4424 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
4425 {
4426         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4427         struct vxlan_dev *vxlan, *next;
4428         struct net_device *dev, *aux;
4429
4430         for_each_netdev_safe(net, dev, aux)
4431                 if (dev->rtnl_link_ops == &vxlan_link_ops)
4432                         unregister_netdevice_queue(dev, head);
4433
4434         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4435                 /* If vxlan->dev is in the same netns, it has already been added
4436                  * to the list by the previous loop.
4437                  */
4438                 if (!net_eq(dev_net(vxlan->dev), net))
4439                         unregister_netdevice_queue(vxlan->dev, head);
4440         }
4441
4442 }
4443
4444 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
4445 {
4446         struct net *net;
4447         LIST_HEAD(list);
4448         unsigned int h;
4449
4450         rtnl_lock();
4451         list_for_each_entry(net, net_list, exit_list)
4452                 vxlan_destroy_tunnels(net, &list);
4453
4454         unregister_netdevice_many(&list);
4455         rtnl_unlock();
4456
4457         list_for_each_entry(net, net_list, exit_list) {
4458                 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4459
4460                 for (h = 0; h < PORT_HASH_SIZE; ++h)
4461                         WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
4462         }
4463 }
4464
4465 static struct pernet_operations vxlan_net_ops = {
4466         .init = vxlan_init_net,
4467         .exit_batch = vxlan_exit_batch_net,
4468         .id   = &vxlan_net_id,
4469         .size = sizeof(struct vxlan_net),
4470 };
4471
4472 static int __init vxlan_init_module(void)
4473 {
4474         int rc;
4475
4476         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
4477
4478         rc = register_pernet_subsys(&vxlan_net_ops);
4479         if (rc)
4480                 goto out1;
4481
4482         rc = register_netdevice_notifier(&vxlan_notifier_block);
4483         if (rc)
4484                 goto out2;
4485
4486         rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block);
4487         if (rc)
4488                 goto out3;
4489
4490         rc = rtnl_link_register(&vxlan_link_ops);
4491         if (rc)
4492                 goto out4;
4493
4494         return 0;
4495 out4:
4496         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4497 out3:
4498         unregister_netdevice_notifier(&vxlan_notifier_block);
4499 out2:
4500         unregister_pernet_subsys(&vxlan_net_ops);
4501 out1:
4502         return rc;
4503 }
4504 late_initcall(vxlan_init_module);
4505
4506 static void __exit vxlan_cleanup_module(void)
4507 {
4508         rtnl_link_unregister(&vxlan_link_ops);
4509         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4510         unregister_netdevice_notifier(&vxlan_notifier_block);
4511         unregister_pernet_subsys(&vxlan_net_ops);
4512         /* rcu_barrier() is called by netns */
4513 }
4514 module_exit(vxlan_cleanup_module);
4515
4516 MODULE_LICENSE("GPL");
4517 MODULE_VERSION(VXLAN_VERSION);
4518 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
4519 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
4520 MODULE_ALIAS_RTNL_LINK("vxlan");