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 #include <net/nexthop.h>
  30
  31 #if IS_ENABLED(CONFIG_IPV6)
  32 #include <net/ip6_tunnel.h>
  33 #include <net/ip6_checksum.h>
  34 #endif
  35
  36 #define VXLAN_VERSION   "0.1"
  37
  38 #define PORT_HASH_BITS  8
  39 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  40 #define FDB_AGE_DEFAULT 300 /* 5 min */
  41 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  42
  43 /* UDP port for VXLAN traffic.
  44  * The IANA assigned port is 4789, but the Linux default is 8472
  45  * for compatibility with early adopters.
  46  */
  47 static unsigned short vxlan_port __read_mostly = 8472;
  48 module_param_named(udp_port, vxlan_port, ushort, 0444);
  49 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  50
  51 static bool log_ecn_error = true;
  52 module_param(log_ecn_error, bool, 0644);
  53 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  54
  55 static unsigned int vxlan_net_id;
  56 static struct rtnl_link_ops vxlan_link_ops;
  57
  58 static const u8 all_zeros_mac[ETH_ALEN + 2];
  59
  60 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  61
  62 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan);
  63
  64 /* per-network namespace private data for this module */
  65 struct vxlan_net {
  66         struct list_head  vxlan_list;
  67         struct hlist_head sock_list[PORT_HASH_SIZE];
  68         spinlock_t        sock_lock;
  69 };
  70
  71 /* Forwarding table entry */
  72 struct vxlan_fdb {
  73         struct hlist_node hlist;        /* linked list of entries */
  74         struct rcu_head   rcu;
  75         unsigned long     updated;      /* jiffies */
  76         unsigned long     used;
  77         struct list_head  remotes;
  78         u8                eth_addr[ETH_ALEN];
  79         u16               state;        /* see ndm_state */
  80         __be32            vni;
  81         u16               flags;        /* see ndm_flags and below */
  82         struct list_head  nh_list;
  83         struct nexthop __rcu *nh;
  84         struct vxlan_dev  __rcu *vdev;
  85 };
  86
  87 #define NTF_VXLAN_ADDED_BY_USER 0x100
  88
  89 /* salt for hash table */
  90 static u32 vxlan_salt __read_mostly;
  91
  92 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  93 {
  94         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  95                ip_tunnel_collect_metadata();
  96 }
  97
  98 #if IS_ENABLED(CONFIG_IPV6)
  99 static inline
 100 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 101 {
 102         if (a->sa.sa_family != b->sa.sa_family)
 103                 return false;
 104         if (a->sa.sa_family == AF_INET6)
 105                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 106         else
 107                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 108 }
 109
 110 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 111 {
 112         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 113                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 114                 ip->sa.sa_family = AF_INET6;
 115                 return 0;
 116         } else if (nla_len(nla) >= sizeof(__be32)) {
 117                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 118                 ip->sa.sa_family = AF_INET;
 119                 return 0;
 120         } else {
 121                 return -EAFNOSUPPORT;
 122         }
 123 }
 124
 125 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 126                               const union vxlan_addr *ip)
 127 {
 128         if (ip->sa.sa_family == AF_INET6)
 129                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 130         else
 131                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 132 }
 133
 134 #else /* !CONFIG_IPV6 */
 135
 136 static inline
 137 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 138 {
 139         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 140 }
 141
 142 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 143 {
 144         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 145                 return -EAFNOSUPPORT;
 146         } else if (nla_len(nla) >= sizeof(__be32)) {
 147                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 148                 ip->sa.sa_family = AF_INET;
 149                 return 0;
 150         } else {
 151                 return -EAFNOSUPPORT;
 152         }
 153 }
 154
 155 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 156                               const union vxlan_addr *ip)
 157 {
 158         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 159 }
 160 #endif
 161
 162 /* Virtual Network hash table head */
 163 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 164 {
 165         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 166 }
 167
 168 /* Socket hash table head */
 169 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 170 {
 171         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 172
 173         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 174 }
 175
 176 /* First remote destination for a forwarding entry.
 177  * Guaranteed to be non-NULL because remotes are never deleted.
 178  */
 179 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 180 {
 181         if (rcu_access_pointer(fdb->nh))
 182                 return NULL;
 183         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 184 }
 185
 186 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 187 {
 188         if (rcu_access_pointer(fdb->nh))
 189                 return NULL;
 190         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 191 }
 192
 193 /* Find VXLAN socket based on network namespace, address family, UDP port,
 194  * enabled unshareable flags and socket device binding (see l3mdev with
 195  * non-default VRF).
 196  */
 197 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 198                                           __be16 port, u32 flags, int ifindex)
 199 {
 200         struct vxlan_sock *vs;
 201
 202         flags &= VXLAN_F_RCV_FLAGS;
 203
 204         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 205                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 206                     vxlan_get_sk_family(vs) == family &&
 207                     vs->flags == flags &&
 208                     vs->sock->sk->sk_bound_dev_if == ifindex)
 209                         return vs;
 210         }
 211         return NULL;
 212 }
 213
 214 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, int ifindex,
 215                                            __be32 vni)
 216 {
 217         struct vxlan_dev_node *node;
 218
 219         /* For flow based devices, map all packets to VNI 0 */
 220         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 221                 vni = 0;
 222
 223         hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) {
 224                 if (node->vxlan->default_dst.remote_vni != vni)
 225                         continue;
 226
 227                 if (IS_ENABLED(CONFIG_IPV6)) {
 228                         const struct vxlan_config *cfg = &node->vxlan->cfg;
 229
 230                         if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) &&
 231                             cfg->remote_ifindex != ifindex)
 232                                 continue;
 233                 }
 234
 235                 return node->vxlan;
 236         }
 237
 238         return NULL;
 239 }
 240
 241 /* Look up VNI in a per net namespace table */
 242 static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex,
 243                                         __be32 vni, sa_family_t family,
 244                                         __be16 port, u32 flags)
 245 {
 246         struct vxlan_sock *vs;
 247
 248         vs = vxlan_find_sock(net, family, port, flags, ifindex);
 249         if (!vs)
 250                 return NULL;
 251
 252         return vxlan_vs_find_vni(vs, ifindex, vni);
 253 }
 254
 255 /* Fill in neighbour message in skbuff. */
 256 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 257                           const struct vxlan_fdb *fdb,
 258                           u32 portid, u32 seq, int type, unsigned int flags,
 259                           const struct vxlan_rdst *rdst)
 260 {
 261         unsigned long now = jiffies;
 262         struct nda_cacheinfo ci;
 263         bool send_ip, send_eth;
 264         struct nlmsghdr *nlh;
 265         struct nexthop *nh;
 266         struct ndmsg *ndm;
 267         int nh_family;
 268         u32 nh_id;
 269
 270         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 271         if (nlh == NULL)
 272                 return -EMSGSIZE;
 273
 274         ndm = nlmsg_data(nlh);
 275         memset(ndm, 0, sizeof(*ndm));
 276
 277         send_eth = send_ip = true;
 278
 279         rcu_read_lock();
 280         nh = rcu_dereference(fdb->nh);
 281         if (nh) {
 282                 nh_family = nexthop_get_family(nh);
 283                 nh_id = nh->id;
 284         }
 285         rcu_read_unlock();
 286
 287         if (type == RTM_GETNEIGH) {
 288                 if (rdst) {
 289                         send_ip = !vxlan_addr_any(&rdst->remote_ip);
 290                         ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
 291                 } else if (nh) {
 292                         ndm->ndm_family = nh_family;
 293                 }
 294                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 295         } else
 296                 ndm->ndm_family = AF_BRIDGE;
 297         ndm->ndm_state = fdb->state;
 298         ndm->ndm_ifindex = vxlan->dev->ifindex;
 299         ndm->ndm_flags = fdb->flags;
 300         if (rdst && rdst->offloaded)
 301                 ndm->ndm_flags |= NTF_OFFLOADED;
 302         ndm->ndm_type = RTN_UNICAST;
 303
 304         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 305             nla_put_s32(skb, NDA_LINK_NETNSID,
 306                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 307                 goto nla_put_failure;
 308
 309         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 310                 goto nla_put_failure;
 311         if (nh) {
 312                 if (nla_put_u32(skb, NDA_NH_ID, nh_id))
 313                         goto nla_put_failure;
 314         } else if (rdst) {
 315                 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST,
 316                                                   &rdst->remote_ip))
 317                         goto nla_put_failure;
 318
 319                 if (rdst->remote_port &&
 320                     rdst->remote_port != vxlan->cfg.dst_port &&
 321                     nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 322                         goto nla_put_failure;
 323                 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 324                     nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 325                         goto nla_put_failure;
 326                 if (rdst->remote_ifindex &&
 327                     nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 328                         goto nla_put_failure;
 329         }
 330
 331         if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
 332             nla_put_u32(skb, NDA_SRC_VNI,
 333                         be32_to_cpu(fdb->vni)))
 334                 goto nla_put_failure;
 335
 336         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 337         ci.ndm_confirmed = 0;
 338         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 339         ci.ndm_refcnt    = 0;
 340
 341         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 342                 goto nla_put_failure;
 343
 344         nlmsg_end(skb, nlh);
 345         return 0;
 346
 347 nla_put_failure:
 348         nlmsg_cancel(skb, nlh);
 349         return -EMSGSIZE;
 350 }
 351
 352 static inline size_t vxlan_nlmsg_size(void)
 353 {
 354         return NLMSG_ALIGN(sizeof(struct ndmsg))
 355                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 356                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 357                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 358                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 359                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 360                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 361                 + nla_total_size(sizeof(struct nda_cacheinfo));
 362 }
 363
 364 static void __vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 365                                struct vxlan_rdst *rd, int type)
 366 {
 367         struct net *net = dev_net(vxlan->dev);
 368         struct sk_buff *skb;
 369         int err = -ENOBUFS;
 370
 371         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 372         if (skb == NULL)
 373                 goto errout;
 374
 375         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 376         if (err < 0) {
 377                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 378                 WARN_ON(err == -EMSGSIZE);
 379                 kfree_skb(skb);
 380                 goto errout;
 381         }
 382
 383         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 384         return;
 385 errout:
 386         if (err < 0)
 387                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 388 }
 389
 390 static void vxlan_fdb_switchdev_notifier_info(const struct vxlan_dev *vxlan,
 391                             const struct vxlan_fdb *fdb,
 392                             const struct vxlan_rdst *rd,
 393                             struct netlink_ext_ack *extack,
 394                             struct switchdev_notifier_vxlan_fdb_info *fdb_info)
 395 {
 396         fdb_info->info.dev = vxlan->dev;
 397         fdb_info->info.extack = extack;
 398         fdb_info->remote_ip = rd->remote_ip;
 399         fdb_info->remote_port = rd->remote_port;
 400         fdb_info->remote_vni = rd->remote_vni;
 401         fdb_info->remote_ifindex = rd->remote_ifindex;
 402         memcpy(fdb_info->eth_addr, fdb->eth_addr, ETH_ALEN);
 403         fdb_info->vni = fdb->vni;
 404         fdb_info->offloaded = rd->offloaded;
 405         fdb_info->added_by_user = fdb->flags & NTF_VXLAN_ADDED_BY_USER;
 406 }
 407
 408 static int vxlan_fdb_switchdev_call_notifiers(struct vxlan_dev *vxlan,
 409                                               struct vxlan_fdb *fdb,
 410                                               struct vxlan_rdst *rd,
 411                                               bool adding,
 412                                               struct netlink_ext_ack *extack)
 413 {
 414         struct switchdev_notifier_vxlan_fdb_info info;
 415         enum switchdev_notifier_type notifier_type;
 416         int ret;
 417
 418         if (WARN_ON(!rd))
 419                 return 0;
 420
 421         notifier_type = adding ? SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE
 422                                : SWITCHDEV_VXLAN_FDB_DEL_TO_DEVICE;
 423         vxlan_fdb_switchdev_notifier_info(vxlan, fdb, rd, NULL, &info);
 424         ret = call_switchdev_notifiers(notifier_type, vxlan->dev,
 425                                        &info.info, extack);
 426         return notifier_to_errno(ret);
 427 }
 428
 429 static int vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 430                             struct vxlan_rdst *rd, int type, bool swdev_notify,
 431                             struct netlink_ext_ack *extack)
 432 {
 433         int err;
 434
 435         if (swdev_notify && rd) {
 436                 switch (type) {
 437                 case RTM_NEWNEIGH:
 438                         err = vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
 439                                                                  true, extack);
 440                         if (err)
 441                                 return err;
 442                         break;
 443                 case RTM_DELNEIGH:
 444                         vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
 445                                                            false, extack);
 446                         break;
 447                 }
 448         }
 449
 450         __vxlan_fdb_notify(vxlan, fdb, rd, type);
 451         return 0;
 452 }
 453
 454 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 455 {
 456         struct vxlan_dev *vxlan = netdev_priv(dev);
 457         struct vxlan_fdb f = {
 458                 .state = NUD_STALE,
 459         };
 460         struct vxlan_rdst remote = {
 461                 .remote_ip = *ipa, /* goes to NDA_DST */
 462                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 463         };
 464
 465         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
 466 }
 467
 468 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 469 {
 470         struct vxlan_fdb f = {
 471                 .state = NUD_STALE,
 472         };
 473         struct vxlan_rdst remote = { };
 474
 475         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 476
 477         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
 478 }
 479
 480 /* Hash Ethernet address */
 481 static u32 eth_hash(const unsigned char *addr)
 482 {
 483         u64 value = get_unaligned((u64 *)addr);
 484
 485         /* only want 6 bytes */
 486 #ifdef __BIG_ENDIAN
 487         value >>= 16;
 488 #else
 489         value <<= 16;
 490 #endif
 491         return hash_64(value, FDB_HASH_BITS);
 492 }
 493
 494 static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
 495 {
 496         /* use 1 byte of OUI and 3 bytes of NIC */
 497         u32 key = get_unaligned((u32 *)(addr + 2));
 498
 499         return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
 500 }
 501
 502 static u32 fdb_head_index(struct vxlan_dev *vxlan, const u8 *mac, __be32 vni)
 503 {
 504         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)
 505                 return eth_vni_hash(mac, vni);
 506         else
 507                 return eth_hash(mac);
 508 }
 509
 510 /* Hash chain to use given mac address */
 511 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 512                                                 const u8 *mac, __be32 vni)
 513 {
 514         return &vxlan->fdb_head[fdb_head_index(vxlan, mac, vni)];
 515 }
 516
 517 /* Look up Ethernet address in forwarding table */
 518 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 519                                           const u8 *mac, __be32 vni)
 520 {
 521         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
 522         struct vxlan_fdb *f;
 523
 524         hlist_for_each_entry_rcu(f, head, hlist) {
 525                 if (ether_addr_equal(mac, f->eth_addr)) {
 526                         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
 527                                 if (vni == f->vni)
 528                                         return f;
 529                         } else {
 530                                 return f;
 531                         }
 532                 }
 533         }
 534
 535         return NULL;
 536 }
 537
 538 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 539                                         const u8 *mac, __be32 vni)
 540 {
 541         struct vxlan_fdb *f;
 542
 543         f = __vxlan_find_mac(vxlan, mac, vni);
 544         if (f && f->used != jiffies)
 545                 f->used = jiffies;
 546
 547         return f;
 548 }
 549
 550 /* caller should hold vxlan->hash_lock */
 551 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 552                                               union vxlan_addr *ip, __be16 port,
 553                                               __be32 vni, __u32 ifindex)
 554 {
 555         struct vxlan_rdst *rd;
 556
 557         list_for_each_entry(rd, &f->remotes, list) {
 558                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 559                     rd->remote_port == port &&
 560                     rd->remote_vni == vni &&
 561                     rd->remote_ifindex == ifindex)
 562                         return rd;
 563         }
 564
 565         return NULL;
 566 }
 567
 568 int vxlan_fdb_find_uc(struct net_device *dev, const u8 *mac, __be32 vni,
 569                       struct switchdev_notifier_vxlan_fdb_info *fdb_info)
 570 {
 571         struct vxlan_dev *vxlan = netdev_priv(dev);
 572         u8 eth_addr[ETH_ALEN + 2] = { 0 };
 573         struct vxlan_rdst *rdst;
 574         struct vxlan_fdb *f;
 575         int rc = 0;
 576
 577         if (is_multicast_ether_addr(mac) ||
 578             is_zero_ether_addr(mac))
 579                 return -EINVAL;
 580
 581         ether_addr_copy(eth_addr, mac);
 582
 583         rcu_read_lock();
 584
 585         f = __vxlan_find_mac(vxlan, eth_addr, vni);
 586         if (!f) {
 587                 rc = -ENOENT;
 588                 goto out;
 589         }
 590
 591         rdst = first_remote_rcu(f);
 592         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, NULL, fdb_info);
 593
 594 out:
 595         rcu_read_unlock();
 596         return rc;
 597 }
 598 EXPORT_SYMBOL_GPL(vxlan_fdb_find_uc);
 599
 600 static int vxlan_fdb_notify_one(struct notifier_block *nb,
 601                                 const struct vxlan_dev *vxlan,
 602                                 const struct vxlan_fdb *f,
 603                                 const struct vxlan_rdst *rdst,
 604                                 struct netlink_ext_ack *extack)
 605 {
 606         struct switchdev_notifier_vxlan_fdb_info fdb_info;
 607         int rc;
 608
 609         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, extack, &fdb_info);
 610         rc = nb->notifier_call(nb, SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE,
 611                                &fdb_info);
 612         return notifier_to_errno(rc);
 613 }
 614
 615 int vxlan_fdb_replay(const struct net_device *dev, __be32 vni,
 616                      struct notifier_block *nb,
 617                      struct netlink_ext_ack *extack)
 618 {
 619         struct vxlan_dev *vxlan;
 620         struct vxlan_rdst *rdst;
 621         struct vxlan_fdb *f;
 622         unsigned int h;
 623         int rc = 0;
 624
 625         if (!netif_is_vxlan(dev))
 626                 return -EINVAL;
 627         vxlan = netdev_priv(dev);
 628
 629         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 630                 spin_lock_bh(&vxlan->hash_lock[h]);
 631                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) {
 632                         if (f->vni == vni) {
 633                                 list_for_each_entry(rdst, &f->remotes, list) {
 634                                         rc = vxlan_fdb_notify_one(nb, vxlan,
 635                                                                   f, rdst,
 636                                                                   extack);
 637                                         if (rc)
 638                                                 goto unlock;
 639                                 }
 640                         }
 641                 }
 642                 spin_unlock_bh(&vxlan->hash_lock[h]);
 643         }
 644         return 0;
 645
 646 unlock:
 647         spin_unlock_bh(&vxlan->hash_lock[h]);
 648         return rc;
 649 }
 650 EXPORT_SYMBOL_GPL(vxlan_fdb_replay);
 651
 652 void vxlan_fdb_clear_offload(const struct net_device *dev, __be32 vni)
 653 {
 654         struct vxlan_dev *vxlan;
 655         struct vxlan_rdst *rdst;
 656         struct vxlan_fdb *f;
 657         unsigned int h;
 658
 659         if (!netif_is_vxlan(dev))
 660                 return;
 661         vxlan = netdev_priv(dev);
 662
 663         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 664                 spin_lock_bh(&vxlan->hash_lock[h]);
 665                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist)
 666                         if (f->vni == vni)
 667                                 list_for_each_entry(rdst, &f->remotes, list)
 668                                         rdst->offloaded = false;
 669                 spin_unlock_bh(&vxlan->hash_lock[h]);
 670         }
 671
 672 }
 673 EXPORT_SYMBOL_GPL(vxlan_fdb_clear_offload);
 674
 675 /* Replace destination of unicast mac */
 676 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 677                              union vxlan_addr *ip, __be16 port, __be32 vni,
 678                              __u32 ifindex, struct vxlan_rdst *oldrd)
 679 {
 680         struct vxlan_rdst *rd;
 681
 682         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 683         if (rd)
 684                 return 0;
 685
 686         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 687         if (!rd)
 688                 return 0;
 689
 690         *oldrd = *rd;
 691         dst_cache_reset(&rd->dst_cache);
 692         rd->remote_ip = *ip;
 693         rd->remote_port = port;
 694         rd->remote_vni = vni;
 695         rd->remote_ifindex = ifindex;
 696         rd->offloaded = false;
 697         return 1;
 698 }
 699
 700 /* Add/update destinations for multicast */
 701 static int vxlan_fdb_append(struct vxlan_fdb *f,
 702                             union vxlan_addr *ip, __be16 port, __be32 vni,
 703                             __u32 ifindex, struct vxlan_rdst **rdp)
 704 {
 705         struct vxlan_rdst *rd;
 706
 707         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 708         if (rd)
 709                 return 0;
 710
 711         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 712         if (rd == NULL)
 713                 return -ENOBUFS;
 714
 715         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 716                 kfree(rd);
 717                 return -ENOBUFS;
 718         }
 719
 720         rd->remote_ip = *ip;
 721         rd->remote_port = port;
 722         rd->offloaded = false;
 723         rd->remote_vni = vni;
 724         rd->remote_ifindex = ifindex;
 725
 726         list_add_tail_rcu(&rd->list, &f->remotes);
 727
 728         *rdp = rd;
 729         return 1;
 730 }
 731
 732 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 733                                           unsigned int off,
 734                                           struct vxlanhdr *vh, size_t hdrlen,
 735                                           __be32 vni_field,
 736                                           struct gro_remcsum *grc,
 737                                           bool nopartial)
 738 {
 739         size_t start, offset;
 740
 741         if (skb->remcsum_offload)
 742                 return vh;
 743
 744         if (!NAPI_GRO_CB(skb)->csum_valid)
 745                 return NULL;
 746
 747         start = vxlan_rco_start(vni_field);
 748         offset = start + vxlan_rco_offset(vni_field);
 749
 750         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 751                                      start, offset, grc, nopartial);
 752
 753         skb->remcsum_offload = 1;
 754
 755         return vh;
 756 }
 757
 758 static struct sk_buff *vxlan_gro_receive(struct sock *sk,
 759                                          struct list_head *head,
 760                                          struct sk_buff *skb)
 761 {
 762         struct sk_buff *pp = NULL;
 763         struct sk_buff *p;
 764         struct vxlanhdr *vh, *vh2;
 765         unsigned int hlen, off_vx;
 766         int flush = 1;
 767         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 768         __be32 flags;
 769         struct gro_remcsum grc;
 770
 771         skb_gro_remcsum_init(&grc);
 772
 773         off_vx = skb_gro_offset(skb);
 774         hlen = off_vx + sizeof(*vh);
 775         vh   = skb_gro_header_fast(skb, off_vx);
 776         if (skb_gro_header_hard(skb, hlen)) {
 777                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 778                 if (unlikely(!vh))
 779                         goto out;
 780         }
 781
 782         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 783
 784         flags = vh->vx_flags;
 785
 786         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 787                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 788                                        vh->vx_vni, &grc,
 789                                        !!(vs->flags &
 790                                           VXLAN_F_REMCSUM_NOPARTIAL));
 791
 792                 if (!vh)
 793                         goto out;
 794         }
 795
 796         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 797
 798         list_for_each_entry(p, head, list) {
 799                 if (!NAPI_GRO_CB(p)->same_flow)
 800                         continue;
 801
 802                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 803                 if (vh->vx_flags != vh2->vx_flags ||
 804                     vh->vx_vni != vh2->vx_vni) {
 805                         NAPI_GRO_CB(p)->same_flow = 0;
 806                         continue;
 807                 }
 808         }
 809
 810         pp = call_gro_receive(eth_gro_receive, head, skb);
 811         flush = 0;
 812
 813 out:
 814         skb_gro_flush_final_remcsum(skb, pp, flush, &grc);
 815
 816         return pp;
 817 }
 818
 819 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 820 {
 821         /* Sets 'skb->inner_mac_header' since we are always called with
 822          * 'skb->encapsulation' set.
 823          */
 824         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 825 }
 826
 827 static struct vxlan_fdb *vxlan_fdb_alloc(struct vxlan_dev *vxlan, const u8 *mac,
 828                                          __u16 state, __be32 src_vni,
 829                                          __u16 ndm_flags)
 830 {
 831         struct vxlan_fdb *f;
 832
 833         f = kmalloc(sizeof(*f), GFP_ATOMIC);
 834         if (!f)
 835                 return NULL;
 836         f->state = state;
 837         f->flags = ndm_flags;
 838         f->updated = f->used = jiffies;
 839         f->vni = src_vni;
 840         f->nh = NULL;
 841         RCU_INIT_POINTER(f->vdev, vxlan);
 842         INIT_LIST_HEAD(&f->nh_list);
 843         INIT_LIST_HEAD(&f->remotes);
 844         memcpy(f->eth_addr, mac, ETH_ALEN);
 845
 846         return f;
 847 }
 848
 849 static void vxlan_fdb_insert(struct vxlan_dev *vxlan, const u8 *mac,
 850                              __be32 src_vni, struct vxlan_fdb *f)
 851 {
 852         ++vxlan->addrcnt;
 853         hlist_add_head_rcu(&f->hlist,
 854                            vxlan_fdb_head(vxlan, mac, src_vni));
 855 }
 856
 857 static int vxlan_fdb_nh_update(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 858                                u32 nhid, struct netlink_ext_ack *extack)
 859 {
 860         struct nexthop *old_nh = rtnl_dereference(fdb->nh);
 861         struct nexthop *nh;
 862         int err = -EINVAL;
 863
 864         if (old_nh && old_nh->id == nhid)
 865                 return 0;
 866
 867         nh = nexthop_find_by_id(vxlan->net, nhid);
 868         if (!nh) {
 869                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
 870                 goto err_inval;
 871         }
 872
 873         if (nh) {
 874                 if (!nexthop_get(nh)) {
 875                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
 876                         nh = NULL;
 877                         goto err_inval;
 878                 }
 879                 if (!nexthop_is_fdb(nh)) {
 880                         NL_SET_ERR_MSG(extack, "Nexthop is not a fdb nexthop");
 881                         goto err_inval;
 882                 }
 883
 884                 if (!nexthop_is_multipath(nh)) {
 885                         NL_SET_ERR_MSG(extack, "Nexthop is not a multipath group");
 886                         goto err_inval;
 887                 }
 888
 889                 /* check nexthop group family */
 890                 switch (vxlan->default_dst.remote_ip.sa.sa_family) {
 891                 case AF_INET:
 892                         if (!nexthop_has_v4(nh)) {
 893                                 err = -EAFNOSUPPORT;
 894                                 NL_SET_ERR_MSG(extack, "Nexthop group family not supported");
 895                                 goto err_inval;
 896                         }
 897                         break;
 898                 case AF_INET6:
 899                         if (nexthop_has_v4(nh)) {
 900                                 err = -EAFNOSUPPORT;
 901                                 NL_SET_ERR_MSG(extack, "Nexthop group family not supported");
 902                                 goto err_inval;
 903                         }
 904                 }
 905         }
 906
 907         if (old_nh) {
 908                 list_del_rcu(&fdb->nh_list);
 909                 nexthop_put(old_nh);
 910         }
 911         rcu_assign_pointer(fdb->nh, nh);
 912         list_add_tail_rcu(&fdb->nh_list, &nh->fdb_list);
 913         return 1;
 914
 915 err_inval:
 916         if (nh)
 917                 nexthop_put(nh);
 918         return err;
 919 }
 920
 921 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 922                             const u8 *mac, union vxlan_addr *ip,
 923                             __u16 state, __be16 port, __be32 src_vni,
 924                             __be32 vni, __u32 ifindex, __u16 ndm_flags,
 925                             u32 nhid, struct vxlan_fdb **fdb,
 926                             struct netlink_ext_ack *extack)
 927 {
 928         struct vxlan_rdst *rd = NULL;
 929         struct vxlan_fdb *f;
 930         int rc;
 931
 932         if (vxlan->cfg.addrmax &&
 933             vxlan->addrcnt >= vxlan->cfg.addrmax)
 934                 return -ENOSPC;
 935
 936         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 937         f = vxlan_fdb_alloc(vxlan, mac, state, src_vni, ndm_flags);
 938         if (!f)
 939                 return -ENOMEM;
 940
 941         if (nhid)
 942                 rc = vxlan_fdb_nh_update(vxlan, f, nhid, extack);
 943         else
 944                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 945         if (rc < 0)
 946                 goto errout;
 947
 948         *fdb = f;
 949
 950         return 0;
 951
 952 errout:
 953         kfree(f);
 954         return rc;
 955 }
 956
 957 static void __vxlan_fdb_free(struct vxlan_fdb *f)
 958 {
 959         struct vxlan_rdst *rd, *nd;
 960         struct nexthop *nh;
 961
 962         nh = rcu_dereference_raw(f->nh);
 963         if (nh) {
 964                 rcu_assign_pointer(f->nh, NULL);
 965                 rcu_assign_pointer(f->vdev, NULL);
 966                 nexthop_put(nh);
 967         }
 968
 969         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 970                 dst_cache_destroy(&rd->dst_cache);
 971                 kfree(rd);
 972         }
 973         kfree(f);
 974 }
 975
 976 static void vxlan_fdb_free(struct rcu_head *head)
 977 {
 978         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 979
 980         __vxlan_fdb_free(f);
 981 }
 982
 983 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 984                               bool do_notify, bool swdev_notify)
 985 {
 986         struct vxlan_rdst *rd;
 987
 988         netdev_dbg(vxlan->dev, "delete %pM\n", f->eth_addr);
 989
 990         --vxlan->addrcnt;
 991         if (do_notify) {
 992                 if (rcu_access_pointer(f->nh))
 993                         vxlan_fdb_notify(vxlan, f, NULL, RTM_DELNEIGH,
 994                                          swdev_notify, NULL);
 995                 else
 996                         list_for_each_entry(rd, &f->remotes, list)
 997                                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH,
 998                                                  swdev_notify, NULL);
 999         }
1000
1001         hlist_del_rcu(&f->hlist);
1002         list_del_rcu(&f->nh_list);
1003         call_rcu(&f->rcu, vxlan_fdb_free);
1004 }
1005
1006 static void vxlan_dst_free(struct rcu_head *head)
1007 {
1008         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
1009
1010         dst_cache_destroy(&rd->dst_cache);
1011         kfree(rd);
1012 }
1013
1014 static int vxlan_fdb_update_existing(struct vxlan_dev *vxlan,
1015                                      union vxlan_addr *ip,
1016                                      __u16 state, __u16 flags,
1017                                      __be16 port, __be32 vni,
1018                                      __u32 ifindex, __u16 ndm_flags,
1019                                      struct vxlan_fdb *f, u32 nhid,
1020                                      bool swdev_notify,
1021                                      struct netlink_ext_ack *extack)
1022 {
1023         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
1024         struct vxlan_rdst *rd = NULL;
1025         struct vxlan_rdst oldrd;
1026         int notify = 0;
1027         int rc = 0;
1028         int err;
1029
1030         if (nhid && !rcu_access_pointer(f->nh)) {
1031                 NL_SET_ERR_MSG(extack,
1032                                "Cannot replace an existing non nexthop fdb with a nexthop");
1033                 return -EOPNOTSUPP;
1034         }
1035
1036         if (nhid && (flags & NLM_F_APPEND)) {
1037                 NL_SET_ERR_MSG(extack,
1038                                "Cannot append to a nexthop fdb");
1039                 return -EOPNOTSUPP;
1040         }
1041
1042         /* Do not allow an externally learned entry to take over an entry added
1043          * by the user.
1044          */
1045         if (!(fdb_flags & NTF_EXT_LEARNED) ||
1046             !(f->flags & NTF_VXLAN_ADDED_BY_USER)) {
1047                 if (f->state != state) {
1048                         f->state = state;
1049                         f->updated = jiffies;
1050                         notify = 1;
1051                 }
1052                 if (f->flags != fdb_flags) {
1053                         f->flags = fdb_flags;
1054                         f->updated = jiffies;
1055                         notify = 1;
1056                 }
1057         }
1058
1059         if ((flags & NLM_F_REPLACE)) {
1060                 /* Only change unicasts */
1061                 if (!(is_multicast_ether_addr(f->eth_addr) ||
1062                       is_zero_ether_addr(f->eth_addr))) {
1063                         if (nhid) {
1064                                 rc = vxlan_fdb_nh_update(vxlan, f, nhid, extack);
1065                                 if (rc < 0)
1066                                         return rc;
1067                         } else {
1068                                 rc = vxlan_fdb_replace(f, ip, port, vni,
1069                                                        ifindex, &oldrd);
1070                         }
1071                         notify |= rc;
1072                 } else {
1073                         NL_SET_ERR_MSG(extack, "Cannot replace non-unicast fdb entries");
1074                         return -EOPNOTSUPP;
1075                 }
1076         }
1077         if ((flags & NLM_F_APPEND) &&
1078             (is_multicast_ether_addr(f->eth_addr) ||
1079              is_zero_ether_addr(f->eth_addr))) {
1080                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
1081
1082                 if (rc < 0)
1083                         return rc;
1084                 notify |= rc;
1085         }
1086
1087         if (ndm_flags & NTF_USE)
1088                 f->used = jiffies;
1089
1090         if (notify) {
1091                 if (rd == NULL)
1092                         rd = first_remote_rtnl(f);
1093
1094                 err = vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH,
1095                                        swdev_notify, extack);
1096                 if (err)
1097                         goto err_notify;
1098         }
1099
1100         return 0;
1101
1102 err_notify:
1103         if (nhid)
1104                 return err;
1105         if ((flags & NLM_F_REPLACE) && rc)
1106                 *rd = oldrd;
1107         else if ((flags & NLM_F_APPEND) && rc) {
1108                 list_del_rcu(&rd->list);
1109                 call_rcu(&rd->rcu, vxlan_dst_free);
1110         }
1111         return err;
1112 }
1113
1114 static int vxlan_fdb_update_create(struct vxlan_dev *vxlan,
1115                                    const u8 *mac, union vxlan_addr *ip,
1116                                    __u16 state, __u16 flags,
1117                                    __be16 port, __be32 src_vni, __be32 vni,
1118                                    __u32 ifindex, __u16 ndm_flags, u32 nhid,
1119                                    bool swdev_notify,
1120                                    struct netlink_ext_ack *extack)
1121 {
1122         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
1123         struct vxlan_fdb *f;
1124         int rc;
1125
1126         /* Disallow replace to add a multicast entry */
1127         if ((flags & NLM_F_REPLACE) &&
1128             (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
1129                 return -EOPNOTSUPP;
1130
1131         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
1132         rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni,
1133                               vni, ifindex, fdb_flags, nhid, &f, extack);
1134         if (rc < 0)
1135                 return rc;
1136
1137         vxlan_fdb_insert(vxlan, mac, src_vni, f);
1138         rc = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH,
1139                               swdev_notify, extack);
1140         if (rc)
1141                 goto err_notify;
1142
1143         return 0;
1144
1145 err_notify:
1146         vxlan_fdb_destroy(vxlan, f, false, false);
1147         return rc;
1148 }
1149
1150 /* Add new entry to forwarding table -- assumes lock held */
1151 static int vxlan_fdb_update(struct vxlan_dev *vxlan,
1152                             const u8 *mac, union vxlan_addr *ip,
1153                             __u16 state, __u16 flags,
1154                             __be16 port, __be32 src_vni, __be32 vni,
1155                             __u32 ifindex, __u16 ndm_flags, u32 nhid,
1156                             bool swdev_notify,
1157                             struct netlink_ext_ack *extack)
1158 {
1159         struct vxlan_fdb *f;
1160
1161         f = __vxlan_find_mac(vxlan, mac, src_vni);
1162         if (f) {
1163                 if (flags & NLM_F_EXCL) {
1164                         netdev_dbg(vxlan->dev,
1165                                    "lost race to create %pM\n", mac);
1166                         return -EEXIST;
1167                 }
1168
1169                 return vxlan_fdb_update_existing(vxlan, ip, state, flags, port,
1170                                                  vni, ifindex, ndm_flags, f,
1171                                                  nhid, swdev_notify, extack);
1172         } else {
1173                 if (!(flags & NLM_F_CREATE))
1174                         return -ENOENT;
1175
1176                 return vxlan_fdb_update_create(vxlan, mac, ip, state, flags,
1177                                                port, src_vni, vni, ifindex,
1178                                                ndm_flags, nhid, swdev_notify,
1179                                                extack);
1180         }
1181 }
1182
1183 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
1184                                   struct vxlan_rdst *rd, bool swdev_notify)
1185 {
1186         list_del_rcu(&rd->list);
1187         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, swdev_notify, NULL);
1188         call_rcu(&rd->rcu, vxlan_dst_free);
1189 }
1190
1191 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
1192                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
1193                            __be32 *vni, u32 *ifindex, u32 *nhid)
1194 {
1195         struct net *net = dev_net(vxlan->dev);
1196         int err;
1197
1198         if (tb[NDA_NH_ID] && (tb[NDA_DST] || tb[NDA_VNI] || tb[NDA_IFINDEX] ||
1199             tb[NDA_PORT]))
1200                 return -EINVAL;
1201
1202         if (tb[NDA_DST]) {
1203                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
1204                 if (err)
1205                         return err;
1206         } else {
1207                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
1208
1209                 if (remote->sa.sa_family == AF_INET) {
1210                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
1211                         ip->sa.sa_family = AF_INET;
1212 #if IS_ENABLED(CONFIG_IPV6)
1213                 } else {
1214                         ip->sin6.sin6_addr = in6addr_any;
1215                         ip->sa.sa_family = AF_INET6;
1216 #endif
1217                 }
1218         }
1219
1220         if (tb[NDA_PORT]) {
1221                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
1222                         return -EINVAL;
1223                 *port = nla_get_be16(tb[NDA_PORT]);
1224         } else {
1225                 *port = vxlan->cfg.dst_port;
1226         }
1227
1228         if (tb[NDA_VNI]) {
1229                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
1230                         return -EINVAL;
1231                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1232         } else {
1233                 *vni = vxlan->default_dst.remote_vni;
1234         }
1235
1236         if (tb[NDA_SRC_VNI]) {
1237                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
1238                         return -EINVAL;
1239                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
1240         } else {
1241                 *src_vni = vxlan->default_dst.remote_vni;
1242         }
1243
1244         if (tb[NDA_IFINDEX]) {
1245                 struct net_device *tdev;
1246
1247                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
1248                         return -EINVAL;
1249                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
1250                 tdev = __dev_get_by_index(net, *ifindex);
1251                 if (!tdev)
1252                         return -EADDRNOTAVAIL;
1253         } else {
1254                 *ifindex = 0;
1255         }
1256
1257         if (tb[NDA_NH_ID])
1258                 *nhid = nla_get_u32(tb[NDA_NH_ID]);
1259         else
1260                 *nhid = 0;
1261
1262         return 0;
1263 }
1264
1265 /* Add static entry (via netlink) */
1266 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
1267                          struct net_device *dev,
1268                          const unsigned char *addr, u16 vid, u16 flags,
1269                          struct netlink_ext_ack *extack)
1270 {
1271         struct vxlan_dev *vxlan = netdev_priv(dev);
1272         /* struct net *net = dev_net(vxlan->dev); */
1273         union vxlan_addr ip;
1274         __be16 port;
1275         __be32 src_vni, vni;
1276         u32 ifindex, nhid;
1277         u32 hash_index;
1278         int err;
1279
1280         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
1281                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
1282                         ndm->ndm_state);
1283                 return -EINVAL;
1284         }
1285
1286         if (!tb || (!tb[NDA_DST] && !tb[NDA_NH_ID]))
1287                 return -EINVAL;
1288
1289         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex,
1290                               &nhid);
1291         if (err)
1292                 return err;
1293
1294         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
1295                 return -EAFNOSUPPORT;
1296
1297         hash_index = fdb_head_index(vxlan, addr, src_vni);
1298         spin_lock_bh(&vxlan->hash_lock[hash_index]);
1299         err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags,
1300                                port, src_vni, vni, ifindex,
1301                                ndm->ndm_flags | NTF_VXLAN_ADDED_BY_USER,
1302                                nhid, true, extack);
1303         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1304
1305         return err;
1306 }
1307
1308 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
1309                               const unsigned char *addr, union vxlan_addr ip,
1310                               __be16 port, __be32 src_vni, __be32 vni,
1311                               u32 ifindex, bool swdev_notify)
1312 {
1313         struct vxlan_rdst *rd = NULL;
1314         struct vxlan_fdb *f;
1315         int err = -ENOENT;
1316
1317         f = vxlan_find_mac(vxlan, addr, src_vni);
1318         if (!f)
1319                 return err;
1320
1321         if (!vxlan_addr_any(&ip)) {
1322                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
1323                 if (!rd)
1324                         goto out;
1325         }
1326
1327         /* remove a destination if it's not the only one on the list,
1328          * otherwise destroy the fdb entry
1329          */
1330         if (rd && !list_is_singular(&f->remotes)) {
1331                 vxlan_fdb_dst_destroy(vxlan, f, rd, swdev_notify);
1332                 goto out;
1333         }
1334
1335         vxlan_fdb_destroy(vxlan, f, true, swdev_notify);
1336
1337 out:
1338         return 0;
1339 }
1340
1341 /* Delete entry (via netlink) */
1342 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
1343                             struct net_device *dev,
1344                             const unsigned char *addr, u16 vid)
1345 {
1346         struct vxlan_dev *vxlan = netdev_priv(dev);
1347         union vxlan_addr ip;
1348         __be32 src_vni, vni;
1349         u32 ifindex, nhid;
1350         u32 hash_index;
1351         __be16 port;
1352         int err;
1353
1354         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex,
1355                               &nhid);
1356         if (err)
1357                 return err;
1358
1359         hash_index = fdb_head_index(vxlan, addr, src_vni);
1360         spin_lock_bh(&vxlan->hash_lock[hash_index]);
1361         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
1362                                  true);
1363         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1364
1365         return err;
1366 }
1367
1368 /* Dump forwarding table */
1369 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1370                           struct net_device *dev,
1371                           struct net_device *filter_dev, int *idx)
1372 {
1373         struct vxlan_dev *vxlan = netdev_priv(dev);
1374         unsigned int h;
1375         int err = 0;
1376
1377         for (h = 0; h < FDB_HASH_SIZE; ++h) {
1378                 struct vxlan_fdb *f;
1379
1380                 rcu_read_lock();
1381                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
1382                         struct vxlan_rdst *rd;
1383
1384                         if (rcu_access_pointer(f->nh)) {
1385                                 if (*idx < cb->args[2])
1386                                         goto skip_nh;
1387                                 err = vxlan_fdb_info(skb, vxlan, f,
1388                                                      NETLINK_CB(cb->skb).portid,
1389                                                      cb->nlh->nlmsg_seq,
1390                                                      RTM_NEWNEIGH,
1391                                                      NLM_F_MULTI, NULL);
1392                                 if (err < 0) {
1393                                         rcu_read_unlock();
1394                                         goto out;
1395                                 }
1396 skip_nh:
1397                                 *idx += 1;
1398                                 continue;
1399                         }
1400
1401                         list_for_each_entry_rcu(rd, &f->remotes, list) {
1402                                 if (*idx < cb->args[2])
1403                                         goto skip;
1404
1405                                 err = vxlan_fdb_info(skb, vxlan, f,
1406                                                      NETLINK_CB(cb->skb).portid,
1407                                                      cb->nlh->nlmsg_seq,
1408                                                      RTM_NEWNEIGH,
1409                                                      NLM_F_MULTI, rd);
1410                                 if (err < 0) {
1411                                         rcu_read_unlock();
1412                                         goto out;
1413                                 }
1414 skip:
1415                                 *idx += 1;
1416                         }
1417                 }
1418                 rcu_read_unlock();
1419         }
1420 out:
1421         return err;
1422 }
1423
1424 static int vxlan_fdb_get(struct sk_buff *skb,
1425                          struct nlattr *tb[],
1426                          struct net_device *dev,
1427                          const unsigned char *addr,
1428                          u16 vid, u32 portid, u32 seq,
1429                          struct netlink_ext_ack *extack)
1430 {
1431         struct vxlan_dev *vxlan = netdev_priv(dev);
1432         struct vxlan_fdb *f;
1433         __be32 vni;
1434         int err;
1435
1436         if (tb[NDA_VNI])
1437                 vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1438         else
1439                 vni = vxlan->default_dst.remote_vni;
1440
1441         rcu_read_lock();
1442
1443         f = __vxlan_find_mac(vxlan, addr, vni);
1444         if (!f) {
1445                 NL_SET_ERR_MSG(extack, "Fdb entry not found");
1446                 err = -ENOENT;
1447                 goto errout;
1448         }
1449
1450         err = vxlan_fdb_info(skb, vxlan, f, portid, seq,
1451                              RTM_NEWNEIGH, 0, first_remote_rcu(f));
1452 errout:
1453         rcu_read_unlock();
1454         return err;
1455 }
1456
1457 /* Watch incoming packets to learn mapping between Ethernet address
1458  * and Tunnel endpoint.
1459  * Return true if packet is bogus and should be dropped.
1460  */
1461 static bool vxlan_snoop(struct net_device *dev,
1462                         union vxlan_addr *src_ip, const u8 *src_mac,
1463                         u32 src_ifindex, __be32 vni)
1464 {
1465         struct vxlan_dev *vxlan = netdev_priv(dev);
1466         struct vxlan_fdb *f;
1467         u32 ifindex = 0;
1468
1469 #if IS_ENABLED(CONFIG_IPV6)
1470         if (src_ip->sa.sa_family == AF_INET6 &&
1471             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
1472                 ifindex = src_ifindex;
1473 #endif
1474
1475         f = vxlan_find_mac(vxlan, src_mac, vni);
1476         if (likely(f)) {
1477                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1478
1479                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
1480                            rdst->remote_ifindex == ifindex))
1481                         return false;
1482
1483                 /* Don't migrate static entries, drop packets */
1484                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
1485                         return true;
1486
1487                 /* Don't override an fdb with nexthop with a learnt entry */
1488                 if (rcu_access_pointer(f->nh))
1489                         return true;
1490
1491                 if (net_ratelimit())
1492                         netdev_info(dev,
1493                                     "%pM migrated from %pIS to %pIS\n",
1494                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1495
1496                 rdst->remote_ip = *src_ip;
1497                 f->updated = jiffies;
1498                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH, true, NULL);
1499         } else {
1500                 u32 hash_index = fdb_head_index(vxlan, src_mac, vni);
1501
1502                 /* learned new entry */
1503                 spin_lock(&vxlan->hash_lock[hash_index]);
1504
1505                 /* close off race between vxlan_flush and incoming packets */
1506                 if (netif_running(dev))
1507                         vxlan_fdb_update(vxlan, src_mac, src_ip,
1508                                          NUD_REACHABLE,
1509                                          NLM_F_EXCL|NLM_F_CREATE,
1510                                          vxlan->cfg.dst_port,
1511                                          vni,
1512                                          vxlan->default_dst.remote_vni,
1513                                          ifindex, NTF_SELF, 0, true, NULL);
1514                 spin_unlock(&vxlan->hash_lock[hash_index]);
1515         }
1516
1517         return false;
1518 }
1519
1520 /* See if multicast group is already in use by other ID */
1521 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1522 {
1523         struct vxlan_dev *vxlan;
1524         struct vxlan_sock *sock4;
1525 #if IS_ENABLED(CONFIG_IPV6)
1526         struct vxlan_sock *sock6;
1527 #endif
1528         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1529
1530         sock4 = rtnl_dereference(dev->vn4_sock);
1531
1532         /* The vxlan_sock is only used by dev, leaving group has
1533          * no effect on other vxlan devices.
1534          */
1535         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1536                 return false;
1537 #if IS_ENABLED(CONFIG_IPV6)
1538         sock6 = rtnl_dereference(dev->vn6_sock);
1539         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1540                 return false;
1541 #endif
1542
1543         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1544                 if (!netif_running(vxlan->dev) || vxlan == dev)
1545                         continue;
1546
1547                 if (family == AF_INET &&
1548                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1549                         continue;
1550 #if IS_ENABLED(CONFIG_IPV6)
1551                 if (family == AF_INET6 &&
1552                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1553                         continue;
1554 #endif
1555
1556                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1557                                       &dev->default_dst.remote_ip))
1558                         continue;
1559
1560                 if (vxlan->default_dst.remote_ifindex !=
1561                     dev->default_dst.remote_ifindex)
1562                         continue;
1563
1564                 return true;
1565         }
1566
1567         return false;
1568 }
1569
1570 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1571 {
1572         struct vxlan_net *vn;
1573
1574         if (!vs)
1575                 return false;
1576         if (!refcount_dec_and_test(&vs->refcnt))
1577                 return false;
1578
1579         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1580         spin_lock(&vn->sock_lock);
1581         hlist_del_rcu(&vs->hlist);
1582         udp_tunnel_notify_del_rx_port(vs->sock,
1583                                       (vs->flags & VXLAN_F_GPE) ?
1584                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1585                                       UDP_TUNNEL_TYPE_VXLAN);
1586         spin_unlock(&vn->sock_lock);
1587
1588         return true;
1589 }
1590
1591 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1592 {
1593         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1594 #if IS_ENABLED(CONFIG_IPV6)
1595         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1596
1597         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1598 #endif
1599
1600         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1601         synchronize_net();
1602
1603         vxlan_vs_del_dev(vxlan);
1604
1605         if (__vxlan_sock_release_prep(sock4)) {
1606                 udp_tunnel_sock_release(sock4->sock);
1607                 kfree(sock4);
1608         }
1609
1610 #if IS_ENABLED(CONFIG_IPV6)
1611         if (__vxlan_sock_release_prep(sock6)) {
1612                 udp_tunnel_sock_release(sock6->sock);
1613                 kfree(sock6);
1614         }
1615 #endif
1616 }
1617
1618 /* Update multicast group membership when first VNI on
1619  * multicast address is brought up
1620  */
1621 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1622 {
1623         struct sock *sk;
1624         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1625         int ifindex = vxlan->default_dst.remote_ifindex;
1626         int ret = -EINVAL;
1627
1628         if (ip->sa.sa_family == AF_INET) {
1629                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1630                 struct ip_mreqn mreq = {
1631                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1632                         .imr_ifindex            = ifindex,
1633                 };
1634
1635                 sk = sock4->sock->sk;
1636                 lock_sock(sk);
1637                 ret = ip_mc_join_group(sk, &mreq);
1638                 release_sock(sk);
1639 #if IS_ENABLED(CONFIG_IPV6)
1640         } else {
1641                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1642
1643                 sk = sock6->sock->sk;
1644                 lock_sock(sk);
1645                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1646                                                    &ip->sin6.sin6_addr);
1647                 release_sock(sk);
1648 #endif
1649         }
1650
1651         return ret;
1652 }
1653
1654 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1655 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1656 {
1657         struct sock *sk;
1658         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1659         int ifindex = vxlan->default_dst.remote_ifindex;
1660         int ret = -EINVAL;
1661
1662         if (ip->sa.sa_family == AF_INET) {
1663                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1664                 struct ip_mreqn mreq = {
1665                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1666                         .imr_ifindex            = ifindex,
1667                 };
1668
1669                 sk = sock4->sock->sk;
1670                 lock_sock(sk);
1671                 ret = ip_mc_leave_group(sk, &mreq);
1672                 release_sock(sk);
1673 #if IS_ENABLED(CONFIG_IPV6)
1674         } else {
1675                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1676
1677                 sk = sock6->sock->sk;
1678                 lock_sock(sk);
1679                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1680                                                    &ip->sin6.sin6_addr);
1681                 release_sock(sk);
1682 #endif
1683         }
1684
1685         return ret;
1686 }
1687
1688 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1689                           struct sk_buff *skb, u32 vxflags)
1690 {
1691         size_t start, offset;
1692
1693         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1694                 goto out;
1695
1696         start = vxlan_rco_start(unparsed->vx_vni);
1697         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1698
1699         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1700                 return false;
1701
1702         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1703                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1704 out:
1705         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1706         unparsed->vx_vni &= VXLAN_VNI_MASK;
1707         return true;
1708 }
1709
1710 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1711                                 struct sk_buff *skb, u32 vxflags,
1712                                 struct vxlan_metadata *md)
1713 {
1714         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1715         struct metadata_dst *tun_dst;
1716
1717         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1718                 goto out;
1719
1720         md->gbp = ntohs(gbp->policy_id);
1721
1722         tun_dst = (struct metadata_dst *)skb_dst(skb);
1723         if (tun_dst) {
1724                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1725                 tun_dst->u.tun_info.options_len = sizeof(*md);
1726         }
1727         if (gbp->dont_learn)
1728                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1729
1730         if (gbp->policy_applied)
1731                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1732
1733         /* In flow-based mode, GBP is carried in dst_metadata */
1734         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1735                 skb->mark = md->gbp;
1736 out:
1737         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1738 }
1739
1740 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1741                                 __be16 *protocol,
1742                                 struct sk_buff *skb, u32 vxflags)
1743 {
1744         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1745
1746         /* Need to have Next Protocol set for interfaces in GPE mode. */
1747         if (!gpe->np_applied)
1748                 return false;
1749         /* "The initial version is 0. If a receiver does not support the
1750          * version indicated it MUST drop the packet.
1751          */
1752         if (gpe->version != 0)
1753                 return false;
1754         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1755          * processing MUST occur." However, we don't implement OAM
1756          * processing, thus drop the packet.
1757          */
1758         if (gpe->oam_flag)
1759                 return false;
1760
1761         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1762         if (!*protocol)
1763                 return false;
1764
1765         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1766         return true;
1767 }
1768
1769 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1770                           struct vxlan_sock *vs,
1771                           struct sk_buff *skb, __be32 vni)
1772 {
1773         union vxlan_addr saddr;
1774         u32 ifindex = skb->dev->ifindex;
1775
1776         skb_reset_mac_header(skb);
1777         skb->protocol = eth_type_trans(skb, vxlan->dev);
1778         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1779
1780         /* Ignore packet loops (and multicast echo) */
1781         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1782                 return false;
1783
1784         /* Get address from the outer IP header */
1785         if (vxlan_get_sk_family(vs) == AF_INET) {
1786                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1787                 saddr.sa.sa_family = AF_INET;
1788 #if IS_ENABLED(CONFIG_IPV6)
1789         } else {
1790                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1791                 saddr.sa.sa_family = AF_INET6;
1792 #endif
1793         }
1794
1795         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1796             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1797                 return false;
1798
1799         return true;
1800 }
1801
1802 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1803                                   struct sk_buff *skb)
1804 {
1805         int err = 0;
1806
1807         if (vxlan_get_sk_family(vs) == AF_INET)
1808                 err = IP_ECN_decapsulate(oiph, skb);
1809 #if IS_ENABLED(CONFIG_IPV6)
1810         else
1811                 err = IP6_ECN_decapsulate(oiph, skb);
1812 #endif
1813
1814         if (unlikely(err) && log_ecn_error) {
1815                 if (vxlan_get_sk_family(vs) == AF_INET)
1816                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1817                                              &((struct iphdr *)oiph)->saddr,
1818                                              ((struct iphdr *)oiph)->tos);
1819                 else
1820                         net_info_ratelimited("non-ECT from %pI6\n",
1821                                              &((struct ipv6hdr *)oiph)->saddr);
1822         }
1823         return err <= 1;
1824 }
1825
1826 /* Callback from net/ipv4/udp.c to receive packets */
1827 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1828 {
1829         struct vxlan_dev *vxlan;
1830         struct vxlan_sock *vs;
1831         struct vxlanhdr unparsed;
1832         struct vxlan_metadata _md;
1833         struct vxlan_metadata *md = &_md;
1834         __be16 protocol = htons(ETH_P_TEB);
1835         bool raw_proto = false;
1836         void *oiph;
1837         __be32 vni = 0;
1838
1839         /* Need UDP and VXLAN header to be present */
1840         if (!pskb_may_pull(skb, VXLAN_HLEN))
1841                 goto drop;
1842
1843         unparsed = *vxlan_hdr(skb);
1844         /* VNI flag always required to be set */
1845         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1846                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1847                            ntohl(vxlan_hdr(skb)->vx_flags),
1848                            ntohl(vxlan_hdr(skb)->vx_vni));
1849                 /* Return non vxlan pkt */
1850                 goto drop;
1851         }
1852         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1853         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1854
1855         vs = rcu_dereference_sk_user_data(sk);
1856         if (!vs)
1857                 goto drop;
1858
1859         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1860
1861         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1862         if (!vxlan)
1863                 goto drop;
1864
1865         /* For backwards compatibility, only allow reserved fields to be
1866          * used by VXLAN extensions if explicitly requested.
1867          */
1868         if (vs->flags & VXLAN_F_GPE) {
1869                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1870                         goto drop;
1871                 raw_proto = true;
1872         }
1873
1874         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1875                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1876                 goto drop;
1877
1878         if (vs->flags & VXLAN_F_REMCSUM_RX)
1879                 if (unlikely(!vxlan_remcsum(&unparsed, skb, vs->flags)))
1880                         goto drop;
1881
1882         if (vxlan_collect_metadata(vs)) {
1883                 struct metadata_dst *tun_dst;
1884
1885                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1886                                          key32_to_tunnel_id(vni), sizeof(*md));
1887
1888                 if (!tun_dst)
1889                         goto drop;
1890
1891                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1892
1893                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1894         } else {
1895                 memset(md, 0, sizeof(*md));
1896         }
1897
1898         if (vs->flags & VXLAN_F_GBP)
1899                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1900         /* Note that GBP and GPE can never be active together. This is
1901          * ensured in vxlan_dev_configure.
1902          */
1903
1904         if (unparsed.vx_flags || unparsed.vx_vni) {
1905                 /* If there are any unprocessed flags remaining treat
1906                  * this as a malformed packet. This behavior diverges from
1907                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1908                  * in reserved fields are to be ignored. The approach here
1909                  * maintains compatibility with previous stack code, and also
1910                  * is more robust and provides a little more security in
1911                  * adding extensions to VXLAN.
1912                  */
1913                 goto drop;
1914         }
1915
1916         if (!raw_proto) {
1917                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1918                         goto drop;
1919         } else {
1920                 skb_reset_mac_header(skb);
1921                 skb->dev = vxlan->dev;
1922                 skb->pkt_type = PACKET_HOST;
1923         }
1924
1925         oiph = skb_network_header(skb);
1926         skb_reset_network_header(skb);
1927
1928         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1929                 ++vxlan->dev->stats.rx_frame_errors;
1930                 ++vxlan->dev->stats.rx_errors;
1931                 goto drop;
1932         }
1933
1934         rcu_read_lock();
1935
1936         if (unlikely(!(vxlan->dev->flags & IFF_UP))) {
1937                 rcu_read_unlock();
1938                 atomic_long_inc(&vxlan->dev->rx_dropped);
1939                 goto drop;
1940         }
1941
1942         dev_sw_netstats_rx_add(vxlan->dev, skb->len);
1943         gro_cells_receive(&vxlan->gro_cells, skb);
1944
1945         rcu_read_unlock();
1946
1947         return 0;
1948
1949 drop:
1950         /* Consume bad packet */
1951         kfree_skb(skb);
1952         return 0;
1953 }
1954
1955 /* Callback from net/ipv{4,6}/udp.c to check that we have a VNI for errors */
1956 static int vxlan_err_lookup(struct sock *sk, struct sk_buff *skb)
1957 {
1958         struct vxlan_dev *vxlan;
1959         struct vxlan_sock *vs;
1960         struct vxlanhdr *hdr;
1961         __be32 vni;
1962
1963         if (!pskb_may_pull(skb, skb_transport_offset(skb) + VXLAN_HLEN))
1964                 return -EINVAL;
1965
1966         hdr = vxlan_hdr(skb);
1967
1968         if (!(hdr->vx_flags & VXLAN_HF_VNI))
1969                 return -EINVAL;
1970
1971         vs = rcu_dereference_sk_user_data(sk);
1972         if (!vs)
1973                 return -ENOENT;
1974
1975         vni = vxlan_vni(hdr->vx_vni);
1976         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1977         if (!vxlan)
1978                 return -ENOENT;
1979
1980         return 0;
1981 }
1982
1983 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1984 {
1985         struct vxlan_dev *vxlan = netdev_priv(dev);
1986         struct arphdr *parp;
1987         u8 *arpptr, *sha;
1988         __be32 sip, tip;
1989         struct neighbour *n;
1990
1991         if (dev->flags & IFF_NOARP)
1992                 goto out;
1993
1994         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1995                 dev->stats.tx_dropped++;
1996                 goto out;
1997         }
1998         parp = arp_hdr(skb);
1999
2000         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
2001              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
2002             parp->ar_pro != htons(ETH_P_IP) ||
2003             parp->ar_op != htons(ARPOP_REQUEST) ||
2004             parp->ar_hln != dev->addr_len ||
2005             parp->ar_pln != 4)
2006                 goto out;
2007         arpptr = (u8 *)parp + sizeof(struct arphdr);
2008         sha = arpptr;
2009         arpptr += dev->addr_len;        /* sha */
2010         memcpy(&sip, arpptr, sizeof(sip));
2011         arpptr += sizeof(sip);
2012         arpptr += dev->addr_len;        /* tha */
2013         memcpy(&tip, arpptr, sizeof(tip));
2014
2015         if (ipv4_is_loopback(tip) ||
2016             ipv4_is_multicast(tip))
2017                 goto out;
2018
2019         n = neigh_lookup(&arp_tbl, &tip, dev);
2020
2021         if (n) {
2022                 struct vxlan_fdb *f;
2023                 struct sk_buff  *reply;
2024
2025                 if (!(n->nud_state & NUD_CONNECTED)) {
2026                         neigh_release(n);
2027                         goto out;
2028                 }
2029
2030                 f = vxlan_find_mac(vxlan, n->ha, vni);
2031                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2032                         /* bridge-local neighbor */
2033                         neigh_release(n);
2034                         goto out;
2035                 }
2036
2037                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
2038                                 n->ha, sha);
2039
2040                 neigh_release(n);
2041
2042                 if (reply == NULL)
2043                         goto out;
2044
2045                 skb_reset_mac_header(reply);
2046                 __skb_pull(reply, skb_network_offset(reply));
2047                 reply->ip_summed = CHECKSUM_UNNECESSARY;
2048                 reply->pkt_type = PACKET_HOST;
2049
2050                 if (netif_rx_ni(reply) == NET_RX_DROP)
2051                         dev->stats.rx_dropped++;
2052         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2053                 union vxlan_addr ipa = {
2054                         .sin.sin_addr.s_addr = tip,
2055                         .sin.sin_family = AF_INET,
2056                 };
2057
2058                 vxlan_ip_miss(dev, &ipa);
2059         }
2060 out:
2061         consume_skb(skb);
2062         return NETDEV_TX_OK;
2063 }
2064
2065 #if IS_ENABLED(CONFIG_IPV6)
2066 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
2067         struct neighbour *n, bool isrouter)
2068 {
2069         struct net_device *dev = request->dev;
2070         struct sk_buff *reply;
2071         struct nd_msg *ns, *na;
2072         struct ipv6hdr *pip6;
2073         u8 *daddr;
2074         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
2075         int ns_olen;
2076         int i, len;
2077
2078         if (dev == NULL || !pskb_may_pull(request, request->len))
2079                 return NULL;
2080
2081         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
2082                 sizeof(*na) + na_olen + dev->needed_tailroom;
2083         reply = alloc_skb(len, GFP_ATOMIC);
2084         if (reply == NULL)
2085                 return NULL;
2086
2087         reply->protocol = htons(ETH_P_IPV6);
2088         reply->dev = dev;
2089         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
2090         skb_push(reply, sizeof(struct ethhdr));
2091         skb_reset_mac_header(reply);
2092
2093         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
2094
2095         daddr = eth_hdr(request)->h_source;
2096         ns_olen = request->len - skb_network_offset(request) -
2097                 sizeof(struct ipv6hdr) - sizeof(*ns);
2098         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
2099                 if (!ns->opt[i + 1]) {
2100                         kfree_skb(reply);
2101                         return NULL;
2102                 }
2103                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
2104                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
2105                         break;
2106                 }
2107         }
2108
2109         /* Ethernet header */
2110         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
2111         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
2112         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
2113         reply->protocol = htons(ETH_P_IPV6);
2114
2115         skb_pull(reply, sizeof(struct ethhdr));
2116         skb_reset_network_header(reply);
2117         skb_put(reply, sizeof(struct ipv6hdr));
2118
2119         /* IPv6 header */
2120
2121         pip6 = ipv6_hdr(reply);
2122         memset(pip6, 0, sizeof(struct ipv6hdr));
2123         pip6->version = 6;
2124         pip6->priority = ipv6_hdr(request)->priority;
2125         pip6->nexthdr = IPPROTO_ICMPV6;
2126         pip6->hop_limit = 255;
2127         pip6->daddr = ipv6_hdr(request)->saddr;
2128         pip6->saddr = *(struct in6_addr *)n->primary_key;
2129
2130         skb_pull(reply, sizeof(struct ipv6hdr));
2131         skb_reset_transport_header(reply);
2132
2133         /* Neighbor Advertisement */
2134         na = skb_put_zero(reply, sizeof(*na) + na_olen);
2135         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
2136         na->icmph.icmp6_router = isrouter;
2137         na->icmph.icmp6_override = 1;
2138         na->icmph.icmp6_solicited = 1;
2139         na->target = ns->target;
2140         ether_addr_copy(&na->opt[2], n->ha);
2141         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
2142         na->opt[1] = na_olen >> 3;
2143
2144         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
2145                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
2146                 csum_partial(na, sizeof(*na)+na_olen, 0));
2147
2148         pip6->payload_len = htons(sizeof(*na)+na_olen);
2149
2150         skb_push(reply, sizeof(struct ipv6hdr));
2151
2152         reply->ip_summed = CHECKSUM_UNNECESSARY;
2153
2154         return reply;
2155 }
2156
2157 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
2158 {
2159         struct vxlan_dev *vxlan = netdev_priv(dev);
2160         const struct in6_addr *daddr;
2161         const struct ipv6hdr *iphdr;
2162         struct inet6_dev *in6_dev;
2163         struct neighbour *n;
2164         struct nd_msg *msg;
2165
2166         in6_dev = __in6_dev_get(dev);
2167         if (!in6_dev)
2168                 goto out;
2169
2170         iphdr = ipv6_hdr(skb);
2171         daddr = &iphdr->daddr;
2172         msg = (struct nd_msg *)(iphdr + 1);
2173
2174         if (ipv6_addr_loopback(daddr) ||
2175             ipv6_addr_is_multicast(&msg->target))
2176                 goto out;
2177
2178         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
2179
2180         if (n) {
2181                 struct vxlan_fdb *f;
2182                 struct sk_buff *reply;
2183
2184                 if (!(n->nud_state & NUD_CONNECTED)) {
2185                         neigh_release(n);
2186                         goto out;
2187                 }
2188
2189                 f = vxlan_find_mac(vxlan, n->ha, vni);
2190                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2191                         /* bridge-local neighbor */
2192                         neigh_release(n);
2193                         goto out;
2194                 }
2195
2196                 reply = vxlan_na_create(skb, n,
2197                                         !!(f ? f->flags & NTF_ROUTER : 0));
2198
2199                 neigh_release(n);
2200
2201                 if (reply == NULL)
2202                         goto out;
2203
2204                 if (netif_rx_ni(reply) == NET_RX_DROP)
2205                         dev->stats.rx_dropped++;
2206
2207         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2208                 union vxlan_addr ipa = {
2209                         .sin6.sin6_addr = msg->target,
2210                         .sin6.sin6_family = AF_INET6,
2211                 };
2212
2213                 vxlan_ip_miss(dev, &ipa);
2214         }
2215
2216 out:
2217         consume_skb(skb);
2218         return NETDEV_TX_OK;
2219 }
2220 #endif
2221
2222 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
2223 {
2224         struct vxlan_dev *vxlan = netdev_priv(dev);
2225         struct neighbour *n;
2226
2227         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
2228                 return false;
2229
2230         n = NULL;
2231         switch (ntohs(eth_hdr(skb)->h_proto)) {
2232         case ETH_P_IP:
2233         {
2234                 struct iphdr *pip;
2235
2236                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
2237                         return false;
2238                 pip = ip_hdr(skb);
2239                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
2240                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2241                         union vxlan_addr ipa = {
2242                                 .sin.sin_addr.s_addr = pip->daddr,
2243                                 .sin.sin_family = AF_INET,
2244                         };
2245
2246                         vxlan_ip_miss(dev, &ipa);
2247                         return false;
2248                 }
2249
2250                 break;
2251         }
2252 #if IS_ENABLED(CONFIG_IPV6)
2253         case ETH_P_IPV6:
2254         {
2255                 struct ipv6hdr *pip6;
2256
2257                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
2258                         return false;
2259                 pip6 = ipv6_hdr(skb);
2260                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
2261                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2262                         union vxlan_addr ipa = {
2263                                 .sin6.sin6_addr = pip6->daddr,
2264                                 .sin6.sin6_family = AF_INET6,
2265                         };
2266
2267                         vxlan_ip_miss(dev, &ipa);
2268                         return false;
2269                 }
2270
2271                 break;
2272         }
2273 #endif
2274         default:
2275                 return false;
2276         }
2277
2278         if (n) {
2279                 bool diff;
2280
2281                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
2282                 if (diff) {
2283                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
2284                                 dev->addr_len);
2285                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
2286                 }
2287                 neigh_release(n);
2288                 return diff;
2289         }
2290
2291         return false;
2292 }
2293
2294 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
2295                                 struct vxlan_metadata *md)
2296 {
2297         struct vxlanhdr_gbp *gbp;
2298
2299         if (!md->gbp)
2300                 return;
2301
2302         gbp = (struct vxlanhdr_gbp *)vxh;
2303         vxh->vx_flags |= VXLAN_HF_GBP;
2304
2305         if (md->gbp & VXLAN_GBP_DONT_LEARN)
2306                 gbp->dont_learn = 1;
2307
2308         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
2309                 gbp->policy_applied = 1;
2310
2311         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
2312 }
2313
2314 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
2315                                __be16 protocol)
2316 {
2317         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
2318
2319         gpe->np_applied = 1;
2320         gpe->next_protocol = tun_p_from_eth_p(protocol);
2321         if (!gpe->next_protocol)
2322                 return -EPFNOSUPPORT;
2323         return 0;
2324 }
2325
2326 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
2327                            int iphdr_len, __be32 vni,
2328                            struct vxlan_metadata *md, u32 vxflags,
2329                            bool udp_sum)
2330 {
2331         struct vxlanhdr *vxh;
2332         int min_headroom;
2333         int err;
2334         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
2335         __be16 inner_protocol = htons(ETH_P_TEB);
2336
2337         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
2338             skb->ip_summed == CHECKSUM_PARTIAL) {
2339                 int csum_start = skb_checksum_start_offset(skb);
2340
2341                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
2342                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
2343                     (skb->csum_offset == offsetof(struct udphdr, check) ||
2344                      skb->csum_offset == offsetof(struct tcphdr, check)))
2345                         type |= SKB_GSO_TUNNEL_REMCSUM;
2346         }
2347
2348         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
2349                         + VXLAN_HLEN + iphdr_len;
2350
2351         /* Need space for new headers (invalidates iph ptr) */
2352         err = skb_cow_head(skb, min_headroom);
2353         if (unlikely(err))
2354                 return err;
2355
2356         err = iptunnel_handle_offloads(skb, type);
2357         if (err)
2358                 return err;
2359
2360         vxh = __skb_push(skb, sizeof(*vxh));
2361         vxh->vx_flags = VXLAN_HF_VNI;
2362         vxh->vx_vni = vxlan_vni_field(vni);
2363
2364         if (type & SKB_GSO_TUNNEL_REMCSUM) {
2365                 unsigned int start;
2366
2367                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
2368                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
2369                 vxh->vx_flags |= VXLAN_HF_RCO;
2370
2371                 if (!skb_is_gso(skb)) {
2372                         skb->ip_summed = CHECKSUM_NONE;
2373                         skb->encapsulation = 0;
2374                 }
2375         }
2376
2377         if (vxflags & VXLAN_F_GBP)
2378                 vxlan_build_gbp_hdr(vxh, vxflags, md);
2379         if (vxflags & VXLAN_F_GPE) {
2380                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
2381                 if (err < 0)
2382                         return err;
2383                 inner_protocol = skb->protocol;
2384         }
2385
2386         skb_set_inner_protocol(skb, inner_protocol);
2387         return 0;
2388 }
2389
2390 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
2391                                       struct vxlan_sock *sock4,
2392                                       struct sk_buff *skb, int oif, u8 tos,
2393                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
2394                                       struct dst_cache *dst_cache,
2395                                       const struct ip_tunnel_info *info)
2396 {
2397         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2398         struct rtable *rt = NULL;
2399         struct flowi4 fl4;
2400
2401         if (!sock4)
2402                 return ERR_PTR(-EIO);
2403
2404         if (tos && !info)
2405                 use_cache = false;
2406         if (use_cache) {
2407                 rt = dst_cache_get_ip4(dst_cache, saddr);
2408                 if (rt)
2409                         return rt;
2410         }
2411
2412         memset(&fl4, 0, sizeof(fl4));
2413         fl4.flowi4_oif = oif;
2414         fl4.flowi4_tos = RT_TOS(tos);
2415         fl4.flowi4_mark = skb->mark;
2416         fl4.flowi4_proto = IPPROTO_UDP;
2417         fl4.daddr = daddr;
2418         fl4.saddr = *saddr;
2419         fl4.fl4_dport = dport;
2420         fl4.fl4_sport = sport;
2421
2422         rt = ip_route_output_key(vxlan->net, &fl4);
2423         if (!IS_ERR(rt)) {
2424                 if (rt->dst.dev == dev) {
2425                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
2426                         ip_rt_put(rt);
2427                         return ERR_PTR(-ELOOP);
2428                 }
2429
2430                 *saddr = fl4.saddr;
2431                 if (use_cache)
2432                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2433         } else {
2434                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
2435                 return ERR_PTR(-ENETUNREACH);
2436         }
2437         return rt;
2438 }
2439
2440 #if IS_ENABLED(CONFIG_IPV6)
2441 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
2442                                           struct net_device *dev,
2443                                           struct vxlan_sock *sock6,
2444                                           struct sk_buff *skb, int oif, u8 tos,
2445                                           __be32 label,
2446                                           const struct in6_addr *daddr,
2447                                           struct in6_addr *saddr,
2448                                           __be16 dport, __be16 sport,
2449                                           struct dst_cache *dst_cache,
2450                                           const struct ip_tunnel_info *info)
2451 {
2452         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2453         struct dst_entry *ndst;
2454         struct flowi6 fl6;
2455
2456         if (!sock6)
2457                 return ERR_PTR(-EIO);
2458
2459         if (tos && !info)
2460                 use_cache = false;
2461         if (use_cache) {
2462                 ndst = dst_cache_get_ip6(dst_cache, saddr);
2463                 if (ndst)
2464                         return ndst;
2465         }
2466
2467         memset(&fl6, 0, sizeof(fl6));
2468         fl6.flowi6_oif = oif;
2469         fl6.daddr = *daddr;
2470         fl6.saddr = *saddr;
2471         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
2472         fl6.flowi6_mark = skb->mark;
2473         fl6.flowi6_proto = IPPROTO_UDP;
2474         fl6.fl6_dport = dport;
2475         fl6.fl6_sport = sport;
2476
2477         ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk,
2478                                                &fl6, NULL);
2479         if (unlikely(IS_ERR(ndst))) {
2480                 netdev_dbg(dev, "no route to %pI6\n", daddr);
2481                 return ERR_PTR(-ENETUNREACH);
2482         }
2483
2484         if (unlikely(ndst->dev == dev)) {
2485                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
2486                 dst_release(ndst);
2487                 return ERR_PTR(-ELOOP);
2488         }
2489
2490         *saddr = fl6.saddr;
2491         if (use_cache)
2492                 dst_cache_set_ip6(dst_cache, ndst, saddr);
2493         return ndst;
2494 }
2495 #endif
2496
2497 /* Bypass encapsulation if the destination is local */
2498 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2499                                struct vxlan_dev *dst_vxlan, __be32 vni,
2500                                bool snoop)
2501 {
2502         struct pcpu_sw_netstats *tx_stats, *rx_stats;
2503         union vxlan_addr loopback;
2504         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2505         struct net_device *dev;
2506         int len = skb->len;
2507
2508         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2509         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2510         skb->pkt_type = PACKET_HOST;
2511         skb->encapsulation = 0;
2512         skb->dev = dst_vxlan->dev;
2513         __skb_pull(skb, skb_network_offset(skb));
2514
2515         if (remote_ip->sa.sa_family == AF_INET) {
2516                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2517                 loopback.sa.sa_family =  AF_INET;
2518 #if IS_ENABLED(CONFIG_IPV6)
2519         } else {
2520                 loopback.sin6.sin6_addr = in6addr_loopback;
2521                 loopback.sa.sa_family =  AF_INET6;
2522 #endif
2523         }
2524
2525         rcu_read_lock();
2526         dev = skb->dev;
2527         if (unlikely(!(dev->flags & IFF_UP))) {
2528                 kfree_skb(skb);
2529                 goto drop;
2530         }
2531
2532         if ((dst_vxlan->cfg.flags & VXLAN_F_LEARN) && snoop)
2533                 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni);
2534
2535         u64_stats_update_begin(&tx_stats->syncp);
2536         tx_stats->tx_packets++;
2537         tx_stats->tx_bytes += len;
2538         u64_stats_update_end(&tx_stats->syncp);
2539
2540         if (netif_rx(skb) == NET_RX_SUCCESS) {
2541                 u64_stats_update_begin(&rx_stats->syncp);
2542                 rx_stats->rx_packets++;
2543                 rx_stats->rx_bytes += len;
2544                 u64_stats_update_end(&rx_stats->syncp);
2545         } else {
2546 drop:
2547                 dev->stats.rx_dropped++;
2548         }
2549         rcu_read_unlock();
2550 }
2551
2552 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2553                                  struct vxlan_dev *vxlan,
2554                                  union vxlan_addr *daddr,
2555                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2556                                  struct dst_entry *dst,
2557                                  u32 rt_flags)
2558 {
2559 #if IS_ENABLED(CONFIG_IPV6)
2560         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2561          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2562          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2563          */
2564         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2565 #endif
2566         /* Bypass encapsulation if the destination is local */
2567         if (rt_flags & RTCF_LOCAL &&
2568             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2569                 struct vxlan_dev *dst_vxlan;
2570
2571                 dst_release(dst);
2572                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2573                                            daddr->sa.sa_family, dst_port,
2574                                            vxlan->cfg.flags);
2575                 if (!dst_vxlan) {
2576                         dev->stats.tx_errors++;
2577                         kfree_skb(skb);
2578
2579                         return -ENOENT;
2580                 }
2581                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni, true);
2582                 return 1;
2583         }
2584
2585         return 0;
2586 }
2587
2588 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2589                            __be32 default_vni, struct vxlan_rdst *rdst,
2590                            bool did_rsc)
2591 {
2592         struct dst_cache *dst_cache;
2593         struct ip_tunnel_info *info;
2594         struct vxlan_dev *vxlan = netdev_priv(dev);
2595         const struct iphdr *old_iph = ip_hdr(skb);
2596         union vxlan_addr *dst;
2597         union vxlan_addr remote_ip, local_ip;
2598         struct vxlan_metadata _md;
2599         struct vxlan_metadata *md = &_md;
2600         __be16 src_port = 0, dst_port;
2601         struct dst_entry *ndst = NULL;
2602         __be32 vni, label;
2603         __u8 tos, ttl;
2604         int ifindex;
2605         int err;
2606         u32 flags = vxlan->cfg.flags;
2607         bool udp_sum = false;
2608         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2609
2610         info = skb_tunnel_info(skb);
2611
2612         if (rdst) {
2613                 dst = &rdst->remote_ip;
2614                 if (vxlan_addr_any(dst)) {
2615                         if (did_rsc) {
2616                                 /* short-circuited back to local bridge */
2617                                 vxlan_encap_bypass(skb, vxlan, vxlan,
2618                                                    default_vni, true);
2619                                 return;
2620                         }
2621                         goto drop;
2622                 }
2623
2624                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2625                 vni = (rdst->remote_vni) ? : default_vni;
2626                 ifindex = rdst->remote_ifindex;
2627                 local_ip = vxlan->cfg.saddr;
2628                 dst_cache = &rdst->dst_cache;
2629                 md->gbp = skb->mark;
2630                 if (flags & VXLAN_F_TTL_INHERIT) {
2631                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2632                 } else {
2633                         ttl = vxlan->cfg.ttl;
2634                         if (!ttl && vxlan_addr_multicast(dst))
2635                                 ttl = 1;
2636                 }
2637
2638                 tos = vxlan->cfg.tos;
2639                 if (tos == 1)
2640                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2641
2642                 if (dst->sa.sa_family == AF_INET)
2643                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2644                 else
2645                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2646                 label = vxlan->cfg.label;
2647         } else {
2648                 if (!info) {
2649                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2650                                   dev->name);
2651                         goto drop;
2652                 }
2653                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2654                 if (remote_ip.sa.sa_family == AF_INET) {
2655                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2656                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2657                 } else {
2658                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2659                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2660                 }
2661                 dst = &remote_ip;
2662                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2663                 vni = tunnel_id_to_key32(info->key.tun_id);
2664                 ifindex = 0;
2665                 dst_cache = &info->dst_cache;
2666                 if (info->key.tun_flags & TUNNEL_VXLAN_OPT) {
2667                         if (info->options_len < sizeof(*md))
2668                                 goto drop;
2669                         md = ip_tunnel_info_opts(info);
2670                 }
2671                 ttl = info->key.ttl;
2672                 tos = info->key.tos;
2673                 label = info->key.label;
2674                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2675         }
2676         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2677                                      vxlan->cfg.port_max, true);
2678
2679         rcu_read_lock();
2680         if (dst->sa.sa_family == AF_INET) {
2681                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2682                 struct rtable *rt;
2683                 __be16 df = 0;
2684
2685                 if (!ifindex)
2686                         ifindex = sock4->sock->sk->sk_bound_dev_if;
2687
2688                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2689                                      dst->sin.sin_addr.s_addr,
2690                                      &local_ip.sin.sin_addr.s_addr,
2691                                      dst_port, src_port,
2692                                      dst_cache, info);
2693                 if (IS_ERR(rt)) {
2694                         err = PTR_ERR(rt);
2695                         goto tx_error;
2696                 }
2697
2698                 if (!info) {
2699                         /* Bypass encapsulation if the destination is local */
2700                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2701                                                     dst_port, ifindex, vni,
2702                                                     &rt->dst, rt->rt_flags);
2703                         if (err)
2704                                 goto out_unlock;
2705
2706                         if (vxlan->cfg.df == VXLAN_DF_SET) {
2707                                 df = htons(IP_DF);
2708                         } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) {
2709                                 struct ethhdr *eth = eth_hdr(skb);
2710
2711                                 if (ntohs(eth->h_proto) == ETH_P_IPV6 ||
2712                                     (ntohs(eth->h_proto) == ETH_P_IP &&
2713                                      old_iph->frag_off & htons(IP_DF)))
2714                                         df = htons(IP_DF);
2715                         }
2716                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2717                         df = htons(IP_DF);
2718                 }
2719
2720                 ndst = &rt->dst;
2721                 err = skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM,
2722                                             netif_is_any_bridge_port(dev));
2723                 if (err < 0) {
2724                         goto tx_error;
2725                 } else if (err) {
2726                         if (info) {
2727                                 struct in_addr src, dst;
2728
2729                                 src = remote_ip.sin.sin_addr;
2730                                 dst = local_ip.sin.sin_addr;
2731                                 info->key.u.ipv4.src = src.s_addr;
2732                                 info->key.u.ipv4.dst = dst.s_addr;
2733                         }
2734                         vxlan_encap_bypass(skb, vxlan, vxlan, vni, false);
2735                         dst_release(ndst);
2736                         goto out_unlock;
2737                 }
2738
2739                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2740                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2741                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2742                                       vni, md, flags, udp_sum);
2743                 if (err < 0)
2744                         goto tx_error;
2745
2746                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2747                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2748                                     src_port, dst_port, xnet, !udp_sum);
2749 #if IS_ENABLED(CONFIG_IPV6)
2750         } else {
2751                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2752
2753                 if (!ifindex)
2754                         ifindex = sock6->sock->sk->sk_bound_dev_if;
2755
2756                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2757                                         label, &dst->sin6.sin6_addr,
2758                                         &local_ip.sin6.sin6_addr,
2759                                         dst_port, src_port,
2760                                         dst_cache, info);
2761                 if (IS_ERR(ndst)) {
2762                         err = PTR_ERR(ndst);
2763                         ndst = NULL;
2764                         goto tx_error;
2765                 }
2766
2767                 if (!info) {
2768                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2769
2770                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2771                                                     dst_port, ifindex, vni,
2772                                                     ndst, rt6i_flags);
2773                         if (err)
2774                                 goto out_unlock;
2775                 }
2776
2777                 err = skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM,
2778                                             netif_is_any_bridge_port(dev));
2779                 if (err < 0) {
2780                         goto tx_error;
2781                 } else if (err) {
2782                         if (info) {
2783                                 struct in6_addr src, dst;
2784
2785                                 src = remote_ip.sin6.sin6_addr;
2786                                 dst = local_ip.sin6.sin6_addr;
2787                                 info->key.u.ipv6.src = src;
2788                                 info->key.u.ipv6.dst = dst;
2789                         }
2790
2791                         vxlan_encap_bypass(skb, vxlan, vxlan, vni, false);
2792                         dst_release(ndst);
2793                         goto out_unlock;
2794                 }
2795
2796                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2797                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2798                 skb_scrub_packet(skb, xnet);
2799                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2800                                       vni, md, flags, udp_sum);
2801                 if (err < 0)
2802                         goto tx_error;
2803
2804                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2805                                      &local_ip.sin6.sin6_addr,
2806                                      &dst->sin6.sin6_addr, tos, ttl,
2807                                      label, src_port, dst_port, !udp_sum);
2808 #endif
2809         }
2810 out_unlock:
2811         rcu_read_unlock();
2812         return;
2813
2814 drop:
2815         dev->stats.tx_dropped++;
2816         dev_kfree_skb(skb);
2817         return;
2818
2819 tx_error:
2820         rcu_read_unlock();
2821         if (err == -ELOOP)
2822                 dev->stats.collisions++;
2823         else if (err == -ENETUNREACH)
2824                 dev->stats.tx_carrier_errors++;
2825         dst_release(ndst);
2826         dev->stats.tx_errors++;
2827         kfree_skb(skb);
2828 }
2829
2830 static void vxlan_xmit_nh(struct sk_buff *skb, struct net_device *dev,
2831                           struct vxlan_fdb *f, __be32 vni, bool did_rsc)
2832 {
2833         struct vxlan_rdst nh_rdst;
2834         struct nexthop *nh;
2835         bool do_xmit;
2836         u32 hash;
2837
2838         memset(&nh_rdst, 0, sizeof(struct vxlan_rdst));
2839         hash = skb_get_hash(skb);
2840
2841         rcu_read_lock();
2842         nh = rcu_dereference(f->nh);
2843         if (!nh) {
2844                 rcu_read_unlock();
2845                 goto drop;
2846         }
2847         do_xmit = vxlan_fdb_nh_path_select(nh, hash, &nh_rdst);
2848         rcu_read_unlock();
2849
2850         if (likely(do_xmit))
2851                 vxlan_xmit_one(skb, dev, vni, &nh_rdst, did_rsc);
2852         else
2853                 goto drop;
2854
2855         return;
2856
2857 drop:
2858         dev->stats.tx_dropped++;
2859         dev_kfree_skb(skb);
2860 }
2861
2862 /* Transmit local packets over Vxlan
2863  *
2864  * Outer IP header inherits ECN and DF from inner header.
2865  * Outer UDP destination is the VXLAN assigned port.
2866  *           source port is based on hash of flow
2867  */
2868 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2869 {
2870         struct vxlan_dev *vxlan = netdev_priv(dev);
2871         struct vxlan_rdst *rdst, *fdst = NULL;
2872         const struct ip_tunnel_info *info;
2873         bool did_rsc = false;
2874         struct vxlan_fdb *f;
2875         struct ethhdr *eth;
2876         __be32 vni = 0;
2877
2878         info = skb_tunnel_info(skb);
2879
2880         skb_reset_mac_header(skb);
2881
2882         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2883                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2884                     info->mode & IP_TUNNEL_INFO_TX) {
2885                         vni = tunnel_id_to_key32(info->key.tun_id);
2886                 } else {
2887                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2888                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2889                         else
2890                                 kfree_skb(skb);
2891                         return NETDEV_TX_OK;
2892                 }
2893         }
2894
2895         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2896                 eth = eth_hdr(skb);
2897                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2898                         return arp_reduce(dev, skb, vni);
2899 #if IS_ENABLED(CONFIG_IPV6)
2900                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2901                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2902                                             sizeof(struct nd_msg)) &&
2903                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2904                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2905
2906                         if (m->icmph.icmp6_code == 0 &&
2907                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2908                                 return neigh_reduce(dev, skb, vni);
2909                 }
2910 #endif
2911         }
2912
2913         eth = eth_hdr(skb);
2914         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2915         did_rsc = false;
2916
2917         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2918             (ntohs(eth->h_proto) == ETH_P_IP ||
2919              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2920                 did_rsc = route_shortcircuit(dev, skb);
2921                 if (did_rsc)
2922                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2923         }
2924
2925         if (f == NULL) {
2926                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2927                 if (f == NULL) {
2928                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2929                             !is_multicast_ether_addr(eth->h_dest))
2930                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2931
2932                         dev->stats.tx_dropped++;
2933                         kfree_skb(skb);
2934                         return NETDEV_TX_OK;
2935                 }
2936         }
2937
2938         if (rcu_access_pointer(f->nh)) {
2939                 vxlan_xmit_nh(skb, dev, f,
2940                               (vni ? : vxlan->default_dst.remote_vni), did_rsc);
2941         } else {
2942                 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2943                         struct sk_buff *skb1;
2944
2945                         if (!fdst) {
2946                                 fdst = rdst;
2947                                 continue;
2948                         }
2949                         skb1 = skb_clone(skb, GFP_ATOMIC);
2950                         if (skb1)
2951                                 vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2952                 }
2953                 if (fdst)
2954                         vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2955                 else
2956                         kfree_skb(skb);
2957         }
2958
2959         return NETDEV_TX_OK;
2960 }
2961
2962 /* Walk the forwarding table and purge stale entries */
2963 static void vxlan_cleanup(struct timer_list *t)
2964 {
2965         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2966         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2967         unsigned int h;
2968
2969         if (!netif_running(vxlan->dev))
2970                 return;
2971
2972         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2973                 struct hlist_node *p, *n;
2974
2975                 spin_lock(&vxlan->hash_lock[h]);
2976                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2977                         struct vxlan_fdb *f
2978                                 = container_of(p, struct vxlan_fdb, hlist);
2979                         unsigned long timeout;
2980
2981                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2982                                 continue;
2983
2984                         if (f->flags & NTF_EXT_LEARNED)
2985                                 continue;
2986
2987                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2988                         if (time_before_eq(timeout, jiffies)) {
2989                                 netdev_dbg(vxlan->dev,
2990                                            "garbage collect %pM\n",
2991                                            f->eth_addr);
2992                                 f->state = NUD_STALE;
2993                                 vxlan_fdb_destroy(vxlan, f, true, true);
2994                         } else if (time_before(timeout, next_timer))
2995                                 next_timer = timeout;
2996                 }
2997                 spin_unlock(&vxlan->hash_lock[h]);
2998         }
2999
3000         mod_timer(&vxlan->age_timer, next_timer);
3001 }
3002
3003 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
3004 {
3005         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3006
3007         spin_lock(&vn->sock_lock);
3008         hlist_del_init_rcu(&vxlan->hlist4.hlist);
3009 #if IS_ENABLED(CONFIG_IPV6)
3010         hlist_del_init_rcu(&vxlan->hlist6.hlist);
3011 #endif
3012         spin_unlock(&vn->sock_lock);
3013 }
3014
3015 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
3016                              struct vxlan_dev_node *node)
3017 {
3018         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3019         __be32 vni = vxlan->default_dst.remote_vni;
3020
3021         node->vxlan = vxlan;
3022         spin_lock(&vn->sock_lock);
3023         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
3024         spin_unlock(&vn->sock_lock);
3025 }
3026
3027 /* Setup stats when device is created */
3028 static int vxlan_init(struct net_device *dev)
3029 {
3030         struct vxlan_dev *vxlan = netdev_priv(dev);
3031         int err;
3032
3033         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
3034         if (!dev->tstats)
3035                 return -ENOMEM;
3036
3037         err = gro_cells_init(&vxlan->gro_cells, dev);
3038         if (err) {
3039                 free_percpu(dev->tstats);
3040                 return err;
3041         }
3042
3043         return 0;
3044 }
3045
3046 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
3047 {
3048         struct vxlan_fdb *f;
3049         u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, vni);
3050
3051         spin_lock_bh(&vxlan->hash_lock[hash_index]);
3052         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
3053         if (f)
3054                 vxlan_fdb_destroy(vxlan, f, true, true);
3055         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
3056 }
3057
3058 static void vxlan_uninit(struct net_device *dev)
3059 {
3060         struct vxlan_dev *vxlan = netdev_priv(dev);
3061
3062         gro_cells_destroy(&vxlan->gro_cells);
3063
3064         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
3065
3066         free_percpu(dev->tstats);
3067 }
3068
3069 /* Start ageing timer and join group when device is brought up */
3070 static int vxlan_open(struct net_device *dev)
3071 {
3072         struct vxlan_dev *vxlan = netdev_priv(dev);
3073         int ret;
3074
3075         ret = vxlan_sock_add(vxlan);
3076         if (ret < 0)
3077                 return ret;
3078
3079         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
3080                 ret = vxlan_igmp_join(vxlan);
3081                 if (ret == -EADDRINUSE)
3082                         ret = 0;
3083                 if (ret) {
3084                         vxlan_sock_release(vxlan);
3085                         return ret;
3086                 }
3087         }
3088
3089         if (vxlan->cfg.age_interval)
3090                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
3091
3092         return ret;
3093 }
3094
3095 /* Purge the forwarding table */
3096 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
3097 {
3098         unsigned int h;
3099
3100         for (h = 0; h < FDB_HASH_SIZE; ++h) {
3101                 struct hlist_node *p, *n;
3102
3103                 spin_lock_bh(&vxlan->hash_lock[h]);
3104                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
3105                         struct vxlan_fdb *f
3106                                 = container_of(p, struct vxlan_fdb, hlist);
3107                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
3108                                 continue;
3109                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
3110                         if (is_zero_ether_addr(f->eth_addr) &&
3111                             f->vni == vxlan->cfg.vni)
3112                                 continue;
3113                         vxlan_fdb_destroy(vxlan, f, true, true);
3114                 }
3115                 spin_unlock_bh(&vxlan->hash_lock[h]);
3116         }
3117 }
3118
3119 /* Cleanup timer and forwarding table on shutdown */
3120 static int vxlan_stop(struct net_device *dev)
3121 {
3122         struct vxlan_dev *vxlan = netdev_priv(dev);
3123         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3124         int ret = 0;
3125
3126         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
3127             !vxlan_group_used(vn, vxlan))
3128                 ret = vxlan_igmp_leave(vxlan);
3129
3130         del_timer_sync(&vxlan->age_timer);
3131
3132         vxlan_flush(vxlan, false);
3133         vxlan_sock_release(vxlan);
3134
3135         return ret;
3136 }
3137
3138 /* Stub, nothing needs to be done. */
3139 static void vxlan_set_multicast_list(struct net_device *dev)
3140 {
3141 }
3142
3143 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
3144 {
3145         struct vxlan_dev *vxlan = netdev_priv(dev);
3146         struct vxlan_rdst *dst = &vxlan->default_dst;
3147         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
3148                                                          dst->remote_ifindex);
3149         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
3150
3151         /* This check is different than dev->max_mtu, because it looks at
3152          * the lowerdev->mtu, rather than the static dev->max_mtu
3153          */
3154         if (lowerdev) {
3155                 int max_mtu = lowerdev->mtu -
3156                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
3157                 if (new_mtu > max_mtu)
3158                         return -EINVAL;
3159         }
3160
3161         dev->mtu = new_mtu;
3162         return 0;
3163 }
3164
3165 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
3166 {
3167         struct vxlan_dev *vxlan = netdev_priv(dev);
3168         struct ip_tunnel_info *info = skb_tunnel_info(skb);
3169         __be16 sport, dport;
3170
3171         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
3172                                   vxlan->cfg.port_max, true);
3173         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
3174
3175         if (ip_tunnel_info_af(info) == AF_INET) {
3176                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
3177                 struct rtable *rt;
3178
3179                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
3180                                      info->key.u.ipv4.dst,
3181                                      &info->key.u.ipv4.src, dport, sport,
3182                                      &info->dst_cache, info);
3183                 if (IS_ERR(rt))
3184                         return PTR_ERR(rt);
3185                 ip_rt_put(rt);
3186         } else {
3187 #if IS_ENABLED(CONFIG_IPV6)
3188                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
3189                 struct dst_entry *ndst;
3190
3191                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
3192                                         info->key.label, &info->key.u.ipv6.dst,
3193                                         &info->key.u.ipv6.src, dport, sport,
3194                                         &info->dst_cache, info);
3195                 if (IS_ERR(ndst))
3196                         return PTR_ERR(ndst);
3197                 dst_release(ndst);
3198 #else /* !CONFIG_IPV6 */
3199                 return -EPFNOSUPPORT;
3200 #endif
3201         }
3202         info->key.tp_src = sport;
3203         info->key.tp_dst = dport;
3204         return 0;
3205 }
3206
3207 static const struct net_device_ops vxlan_netdev_ether_ops = {
3208         .ndo_init               = vxlan_init,
3209         .ndo_uninit             = vxlan_uninit,
3210         .ndo_open               = vxlan_open,
3211         .ndo_stop               = vxlan_stop,
3212         .ndo_start_xmit         = vxlan_xmit,
3213         .ndo_get_stats64        = ip_tunnel_get_stats64,
3214         .ndo_set_rx_mode        = vxlan_set_multicast_list,
3215         .ndo_change_mtu         = vxlan_change_mtu,
3216         .ndo_validate_addr      = eth_validate_addr,
3217         .ndo_set_mac_address    = eth_mac_addr,
3218         .ndo_fdb_add            = vxlan_fdb_add,
3219         .ndo_fdb_del            = vxlan_fdb_delete,
3220         .ndo_fdb_dump           = vxlan_fdb_dump,
3221         .ndo_fdb_get            = vxlan_fdb_get,
3222         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
3223         .ndo_change_proto_down  = dev_change_proto_down_generic,
3224 };
3225
3226 static const struct net_device_ops vxlan_netdev_raw_ops = {
3227         .ndo_init               = vxlan_init,
3228         .ndo_uninit             = vxlan_uninit,
3229         .ndo_open               = vxlan_open,
3230         .ndo_stop               = vxlan_stop,
3231         .ndo_start_xmit         = vxlan_xmit,
3232         .ndo_get_stats64        = ip_tunnel_get_stats64,
3233         .ndo_change_mtu         = vxlan_change_mtu,
3234         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
3235 };
3236
3237 /* Info for udev, that this is a virtual tunnel endpoint */
3238 static struct device_type vxlan_type = {
3239         .name = "vxlan",
3240 };
3241
3242 /* Calls the ndo_udp_tunnel_add of the caller in order to
3243  * supply the listening VXLAN udp ports. Callers are expected
3244  * to implement the ndo_udp_tunnel_add.
3245  */
3246 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
3247 {
3248         struct vxlan_sock *vs;
3249         struct net *net = dev_net(dev);
3250         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3251         unsigned int i;
3252
3253         spin_lock(&vn->sock_lock);
3254         for (i = 0; i < PORT_HASH_SIZE; ++i) {
3255                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
3256                         unsigned short type;
3257
3258                         if (vs->flags & VXLAN_F_GPE)
3259                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
3260                         else
3261                                 type = UDP_TUNNEL_TYPE_VXLAN;
3262
3263                         if (push)
3264                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
3265                         else
3266                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
3267                 }
3268         }
3269         spin_unlock(&vn->sock_lock);
3270 }
3271
3272 /* Initialize the device structure. */
3273 static void vxlan_setup(struct net_device *dev)
3274 {
3275         struct vxlan_dev *vxlan = netdev_priv(dev);
3276         unsigned int h;
3277
3278         eth_hw_addr_random(dev);
3279         ether_setup(dev);
3280
3281         dev->needs_free_netdev = true;
3282         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
3283
3284         dev->features   |= NETIF_F_LLTX;
3285         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
3286         dev->features   |= NETIF_F_RXCSUM;
3287         dev->features   |= NETIF_F_GSO_SOFTWARE;
3288
3289         dev->vlan_features = dev->features;
3290         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
3291         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
3292         netif_keep_dst(dev);
3293         dev->priv_flags |= IFF_NO_QUEUE;
3294
3295         /* MTU range: 68 - 65535 */
3296         dev->min_mtu = ETH_MIN_MTU;
3297         dev->max_mtu = ETH_MAX_MTU;
3298
3299         INIT_LIST_HEAD(&vxlan->next);
3300
3301         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
3302
3303         vxlan->dev = dev;
3304
3305         for (h = 0; h < FDB_HASH_SIZE; ++h) {
3306                 spin_lock_init(&vxlan->hash_lock[h]);
3307                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
3308         }
3309 }
3310
3311 static void vxlan_ether_setup(struct net_device *dev)
3312 {
3313         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
3314         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3315         dev->netdev_ops = &vxlan_netdev_ether_ops;
3316 }
3317
3318 static void vxlan_raw_setup(struct net_device *dev)
3319 {
3320         dev->header_ops = NULL;
3321         dev->type = ARPHRD_NONE;
3322         dev->hard_header_len = 0;
3323         dev->addr_len = 0;
3324         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
3325         dev->netdev_ops = &vxlan_netdev_raw_ops;
3326 }
3327
3328 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
3329         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
3330         [IFLA_VXLAN_GROUP]      = { .len = sizeof_field(struct iphdr, daddr) },
3331         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
3332         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
3333         [IFLA_VXLAN_LOCAL]      = { .len = sizeof_field(struct iphdr, saddr) },
3334         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
3335         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
3336         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
3337         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
3338         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
3339         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
3340         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
3341         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
3342         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
3343         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
3344         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
3345         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
3346         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
3347         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
3348         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
3349         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
3350         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
3351         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
3352         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
3353         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
3354         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
3355         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
3356         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
3357         [IFLA_VXLAN_DF]         = { .type = NLA_U8 },
3358 };
3359
3360 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
3361                           struct netlink_ext_ack *extack)
3362 {
3363         if (tb[IFLA_ADDRESS]) {
3364                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
3365                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3366                                             "Provided link layer address is not Ethernet");
3367                         return -EINVAL;
3368                 }
3369
3370                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
3371                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3372                                             "Provided Ethernet address is not unicast");
3373                         return -EADDRNOTAVAIL;
3374                 }
3375         }
3376
3377         if (tb[IFLA_MTU]) {
3378                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
3379
3380                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
3381                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3382                                             "MTU must be between 68 and 65535");
3383                         return -EINVAL;
3384                 }
3385         }
3386
3387         if (!data) {
3388                 NL_SET_ERR_MSG(extack,
3389                                "Required attributes not provided to perform the operation");
3390                 return -EINVAL;
3391         }
3392
3393         if (data[IFLA_VXLAN_ID]) {
3394                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
3395
3396                 if (id >= VXLAN_N_VID) {
3397                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_ID],
3398                                             "VXLAN ID must be lower than 16777216");
3399                         return -ERANGE;
3400                 }
3401         }
3402
3403         if (data[IFLA_VXLAN_PORT_RANGE]) {
3404                 const struct ifla_vxlan_port_range *p
3405                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3406
3407                 if (ntohs(p->high) < ntohs(p->low)) {
3408                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_PORT_RANGE],
3409                                             "Invalid source port range");
3410                         return -EINVAL;
3411                 }
3412         }
3413
3414         if (data[IFLA_VXLAN_DF]) {
3415                 enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]);
3416
3417                 if (df < 0 || df > VXLAN_DF_MAX) {
3418                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_DF],
3419                                             "Invalid DF attribute");
3420                         return -EINVAL;
3421                 }
3422         }
3423
3424         return 0;
3425 }
3426
3427 static void vxlan_get_drvinfo(struct net_device *netdev,
3428                               struct ethtool_drvinfo *drvinfo)
3429 {
3430         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
3431         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
3432 }
3433
3434 static int vxlan_get_link_ksettings(struct net_device *dev,
3435                                     struct ethtool_link_ksettings *cmd)
3436 {
3437         struct vxlan_dev *vxlan = netdev_priv(dev);
3438         struct vxlan_rdst *dst = &vxlan->default_dst;
3439         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
3440                                                          dst->remote_ifindex);
3441
3442         if (!lowerdev) {
3443                 cmd->base.duplex = DUPLEX_UNKNOWN;
3444                 cmd->base.port = PORT_OTHER;
3445                 cmd->base.speed = SPEED_UNKNOWN;
3446
3447                 return 0;
3448         }
3449
3450         return __ethtool_get_link_ksettings(lowerdev, cmd);
3451 }
3452
3453 static const struct ethtool_ops vxlan_ethtool_ops = {
3454         .get_drvinfo            = vxlan_get_drvinfo,
3455         .get_link               = ethtool_op_get_link,
3456         .get_link_ksettings     = vxlan_get_link_ksettings,
3457 };
3458
3459 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
3460                                         __be16 port, u32 flags, int ifindex)
3461 {
3462         struct socket *sock;
3463         struct udp_port_cfg udp_conf;
3464         int err;
3465
3466         memset(&udp_conf, 0, sizeof(udp_conf));
3467
3468         if (ipv6) {
3469                 udp_conf.family = AF_INET6;
3470                 udp_conf.use_udp6_rx_checksums =
3471                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
3472                 udp_conf.ipv6_v6only = 1;
3473         } else {
3474                 udp_conf.family = AF_INET;
3475         }
3476
3477         udp_conf.local_udp_port = port;
3478         udp_conf.bind_ifindex = ifindex;
3479
3480         /* Open UDP socket */
3481         err = udp_sock_create(net, &udp_conf, &sock);
3482         if (err < 0)
3483                 return ERR_PTR(err);
3484
3485         return sock;
3486 }
3487
3488 /* Create new listen socket if needed */
3489 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
3490                                               __be16 port, u32 flags,
3491                                               int ifindex)
3492 {
3493         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3494         struct vxlan_sock *vs;
3495         struct socket *sock;
3496         unsigned int h;
3497         struct udp_tunnel_sock_cfg tunnel_cfg;
3498
3499         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
3500         if (!vs)
3501                 return ERR_PTR(-ENOMEM);
3502
3503         for (h = 0; h < VNI_HASH_SIZE; ++h)
3504                 INIT_HLIST_HEAD(&vs->vni_list[h]);
3505
3506         sock = vxlan_create_sock(net, ipv6, port, flags, ifindex);
3507         if (IS_ERR(sock)) {
3508                 kfree(vs);
3509                 return ERR_CAST(sock);
3510         }
3511
3512         vs->sock = sock;
3513         refcount_set(&vs->refcnt, 1);
3514         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
3515
3516         spin_lock(&vn->sock_lock);
3517         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
3518         udp_tunnel_notify_add_rx_port(sock,
3519                                       (vs->flags & VXLAN_F_GPE) ?
3520                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
3521                                       UDP_TUNNEL_TYPE_VXLAN);
3522         spin_unlock(&vn->sock_lock);
3523
3524         /* Mark socket as an encapsulation socket. */
3525         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
3526         tunnel_cfg.sk_user_data = vs;
3527         tunnel_cfg.encap_type = 1;
3528         tunnel_cfg.encap_rcv = vxlan_rcv;
3529         tunnel_cfg.encap_err_lookup = vxlan_err_lookup;
3530         tunnel_cfg.encap_destroy = NULL;
3531         tunnel_cfg.gro_receive = vxlan_gro_receive;
3532         tunnel_cfg.gro_complete = vxlan_gro_complete;
3533
3534         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
3535
3536         return vs;
3537 }
3538
3539 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
3540 {
3541         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3542         struct vxlan_sock *vs = NULL;
3543         struct vxlan_dev_node *node;
3544         int l3mdev_index = 0;
3545
3546         if (vxlan->cfg.remote_ifindex)
3547                 l3mdev_index = l3mdev_master_upper_ifindex_by_index(
3548                         vxlan->net, vxlan->cfg.remote_ifindex);
3549
3550         if (!vxlan->cfg.no_share) {
3551                 spin_lock(&vn->sock_lock);
3552                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
3553                                      vxlan->cfg.dst_port, vxlan->cfg.flags,
3554                                      l3mdev_index);
3555                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
3556                         spin_unlock(&vn->sock_lock);
3557                         return -EBUSY;
3558                 }
3559                 spin_unlock(&vn->sock_lock);
3560         }
3561         if (!vs)
3562                 vs = vxlan_socket_create(vxlan->net, ipv6,
3563                                          vxlan->cfg.dst_port, vxlan->cfg.flags,
3564                                          l3mdev_index);
3565         if (IS_ERR(vs))
3566                 return PTR_ERR(vs);
3567 #if IS_ENABLED(CONFIG_IPV6)
3568         if (ipv6) {
3569                 rcu_assign_pointer(vxlan->vn6_sock, vs);
3570                 node = &vxlan->hlist6;
3571         } else
3572 #endif
3573         {
3574                 rcu_assign_pointer(vxlan->vn4_sock, vs);
3575                 node = &vxlan->hlist4;
3576         }
3577         vxlan_vs_add_dev(vs, vxlan, node);
3578         return 0;
3579 }
3580
3581 static int vxlan_sock_add(struct vxlan_dev *vxlan)
3582 {
3583         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
3584         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
3585         bool ipv4 = !ipv6 || metadata;
3586         int ret = 0;
3587
3588         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
3589 #if IS_ENABLED(CONFIG_IPV6)
3590         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
3591         if (ipv6) {
3592                 ret = __vxlan_sock_add(vxlan, true);
3593                 if (ret < 0 && ret != -EAFNOSUPPORT)
3594                         ipv4 = false;
3595         }
3596 #endif
3597         if (ipv4)
3598                 ret = __vxlan_sock_add(vxlan, false);
3599         if (ret < 0)
3600                 vxlan_sock_release(vxlan);
3601         return ret;
3602 }
3603
3604 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
3605                                  struct net_device **lower,
3606                                  struct vxlan_dev *old,
3607                                  struct netlink_ext_ack *extack)
3608 {
3609         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
3610         struct vxlan_dev *tmp;
3611         bool use_ipv6 = false;
3612
3613         if (conf->flags & VXLAN_F_GPE) {
3614                 /* For now, allow GPE only together with
3615                  * COLLECT_METADATA. This can be relaxed later; in such
3616                  * case, the other side of the PtP link will have to be
3617                  * provided.
3618                  */
3619                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
3620                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3621                         NL_SET_ERR_MSG(extack,
3622                                        "VXLAN GPE does not support this combination of attributes");
3623                         return -EINVAL;
3624                 }
3625         }
3626
3627         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
3628                 /* Unless IPv6 is explicitly requested, assume IPv4 */
3629                 conf->remote_ip.sa.sa_family = AF_INET;
3630                 conf->saddr.sa.sa_family = AF_INET;
3631         } else if (!conf->remote_ip.sa.sa_family) {
3632                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
3633         } else if (!conf->saddr.sa.sa_family) {
3634                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
3635         }
3636
3637         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
3638                 NL_SET_ERR_MSG(extack,
3639                                "Local and remote address must be from the same family");
3640                 return -EINVAL;
3641         }
3642
3643         if (vxlan_addr_multicast(&conf->saddr)) {
3644                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
3645                 return -EINVAL;
3646         }
3647
3648         if (conf->saddr.sa.sa_family == AF_INET6) {
3649                 if (!IS_ENABLED(CONFIG_IPV6)) {
3650                         NL_SET_ERR_MSG(extack,
3651                                        "IPv6 support not enabled in the kernel");
3652                         return -EPFNOSUPPORT;
3653                 }
3654                 use_ipv6 = true;
3655                 conf->flags |= VXLAN_F_IPV6;
3656
3657                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3658                         int local_type =
3659                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3660                         int remote_type =
3661                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3662
3663                         if (local_type & IPV6_ADDR_LINKLOCAL) {
3664                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3665                                     (remote_type != IPV6_ADDR_ANY)) {
3666                                         NL_SET_ERR_MSG(extack,
3667                                                        "Invalid combination of local and remote address scopes");
3668                                         return -EINVAL;
3669                                 }
3670
3671                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3672                         } else {
3673                                 if (remote_type ==
3674                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3675                                         NL_SET_ERR_MSG(extack,
3676                                                        "Invalid combination of local and remote address scopes");
3677                                         return -EINVAL;
3678                                 }
3679
3680                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3681                         }
3682                 }
3683         }
3684
3685         if (conf->label && !use_ipv6) {
3686                 NL_SET_ERR_MSG(extack,
3687                                "Label attribute only applies to IPv6 VXLAN devices");
3688                 return -EINVAL;
3689         }
3690
3691         if (conf->remote_ifindex) {
3692                 struct net_device *lowerdev;
3693
3694                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3695                 if (!lowerdev) {
3696                         NL_SET_ERR_MSG(extack,
3697                                        "Invalid local interface, device not found");
3698                         return -ENODEV;
3699                 }
3700
3701 #if IS_ENABLED(CONFIG_IPV6)
3702                 if (use_ipv6) {
3703                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3704                         if (idev && idev->cnf.disable_ipv6) {
3705                                 NL_SET_ERR_MSG(extack,
3706                                                "IPv6 support disabled by administrator");
3707                                 return -EPERM;
3708                         }
3709                 }
3710 #endif
3711
3712                 *lower = lowerdev;
3713         } else {
3714                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3715                         NL_SET_ERR_MSG(extack,
3716                                        "Local interface required for multicast remote destination");
3717
3718                         return -EINVAL;
3719                 }
3720
3721 #if IS_ENABLED(CONFIG_IPV6)
3722                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3723                         NL_SET_ERR_MSG(extack,
3724                                        "Local interface required for link-local local/remote addresses");
3725                         return -EINVAL;
3726                 }
3727 #endif
3728
3729                 *lower = NULL;
3730         }
3731
3732         if (!conf->dst_port) {
3733                 if (conf->flags & VXLAN_F_GPE)
3734                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3735                 else
3736                         conf->dst_port = htons(vxlan_port);
3737         }
3738
3739         if (!conf->age_interval)
3740                 conf->age_interval = FDB_AGE_DEFAULT;
3741
3742         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3743                 if (tmp == old)
3744                         continue;
3745
3746                 if (tmp->cfg.vni != conf->vni)
3747                         continue;
3748                 if (tmp->cfg.dst_port != conf->dst_port)
3749                         continue;
3750                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3751                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3752                         continue;
3753
3754                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3755                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3756                         continue;
3757
3758                 NL_SET_ERR_MSG(extack,
3759                                "A VXLAN device with the specified VNI already exists");
3760                 return -EEXIST;
3761         }
3762
3763         return 0;
3764 }
3765
3766 static void vxlan_config_apply(struct net_device *dev,
3767                                struct vxlan_config *conf,
3768                                struct net_device *lowerdev,
3769                                struct net *src_net,
3770                                bool changelink)
3771 {
3772         struct vxlan_dev *vxlan = netdev_priv(dev);
3773         struct vxlan_rdst *dst = &vxlan->default_dst;
3774         unsigned short needed_headroom = ETH_HLEN;
3775         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3776         int max_mtu = ETH_MAX_MTU;
3777
3778         if (!changelink) {
3779                 if (conf->flags & VXLAN_F_GPE)
3780                         vxlan_raw_setup(dev);
3781                 else
3782                         vxlan_ether_setup(dev);
3783
3784                 if (conf->mtu)
3785                         dev->mtu = conf->mtu;
3786
3787                 vxlan->net = src_net;
3788         }
3789
3790         dst->remote_vni = conf->vni;
3791
3792         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3793
3794         if (lowerdev) {
3795                 dst->remote_ifindex = conf->remote_ifindex;
3796
3797                 dev->gso_max_size = lowerdev->gso_max_size;
3798                 dev->gso_max_segs = lowerdev->gso_max_segs;
3799
3800                 needed_headroom = lowerdev->hard_header_len;
3801
3802                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3803                                            VXLAN_HEADROOM);
3804                 if (max_mtu < ETH_MIN_MTU)
3805                         max_mtu = ETH_MIN_MTU;
3806
3807                 if (!changelink && !conf->mtu)
3808                         dev->mtu = max_mtu;
3809         }
3810
3811         if (dev->mtu > max_mtu)
3812                 dev->mtu = max_mtu;
3813
3814         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3815                 needed_headroom += VXLAN6_HEADROOM;
3816         else
3817                 needed_headroom += VXLAN_HEADROOM;
3818         dev->needed_headroom = needed_headroom;
3819
3820         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3821 }
3822
3823 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3824                                struct vxlan_config *conf, bool changelink,
3825                                struct netlink_ext_ack *extack)
3826 {
3827         struct vxlan_dev *vxlan = netdev_priv(dev);
3828         struct net_device *lowerdev;
3829         int ret;
3830
3831         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3832         if (ret)
3833                 return ret;
3834
3835         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3836
3837         return 0;
3838 }
3839
3840 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3841                               struct vxlan_config *conf,
3842                               struct netlink_ext_ack *extack)
3843 {
3844         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3845         struct vxlan_dev *vxlan = netdev_priv(dev);
3846         struct net_device *remote_dev = NULL;
3847         struct vxlan_fdb *f = NULL;
3848         bool unregister = false;
3849         struct vxlan_rdst *dst;
3850         int err;
3851
3852         dst = &vxlan->default_dst;
3853         err = vxlan_dev_configure(net, dev, conf, false, extack);
3854         if (err)
3855                 return err;
3856
3857         dev->ethtool_ops = &vxlan_ethtool_ops;
3858
3859         /* create an fdb entry for a valid default destination */
3860         if (!vxlan_addr_any(&dst->remote_ip)) {
3861                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3862                                        &dst->remote_ip,
3863                                        NUD_REACHABLE | NUD_PERMANENT,
3864                                        vxlan->cfg.dst_port,
3865                                        dst->remote_vni,
3866                                        dst->remote_vni,
3867                                        dst->remote_ifindex,
3868                                        NTF_SELF, 0, &f, extack);
3869                 if (err)
3870                         return err;
3871         }
3872
3873         err = register_netdevice(dev);
3874         if (err)
3875                 goto errout;
3876         unregister = true;
3877
3878         if (dst->remote_ifindex) {
3879                 remote_dev = __dev_get_by_index(net, dst->remote_ifindex);
3880                 if (!remote_dev)
3881                         goto errout;
3882
3883                 err = netdev_upper_dev_link(remote_dev, dev, extack);
3884                 if (err)
3885                         goto errout;
3886         }
3887
3888         err = rtnl_configure_link(dev, NULL);
3889         if (err < 0)
3890                 goto unlink;
3891
3892         if (f) {
3893                 vxlan_fdb_insert(vxlan, all_zeros_mac, dst->remote_vni, f);
3894
3895                 /* notify default fdb entry */
3896                 err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f),
3897                                        RTM_NEWNEIGH, true, extack);
3898                 if (err) {
3899                         vxlan_fdb_destroy(vxlan, f, false, false);
3900                         if (remote_dev)
3901                                 netdev_upper_dev_unlink(remote_dev, dev);
3902                         goto unregister;
3903                 }
3904         }
3905
3906         list_add(&vxlan->next, &vn->vxlan_list);
3907         if (remote_dev)
3908                 dst->remote_dev = remote_dev;
3909         return 0;
3910 unlink:
3911         if (remote_dev)
3912                 netdev_upper_dev_unlink(remote_dev, dev);
3913 errout:
3914         /* unregister_netdevice() destroys the default FDB entry with deletion
3915          * notification. But the addition notification was not sent yet, so
3916          * destroy the entry by hand here.
3917          */
3918         if (f)
3919                 __vxlan_fdb_free(f);
3920 unregister:
3921         if (unregister)
3922                 unregister_netdevice(dev);
3923         return err;
3924 }
3925
3926 /* Set/clear flags based on attribute */
3927 static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[],
3928                           int attrtype, unsigned long mask, bool changelink,
3929                           bool changelink_supported,
3930                           struct netlink_ext_ack *extack)
3931 {
3932         unsigned long flags;
3933
3934         if (!tb[attrtype])
3935                 return 0;
3936
3937         if (changelink && !changelink_supported) {
3938                 vxlan_flag_attr_error(attrtype, extack);
3939                 return -EOPNOTSUPP;
3940         }
3941
3942         if (vxlan_policy[attrtype].type == NLA_FLAG)
3943                 flags = conf->flags | mask;
3944         else if (nla_get_u8(tb[attrtype]))
3945                 flags = conf->flags | mask;
3946         else
3947                 flags = conf->flags & ~mask;
3948
3949         conf->flags = flags;
3950
3951         return 0;
3952 }
3953
3954 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3955                          struct net_device *dev, struct vxlan_config *conf,
3956                          bool changelink, struct netlink_ext_ack *extack)
3957 {
3958         struct vxlan_dev *vxlan = netdev_priv(dev);
3959         int err = 0;
3960
3961         memset(conf, 0, sizeof(*conf));
3962
3963         /* if changelink operation, start with old existing cfg */
3964         if (changelink)
3965                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3966
3967         if (data[IFLA_VXLAN_ID]) {
3968                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3969
3970                 if (changelink && (vni != conf->vni)) {
3971                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI");
3972                         return -EOPNOTSUPP;
3973                 }
3974                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3975         }
3976
3977         if (data[IFLA_VXLAN_GROUP]) {
3978                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) {
3979                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group");
3980                         return -EOPNOTSUPP;
3981                 }
3982
3983                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3984                 conf->remote_ip.sa.sa_family = AF_INET;
3985         } else if (data[IFLA_VXLAN_GROUP6]) {
3986                 if (!IS_ENABLED(CONFIG_IPV6)) {
3987                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel");
3988                         return -EPFNOSUPPORT;
3989                 }
3990
3991                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) {
3992                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group");
3993                         return -EOPNOTSUPP;
3994                 }
3995
3996                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3997                 conf->remote_ip.sa.sa_family = AF_INET6;
3998         }
3999
4000         if (data[IFLA_VXLAN_LOCAL]) {
4001                 if (changelink && (conf->saddr.sa.sa_family != AF_INET)) {
4002                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old");
4003                         return -EOPNOTSUPP;
4004                 }
4005
4006                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
4007                 conf->saddr.sa.sa_family = AF_INET;
4008         } else if (data[IFLA_VXLAN_LOCAL6]) {
4009                 if (!IS_ENABLED(CONFIG_IPV6)) {
4010                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel");
4011                         return -EPFNOSUPPORT;
4012                 }
4013
4014                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) {
4015                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old");
4016                         return -EOPNOTSUPP;
4017                 }
4018
4019                 /* TODO: respect scope id */
4020                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
4021                 conf->saddr.sa.sa_family = AF_INET6;
4022         }
4023
4024         if (data[IFLA_VXLAN_LINK])
4025                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
4026
4027         if (data[IFLA_VXLAN_TOS])
4028                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
4029
4030         if (data[IFLA_VXLAN_TTL])
4031                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
4032
4033         if (data[IFLA_VXLAN_TTL_INHERIT]) {
4034                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT,
4035                                     VXLAN_F_TTL_INHERIT, changelink, false,
4036                                     extack);
4037                 if (err)
4038                         return err;
4039
4040         }
4041
4042         if (data[IFLA_VXLAN_LABEL])
4043                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
4044                              IPV6_FLOWLABEL_MASK;
4045
4046         if (data[IFLA_VXLAN_LEARNING]) {
4047                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING,
4048                                     VXLAN_F_LEARN, changelink, true,
4049                                     extack);
4050                 if (err)
4051                         return err;
4052         } else if (!changelink) {
4053                 /* default to learn on a new device */
4054                 conf->flags |= VXLAN_F_LEARN;
4055         }
4056
4057         if (data[IFLA_VXLAN_AGEING])
4058                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
4059
4060         if (data[IFLA_VXLAN_PROXY]) {
4061                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY,
4062                                     VXLAN_F_PROXY, changelink, false,
4063                                     extack);
4064                 if (err)
4065                         return err;
4066         }
4067
4068         if (data[IFLA_VXLAN_RSC]) {
4069                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC,
4070                                     VXLAN_F_RSC, changelink, false,
4071                                     extack);
4072                 if (err)
4073                         return err;
4074         }
4075
4076         if (data[IFLA_VXLAN_L2MISS]) {
4077                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS,
4078                                     VXLAN_F_L2MISS, changelink, false,
4079                                     extack);
4080                 if (err)
4081                         return err;
4082         }
4083
4084         if (data[IFLA_VXLAN_L3MISS]) {
4085                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS,
4086                                     VXLAN_F_L3MISS, changelink, false,
4087                                     extack);
4088                 if (err)
4089                         return err;
4090         }
4091
4092         if (data[IFLA_VXLAN_LIMIT]) {
4093                 if (changelink) {
4094                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT],
4095                                             "Cannot change limit");
4096                         return -EOPNOTSUPP;
4097                 }
4098                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
4099         }
4100
4101         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
4102                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA,
4103                                     VXLAN_F_COLLECT_METADATA, changelink, false,
4104                                     extack);
4105                 if (err)
4106                         return err;
4107         }
4108
4109         if (data[IFLA_VXLAN_PORT_RANGE]) {
4110                 if (!changelink) {
4111                         const struct ifla_vxlan_port_range *p
4112                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
4113                         conf->port_min = ntohs(p->low);
4114                         conf->port_max = ntohs(p->high);
4115                 } else {
4116                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
4117                                             "Cannot change port range");
4118                         return -EOPNOTSUPP;
4119                 }
4120         }
4121
4122         if (data[IFLA_VXLAN_PORT]) {
4123                 if (changelink) {
4124                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT],
4125                                             "Cannot change port");
4126                         return -EOPNOTSUPP;
4127                 }
4128                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
4129         }
4130
4131         if (data[IFLA_VXLAN_UDP_CSUM]) {
4132                 if (changelink) {
4133                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM],
4134                                             "Cannot change UDP_CSUM flag");
4135                         return -EOPNOTSUPP;
4136                 }
4137                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
4138                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
4139         }
4140
4141         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
4142                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4143                                     VXLAN_F_UDP_ZERO_CSUM6_TX, changelink,
4144                                     false, extack);
4145                 if (err)
4146                         return err;
4147         }
4148
4149         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
4150                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4151                                     VXLAN_F_UDP_ZERO_CSUM6_RX, changelink,
4152                                     false, extack);
4153                 if (err)
4154                         return err;
4155         }
4156
4157         if (data[IFLA_VXLAN_REMCSUM_TX]) {
4158                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX,
4159                                     VXLAN_F_REMCSUM_TX, changelink, false,
4160                                     extack);
4161                 if (err)
4162                         return err;
4163         }
4164
4165         if (data[IFLA_VXLAN_REMCSUM_RX]) {
4166                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX,
4167                                     VXLAN_F_REMCSUM_RX, changelink, false,
4168                                     extack);
4169                 if (err)
4170                         return err;
4171         }
4172
4173         if (data[IFLA_VXLAN_GBP]) {
4174                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP,
4175                                     VXLAN_F_GBP, changelink, false, extack);
4176                 if (err)
4177                         return err;
4178         }
4179
4180         if (data[IFLA_VXLAN_GPE]) {
4181                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE,
4182                                     VXLAN_F_GPE, changelink, false,
4183                                     extack);
4184                 if (err)
4185                         return err;
4186         }
4187
4188         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
4189                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL,
4190                                     VXLAN_F_REMCSUM_NOPARTIAL, changelink,
4191                                     false, extack);
4192                 if (err)
4193                         return err;
4194         }
4195
4196         if (tb[IFLA_MTU]) {
4197                 if (changelink) {
4198                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
4199                                             "Cannot change mtu");
4200                         return -EOPNOTSUPP;
4201                 }
4202                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
4203         }
4204
4205         if (data[IFLA_VXLAN_DF])
4206                 conf->df = nla_get_u8(data[IFLA_VXLAN_DF]);
4207
4208         return 0;
4209 }
4210
4211 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
4212                          struct nlattr *tb[], struct nlattr *data[],
4213                          struct netlink_ext_ack *extack)
4214 {
4215         struct vxlan_config conf;
4216         int err;
4217
4218         err = vxlan_nl2conf(tb, data, dev, &conf, false, extack);
4219         if (err)
4220                 return err;
4221
4222         return __vxlan_dev_create(src_net, dev, &conf, extack);
4223 }
4224
4225 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
4226                             struct nlattr *data[],
4227                             struct netlink_ext_ack *extack)
4228 {
4229         struct vxlan_dev *vxlan = netdev_priv(dev);
4230         struct net_device *lowerdev;
4231         struct vxlan_config conf;
4232         struct vxlan_rdst *dst;
4233         int err;
4234
4235         dst = &vxlan->default_dst;
4236         err = vxlan_nl2conf(tb, data, dev, &conf, true, extack);
4237         if (err)
4238                 return err;
4239
4240         err = vxlan_config_validate(vxlan->net, &conf, &lowerdev,
4241                                     vxlan, extack);
4242         if (err)
4243                 return err;
4244
4245         if (dst->remote_dev == lowerdev)
4246                 lowerdev = NULL;
4247
4248         err = netdev_adjacent_change_prepare(dst->remote_dev, lowerdev, dev,
4249                                              extack);
4250         if (err)
4251                 return err;
4252
4253         /* handle default dst entry */
4254         if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) {
4255                 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, conf.vni);
4256
4257                 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4258                 if (!vxlan_addr_any(&conf.remote_ip)) {
4259                         err = vxlan_fdb_update(vxlan, all_zeros_mac,
4260                                                &conf.remote_ip,
4261                                                NUD_REACHABLE | NUD_PERMANENT,
4262                                                NLM_F_APPEND | NLM_F_CREATE,
4263                                                vxlan->cfg.dst_port,
4264                                                conf.vni, conf.vni,
4265                                                conf.remote_ifindex,
4266                                                NTF_SELF, 0, true, extack);
4267                         if (err) {
4268                                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4269                                 netdev_adjacent_change_abort(dst->remote_dev,
4270                                                              lowerdev, dev);
4271                                 return err;
4272                         }
4273                 }
4274                 if (!vxlan_addr_any(&dst->remote_ip))
4275                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
4276                                            dst->remote_ip,
4277                                            vxlan->cfg.dst_port,
4278                                            dst->remote_vni,
4279                                            dst->remote_vni,
4280                                            dst->remote_ifindex,
4281                                            true);
4282                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4283         }
4284
4285         if (conf.age_interval != vxlan->cfg.age_interval)
4286                 mod_timer(&vxlan->age_timer, jiffies);
4287
4288         netdev_adjacent_change_commit(dst->remote_dev, lowerdev, dev);
4289         if (lowerdev && lowerdev != dst->remote_dev)
4290                 dst->remote_dev = lowerdev;
4291         vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true);
4292         return 0;
4293 }
4294
4295 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
4296 {
4297         struct vxlan_dev *vxlan = netdev_priv(dev);
4298
4299         vxlan_flush(vxlan, true);
4300
4301         list_del(&vxlan->next);
4302         unregister_netdevice_queue(dev, head);
4303         if (vxlan->default_dst.remote_dev)
4304                 netdev_upper_dev_unlink(vxlan->default_dst.remote_dev, dev);
4305 }
4306
4307 static size_t vxlan_get_size(const struct net_device *dev)
4308 {
4309
4310         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
4311                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
4312                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
4313                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
4314                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
4315                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL_INHERIT */
4316                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
4317                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_DF */
4318                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
4319                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
4320                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
4321                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
4322                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
4323                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
4324                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
4325                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
4326                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
4327                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
4328                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
4329                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
4330                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
4331                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
4332                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
4333                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
4334                 0;
4335 }
4336
4337 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
4338 {
4339         const struct vxlan_dev *vxlan = netdev_priv(dev);
4340         const struct vxlan_rdst *dst = &vxlan->default_dst;
4341         struct ifla_vxlan_port_range ports = {
4342                 .low =  htons(vxlan->cfg.port_min),
4343                 .high = htons(vxlan->cfg.port_max),
4344         };
4345
4346         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
4347                 goto nla_put_failure;
4348
4349         if (!vxlan_addr_any(&dst->remote_ip)) {
4350                 if (dst->remote_ip.sa.sa_family == AF_INET) {
4351                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
4352                                             dst->remote_ip.sin.sin_addr.s_addr))
4353                                 goto nla_put_failure;
4354 #if IS_ENABLED(CONFIG_IPV6)
4355                 } else {
4356                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
4357                                              &dst->remote_ip.sin6.sin6_addr))
4358                                 goto nla_put_failure;
4359 #endif
4360                 }
4361         }
4362
4363         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
4364                 goto nla_put_failure;
4365
4366         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
4367                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
4368                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
4369                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
4370                                 goto nla_put_failure;
4371 #if IS_ENABLED(CONFIG_IPV6)
4372                 } else {
4373                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
4374                                              &vxlan->cfg.saddr.sin6.sin6_addr))
4375                                 goto nla_put_failure;
4376 #endif
4377                 }
4378         }
4379
4380         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
4381             nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT,
4382                        !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) ||
4383             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
4384             nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) ||
4385             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
4386             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
4387                        !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
4388             nla_put_u8(skb, IFLA_VXLAN_PROXY,
4389                        !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
4390             nla_put_u8(skb, IFLA_VXLAN_RSC,
4391                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
4392             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
4393                        !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
4394             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
4395                        !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
4396             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
4397                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
4398             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
4399             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
4400             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
4401             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
4402                        !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
4403             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4404                        !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
4405             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4406                        !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
4407             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
4408                        !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
4409             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
4410                        !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
4411                 goto nla_put_failure;
4412
4413         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
4414                 goto nla_put_failure;
4415
4416         if (vxlan->cfg.flags & VXLAN_F_GBP &&
4417             nla_put_flag(skb, IFLA_VXLAN_GBP))
4418                 goto nla_put_failure;
4419
4420         if (vxlan->cfg.flags & VXLAN_F_GPE &&
4421             nla_put_flag(skb, IFLA_VXLAN_GPE))
4422                 goto nla_put_failure;
4423
4424         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
4425             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
4426                 goto nla_put_failure;
4427
4428         return 0;
4429
4430 nla_put_failure:
4431         return -EMSGSIZE;
4432 }
4433
4434 static struct net *vxlan_get_link_net(const struct net_device *dev)
4435 {
4436         struct vxlan_dev *vxlan = netdev_priv(dev);
4437
4438         return vxlan->net;
4439 }
4440
4441 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
4442         .kind           = "vxlan",
4443         .maxtype        = IFLA_VXLAN_MAX,
4444         .policy         = vxlan_policy,
4445         .priv_size      = sizeof(struct vxlan_dev),
4446         .setup          = vxlan_setup,
4447         .validate       = vxlan_validate,
4448         .newlink        = vxlan_newlink,
4449         .changelink     = vxlan_changelink,
4450         .dellink        = vxlan_dellink,
4451         .get_size       = vxlan_get_size,
4452         .fill_info      = vxlan_fill_info,
4453         .get_link_net   = vxlan_get_link_net,
4454 };
4455
4456 struct net_device *vxlan_dev_create(struct net *net, const char *name,
4457                                     u8 name_assign_type,
4458                                     struct vxlan_config *conf)
4459 {
4460         struct nlattr *tb[IFLA_MAX + 1];
4461         struct net_device *dev;
4462         int err;
4463
4464         memset(&tb, 0, sizeof(tb));
4465
4466         dev = rtnl_create_link(net, name, name_assign_type,
4467                                &vxlan_link_ops, tb, NULL);
4468         if (IS_ERR(dev))
4469                 return dev;
4470
4471         err = __vxlan_dev_create(net, dev, conf, NULL);
4472         if (err < 0) {
4473                 free_netdev(dev);
4474                 return ERR_PTR(err);
4475         }
4476
4477         err = rtnl_configure_link(dev, NULL);
4478         if (err < 0) {
4479                 LIST_HEAD(list_kill);
4480
4481                 vxlan_dellink(dev, &list_kill);
4482                 unregister_netdevice_many(&list_kill);
4483                 return ERR_PTR(err);
4484         }
4485
4486         return dev;
4487 }
4488 EXPORT_SYMBOL_GPL(vxlan_dev_create);
4489
4490 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
4491                                              struct net_device *dev)
4492 {
4493         struct vxlan_dev *vxlan, *next;
4494         LIST_HEAD(list_kill);
4495
4496         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4497                 struct vxlan_rdst *dst = &vxlan->default_dst;
4498
4499                 /* In case we created vxlan device with carrier
4500                  * and we loose the carrier due to module unload
4501                  * we also need to remove vxlan device. In other
4502                  * cases, it's not necessary and remote_ifindex
4503                  * is 0 here, so no matches.
4504                  */
4505                 if (dst->remote_ifindex == dev->ifindex)
4506                         vxlan_dellink(vxlan->dev, &list_kill);
4507         }
4508
4509         unregister_netdevice_many(&list_kill);
4510 }
4511
4512 static int vxlan_netdevice_event(struct notifier_block *unused,
4513                                  unsigned long event, void *ptr)
4514 {
4515         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4516         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
4517
4518         if (event == NETDEV_UNREGISTER) {
4519                 if (!dev->udp_tunnel_nic_info)
4520                         vxlan_offload_rx_ports(dev, false);
4521                 vxlan_handle_lowerdev_unregister(vn, dev);
4522         } else if (event == NETDEV_REGISTER) {
4523                 if (!dev->udp_tunnel_nic_info)
4524                         vxlan_offload_rx_ports(dev, true);
4525         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
4526                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
4527                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
4528         }
4529
4530         return NOTIFY_DONE;
4531 }
4532
4533 static struct notifier_block vxlan_notifier_block __read_mostly = {
4534         .notifier_call = vxlan_netdevice_event,
4535 };
4536
4537 static void
4538 vxlan_fdb_offloaded_set(struct net_device *dev,
4539                         struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4540 {
4541         struct vxlan_dev *vxlan = netdev_priv(dev);
4542         struct vxlan_rdst *rdst;
4543         struct vxlan_fdb *f;
4544         u32 hash_index;
4545
4546         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4547
4548         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4549
4550         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4551         if (!f)
4552                 goto out;
4553
4554         rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip,
4555                                    fdb_info->remote_port,
4556                                    fdb_info->remote_vni,
4557                                    fdb_info->remote_ifindex);
4558         if (!rdst)
4559                 goto out;
4560
4561         rdst->offloaded = fdb_info->offloaded;
4562
4563 out:
4564         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4565 }
4566
4567 static int
4568 vxlan_fdb_external_learn_add(struct net_device *dev,
4569                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4570 {
4571         struct vxlan_dev *vxlan = netdev_priv(dev);
4572         struct netlink_ext_ack *extack;
4573         u32 hash_index;
4574         int err;
4575
4576         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4577         extack = switchdev_notifier_info_to_extack(&fdb_info->info);
4578
4579         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4580         err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip,
4581                                NUD_REACHABLE,
4582                                NLM_F_CREATE | NLM_F_REPLACE,
4583                                fdb_info->remote_port,
4584                                fdb_info->vni,
4585                                fdb_info->remote_vni,
4586                                fdb_info->remote_ifindex,
4587                                NTF_USE | NTF_SELF | NTF_EXT_LEARNED,
4588                                0, false, extack);
4589         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4590
4591         return err;
4592 }
4593
4594 static int
4595 vxlan_fdb_external_learn_del(struct net_device *dev,
4596                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4597 {
4598         struct vxlan_dev *vxlan = netdev_priv(dev);
4599         struct vxlan_fdb *f;
4600         u32 hash_index;
4601         int err = 0;
4602
4603         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4604         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4605
4606         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4607         if (!f)
4608                 err = -ENOENT;
4609         else if (f->flags & NTF_EXT_LEARNED)
4610                 err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr,
4611                                          fdb_info->remote_ip,
4612                                          fdb_info->remote_port,
4613                                          fdb_info->vni,
4614                                          fdb_info->remote_vni,
4615                                          fdb_info->remote_ifindex,
4616                                          false);
4617
4618         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4619
4620         return err;
4621 }
4622
4623 static int vxlan_switchdev_event(struct notifier_block *unused,
4624                                  unsigned long event, void *ptr)
4625 {
4626         struct net_device *dev = switchdev_notifier_info_to_dev(ptr);
4627         struct switchdev_notifier_vxlan_fdb_info *fdb_info;
4628         int err = 0;
4629
4630         switch (event) {
4631         case SWITCHDEV_VXLAN_FDB_OFFLOADED:
4632                 vxlan_fdb_offloaded_set(dev, ptr);
4633                 break;
4634         case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE:
4635                 fdb_info = ptr;
4636                 err = vxlan_fdb_external_learn_add(dev, fdb_info);
4637                 if (err) {
4638                         err = notifier_from_errno(err);
4639                         break;
4640                 }
4641                 fdb_info->offloaded = true;
4642                 vxlan_fdb_offloaded_set(dev, fdb_info);
4643                 break;
4644         case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE:
4645                 fdb_info = ptr;
4646                 err = vxlan_fdb_external_learn_del(dev, fdb_info);
4647                 if (err) {
4648                         err = notifier_from_errno(err);
4649                         break;
4650                 }
4651                 fdb_info->offloaded = false;
4652                 vxlan_fdb_offloaded_set(dev, fdb_info);
4653                 break;
4654         }
4655
4656         return err;
4657 }
4658
4659 static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = {
4660         .notifier_call = vxlan_switchdev_event,
4661 };
4662
4663 static void vxlan_fdb_nh_flush(struct nexthop *nh)
4664 {
4665         struct vxlan_fdb *fdb;
4666         struct vxlan_dev *vxlan;
4667         u32 hash_index;
4668
4669         rcu_read_lock();
4670         list_for_each_entry_rcu(fdb, &nh->fdb_list, nh_list) {
4671                 vxlan = rcu_dereference(fdb->vdev);
4672                 WARN_ON(!vxlan);
4673                 hash_index = fdb_head_index(vxlan, fdb->eth_addr,
4674                                             vxlan->default_dst.remote_vni);
4675                 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4676                 if (!hlist_unhashed(&fdb->hlist))
4677                         vxlan_fdb_destroy(vxlan, fdb, false, false);
4678                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4679         }
4680         rcu_read_unlock();
4681 }
4682
4683 static int vxlan_nexthop_event(struct notifier_block *nb,
4684                                unsigned long event, void *ptr)
4685 {
4686         struct nexthop *nh = ptr;
4687
4688         if (!nh || event != NEXTHOP_EVENT_DEL)
4689                 return NOTIFY_DONE;
4690
4691         vxlan_fdb_nh_flush(nh);
4692
4693         return NOTIFY_DONE;
4694 }
4695
4696 static struct notifier_block vxlan_nexthop_notifier_block __read_mostly = {
4697         .notifier_call = vxlan_nexthop_event,
4698 };
4699
4700 static __net_init int vxlan_init_net(struct net *net)
4701 {
4702         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4703         unsigned int h;
4704
4705         INIT_LIST_HEAD(&vn->vxlan_list);
4706         spin_lock_init(&vn->sock_lock);
4707
4708         for (h = 0; h < PORT_HASH_SIZE; ++h)
4709                 INIT_HLIST_HEAD(&vn->sock_list[h]);
4710
4711         return register_nexthop_notifier(net, &vxlan_nexthop_notifier_block);
4712 }
4713
4714 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
4715 {
4716         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4717         struct vxlan_dev *vxlan, *next;
4718         struct net_device *dev, *aux;
4719         unsigned int h;
4720
4721         for_each_netdev_safe(net, dev, aux)
4722                 if (dev->rtnl_link_ops == &vxlan_link_ops)
4723                         unregister_netdevice_queue(dev, head);
4724
4725         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4726                 /* If vxlan->dev is in the same netns, it has already been added
4727                  * to the list by the previous loop.
4728                  */
4729                 if (!net_eq(dev_net(vxlan->dev), net))
4730                         unregister_netdevice_queue(vxlan->dev, head);
4731         }
4732
4733         for (h = 0; h < PORT_HASH_SIZE; ++h)
4734                 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
4735 }
4736
4737 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
4738 {
4739         struct net *net;
4740         LIST_HEAD(list);
4741
4742         rtnl_lock();
4743         list_for_each_entry(net, net_list, exit_list)
4744                 unregister_nexthop_notifier(net, &vxlan_nexthop_notifier_block);
4745         list_for_each_entry(net, net_list, exit_list)
4746                 vxlan_destroy_tunnels(net, &list);
4747
4748         unregister_netdevice_many(&list);
4749         rtnl_unlock();
4750 }
4751
4752 static struct pernet_operations vxlan_net_ops = {
4753         .init = vxlan_init_net,
4754         .exit_batch = vxlan_exit_batch_net,
4755         .id   = &vxlan_net_id,
4756         .size = sizeof(struct vxlan_net),
4757 };
4758
4759 static int __init vxlan_init_module(void)
4760 {
4761         int rc;
4762
4763         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
4764
4765         rc = register_pernet_subsys(&vxlan_net_ops);
4766         if (rc)
4767                 goto out1;
4768
4769         rc = register_netdevice_notifier(&vxlan_notifier_block);
4770         if (rc)
4771                 goto out2;
4772
4773         rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block);
4774         if (rc)
4775                 goto out3;
4776
4777         rc = rtnl_link_register(&vxlan_link_ops);
4778         if (rc)
4779                 goto out4;
4780
4781         return 0;
4782 out4:
4783         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4784 out3:
4785         unregister_netdevice_notifier(&vxlan_notifier_block);
4786 out2:
4787         unregister_pernet_subsys(&vxlan_net_ops);
4788 out1:
4789         return rc;
4790 }
4791 late_initcall(vxlan_init_module);
4792
4793 static void __exit vxlan_cleanup_module(void)
4794 {
4795         rtnl_link_unregister(&vxlan_link_ops);
4796         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4797         unregister_netdevice_notifier(&vxlan_notifier_block);
4798         unregister_pernet_subsys(&vxlan_net_ops);
4799         /* rcu_barrier() is called by netns */
4800 }
4801 module_exit(vxlan_cleanup_module);
4802
4803 MODULE_LICENSE("GPL");
4804 MODULE_VERSION(VXLAN_VERSION);
4805 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
4806 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
4807 MODULE_ALIAS_RTNL_LINK("vxlan");