drivers/net/vxlan.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * VXLAN: Virtual eXtensible Local Area Network
   4  *
   5  * Copyright (c) 2012-2013 Vyatta Inc.
   6  */
   7
   8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   9
  10 #include <linux/kernel.h>
  11 #include <linux/module.h>
  12 #include <linux/errno.h>
  13 #include <linux/slab.h>
  14 #include <linux/udp.h>
  15 #include <linux/igmp.h>
  16 #include <linux/if_ether.h>
  17 #include <linux/ethtool.h>
  18 #include <net/arp.h>
  19 #include <net/ndisc.h>
  20 #include <net/ipv6_stubs.h>
  21 #include <net/ip.h>
  22 #include <net/icmp.h>
  23 #include <net/rtnetlink.h>
  24 #include <net/inet_ecn.h>
  25 #include <net/net_namespace.h>
  26 #include <net/netns/generic.h>
  27 #include <net/tun_proto.h>
  28 #include <net/vxlan.h>
  29
  30 #if IS_ENABLED(CONFIG_IPV6)
  31 #include <net/ip6_tunnel.h>
  32 #include <net/ip6_checksum.h>
  33 #endif
  34
  35 #define VXLAN_VERSION   "0.1"
  36
  37 #define PORT_HASH_BITS  8
  38 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  39 #define FDB_AGE_DEFAULT 300 /* 5 min */
  40 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  41
  42 /* UDP port for VXLAN traffic.
  43  * The IANA assigned port is 4789, but the Linux default is 8472
  44  * for compatibility with early adopters.
  45  */
  46 static unsigned short vxlan_port __read_mostly = 8472;
  47 module_param_named(udp_port, vxlan_port, ushort, 0444);
  48 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  49
  50 static bool log_ecn_error = true;
  51 module_param(log_ecn_error, bool, 0644);
  52 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  53
  54 static unsigned int vxlan_net_id;
  55 static struct rtnl_link_ops vxlan_link_ops;
  56
  57 static const u8 all_zeros_mac[ETH_ALEN + 2];
  58
  59 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  60
  61 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan);
  62
  63 /* per-network namespace private data for this module */
  64 struct vxlan_net {
  65         struct list_head  vxlan_list;
  66         struct hlist_head sock_list[PORT_HASH_SIZE];
  67         spinlock_t        sock_lock;
  68 };
  69
  70 /* Forwarding table entry */
  71 struct vxlan_fdb {
  72         struct hlist_node hlist;        /* linked list of entries */
  73         struct rcu_head   rcu;
  74         unsigned long     updated;      /* jiffies */
  75         unsigned long     used;
  76         struct list_head  remotes;
  77         u8                eth_addr[ETH_ALEN];
  78         u16               state;        /* see ndm_state */
  79         __be32            vni;
  80         u16               flags;        /* see ndm_flags and below */
  81 };
  82
  83 #define NTF_VXLAN_ADDED_BY_USER 0x100
  84
  85 /* salt for hash table */
  86 static u32 vxlan_salt __read_mostly;
  87
  88 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  89 {
  90         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  91                ip_tunnel_collect_metadata();
  92 }
  93
  94 #if IS_ENABLED(CONFIG_IPV6)
  95 static inline
  96 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
  97 {
  98         if (a->sa.sa_family != b->sa.sa_family)
  99                 return false;
 100         if (a->sa.sa_family == AF_INET6)
 101                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 102         else
 103                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 104 }
 105
 106 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 107 {
 108         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 109                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 110                 ip->sa.sa_family = AF_INET6;
 111                 return 0;
 112         } else if (nla_len(nla) >= sizeof(__be32)) {
 113                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 114                 ip->sa.sa_family = AF_INET;
 115                 return 0;
 116         } else {
 117                 return -EAFNOSUPPORT;
 118         }
 119 }
 120
 121 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 122                               const union vxlan_addr *ip)
 123 {
 124         if (ip->sa.sa_family == AF_INET6)
 125                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 126         else
 127                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 128 }
 129
 130 #else /* !CONFIG_IPV6 */
 131
 132 static inline
 133 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 134 {
 135         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 136 }
 137
 138 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 139 {
 140         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 141                 return -EAFNOSUPPORT;
 142         } else if (nla_len(nla) >= sizeof(__be32)) {
 143                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 144                 ip->sa.sa_family = AF_INET;
 145                 return 0;
 146         } else {
 147                 return -EAFNOSUPPORT;
 148         }
 149 }
 150
 151 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 152                               const union vxlan_addr *ip)
 153 {
 154         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 155 }
 156 #endif
 157
 158 /* Virtual Network hash table head */
 159 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 160 {
 161         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 162 }
 163
 164 /* Socket hash table head */
 165 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 166 {
 167         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 168
 169         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 170 }
 171
 172 /* First remote destination for a forwarding entry.
 173  * Guaranteed to be non-NULL because remotes are never deleted.
 174  */
 175 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 176 {
 177         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 178 }
 179
 180 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 181 {
 182         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 183 }
 184
 185 /* Find VXLAN socket based on network namespace, address family and UDP port
 186  * and enabled unshareable flags.
 187  */
 188 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 189                                           __be16 port, u32 flags, int ifindex)
 190 {
 191         struct vxlan_sock *vs;
 192
 193         flags &= VXLAN_F_RCV_FLAGS;
 194
 195         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 196                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 197                     vxlan_get_sk_family(vs) == family &&
 198                     vs->flags == flags &&
 199                     vs->sock->sk->sk_bound_dev_if == ifindex)
 200                         return vs;
 201         }
 202         return NULL;
 203 }
 204
 205 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, int ifindex,
 206                                            __be32 vni)
 207 {
 208         struct vxlan_dev_node *node;
 209
 210         /* For flow based devices, map all packets to VNI 0 */
 211         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 212                 vni = 0;
 213
 214         hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) {
 215                 if (node->vxlan->default_dst.remote_vni != vni)
 216                         continue;
 217
 218                 if (IS_ENABLED(CONFIG_IPV6)) {
 219                         const struct vxlan_config *cfg = &node->vxlan->cfg;
 220
 221                         if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) &&
 222                             cfg->remote_ifindex != ifindex)
 223                                 continue;
 224                 }
 225
 226                 return node->vxlan;
 227         }
 228
 229         return NULL;
 230 }
 231
 232 /* Look up VNI in a per net namespace table */
 233 static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex,
 234                                         __be32 vni, sa_family_t family,
 235                                         __be16 port, u32 flags)
 236 {
 237         struct vxlan_sock *vs;
 238
 239         vs = vxlan_find_sock(net, family, port, flags, ifindex);
 240         if (!vs)
 241                 return NULL;
 242
 243         return vxlan_vs_find_vni(vs, ifindex, vni);
 244 }
 245
 246 /* Fill in neighbour message in skbuff. */
 247 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 248                           const struct vxlan_fdb *fdb,
 249                           u32 portid, u32 seq, int type, unsigned int flags,
 250                           const struct vxlan_rdst *rdst)
 251 {
 252         unsigned long now = jiffies;
 253         struct nda_cacheinfo ci;
 254         struct nlmsghdr *nlh;
 255         struct ndmsg *ndm;
 256         bool send_ip, send_eth;
 257
 258         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 259         if (nlh == NULL)
 260                 return -EMSGSIZE;
 261
 262         ndm = nlmsg_data(nlh);
 263         memset(ndm, 0, sizeof(*ndm));
 264
 265         send_eth = send_ip = true;
 266
 267         if (type == RTM_GETNEIGH) {
 268                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 269                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 270                 ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
 271         } else
 272                 ndm->ndm_family = AF_BRIDGE;
 273         ndm->ndm_state = fdb->state;
 274         ndm->ndm_ifindex = vxlan->dev->ifindex;
 275         ndm->ndm_flags = fdb->flags;
 276         if (rdst->offloaded)
 277                 ndm->ndm_flags |= NTF_OFFLOADED;
 278         ndm->ndm_type = RTN_UNICAST;
 279
 280         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 281             nla_put_s32(skb, NDA_LINK_NETNSID,
 282                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 283                 goto nla_put_failure;
 284
 285         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 286                 goto nla_put_failure;
 287
 288         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 289                 goto nla_put_failure;
 290
 291         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 292             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 293                 goto nla_put_failure;
 294         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 295             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 296                 goto nla_put_failure;
 297         if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
 298             nla_put_u32(skb, NDA_SRC_VNI,
 299                         be32_to_cpu(fdb->vni)))
 300                 goto nla_put_failure;
 301         if (rdst->remote_ifindex &&
 302             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 303                 goto nla_put_failure;
 304
 305         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 306         ci.ndm_confirmed = 0;
 307         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 308         ci.ndm_refcnt    = 0;
 309
 310         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 311                 goto nla_put_failure;
 312
 313         nlmsg_end(skb, nlh);
 314         return 0;
 315
 316 nla_put_failure:
 317         nlmsg_cancel(skb, nlh);
 318         return -EMSGSIZE;
 319 }
 320
 321 static inline size_t vxlan_nlmsg_size(void)
 322 {
 323         return NLMSG_ALIGN(sizeof(struct ndmsg))
 324                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 325                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 326                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 327                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 328                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 329                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 330                 + nla_total_size(sizeof(struct nda_cacheinfo));
 331 }
 332
 333 static void __vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 334                                struct vxlan_rdst *rd, int type)
 335 {
 336         struct net *net = dev_net(vxlan->dev);
 337         struct sk_buff *skb;
 338         int err = -ENOBUFS;
 339
 340         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 341         if (skb == NULL)
 342                 goto errout;
 343
 344         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 345         if (err < 0) {
 346                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 347                 WARN_ON(err == -EMSGSIZE);
 348                 kfree_skb(skb);
 349                 goto errout;
 350         }
 351
 352         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 353         return;
 354 errout:
 355         if (err < 0)
 356                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 357 }
 358
 359 static void vxlan_fdb_switchdev_notifier_info(const struct vxlan_dev *vxlan,
 360                             const struct vxlan_fdb *fdb,
 361                             const struct vxlan_rdst *rd,
 362                             struct netlink_ext_ack *extack,
 363                             struct switchdev_notifier_vxlan_fdb_info *fdb_info)
 364 {
 365         fdb_info->info.dev = vxlan->dev;
 366         fdb_info->info.extack = extack;
 367         fdb_info->remote_ip = rd->remote_ip;
 368         fdb_info->remote_port = rd->remote_port;
 369         fdb_info->remote_vni = rd->remote_vni;
 370         fdb_info->remote_ifindex = rd->remote_ifindex;
 371         memcpy(fdb_info->eth_addr, fdb->eth_addr, ETH_ALEN);
 372         fdb_info->vni = fdb->vni;
 373         fdb_info->offloaded = rd->offloaded;
 374         fdb_info->added_by_user = fdb->flags & NTF_VXLAN_ADDED_BY_USER;
 375 }
 376
 377 static int vxlan_fdb_switchdev_call_notifiers(struct vxlan_dev *vxlan,
 378                                               struct vxlan_fdb *fdb,
 379                                               struct vxlan_rdst *rd,
 380                                               bool adding,
 381                                               struct netlink_ext_ack *extack)
 382 {
 383         struct switchdev_notifier_vxlan_fdb_info info;
 384         enum switchdev_notifier_type notifier_type;
 385         int ret;
 386
 387         if (WARN_ON(!rd))
 388                 return 0;
 389
 390         notifier_type = adding ? SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE
 391                                : SWITCHDEV_VXLAN_FDB_DEL_TO_DEVICE;
 392         vxlan_fdb_switchdev_notifier_info(vxlan, fdb, rd, NULL, &info);
 393         ret = call_switchdev_notifiers(notifier_type, vxlan->dev,
 394                                        &info.info, extack);
 395         return notifier_to_errno(ret);
 396 }
 397
 398 static int vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 399                             struct vxlan_rdst *rd, int type, bool swdev_notify,
 400                             struct netlink_ext_ack *extack)
 401 {
 402         int err;
 403
 404         if (swdev_notify) {
 405                 switch (type) {
 406                 case RTM_NEWNEIGH:
 407                         err = vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
 408                                                                  true, extack);
 409                         if (err)
 410                                 return err;
 411                         break;
 412                 case RTM_DELNEIGH:
 413                         vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
 414                                                            false, extack);
 415                         break;
 416                 }
 417         }
 418
 419         __vxlan_fdb_notify(vxlan, fdb, rd, type);
 420         return 0;
 421 }
 422
 423 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 424 {
 425         struct vxlan_dev *vxlan = netdev_priv(dev);
 426         struct vxlan_fdb f = {
 427                 .state = NUD_STALE,
 428         };
 429         struct vxlan_rdst remote = {
 430                 .remote_ip = *ipa, /* goes to NDA_DST */
 431                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 432         };
 433
 434         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
 435 }
 436
 437 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 438 {
 439         struct vxlan_fdb f = {
 440                 .state = NUD_STALE,
 441         };
 442         struct vxlan_rdst remote = { };
 443
 444         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 445
 446         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
 447 }
 448
 449 /* Hash Ethernet address */
 450 static u32 eth_hash(const unsigned char *addr)
 451 {
 452         u64 value = get_unaligned((u64 *)addr);
 453
 454         /* only want 6 bytes */
 455 #ifdef __BIG_ENDIAN
 456         value >>= 16;
 457 #else
 458         value <<= 16;
 459 #endif
 460         return hash_64(value, FDB_HASH_BITS);
 461 }
 462
 463 static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
 464 {
 465         /* use 1 byte of OUI and 3 bytes of NIC */
 466         u32 key = get_unaligned((u32 *)(addr + 2));
 467
 468         return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
 469 }
 470
 471 static u32 fdb_head_index(struct vxlan_dev *vxlan, const u8 *mac, __be32 vni)
 472 {
 473         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)
 474                 return eth_vni_hash(mac, vni);
 475         else
 476                 return eth_hash(mac);
 477 }
 478
 479 /* Hash chain to use given mac address */
 480 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 481                                                 const u8 *mac, __be32 vni)
 482 {
 483         return &vxlan->fdb_head[fdb_head_index(vxlan, mac, vni)];
 484 }
 485
 486 /* Look up Ethernet address in forwarding table */
 487 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 488                                           const u8 *mac, __be32 vni)
 489 {
 490         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
 491         struct vxlan_fdb *f;
 492
 493         hlist_for_each_entry_rcu(f, head, hlist) {
 494                 if (ether_addr_equal(mac, f->eth_addr)) {
 495                         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
 496                                 if (vni == f->vni)
 497                                         return f;
 498                         } else {
 499                                 return f;
 500                         }
 501                 }
 502         }
 503
 504         return NULL;
 505 }
 506
 507 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 508                                         const u8 *mac, __be32 vni)
 509 {
 510         struct vxlan_fdb *f;
 511
 512         f = __vxlan_find_mac(vxlan, mac, vni);
 513         if (f && f->used != jiffies)
 514                 f->used = jiffies;
 515
 516         return f;
 517 }
 518
 519 /* caller should hold vxlan->hash_lock */
 520 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 521                                               union vxlan_addr *ip, __be16 port,
 522                                               __be32 vni, __u32 ifindex)
 523 {
 524         struct vxlan_rdst *rd;
 525
 526         list_for_each_entry(rd, &f->remotes, list) {
 527                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 528                     rd->remote_port == port &&
 529                     rd->remote_vni == vni &&
 530                     rd->remote_ifindex == ifindex)
 531                         return rd;
 532         }
 533
 534         return NULL;
 535 }
 536
 537 int vxlan_fdb_find_uc(struct net_device *dev, const u8 *mac, __be32 vni,
 538                       struct switchdev_notifier_vxlan_fdb_info *fdb_info)
 539 {
 540         struct vxlan_dev *vxlan = netdev_priv(dev);
 541         u8 eth_addr[ETH_ALEN + 2] = { 0 };
 542         struct vxlan_rdst *rdst;
 543         struct vxlan_fdb *f;
 544         int rc = 0;
 545
 546         if (is_multicast_ether_addr(mac) ||
 547             is_zero_ether_addr(mac))
 548                 return -EINVAL;
 549
 550         ether_addr_copy(eth_addr, mac);
 551
 552         rcu_read_lock();
 553
 554         f = __vxlan_find_mac(vxlan, eth_addr, vni);
 555         if (!f) {
 556                 rc = -ENOENT;
 557                 goto out;
 558         }
 559
 560         rdst = first_remote_rcu(f);
 561         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, NULL, fdb_info);
 562
 563 out:
 564         rcu_read_unlock();
 565         return rc;
 566 }
 567 EXPORT_SYMBOL_GPL(vxlan_fdb_find_uc);
 568
 569 static int vxlan_fdb_notify_one(struct notifier_block *nb,
 570                                 const struct vxlan_dev *vxlan,
 571                                 const struct vxlan_fdb *f,
 572                                 const struct vxlan_rdst *rdst,
 573                                 struct netlink_ext_ack *extack)
 574 {
 575         struct switchdev_notifier_vxlan_fdb_info fdb_info;
 576         int rc;
 577
 578         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, extack, &fdb_info);
 579         rc = nb->notifier_call(nb, SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE,
 580                                &fdb_info);
 581         return notifier_to_errno(rc);
 582 }
 583
 584 int vxlan_fdb_replay(const struct net_device *dev, __be32 vni,
 585                      struct notifier_block *nb,
 586                      struct netlink_ext_ack *extack)
 587 {
 588         struct vxlan_dev *vxlan;
 589         struct vxlan_rdst *rdst;
 590         struct vxlan_fdb *f;
 591         unsigned int h;
 592         int rc = 0;
 593
 594         if (!netif_is_vxlan(dev))
 595                 return -EINVAL;
 596         vxlan = netdev_priv(dev);
 597
 598         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 599                 spin_lock_bh(&vxlan->hash_lock[h]);
 600                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) {
 601                         if (f->vni == vni) {
 602                                 list_for_each_entry(rdst, &f->remotes, list) {
 603                                         rc = vxlan_fdb_notify_one(nb, vxlan,
 604                                                                   f, rdst,
 605                                                                   extack);
 606                                         if (rc)
 607                                                 goto unlock;
 608                                 }
 609                         }
 610                 }
 611                 spin_unlock_bh(&vxlan->hash_lock[h]);
 612         }
 613         return 0;
 614
 615 unlock:
 616         spin_unlock_bh(&vxlan->hash_lock[h]);
 617         return rc;
 618 }
 619 EXPORT_SYMBOL_GPL(vxlan_fdb_replay);
 620
 621 void vxlan_fdb_clear_offload(const struct net_device *dev, __be32 vni)
 622 {
 623         struct vxlan_dev *vxlan;
 624         struct vxlan_rdst *rdst;
 625         struct vxlan_fdb *f;
 626         unsigned int h;
 627
 628         if (!netif_is_vxlan(dev))
 629                 return;
 630         vxlan = netdev_priv(dev);
 631
 632         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 633                 spin_lock_bh(&vxlan->hash_lock[h]);
 634                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist)
 635                         if (f->vni == vni)
 636                                 list_for_each_entry(rdst, &f->remotes, list)
 637                                         rdst->offloaded = false;
 638                 spin_unlock_bh(&vxlan->hash_lock[h]);
 639         }
 640
 641 }
 642 EXPORT_SYMBOL_GPL(vxlan_fdb_clear_offload);
 643
 644 /* Replace destination of unicast mac */
 645 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 646                              union vxlan_addr *ip, __be16 port, __be32 vni,
 647                              __u32 ifindex, struct vxlan_rdst *oldrd)
 648 {
 649         struct vxlan_rdst *rd;
 650
 651         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 652         if (rd)
 653                 return 0;
 654
 655         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 656         if (!rd)
 657                 return 0;
 658
 659         *oldrd = *rd;
 660         dst_cache_reset(&rd->dst_cache);
 661         rd->remote_ip = *ip;
 662         rd->remote_port = port;
 663         rd->remote_vni = vni;
 664         rd->remote_ifindex = ifindex;
 665         rd->offloaded = false;
 666         return 1;
 667 }
 668
 669 /* Add/update destinations for multicast */
 670 static int vxlan_fdb_append(struct vxlan_fdb *f,
 671                             union vxlan_addr *ip, __be16 port, __be32 vni,
 672                             __u32 ifindex, struct vxlan_rdst **rdp)
 673 {
 674         struct vxlan_rdst *rd;
 675
 676         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 677         if (rd)
 678                 return 0;
 679
 680         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 681         if (rd == NULL)
 682                 return -ENOBUFS;
 683
 684         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 685                 kfree(rd);
 686                 return -ENOBUFS;
 687         }
 688
 689         rd->remote_ip = *ip;
 690         rd->remote_port = port;
 691         rd->offloaded = false;
 692         rd->remote_vni = vni;
 693         rd->remote_ifindex = ifindex;
 694
 695         list_add_tail_rcu(&rd->list, &f->remotes);
 696
 697         *rdp = rd;
 698         return 1;
 699 }
 700
 701 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 702                                           unsigned int off,
 703                                           struct vxlanhdr *vh, size_t hdrlen,
 704                                           __be32 vni_field,
 705                                           struct gro_remcsum *grc,
 706                                           bool nopartial)
 707 {
 708         size_t start, offset;
 709
 710         if (skb->remcsum_offload)
 711                 return vh;
 712
 713         if (!NAPI_GRO_CB(skb)->csum_valid)
 714                 return NULL;
 715
 716         start = vxlan_rco_start(vni_field);
 717         offset = start + vxlan_rco_offset(vni_field);
 718
 719         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 720                                      start, offset, grc, nopartial);
 721
 722         skb->remcsum_offload = 1;
 723
 724         return vh;
 725 }
 726
 727 static struct sk_buff *vxlan_gro_receive(struct sock *sk,
 728                                          struct list_head *head,
 729                                          struct sk_buff *skb)
 730 {
 731         struct sk_buff *pp = NULL;
 732         struct sk_buff *p;
 733         struct vxlanhdr *vh, *vh2;
 734         unsigned int hlen, off_vx;
 735         int flush = 1;
 736         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 737         __be32 flags;
 738         struct gro_remcsum grc;
 739
 740         skb_gro_remcsum_init(&grc);
 741
 742         off_vx = skb_gro_offset(skb);
 743         hlen = off_vx + sizeof(*vh);
 744         vh   = skb_gro_header_fast(skb, off_vx);
 745         if (skb_gro_header_hard(skb, hlen)) {
 746                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 747                 if (unlikely(!vh))
 748                         goto out;
 749         }
 750
 751         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 752
 753         flags = vh->vx_flags;
 754
 755         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 756                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 757                                        vh->vx_vni, &grc,
 758                                        !!(vs->flags &
 759                                           VXLAN_F_REMCSUM_NOPARTIAL));
 760
 761                 if (!vh)
 762                         goto out;
 763         }
 764
 765         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 766
 767         list_for_each_entry(p, head, list) {
 768                 if (!NAPI_GRO_CB(p)->same_flow)
 769                         continue;
 770
 771                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 772                 if (vh->vx_flags != vh2->vx_flags ||
 773                     vh->vx_vni != vh2->vx_vni) {
 774                         NAPI_GRO_CB(p)->same_flow = 0;
 775                         continue;
 776                 }
 777         }
 778
 779         pp = call_gro_receive(eth_gro_receive, head, skb);
 780         flush = 0;
 781
 782 out:
 783         skb_gro_flush_final_remcsum(skb, pp, flush, &grc);
 784
 785         return pp;
 786 }
 787
 788 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 789 {
 790         /* Sets 'skb->inner_mac_header' since we are always called with
 791          * 'skb->encapsulation' set.
 792          */
 793         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 794 }
 795
 796 static struct vxlan_fdb *vxlan_fdb_alloc(const u8 *mac, __u16 state,
 797                                          __be32 src_vni, __u16 ndm_flags)
 798 {
 799         struct vxlan_fdb *f;
 800
 801         f = kmalloc(sizeof(*f), GFP_ATOMIC);
 802         if (!f)
 803                 return NULL;
 804         f->state = state;
 805         f->flags = ndm_flags;
 806         f->updated = f->used = jiffies;
 807         f->vni = src_vni;
 808         INIT_LIST_HEAD(&f->remotes);
 809         memcpy(f->eth_addr, mac, ETH_ALEN);
 810
 811         return f;
 812 }
 813
 814 static void vxlan_fdb_insert(struct vxlan_dev *vxlan, const u8 *mac,
 815                              __be32 src_vni, struct vxlan_fdb *f)
 816 {
 817         ++vxlan->addrcnt;
 818         hlist_add_head_rcu(&f->hlist,
 819                            vxlan_fdb_head(vxlan, mac, src_vni));
 820 }
 821
 822 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 823                             const u8 *mac, union vxlan_addr *ip,
 824                             __u16 state, __be16 port, __be32 src_vni,
 825                             __be32 vni, __u32 ifindex, __u16 ndm_flags,
 826                             struct vxlan_fdb **fdb)
 827 {
 828         struct vxlan_rdst *rd = NULL;
 829         struct vxlan_fdb *f;
 830         int rc;
 831
 832         if (vxlan->cfg.addrmax &&
 833             vxlan->addrcnt >= vxlan->cfg.addrmax)
 834                 return -ENOSPC;
 835
 836         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 837         f = vxlan_fdb_alloc(mac, state, src_vni, ndm_flags);
 838         if (!f)
 839                 return -ENOMEM;
 840
 841         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 842         if (rc < 0) {
 843                 kfree(f);
 844                 return rc;
 845         }
 846
 847         *fdb = f;
 848
 849         return 0;
 850 }
 851
 852 static void __vxlan_fdb_free(struct vxlan_fdb *f)
 853 {
 854         struct vxlan_rdst *rd, *nd;
 855
 856         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 857                 dst_cache_destroy(&rd->dst_cache);
 858                 kfree(rd);
 859         }
 860         kfree(f);
 861 }
 862
 863 static void vxlan_fdb_free(struct rcu_head *head)
 864 {
 865         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 866
 867         __vxlan_fdb_free(f);
 868 }
 869
 870 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 871                               bool do_notify, bool swdev_notify)
 872 {
 873         struct vxlan_rdst *rd;
 874
 875         netdev_dbg(vxlan->dev, "delete %pM\n", f->eth_addr);
 876
 877         --vxlan->addrcnt;
 878         if (do_notify)
 879                 list_for_each_entry(rd, &f->remotes, list)
 880                         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH,
 881                                          swdev_notify, NULL);
 882
 883         hlist_del_rcu(&f->hlist);
 884         call_rcu(&f->rcu, vxlan_fdb_free);
 885 }
 886
 887 static void vxlan_dst_free(struct rcu_head *head)
 888 {
 889         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 890
 891         dst_cache_destroy(&rd->dst_cache);
 892         kfree(rd);
 893 }
 894
 895 static int vxlan_fdb_update_existing(struct vxlan_dev *vxlan,
 896                                      union vxlan_addr *ip,
 897                                      __u16 state, __u16 flags,
 898                                      __be16 port, __be32 vni,
 899                                      __u32 ifindex, __u16 ndm_flags,
 900                                      struct vxlan_fdb *f,
 901                                      bool swdev_notify,
 902                                      struct netlink_ext_ack *extack)
 903 {
 904         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
 905         struct vxlan_rdst *rd = NULL;
 906         struct vxlan_rdst oldrd;
 907         int notify = 0;
 908         int rc = 0;
 909         int err;
 910
 911         /* Do not allow an externally learned entry to take over an entry added
 912          * by the user.
 913          */
 914         if (!(fdb_flags & NTF_EXT_LEARNED) ||
 915             !(f->flags & NTF_VXLAN_ADDED_BY_USER)) {
 916                 if (f->state != state) {
 917                         f->state = state;
 918                         f->updated = jiffies;
 919                         notify = 1;
 920                 }
 921                 if (f->flags != fdb_flags) {
 922                         f->flags = fdb_flags;
 923                         f->updated = jiffies;
 924                         notify = 1;
 925                 }
 926         }
 927
 928         if ((flags & NLM_F_REPLACE)) {
 929                 /* Only change unicasts */
 930                 if (!(is_multicast_ether_addr(f->eth_addr) ||
 931                       is_zero_ether_addr(f->eth_addr))) {
 932                         rc = vxlan_fdb_replace(f, ip, port, vni,
 933                                                ifindex, &oldrd);
 934                         notify |= rc;
 935                 } else {
 936                         return -EOPNOTSUPP;
 937                 }
 938         }
 939         if ((flags & NLM_F_APPEND) &&
 940             (is_multicast_ether_addr(f->eth_addr) ||
 941              is_zero_ether_addr(f->eth_addr))) {
 942                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 943
 944                 if (rc < 0)
 945                         return rc;
 946                 notify |= rc;
 947         }
 948
 949         if (ndm_flags & NTF_USE)
 950                 f->used = jiffies;
 951
 952         if (notify) {
 953                 if (rd == NULL)
 954                         rd = first_remote_rtnl(f);
 955
 956                 err = vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH,
 957                                        swdev_notify, extack);
 958                 if (err)
 959                         goto err_notify;
 960         }
 961
 962         return 0;
 963
 964 err_notify:
 965         if ((flags & NLM_F_REPLACE) && rc)
 966                 *rd = oldrd;
 967         else if ((flags & NLM_F_APPEND) && rc) {
 968                 list_del_rcu(&rd->list);
 969                 call_rcu(&rd->rcu, vxlan_dst_free);
 970         }
 971         return err;
 972 }
 973
 974 static int vxlan_fdb_update_create(struct vxlan_dev *vxlan,
 975                                    const u8 *mac, union vxlan_addr *ip,
 976                                    __u16 state, __u16 flags,
 977                                    __be16 port, __be32 src_vni, __be32 vni,
 978                                    __u32 ifindex, __u16 ndm_flags,
 979                                    bool swdev_notify,
 980                                    struct netlink_ext_ack *extack)
 981 {
 982         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
 983         struct vxlan_fdb *f;
 984         int rc;
 985
 986         /* Disallow replace to add a multicast entry */
 987         if ((flags & NLM_F_REPLACE) &&
 988             (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 989                 return -EOPNOTSUPP;
 990
 991         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 992         rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni,
 993                               vni, ifindex, fdb_flags, &f);
 994         if (rc < 0)
 995                 return rc;
 996
 997         vxlan_fdb_insert(vxlan, mac, src_vni, f);
 998         rc = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH,
 999                               swdev_notify, extack);
1000         if (rc)
1001                 goto err_notify;
1002
1003         return 0;
1004
1005 err_notify:
1006         vxlan_fdb_destroy(vxlan, f, false, false);
1007         return rc;
1008 }
1009
1010 /* Add new entry to forwarding table -- assumes lock held */
1011 static int vxlan_fdb_update(struct vxlan_dev *vxlan,
1012                             const u8 *mac, union vxlan_addr *ip,
1013                             __u16 state, __u16 flags,
1014                             __be16 port, __be32 src_vni, __be32 vni,
1015                             __u32 ifindex, __u16 ndm_flags,
1016                             bool swdev_notify,
1017                             struct netlink_ext_ack *extack)
1018 {
1019         struct vxlan_fdb *f;
1020
1021         f = __vxlan_find_mac(vxlan, mac, src_vni);
1022         if (f) {
1023                 if (flags & NLM_F_EXCL) {
1024                         netdev_dbg(vxlan->dev,
1025                                    "lost race to create %pM\n", mac);
1026                         return -EEXIST;
1027                 }
1028
1029                 return vxlan_fdb_update_existing(vxlan, ip, state, flags, port,
1030                                                  vni, ifindex, ndm_flags, f,
1031                                                  swdev_notify, extack);
1032         } else {
1033                 if (!(flags & NLM_F_CREATE))
1034                         return -ENOENT;
1035
1036                 return vxlan_fdb_update_create(vxlan, mac, ip, state, flags,
1037                                                port, src_vni, vni, ifindex,
1038                                                ndm_flags, swdev_notify, extack);
1039         }
1040 }
1041
1042 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
1043                                   struct vxlan_rdst *rd, bool swdev_notify)
1044 {
1045         list_del_rcu(&rd->list);
1046         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, swdev_notify, NULL);
1047         call_rcu(&rd->rcu, vxlan_dst_free);
1048 }
1049
1050 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
1051                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
1052                            __be32 *vni, u32 *ifindex)
1053 {
1054         struct net *net = dev_net(vxlan->dev);
1055         int err;
1056
1057         if (tb[NDA_DST]) {
1058                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
1059                 if (err)
1060                         return err;
1061         } else {
1062                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
1063
1064                 if (remote->sa.sa_family == AF_INET) {
1065                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
1066                         ip->sa.sa_family = AF_INET;
1067 #if IS_ENABLED(CONFIG_IPV6)
1068                 } else {
1069                         ip->sin6.sin6_addr = in6addr_any;
1070                         ip->sa.sa_family = AF_INET6;
1071 #endif
1072                 }
1073         }
1074
1075         if (tb[NDA_PORT]) {
1076                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
1077                         return -EINVAL;
1078                 *port = nla_get_be16(tb[NDA_PORT]);
1079         } else {
1080                 *port = vxlan->cfg.dst_port;
1081         }
1082
1083         if (tb[NDA_VNI]) {
1084                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
1085                         return -EINVAL;
1086                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1087         } else {
1088                 *vni = vxlan->default_dst.remote_vni;
1089         }
1090
1091         if (tb[NDA_SRC_VNI]) {
1092                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
1093                         return -EINVAL;
1094                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
1095         } else {
1096                 *src_vni = vxlan->default_dst.remote_vni;
1097         }
1098
1099         if (tb[NDA_IFINDEX]) {
1100                 struct net_device *tdev;
1101
1102                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
1103                         return -EINVAL;
1104                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
1105                 tdev = __dev_get_by_index(net, *ifindex);
1106                 if (!tdev)
1107                         return -EADDRNOTAVAIL;
1108         } else {
1109                 *ifindex = 0;
1110         }
1111
1112         return 0;
1113 }
1114
1115 /* Add static entry (via netlink) */
1116 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
1117                          struct net_device *dev,
1118                          const unsigned char *addr, u16 vid, u16 flags,
1119                          struct netlink_ext_ack *extack)
1120 {
1121         struct vxlan_dev *vxlan = netdev_priv(dev);
1122         /* struct net *net = dev_net(vxlan->dev); */
1123         union vxlan_addr ip;
1124         __be16 port;
1125         __be32 src_vni, vni;
1126         u32 ifindex;
1127         u32 hash_index;
1128         int err;
1129
1130         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
1131                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
1132                         ndm->ndm_state);
1133                 return -EINVAL;
1134         }
1135
1136         if (tb[NDA_DST] == NULL)
1137                 return -EINVAL;
1138
1139         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
1140         if (err)
1141                 return err;
1142
1143         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
1144                 return -EAFNOSUPPORT;
1145
1146         hash_index = fdb_head_index(vxlan, addr, src_vni);
1147         spin_lock_bh(&vxlan->hash_lock[hash_index]);
1148         err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags,
1149                                port, src_vni, vni, ifindex,
1150                                ndm->ndm_flags | NTF_VXLAN_ADDED_BY_USER,
1151                                true, extack);
1152         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1153
1154         return err;
1155 }
1156
1157 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
1158                               const unsigned char *addr, union vxlan_addr ip,
1159                               __be16 port, __be32 src_vni, __be32 vni,
1160                               u32 ifindex, bool swdev_notify)
1161 {
1162         struct vxlan_fdb *f;
1163         struct vxlan_rdst *rd = NULL;
1164         int err = -ENOENT;
1165
1166         f = vxlan_find_mac(vxlan, addr, src_vni);
1167         if (!f)
1168                 return err;
1169
1170         if (!vxlan_addr_any(&ip)) {
1171                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
1172                 if (!rd)
1173                         goto out;
1174         }
1175
1176         /* remove a destination if it's not the only one on the list,
1177          * otherwise destroy the fdb entry
1178          */
1179         if (rd && !list_is_singular(&f->remotes)) {
1180                 vxlan_fdb_dst_destroy(vxlan, f, rd, swdev_notify);
1181                 goto out;
1182         }
1183
1184         vxlan_fdb_destroy(vxlan, f, true, swdev_notify);
1185
1186 out:
1187         return 0;
1188 }
1189
1190 /* Delete entry (via netlink) */
1191 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
1192                             struct net_device *dev,
1193                             const unsigned char *addr, u16 vid)
1194 {
1195         struct vxlan_dev *vxlan = netdev_priv(dev);
1196         union vxlan_addr ip;
1197         __be32 src_vni, vni;
1198         __be16 port;
1199         u32 ifindex;
1200         u32 hash_index;
1201         int err;
1202
1203         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
1204         if (err)
1205                 return err;
1206
1207         hash_index = fdb_head_index(vxlan, addr, src_vni);
1208         spin_lock_bh(&vxlan->hash_lock[hash_index]);
1209         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
1210                                  true);
1211         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1212
1213         return err;
1214 }
1215
1216 /* Dump forwarding table */
1217 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1218                           struct net_device *dev,
1219                           struct net_device *filter_dev, int *idx)
1220 {
1221         struct vxlan_dev *vxlan = netdev_priv(dev);
1222         unsigned int h;
1223         int err = 0;
1224
1225         for (h = 0; h < FDB_HASH_SIZE; ++h) {
1226                 struct vxlan_fdb *f;
1227
1228                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
1229                         struct vxlan_rdst *rd;
1230
1231                         list_for_each_entry_rcu(rd, &f->remotes, list) {
1232                                 if (*idx < cb->args[2])
1233                                         goto skip;
1234
1235                                 err = vxlan_fdb_info(skb, vxlan, f,
1236                                                      NETLINK_CB(cb->skb).portid,
1237                                                      cb->nlh->nlmsg_seq,
1238                                                      RTM_NEWNEIGH,
1239                                                      NLM_F_MULTI, rd);
1240                                 if (err < 0)
1241                                         goto out;
1242 skip:
1243                                 *idx += 1;
1244                         }
1245                 }
1246         }
1247 out:
1248         return err;
1249 }
1250
1251 static int vxlan_fdb_get(struct sk_buff *skb,
1252                          struct nlattr *tb[],
1253                          struct net_device *dev,
1254                          const unsigned char *addr,
1255                          u16 vid, u32 portid, u32 seq,
1256                          struct netlink_ext_ack *extack)
1257 {
1258         struct vxlan_dev *vxlan = netdev_priv(dev);
1259         struct vxlan_fdb *f;
1260         __be32 vni;
1261         int err;
1262
1263         if (tb[NDA_VNI])
1264                 vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1265         else
1266                 vni = vxlan->default_dst.remote_vni;
1267
1268         rcu_read_lock();
1269
1270         f = __vxlan_find_mac(vxlan, addr, vni);
1271         if (!f) {
1272                 NL_SET_ERR_MSG(extack, "Fdb entry not found");
1273                 err = -ENOENT;
1274                 goto errout;
1275         }
1276
1277         err = vxlan_fdb_info(skb, vxlan, f, portid, seq,
1278                              RTM_NEWNEIGH, 0, first_remote_rcu(f));
1279 errout:
1280         rcu_read_unlock();
1281         return err;
1282 }
1283
1284 /* Watch incoming packets to learn mapping between Ethernet address
1285  * and Tunnel endpoint.
1286  * Return true if packet is bogus and should be dropped.
1287  */
1288 static bool vxlan_snoop(struct net_device *dev,
1289                         union vxlan_addr *src_ip, const u8 *src_mac,
1290                         u32 src_ifindex, __be32 vni)
1291 {
1292         struct vxlan_dev *vxlan = netdev_priv(dev);
1293         struct vxlan_fdb *f;
1294         u32 ifindex = 0;
1295
1296 #if IS_ENABLED(CONFIG_IPV6)
1297         if (src_ip->sa.sa_family == AF_INET6 &&
1298             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
1299                 ifindex = src_ifindex;
1300 #endif
1301
1302         f = vxlan_find_mac(vxlan, src_mac, vni);
1303         if (likely(f)) {
1304                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1305
1306                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
1307                            rdst->remote_ifindex == ifindex))
1308                         return false;
1309
1310                 /* Don't migrate static entries, drop packets */
1311                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
1312                         return true;
1313
1314                 if (net_ratelimit())
1315                         netdev_info(dev,
1316                                     "%pM migrated from %pIS to %pIS\n",
1317                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1318
1319                 rdst->remote_ip = *src_ip;
1320                 f->updated = jiffies;
1321                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH, true, NULL);
1322         } else {
1323                 u32 hash_index = fdb_head_index(vxlan, src_mac, vni);
1324
1325                 /* learned new entry */
1326                 spin_lock(&vxlan->hash_lock[hash_index]);
1327
1328                 /* close off race between vxlan_flush and incoming packets */
1329                 if (netif_running(dev))
1330                         vxlan_fdb_update(vxlan, src_mac, src_ip,
1331                                          NUD_REACHABLE,
1332                                          NLM_F_EXCL|NLM_F_CREATE,
1333                                          vxlan->cfg.dst_port,
1334                                          vni,
1335                                          vxlan->default_dst.remote_vni,
1336                                          ifindex, NTF_SELF, true, NULL);
1337                 spin_unlock(&vxlan->hash_lock[hash_index]);
1338         }
1339
1340         return false;
1341 }
1342
1343 /* See if multicast group is already in use by other ID */
1344 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1345 {
1346         struct vxlan_dev *vxlan;
1347         struct vxlan_sock *sock4;
1348 #if IS_ENABLED(CONFIG_IPV6)
1349         struct vxlan_sock *sock6;
1350 #endif
1351         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1352
1353         sock4 = rtnl_dereference(dev->vn4_sock);
1354
1355         /* The vxlan_sock is only used by dev, leaving group has
1356          * no effect on other vxlan devices.
1357          */
1358         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1359                 return false;
1360 #if IS_ENABLED(CONFIG_IPV6)
1361         sock6 = rtnl_dereference(dev->vn6_sock);
1362         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1363                 return false;
1364 #endif
1365
1366         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1367                 if (!netif_running(vxlan->dev) || vxlan == dev)
1368                         continue;
1369
1370                 if (family == AF_INET &&
1371                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1372                         continue;
1373 #if IS_ENABLED(CONFIG_IPV6)
1374                 if (family == AF_INET6 &&
1375                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1376                         continue;
1377 #endif
1378
1379                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1380                                       &dev->default_dst.remote_ip))
1381                         continue;
1382
1383                 if (vxlan->default_dst.remote_ifindex !=
1384                     dev->default_dst.remote_ifindex)
1385                         continue;
1386
1387                 return true;
1388         }
1389
1390         return false;
1391 }
1392
1393 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1394 {
1395         struct vxlan_net *vn;
1396
1397         if (!vs)
1398                 return false;
1399         if (!refcount_dec_and_test(&vs->refcnt))
1400                 return false;
1401
1402         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1403         spin_lock(&vn->sock_lock);
1404         hlist_del_rcu(&vs->hlist);
1405         udp_tunnel_notify_del_rx_port(vs->sock,
1406                                       (vs->flags & VXLAN_F_GPE) ?
1407                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1408                                       UDP_TUNNEL_TYPE_VXLAN);
1409         spin_unlock(&vn->sock_lock);
1410
1411         return true;
1412 }
1413
1414 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1415 {
1416         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1417 #if IS_ENABLED(CONFIG_IPV6)
1418         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1419
1420         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1421 #endif
1422
1423         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1424         synchronize_net();
1425
1426         vxlan_vs_del_dev(vxlan);
1427
1428         if (__vxlan_sock_release_prep(sock4)) {
1429                 udp_tunnel_sock_release(sock4->sock);
1430                 kfree(sock4);
1431         }
1432
1433 #if IS_ENABLED(CONFIG_IPV6)
1434         if (__vxlan_sock_release_prep(sock6)) {
1435                 udp_tunnel_sock_release(sock6->sock);
1436                 kfree(sock6);
1437         }
1438 #endif
1439 }
1440
1441 /* Update multicast group membership when first VNI on
1442  * multicast address is brought up
1443  */
1444 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1445 {
1446         struct sock *sk;
1447         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1448         int ifindex = vxlan->default_dst.remote_ifindex;
1449         int ret = -EINVAL;
1450
1451         if (ip->sa.sa_family == AF_INET) {
1452                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1453                 struct ip_mreqn mreq = {
1454                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1455                         .imr_ifindex            = ifindex,
1456                 };
1457
1458                 sk = sock4->sock->sk;
1459                 lock_sock(sk);
1460                 ret = ip_mc_join_group(sk, &mreq);
1461                 release_sock(sk);
1462 #if IS_ENABLED(CONFIG_IPV6)
1463         } else {
1464                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1465
1466                 sk = sock6->sock->sk;
1467                 lock_sock(sk);
1468                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1469                                                    &ip->sin6.sin6_addr);
1470                 release_sock(sk);
1471 #endif
1472         }
1473
1474         return ret;
1475 }
1476
1477 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1478 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1479 {
1480         struct sock *sk;
1481         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1482         int ifindex = vxlan->default_dst.remote_ifindex;
1483         int ret = -EINVAL;
1484
1485         if (ip->sa.sa_family == AF_INET) {
1486                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1487                 struct ip_mreqn mreq = {
1488                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1489                         .imr_ifindex            = ifindex,
1490                 };
1491
1492                 sk = sock4->sock->sk;
1493                 lock_sock(sk);
1494                 ret = ip_mc_leave_group(sk, &mreq);
1495                 release_sock(sk);
1496 #if IS_ENABLED(CONFIG_IPV6)
1497         } else {
1498                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1499
1500                 sk = sock6->sock->sk;
1501                 lock_sock(sk);
1502                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1503                                                    &ip->sin6.sin6_addr);
1504                 release_sock(sk);
1505 #endif
1506         }
1507
1508         return ret;
1509 }
1510
1511 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1512                           struct sk_buff *skb, u32 vxflags)
1513 {
1514         size_t start, offset;
1515
1516         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1517                 goto out;
1518
1519         start = vxlan_rco_start(unparsed->vx_vni);
1520         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1521
1522         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1523                 return false;
1524
1525         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1526                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1527 out:
1528         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1529         unparsed->vx_vni &= VXLAN_VNI_MASK;
1530         return true;
1531 }
1532
1533 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1534                                 struct sk_buff *skb, u32 vxflags,
1535                                 struct vxlan_metadata *md)
1536 {
1537         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1538         struct metadata_dst *tun_dst;
1539
1540         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1541                 goto out;
1542
1543         md->gbp = ntohs(gbp->policy_id);
1544
1545         tun_dst = (struct metadata_dst *)skb_dst(skb);
1546         if (tun_dst) {
1547                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1548                 tun_dst->u.tun_info.options_len = sizeof(*md);
1549         }
1550         if (gbp->dont_learn)
1551                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1552
1553         if (gbp->policy_applied)
1554                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1555
1556         /* In flow-based mode, GBP is carried in dst_metadata */
1557         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1558                 skb->mark = md->gbp;
1559 out:
1560         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1561 }
1562
1563 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1564                                 __be16 *protocol,
1565                                 struct sk_buff *skb, u32 vxflags)
1566 {
1567         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1568
1569         /* Need to have Next Protocol set for interfaces in GPE mode. */
1570         if (!gpe->np_applied)
1571                 return false;
1572         /* "The initial version is 0. If a receiver does not support the
1573          * version indicated it MUST drop the packet.
1574          */
1575         if (gpe->version != 0)
1576                 return false;
1577         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1578          * processing MUST occur." However, we don't implement OAM
1579          * processing, thus drop the packet.
1580          */
1581         if (gpe->oam_flag)
1582                 return false;
1583
1584         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1585         if (!*protocol)
1586                 return false;
1587
1588         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1589         return true;
1590 }
1591
1592 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1593                           struct vxlan_sock *vs,
1594                           struct sk_buff *skb, __be32 vni)
1595 {
1596         union vxlan_addr saddr;
1597         u32 ifindex = skb->dev->ifindex;
1598
1599         skb_reset_mac_header(skb);
1600         skb->protocol = eth_type_trans(skb, vxlan->dev);
1601         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1602
1603         /* Ignore packet loops (and multicast echo) */
1604         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1605                 return false;
1606
1607         /* Get address from the outer IP header */
1608         if (vxlan_get_sk_family(vs) == AF_INET) {
1609                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1610                 saddr.sa.sa_family = AF_INET;
1611 #if IS_ENABLED(CONFIG_IPV6)
1612         } else {
1613                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1614                 saddr.sa.sa_family = AF_INET6;
1615 #endif
1616         }
1617
1618         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1619             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1620                 return false;
1621
1622         return true;
1623 }
1624
1625 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1626                                   struct sk_buff *skb)
1627 {
1628         int err = 0;
1629
1630         if (vxlan_get_sk_family(vs) == AF_INET)
1631                 err = IP_ECN_decapsulate(oiph, skb);
1632 #if IS_ENABLED(CONFIG_IPV6)
1633         else
1634                 err = IP6_ECN_decapsulate(oiph, skb);
1635 #endif
1636
1637         if (unlikely(err) && log_ecn_error) {
1638                 if (vxlan_get_sk_family(vs) == AF_INET)
1639                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1640                                              &((struct iphdr *)oiph)->saddr,
1641                                              ((struct iphdr *)oiph)->tos);
1642                 else
1643                         net_info_ratelimited("non-ECT from %pI6\n",
1644                                              &((struct ipv6hdr *)oiph)->saddr);
1645         }
1646         return err <= 1;
1647 }
1648
1649 /* Callback from net/ipv4/udp.c to receive packets */
1650 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1651 {
1652         struct pcpu_sw_netstats *stats;
1653         struct vxlan_dev *vxlan;
1654         struct vxlan_sock *vs;
1655         struct vxlanhdr unparsed;
1656         struct vxlan_metadata _md;
1657         struct vxlan_metadata *md = &_md;
1658         __be16 protocol = htons(ETH_P_TEB);
1659         bool raw_proto = false;
1660         void *oiph;
1661         __be32 vni = 0;
1662
1663         /* Need UDP and VXLAN header to be present */
1664         if (!pskb_may_pull(skb, VXLAN_HLEN))
1665                 goto drop;
1666
1667         unparsed = *vxlan_hdr(skb);
1668         /* VNI flag always required to be set */
1669         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1670                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1671                            ntohl(vxlan_hdr(skb)->vx_flags),
1672                            ntohl(vxlan_hdr(skb)->vx_vni));
1673                 /* Return non vxlan pkt */
1674                 goto drop;
1675         }
1676         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1677         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1678
1679         vs = rcu_dereference_sk_user_data(sk);
1680         if (!vs)
1681                 goto drop;
1682
1683         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1684
1685         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1686         if (!vxlan)
1687                 goto drop;
1688
1689         /* For backwards compatibility, only allow reserved fields to be
1690          * used by VXLAN extensions if explicitly requested.
1691          */
1692         if (vs->flags & VXLAN_F_GPE) {
1693                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1694                         goto drop;
1695                 raw_proto = true;
1696         }
1697
1698         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1699                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1700                 goto drop;
1701
1702         if (vxlan_collect_metadata(vs)) {
1703                 struct metadata_dst *tun_dst;
1704
1705                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1706                                          key32_to_tunnel_id(vni), sizeof(*md));
1707
1708                 if (!tun_dst)
1709                         goto drop;
1710
1711                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1712
1713                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1714         } else {
1715                 memset(md, 0, sizeof(*md));
1716         }
1717
1718         if (vs->flags & VXLAN_F_REMCSUM_RX)
1719                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1720                         goto drop;
1721         if (vs->flags & VXLAN_F_GBP)
1722                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1723         /* Note that GBP and GPE can never be active together. This is
1724          * ensured in vxlan_dev_configure.
1725          */
1726
1727         if (unparsed.vx_flags || unparsed.vx_vni) {
1728                 /* If there are any unprocessed flags remaining treat
1729                  * this as a malformed packet. This behavior diverges from
1730                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1731                  * in reserved fields are to be ignored. The approach here
1732                  * maintains compatibility with previous stack code, and also
1733                  * is more robust and provides a little more security in
1734                  * adding extensions to VXLAN.
1735                  */
1736                 goto drop;
1737         }
1738
1739         if (!raw_proto) {
1740                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1741                         goto drop;
1742         } else {
1743                 skb_reset_mac_header(skb);
1744                 skb->dev = vxlan->dev;
1745                 skb->pkt_type = PACKET_HOST;
1746         }
1747
1748         oiph = skb_network_header(skb);
1749         skb_reset_network_header(skb);
1750
1751         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1752                 ++vxlan->dev->stats.rx_frame_errors;
1753                 ++vxlan->dev->stats.rx_errors;
1754                 goto drop;
1755         }
1756
1757         rcu_read_lock();
1758
1759         if (unlikely(!(vxlan->dev->flags & IFF_UP))) {
1760                 rcu_read_unlock();
1761                 atomic_long_inc(&vxlan->dev->rx_dropped);
1762                 goto drop;
1763         }
1764
1765         stats = this_cpu_ptr(vxlan->dev->tstats);
1766         u64_stats_update_begin(&stats->syncp);
1767         stats->rx_packets++;
1768         stats->rx_bytes += skb->len;
1769         u64_stats_update_end(&stats->syncp);
1770
1771         gro_cells_receive(&vxlan->gro_cells, skb);
1772
1773         rcu_read_unlock();
1774
1775         return 0;
1776
1777 drop:
1778         /* Consume bad packet */
1779         kfree_skb(skb);
1780         return 0;
1781 }
1782
1783 /* Callback from net/ipv{4,6}/udp.c to check that we have a VNI for errors */
1784 static int vxlan_err_lookup(struct sock *sk, struct sk_buff *skb)
1785 {
1786         struct vxlan_dev *vxlan;
1787         struct vxlan_sock *vs;
1788         struct vxlanhdr *hdr;
1789         __be32 vni;
1790
1791         if (!pskb_may_pull(skb, skb_transport_offset(skb) + VXLAN_HLEN))
1792                 return -EINVAL;
1793
1794         hdr = vxlan_hdr(skb);
1795
1796         if (!(hdr->vx_flags & VXLAN_HF_VNI))
1797                 return -EINVAL;
1798
1799         vs = rcu_dereference_sk_user_data(sk);
1800         if (!vs)
1801                 return -ENOENT;
1802
1803         vni = vxlan_vni(hdr->vx_vni);
1804         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1805         if (!vxlan)
1806                 return -ENOENT;
1807
1808         return 0;
1809 }
1810
1811 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1812 {
1813         struct vxlan_dev *vxlan = netdev_priv(dev);
1814         struct arphdr *parp;
1815         u8 *arpptr, *sha;
1816         __be32 sip, tip;
1817         struct neighbour *n;
1818
1819         if (dev->flags & IFF_NOARP)
1820                 goto out;
1821
1822         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1823                 dev->stats.tx_dropped++;
1824                 goto out;
1825         }
1826         parp = arp_hdr(skb);
1827
1828         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1829              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1830             parp->ar_pro != htons(ETH_P_IP) ||
1831             parp->ar_op != htons(ARPOP_REQUEST) ||
1832             parp->ar_hln != dev->addr_len ||
1833             parp->ar_pln != 4)
1834                 goto out;
1835         arpptr = (u8 *)parp + sizeof(struct arphdr);
1836         sha = arpptr;
1837         arpptr += dev->addr_len;        /* sha */
1838         memcpy(&sip, arpptr, sizeof(sip));
1839         arpptr += sizeof(sip);
1840         arpptr += dev->addr_len;        /* tha */
1841         memcpy(&tip, arpptr, sizeof(tip));
1842
1843         if (ipv4_is_loopback(tip) ||
1844             ipv4_is_multicast(tip))
1845                 goto out;
1846
1847         n = neigh_lookup(&arp_tbl, &tip, dev);
1848
1849         if (n) {
1850                 struct vxlan_fdb *f;
1851                 struct sk_buff  *reply;
1852
1853                 if (!(n->nud_state & NUD_CONNECTED)) {
1854                         neigh_release(n);
1855                         goto out;
1856                 }
1857
1858                 f = vxlan_find_mac(vxlan, n->ha, vni);
1859                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1860                         /* bridge-local neighbor */
1861                         neigh_release(n);
1862                         goto out;
1863                 }
1864
1865                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1866                                 n->ha, sha);
1867
1868                 neigh_release(n);
1869
1870                 if (reply == NULL)
1871                         goto out;
1872
1873                 skb_reset_mac_header(reply);
1874                 __skb_pull(reply, skb_network_offset(reply));
1875                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1876                 reply->pkt_type = PACKET_HOST;
1877
1878                 if (netif_rx_ni(reply) == NET_RX_DROP)
1879                         dev->stats.rx_dropped++;
1880         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1881                 union vxlan_addr ipa = {
1882                         .sin.sin_addr.s_addr = tip,
1883                         .sin.sin_family = AF_INET,
1884                 };
1885
1886                 vxlan_ip_miss(dev, &ipa);
1887         }
1888 out:
1889         consume_skb(skb);
1890         return NETDEV_TX_OK;
1891 }
1892
1893 #if IS_ENABLED(CONFIG_IPV6)
1894 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1895         struct neighbour *n, bool isrouter)
1896 {
1897         struct net_device *dev = request->dev;
1898         struct sk_buff *reply;
1899         struct nd_msg *ns, *na;
1900         struct ipv6hdr *pip6;
1901         u8 *daddr;
1902         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1903         int ns_olen;
1904         int i, len;
1905
1906         if (dev == NULL || !pskb_may_pull(request, request->len))
1907                 return NULL;
1908
1909         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1910                 sizeof(*na) + na_olen + dev->needed_tailroom;
1911         reply = alloc_skb(len, GFP_ATOMIC);
1912         if (reply == NULL)
1913                 return NULL;
1914
1915         reply->protocol = htons(ETH_P_IPV6);
1916         reply->dev = dev;
1917         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1918         skb_push(reply, sizeof(struct ethhdr));
1919         skb_reset_mac_header(reply);
1920
1921         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
1922
1923         daddr = eth_hdr(request)->h_source;
1924         ns_olen = request->len - skb_network_offset(request) -
1925                 sizeof(struct ipv6hdr) - sizeof(*ns);
1926         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1927                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1928                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1929                         break;
1930                 }
1931         }
1932
1933         /* Ethernet header */
1934         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1935         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1936         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1937         reply->protocol = htons(ETH_P_IPV6);
1938
1939         skb_pull(reply, sizeof(struct ethhdr));
1940         skb_reset_network_header(reply);
1941         skb_put(reply, sizeof(struct ipv6hdr));
1942
1943         /* IPv6 header */
1944
1945         pip6 = ipv6_hdr(reply);
1946         memset(pip6, 0, sizeof(struct ipv6hdr));
1947         pip6->version = 6;
1948         pip6->priority = ipv6_hdr(request)->priority;
1949         pip6->nexthdr = IPPROTO_ICMPV6;
1950         pip6->hop_limit = 255;
1951         pip6->daddr = ipv6_hdr(request)->saddr;
1952         pip6->saddr = *(struct in6_addr *)n->primary_key;
1953
1954         skb_pull(reply, sizeof(struct ipv6hdr));
1955         skb_reset_transport_header(reply);
1956
1957         /* Neighbor Advertisement */
1958         na = skb_put_zero(reply, sizeof(*na) + na_olen);
1959         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1960         na->icmph.icmp6_router = isrouter;
1961         na->icmph.icmp6_override = 1;
1962         na->icmph.icmp6_solicited = 1;
1963         na->target = ns->target;
1964         ether_addr_copy(&na->opt[2], n->ha);
1965         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1966         na->opt[1] = na_olen >> 3;
1967
1968         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1969                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1970                 csum_partial(na, sizeof(*na)+na_olen, 0));
1971
1972         pip6->payload_len = htons(sizeof(*na)+na_olen);
1973
1974         skb_push(reply, sizeof(struct ipv6hdr));
1975
1976         reply->ip_summed = CHECKSUM_UNNECESSARY;
1977
1978         return reply;
1979 }
1980
1981 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1982 {
1983         struct vxlan_dev *vxlan = netdev_priv(dev);
1984         const struct in6_addr *daddr;
1985         const struct ipv6hdr *iphdr;
1986         struct inet6_dev *in6_dev;
1987         struct neighbour *n;
1988         struct nd_msg *msg;
1989
1990         in6_dev = __in6_dev_get(dev);
1991         if (!in6_dev)
1992                 goto out;
1993
1994         iphdr = ipv6_hdr(skb);
1995         daddr = &iphdr->daddr;
1996         msg = (struct nd_msg *)(iphdr + 1);
1997
1998         if (ipv6_addr_loopback(daddr) ||
1999             ipv6_addr_is_multicast(&msg->target))
2000                 goto out;
2001
2002         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
2003
2004         if (n) {
2005                 struct vxlan_fdb *f;
2006                 struct sk_buff *reply;
2007
2008                 if (!(n->nud_state & NUD_CONNECTED)) {
2009                         neigh_release(n);
2010                         goto out;
2011                 }
2012
2013                 f = vxlan_find_mac(vxlan, n->ha, vni);
2014                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2015                         /* bridge-local neighbor */
2016                         neigh_release(n);
2017                         goto out;
2018                 }
2019
2020                 reply = vxlan_na_create(skb, n,
2021                                         !!(f ? f->flags & NTF_ROUTER : 0));
2022
2023                 neigh_release(n);
2024
2025                 if (reply == NULL)
2026                         goto out;
2027
2028                 if (netif_rx_ni(reply) == NET_RX_DROP)
2029                         dev->stats.rx_dropped++;
2030
2031         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2032                 union vxlan_addr ipa = {
2033                         .sin6.sin6_addr = msg->target,
2034                         .sin6.sin6_family = AF_INET6,
2035                 };
2036
2037                 vxlan_ip_miss(dev, &ipa);
2038         }
2039
2040 out:
2041         consume_skb(skb);
2042         return NETDEV_TX_OK;
2043 }
2044 #endif
2045
2046 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
2047 {
2048         struct vxlan_dev *vxlan = netdev_priv(dev);
2049         struct neighbour *n;
2050
2051         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
2052                 return false;
2053
2054         n = NULL;
2055         switch (ntohs(eth_hdr(skb)->h_proto)) {
2056         case ETH_P_IP:
2057         {
2058                 struct iphdr *pip;
2059
2060                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
2061                         return false;
2062                 pip = ip_hdr(skb);
2063                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
2064                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2065                         union vxlan_addr ipa = {
2066                                 .sin.sin_addr.s_addr = pip->daddr,
2067                                 .sin.sin_family = AF_INET,
2068                         };
2069
2070                         vxlan_ip_miss(dev, &ipa);
2071                         return false;
2072                 }
2073
2074                 break;
2075         }
2076 #if IS_ENABLED(CONFIG_IPV6)
2077         case ETH_P_IPV6:
2078         {
2079                 struct ipv6hdr *pip6;
2080
2081                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
2082                         return false;
2083                 pip6 = ipv6_hdr(skb);
2084                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
2085                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2086                         union vxlan_addr ipa = {
2087                                 .sin6.sin6_addr = pip6->daddr,
2088                                 .sin6.sin6_family = AF_INET6,
2089                         };
2090
2091                         vxlan_ip_miss(dev, &ipa);
2092                         return false;
2093                 }
2094
2095                 break;
2096         }
2097 #endif
2098         default:
2099                 return false;
2100         }
2101
2102         if (n) {
2103                 bool diff;
2104
2105                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
2106                 if (diff) {
2107                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
2108                                 dev->addr_len);
2109                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
2110                 }
2111                 neigh_release(n);
2112                 return diff;
2113         }
2114
2115         return false;
2116 }
2117
2118 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
2119                                 struct vxlan_metadata *md)
2120 {
2121         struct vxlanhdr_gbp *gbp;
2122
2123         if (!md->gbp)
2124                 return;
2125
2126         gbp = (struct vxlanhdr_gbp *)vxh;
2127         vxh->vx_flags |= VXLAN_HF_GBP;
2128
2129         if (md->gbp & VXLAN_GBP_DONT_LEARN)
2130                 gbp->dont_learn = 1;
2131
2132         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
2133                 gbp->policy_applied = 1;
2134
2135         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
2136 }
2137
2138 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
2139                                __be16 protocol)
2140 {
2141         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
2142
2143         gpe->np_applied = 1;
2144         gpe->next_protocol = tun_p_from_eth_p(protocol);
2145         if (!gpe->next_protocol)
2146                 return -EPFNOSUPPORT;
2147         return 0;
2148 }
2149
2150 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
2151                            int iphdr_len, __be32 vni,
2152                            struct vxlan_metadata *md, u32 vxflags,
2153                            bool udp_sum)
2154 {
2155         struct vxlanhdr *vxh;
2156         int min_headroom;
2157         int err;
2158         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
2159         __be16 inner_protocol = htons(ETH_P_TEB);
2160
2161         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
2162             skb->ip_summed == CHECKSUM_PARTIAL) {
2163                 int csum_start = skb_checksum_start_offset(skb);
2164
2165                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
2166                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
2167                     (skb->csum_offset == offsetof(struct udphdr, check) ||
2168                      skb->csum_offset == offsetof(struct tcphdr, check)))
2169                         type |= SKB_GSO_TUNNEL_REMCSUM;
2170         }
2171
2172         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
2173                         + VXLAN_HLEN + iphdr_len;
2174
2175         /* Need space for new headers (invalidates iph ptr) */
2176         err = skb_cow_head(skb, min_headroom);
2177         if (unlikely(err))
2178                 return err;
2179
2180         err = iptunnel_handle_offloads(skb, type);
2181         if (err)
2182                 return err;
2183
2184         vxh = __skb_push(skb, sizeof(*vxh));
2185         vxh->vx_flags = VXLAN_HF_VNI;
2186         vxh->vx_vni = vxlan_vni_field(vni);
2187
2188         if (type & SKB_GSO_TUNNEL_REMCSUM) {
2189                 unsigned int start;
2190
2191                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
2192                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
2193                 vxh->vx_flags |= VXLAN_HF_RCO;
2194
2195                 if (!skb_is_gso(skb)) {
2196                         skb->ip_summed = CHECKSUM_NONE;
2197                         skb->encapsulation = 0;
2198                 }
2199         }
2200
2201         if (vxflags & VXLAN_F_GBP)
2202                 vxlan_build_gbp_hdr(vxh, vxflags, md);
2203         if (vxflags & VXLAN_F_GPE) {
2204                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
2205                 if (err < 0)
2206                         return err;
2207                 inner_protocol = skb->protocol;
2208         }
2209
2210         skb_set_inner_protocol(skb, inner_protocol);
2211         return 0;
2212 }
2213
2214 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
2215                                       struct vxlan_sock *sock4,
2216                                       struct sk_buff *skb, int oif, u8 tos,
2217                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
2218                                       struct dst_cache *dst_cache,
2219                                       const struct ip_tunnel_info *info)
2220 {
2221         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2222         struct rtable *rt = NULL;
2223         struct flowi4 fl4;
2224
2225         if (!sock4)
2226                 return ERR_PTR(-EIO);
2227
2228         if (tos && !info)
2229                 use_cache = false;
2230         if (use_cache) {
2231                 rt = dst_cache_get_ip4(dst_cache, saddr);
2232                 if (rt)
2233                         return rt;
2234         }
2235
2236         memset(&fl4, 0, sizeof(fl4));
2237         fl4.flowi4_oif = oif;
2238         fl4.flowi4_tos = RT_TOS(tos);
2239         fl4.flowi4_mark = skb->mark;
2240         fl4.flowi4_proto = IPPROTO_UDP;
2241         fl4.daddr = daddr;
2242         fl4.saddr = *saddr;
2243         fl4.fl4_dport = dport;
2244         fl4.fl4_sport = sport;
2245
2246         rt = ip_route_output_key(vxlan->net, &fl4);
2247         if (!IS_ERR(rt)) {
2248                 if (rt->dst.dev == dev) {
2249                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
2250                         ip_rt_put(rt);
2251                         return ERR_PTR(-ELOOP);
2252                 }
2253
2254                 *saddr = fl4.saddr;
2255                 if (use_cache)
2256                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2257         } else {
2258                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
2259                 return ERR_PTR(-ENETUNREACH);
2260         }
2261         return rt;
2262 }
2263
2264 #if IS_ENABLED(CONFIG_IPV6)
2265 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
2266                                           struct net_device *dev,
2267                                           struct vxlan_sock *sock6,
2268                                           struct sk_buff *skb, int oif, u8 tos,
2269                                           __be32 label,
2270                                           const struct in6_addr *daddr,
2271                                           struct in6_addr *saddr,
2272                                           __be16 dport, __be16 sport,
2273                                           struct dst_cache *dst_cache,
2274                                           const struct ip_tunnel_info *info)
2275 {
2276         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2277         struct dst_entry *ndst;
2278         struct flowi6 fl6;
2279
2280         if (!sock6)
2281                 return ERR_PTR(-EIO);
2282
2283         if (tos && !info)
2284                 use_cache = false;
2285         if (use_cache) {
2286                 ndst = dst_cache_get_ip6(dst_cache, saddr);
2287                 if (ndst)
2288                         return ndst;
2289         }
2290
2291         memset(&fl6, 0, sizeof(fl6));
2292         fl6.flowi6_oif = oif;
2293         fl6.daddr = *daddr;
2294         fl6.saddr = *saddr;
2295         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
2296         fl6.flowi6_mark = skb->mark;
2297         fl6.flowi6_proto = IPPROTO_UDP;
2298         fl6.fl6_dport = dport;
2299         fl6.fl6_sport = sport;
2300
2301         ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk,
2302                                                &fl6, NULL);
2303         if (unlikely(IS_ERR(ndst))) {
2304                 netdev_dbg(dev, "no route to %pI6\n", daddr);
2305                 return ERR_PTR(-ENETUNREACH);
2306         }
2307
2308         if (unlikely(ndst->dev == dev)) {
2309                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
2310                 dst_release(ndst);
2311                 return ERR_PTR(-ELOOP);
2312         }
2313
2314         *saddr = fl6.saddr;
2315         if (use_cache)
2316                 dst_cache_set_ip6(dst_cache, ndst, saddr);
2317         return ndst;
2318 }
2319 #endif
2320
2321 /* Bypass encapsulation if the destination is local */
2322 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2323                                struct vxlan_dev *dst_vxlan, __be32 vni)
2324 {
2325         struct pcpu_sw_netstats *tx_stats, *rx_stats;
2326         union vxlan_addr loopback;
2327         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2328         struct net_device *dev;
2329         int len = skb->len;
2330
2331         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2332         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2333         skb->pkt_type = PACKET_HOST;
2334         skb->encapsulation = 0;
2335         skb->dev = dst_vxlan->dev;
2336         __skb_pull(skb, skb_network_offset(skb));
2337
2338         if (remote_ip->sa.sa_family == AF_INET) {
2339                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2340                 loopback.sa.sa_family =  AF_INET;
2341 #if IS_ENABLED(CONFIG_IPV6)
2342         } else {
2343                 loopback.sin6.sin6_addr = in6addr_loopback;
2344                 loopback.sa.sa_family =  AF_INET6;
2345 #endif
2346         }
2347
2348         rcu_read_lock();
2349         dev = skb->dev;
2350         if (unlikely(!(dev->flags & IFF_UP))) {
2351                 kfree_skb(skb);
2352                 goto drop;
2353         }
2354
2355         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
2356                 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni);
2357
2358         u64_stats_update_begin(&tx_stats->syncp);
2359         tx_stats->tx_packets++;
2360         tx_stats->tx_bytes += len;
2361         u64_stats_update_end(&tx_stats->syncp);
2362
2363         if (netif_rx(skb) == NET_RX_SUCCESS) {
2364                 u64_stats_update_begin(&rx_stats->syncp);
2365                 rx_stats->rx_packets++;
2366                 rx_stats->rx_bytes += len;
2367                 u64_stats_update_end(&rx_stats->syncp);
2368         } else {
2369 drop:
2370                 dev->stats.rx_dropped++;
2371         }
2372         rcu_read_unlock();
2373 }
2374
2375 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2376                                  struct vxlan_dev *vxlan,
2377                                  union vxlan_addr *daddr,
2378                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2379                                  struct dst_entry *dst,
2380                                  u32 rt_flags)
2381 {
2382 #if IS_ENABLED(CONFIG_IPV6)
2383         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2384          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2385          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2386          */
2387         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2388 #endif
2389         /* Bypass encapsulation if the destination is local */
2390         if (rt_flags & RTCF_LOCAL &&
2391             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2392                 struct vxlan_dev *dst_vxlan;
2393
2394                 dst_release(dst);
2395                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2396                                            daddr->sa.sa_family, dst_port,
2397                                            vxlan->cfg.flags);
2398                 if (!dst_vxlan) {
2399                         dev->stats.tx_errors++;
2400                         kfree_skb(skb);
2401
2402                         return -ENOENT;
2403                 }
2404                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2405                 return 1;
2406         }
2407
2408         return 0;
2409 }
2410
2411 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2412                            __be32 default_vni, struct vxlan_rdst *rdst,
2413                            bool did_rsc)
2414 {
2415         struct dst_cache *dst_cache;
2416         struct ip_tunnel_info *info;
2417         struct vxlan_dev *vxlan = netdev_priv(dev);
2418         const struct iphdr *old_iph = ip_hdr(skb);
2419         union vxlan_addr *dst;
2420         union vxlan_addr remote_ip, local_ip;
2421         struct vxlan_metadata _md;
2422         struct vxlan_metadata *md = &_md;
2423         __be16 src_port = 0, dst_port;
2424         struct dst_entry *ndst = NULL;
2425         __be32 vni, label;
2426         __u8 tos, ttl;
2427         int ifindex;
2428         int err;
2429         u32 flags = vxlan->cfg.flags;
2430         bool udp_sum = false;
2431         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2432
2433         info = skb_tunnel_info(skb);
2434
2435         if (rdst) {
2436                 dst = &rdst->remote_ip;
2437                 if (vxlan_addr_any(dst)) {
2438                         if (did_rsc) {
2439                                 /* short-circuited back to local bridge */
2440                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2441                                 return;
2442                         }
2443                         goto drop;
2444                 }
2445
2446                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2447                 vni = (rdst->remote_vni) ? : default_vni;
2448                 ifindex = rdst->remote_ifindex;
2449                 local_ip = vxlan->cfg.saddr;
2450                 dst_cache = &rdst->dst_cache;
2451                 md->gbp = skb->mark;
2452                 if (flags & VXLAN_F_TTL_INHERIT) {
2453                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2454                 } else {
2455                         ttl = vxlan->cfg.ttl;
2456                         if (!ttl && vxlan_addr_multicast(dst))
2457                                 ttl = 1;
2458                 }
2459
2460                 tos = vxlan->cfg.tos;
2461                 if (tos == 1)
2462                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2463
2464                 if (dst->sa.sa_family == AF_INET)
2465                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2466                 else
2467                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2468                 label = vxlan->cfg.label;
2469         } else {
2470                 if (!info) {
2471                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2472                                   dev->name);
2473                         goto drop;
2474                 }
2475                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2476                 if (remote_ip.sa.sa_family == AF_INET) {
2477                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2478                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2479                 } else {
2480                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2481                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2482                 }
2483                 dst = &remote_ip;
2484                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2485                 vni = tunnel_id_to_key32(info->key.tun_id);
2486                 ifindex = 0;
2487                 dst_cache = &info->dst_cache;
2488                 if (info->key.tun_flags & TUNNEL_VXLAN_OPT) {
2489                         if (info->options_len < sizeof(*md))
2490                                 goto drop;
2491                         md = ip_tunnel_info_opts(info);
2492                 }
2493                 ttl = info->key.ttl;
2494                 tos = info->key.tos;
2495                 label = info->key.label;
2496                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2497         }
2498         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2499                                      vxlan->cfg.port_max, true);
2500
2501         rcu_read_lock();
2502         if (dst->sa.sa_family == AF_INET) {
2503                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2504                 struct rtable *rt;
2505                 __be16 df = 0;
2506
2507                 if (!ifindex)
2508                         ifindex = sock4->sock->sk->sk_bound_dev_if;
2509
2510                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2511                                      dst->sin.sin_addr.s_addr,
2512                                      &local_ip.sin.sin_addr.s_addr,
2513                                      dst_port, src_port,
2514                                      dst_cache, info);
2515                 if (IS_ERR(rt)) {
2516                         err = PTR_ERR(rt);
2517                         goto tx_error;
2518                 }
2519
2520                 if (!info) {
2521                         /* Bypass encapsulation if the destination is local */
2522                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2523                                                     dst_port, ifindex, vni,
2524                                                     &rt->dst, rt->rt_flags);
2525                         if (err)
2526                                 goto out_unlock;
2527
2528                         if (vxlan->cfg.df == VXLAN_DF_SET) {
2529                                 df = htons(IP_DF);
2530                         } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) {
2531                                 struct ethhdr *eth = eth_hdr(skb);
2532
2533                                 if (ntohs(eth->h_proto) == ETH_P_IPV6 ||
2534                                     (ntohs(eth->h_proto) == ETH_P_IP &&
2535                                      old_iph->frag_off & htons(IP_DF)))
2536                                         df = htons(IP_DF);
2537                         }
2538                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2539                         df = htons(IP_DF);
2540                 }
2541
2542                 ndst = &rt->dst;
2543                 skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM);
2544
2545                 tos = ip_tunnel_ecn_encap(RT_TOS(tos), old_iph, skb);
2546                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2547                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2548                                       vni, md, flags, udp_sum);
2549                 if (err < 0)
2550                         goto tx_error;
2551
2552                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2553                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2554                                     src_port, dst_port, xnet, !udp_sum);
2555 #if IS_ENABLED(CONFIG_IPV6)
2556         } else {
2557                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2558
2559                 if (!ifindex)
2560                         ifindex = sock6->sock->sk->sk_bound_dev_if;
2561
2562                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2563                                         label, &dst->sin6.sin6_addr,
2564                                         &local_ip.sin6.sin6_addr,
2565                                         dst_port, src_port,
2566                                         dst_cache, info);
2567                 if (IS_ERR(ndst)) {
2568                         err = PTR_ERR(ndst);
2569                         ndst = NULL;
2570                         goto tx_error;
2571                 }
2572
2573                 if (!info) {
2574                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2575
2576                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2577                                                     dst_port, ifindex, vni,
2578                                                     ndst, rt6i_flags);
2579                         if (err)
2580                                 goto out_unlock;
2581                 }
2582
2583                 skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM);
2584
2585                 tos = ip_tunnel_ecn_encap(RT_TOS(tos), old_iph, skb);
2586                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2587                 skb_scrub_packet(skb, xnet);
2588                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2589                                       vni, md, flags, udp_sum);
2590                 if (err < 0)
2591                         goto tx_error;
2592
2593                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2594                                      &local_ip.sin6.sin6_addr,
2595                                      &dst->sin6.sin6_addr, tos, ttl,
2596                                      label, src_port, dst_port, !udp_sum);
2597 #endif
2598         }
2599 out_unlock:
2600         rcu_read_unlock();
2601         return;
2602
2603 drop:
2604         dev->stats.tx_dropped++;
2605         dev_kfree_skb(skb);
2606         return;
2607
2608 tx_error:
2609         rcu_read_unlock();
2610         if (err == -ELOOP)
2611                 dev->stats.collisions++;
2612         else if (err == -ENETUNREACH)
2613                 dev->stats.tx_carrier_errors++;
2614         dst_release(ndst);
2615         dev->stats.tx_errors++;
2616         kfree_skb(skb);
2617 }
2618
2619 /* Transmit local packets over Vxlan
2620  *
2621  * Outer IP header inherits ECN and DF from inner header.
2622  * Outer UDP destination is the VXLAN assigned port.
2623  *           source port is based on hash of flow
2624  */
2625 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2626 {
2627         struct vxlan_dev *vxlan = netdev_priv(dev);
2628         struct vxlan_rdst *rdst, *fdst = NULL;
2629         const struct ip_tunnel_info *info;
2630         bool did_rsc = false;
2631         struct vxlan_fdb *f;
2632         struct ethhdr *eth;
2633         __be32 vni = 0;
2634
2635         info = skb_tunnel_info(skb);
2636
2637         skb_reset_mac_header(skb);
2638
2639         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2640                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2641                     info->mode & IP_TUNNEL_INFO_TX) {
2642                         vni = tunnel_id_to_key32(info->key.tun_id);
2643                 } else {
2644                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2645                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2646                         else
2647                                 kfree_skb(skb);
2648                         return NETDEV_TX_OK;
2649                 }
2650         }
2651
2652         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2653                 eth = eth_hdr(skb);
2654                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2655                         return arp_reduce(dev, skb, vni);
2656 #if IS_ENABLED(CONFIG_IPV6)
2657                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2658                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2659                                             sizeof(struct nd_msg)) &&
2660                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2661                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2662
2663                         if (m->icmph.icmp6_code == 0 &&
2664                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2665                                 return neigh_reduce(dev, skb, vni);
2666                 }
2667 #endif
2668         }
2669
2670         eth = eth_hdr(skb);
2671         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2672         did_rsc = false;
2673
2674         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2675             (ntohs(eth->h_proto) == ETH_P_IP ||
2676              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2677                 did_rsc = route_shortcircuit(dev, skb);
2678                 if (did_rsc)
2679                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2680         }
2681
2682         if (f == NULL) {
2683                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2684                 if (f == NULL) {
2685                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2686                             !is_multicast_ether_addr(eth->h_dest))
2687                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2688
2689                         dev->stats.tx_dropped++;
2690                         kfree_skb(skb);
2691                         return NETDEV_TX_OK;
2692                 }
2693         }
2694
2695         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2696                 struct sk_buff *skb1;
2697
2698                 if (!fdst) {
2699                         fdst = rdst;
2700                         continue;
2701                 }
2702                 skb1 = skb_clone(skb, GFP_ATOMIC);
2703                 if (skb1)
2704                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2705         }
2706
2707         if (fdst)
2708                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2709         else
2710                 kfree_skb(skb);
2711         return NETDEV_TX_OK;
2712 }
2713
2714 /* Walk the forwarding table and purge stale entries */
2715 static void vxlan_cleanup(struct timer_list *t)
2716 {
2717         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2718         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2719         unsigned int h;
2720
2721         if (!netif_running(vxlan->dev))
2722                 return;
2723
2724         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2725                 struct hlist_node *p, *n;
2726
2727                 spin_lock(&vxlan->hash_lock[h]);
2728                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2729                         struct vxlan_fdb *f
2730                                 = container_of(p, struct vxlan_fdb, hlist);
2731                         unsigned long timeout;
2732
2733                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2734                                 continue;
2735
2736                         if (f->flags & NTF_EXT_LEARNED)
2737                                 continue;
2738
2739                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2740                         if (time_before_eq(timeout, jiffies)) {
2741                                 netdev_dbg(vxlan->dev,
2742                                            "garbage collect %pM\n",
2743                                            f->eth_addr);
2744                                 f->state = NUD_STALE;
2745                                 vxlan_fdb_destroy(vxlan, f, true, true);
2746                         } else if (time_before(timeout, next_timer))
2747                                 next_timer = timeout;
2748                 }
2749                 spin_unlock(&vxlan->hash_lock[h]);
2750         }
2751
2752         mod_timer(&vxlan->age_timer, next_timer);
2753 }
2754
2755 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2756 {
2757         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2758
2759         spin_lock(&vn->sock_lock);
2760         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2761 #if IS_ENABLED(CONFIG_IPV6)
2762         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2763 #endif
2764         spin_unlock(&vn->sock_lock);
2765 }
2766
2767 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2768                              struct vxlan_dev_node *node)
2769 {
2770         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2771         __be32 vni = vxlan->default_dst.remote_vni;
2772
2773         node->vxlan = vxlan;
2774         spin_lock(&vn->sock_lock);
2775         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2776         spin_unlock(&vn->sock_lock);
2777 }
2778
2779 /* Setup stats when device is created */
2780 static int vxlan_init(struct net_device *dev)
2781 {
2782         struct vxlan_dev *vxlan = netdev_priv(dev);
2783         int err;
2784
2785         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2786         if (!dev->tstats)
2787                 return -ENOMEM;
2788
2789         err = gro_cells_init(&vxlan->gro_cells, dev);
2790         if (err) {
2791                 free_percpu(dev->tstats);
2792                 return err;
2793         }
2794
2795         return 0;
2796 }
2797
2798 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2799 {
2800         struct vxlan_fdb *f;
2801         u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, vni);
2802
2803         spin_lock_bh(&vxlan->hash_lock[hash_index]);
2804         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2805         if (f)
2806                 vxlan_fdb_destroy(vxlan, f, true, true);
2807         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
2808 }
2809
2810 static void vxlan_uninit(struct net_device *dev)
2811 {
2812         struct vxlan_dev *vxlan = netdev_priv(dev);
2813
2814         gro_cells_destroy(&vxlan->gro_cells);
2815
2816         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2817
2818         free_percpu(dev->tstats);
2819 }
2820
2821 /* Start ageing timer and join group when device is brought up */
2822 static int vxlan_open(struct net_device *dev)
2823 {
2824         struct vxlan_dev *vxlan = netdev_priv(dev);
2825         int ret;
2826
2827         ret = vxlan_sock_add(vxlan);
2828         if (ret < 0)
2829                 return ret;
2830
2831         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2832                 ret = vxlan_igmp_join(vxlan);
2833                 if (ret == -EADDRINUSE)
2834                         ret = 0;
2835                 if (ret) {
2836                         vxlan_sock_release(vxlan);
2837                         return ret;
2838                 }
2839         }
2840
2841         if (vxlan->cfg.age_interval)
2842                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2843
2844         return ret;
2845 }
2846
2847 /* Purge the forwarding table */
2848 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2849 {
2850         unsigned int h;
2851
2852         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2853                 struct hlist_node *p, *n;
2854
2855                 spin_lock_bh(&vxlan->hash_lock[h]);
2856                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2857                         struct vxlan_fdb *f
2858                                 = container_of(p, struct vxlan_fdb, hlist);
2859                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2860                                 continue;
2861                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2862                         if (!is_zero_ether_addr(f->eth_addr))
2863                                 vxlan_fdb_destroy(vxlan, f, true, true);
2864                 }
2865                 spin_unlock_bh(&vxlan->hash_lock[h]);
2866         }
2867 }
2868
2869 /* Cleanup timer and forwarding table on shutdown */
2870 static int vxlan_stop(struct net_device *dev)
2871 {
2872         struct vxlan_dev *vxlan = netdev_priv(dev);
2873         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2874         int ret = 0;
2875
2876         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2877             !vxlan_group_used(vn, vxlan))
2878                 ret = vxlan_igmp_leave(vxlan);
2879
2880         del_timer_sync(&vxlan->age_timer);
2881
2882         vxlan_flush(vxlan, false);
2883         vxlan_sock_release(vxlan);
2884
2885         return ret;
2886 }
2887
2888 /* Stub, nothing needs to be done. */
2889 static void vxlan_set_multicast_list(struct net_device *dev)
2890 {
2891 }
2892
2893 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2894 {
2895         struct vxlan_dev *vxlan = netdev_priv(dev);
2896         struct vxlan_rdst *dst = &vxlan->default_dst;
2897         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2898                                                          dst->remote_ifindex);
2899         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
2900
2901         /* This check is different than dev->max_mtu, because it looks at
2902          * the lowerdev->mtu, rather than the static dev->max_mtu
2903          */
2904         if (lowerdev) {
2905                 int max_mtu = lowerdev->mtu -
2906                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2907                 if (new_mtu > max_mtu)
2908                         return -EINVAL;
2909         }
2910
2911         dev->mtu = new_mtu;
2912         return 0;
2913 }
2914
2915 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2916 {
2917         struct vxlan_dev *vxlan = netdev_priv(dev);
2918         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2919         __be16 sport, dport;
2920
2921         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2922                                   vxlan->cfg.port_max, true);
2923         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2924
2925         if (ip_tunnel_info_af(info) == AF_INET) {
2926                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2927                 struct rtable *rt;
2928
2929                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2930                                      info->key.u.ipv4.dst,
2931                                      &info->key.u.ipv4.src, dport, sport,
2932                                      &info->dst_cache, info);
2933                 if (IS_ERR(rt))
2934                         return PTR_ERR(rt);
2935                 ip_rt_put(rt);
2936         } else {
2937 #if IS_ENABLED(CONFIG_IPV6)
2938                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2939                 struct dst_entry *ndst;
2940
2941                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2942                                         info->key.label, &info->key.u.ipv6.dst,
2943                                         &info->key.u.ipv6.src, dport, sport,
2944                                         &info->dst_cache, info);
2945                 if (IS_ERR(ndst))
2946                         return PTR_ERR(ndst);
2947                 dst_release(ndst);
2948 #else /* !CONFIG_IPV6 */
2949                 return -EPFNOSUPPORT;
2950 #endif
2951         }
2952         info->key.tp_src = sport;
2953         info->key.tp_dst = dport;
2954         return 0;
2955 }
2956
2957 static const struct net_device_ops vxlan_netdev_ether_ops = {
2958         .ndo_init               = vxlan_init,
2959         .ndo_uninit             = vxlan_uninit,
2960         .ndo_open               = vxlan_open,
2961         .ndo_stop               = vxlan_stop,
2962         .ndo_start_xmit         = vxlan_xmit,
2963         .ndo_get_stats64        = ip_tunnel_get_stats64,
2964         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2965         .ndo_change_mtu         = vxlan_change_mtu,
2966         .ndo_validate_addr      = eth_validate_addr,
2967         .ndo_set_mac_address    = eth_mac_addr,
2968         .ndo_fdb_add            = vxlan_fdb_add,
2969         .ndo_fdb_del            = vxlan_fdb_delete,
2970         .ndo_fdb_dump           = vxlan_fdb_dump,
2971         .ndo_fdb_get            = vxlan_fdb_get,
2972         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2973         .ndo_change_proto_down  = dev_change_proto_down_generic,
2974 };
2975
2976 static const struct net_device_ops vxlan_netdev_raw_ops = {
2977         .ndo_init               = vxlan_init,
2978         .ndo_uninit             = vxlan_uninit,
2979         .ndo_open               = vxlan_open,
2980         .ndo_stop               = vxlan_stop,
2981         .ndo_start_xmit         = vxlan_xmit,
2982         .ndo_get_stats64        = ip_tunnel_get_stats64,
2983         .ndo_change_mtu         = vxlan_change_mtu,
2984         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2985 };
2986
2987 /* Info for udev, that this is a virtual tunnel endpoint */
2988 static struct device_type vxlan_type = {
2989         .name = "vxlan",
2990 };
2991
2992 /* Calls the ndo_udp_tunnel_add of the caller in order to
2993  * supply the listening VXLAN udp ports. Callers are expected
2994  * to implement the ndo_udp_tunnel_add.
2995  */
2996 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
2997 {
2998         struct vxlan_sock *vs;
2999         struct net *net = dev_net(dev);
3000         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3001         unsigned int i;
3002
3003         spin_lock(&vn->sock_lock);
3004         for (i = 0; i < PORT_HASH_SIZE; ++i) {
3005                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
3006                         unsigned short type;
3007
3008                         if (vs->flags & VXLAN_F_GPE)
3009                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
3010                         else
3011                                 type = UDP_TUNNEL_TYPE_VXLAN;
3012
3013                         if (push)
3014                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
3015                         else
3016                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
3017                 }
3018         }
3019         spin_unlock(&vn->sock_lock);
3020 }
3021
3022 /* Initialize the device structure. */
3023 static void vxlan_setup(struct net_device *dev)
3024 {
3025         struct vxlan_dev *vxlan = netdev_priv(dev);
3026         unsigned int h;
3027
3028         eth_hw_addr_random(dev);
3029         ether_setup(dev);
3030
3031         dev->needs_free_netdev = true;
3032         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
3033
3034         dev->features   |= NETIF_F_LLTX;
3035         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
3036         dev->features   |= NETIF_F_RXCSUM;
3037         dev->features   |= NETIF_F_GSO_SOFTWARE;
3038
3039         dev->vlan_features = dev->features;
3040         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
3041         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
3042         netif_keep_dst(dev);
3043         dev->priv_flags |= IFF_NO_QUEUE;
3044
3045         /* MTU range: 68 - 65535 */
3046         dev->min_mtu = ETH_MIN_MTU;
3047         dev->max_mtu = ETH_MAX_MTU;
3048
3049         INIT_LIST_HEAD(&vxlan->next);
3050
3051         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
3052
3053         vxlan->dev = dev;
3054
3055         for (h = 0; h < FDB_HASH_SIZE; ++h) {
3056                 spin_lock_init(&vxlan->hash_lock[h]);
3057                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
3058         }
3059 }
3060
3061 static void vxlan_ether_setup(struct net_device *dev)
3062 {
3063         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
3064         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3065         dev->netdev_ops = &vxlan_netdev_ether_ops;
3066 }
3067
3068 static void vxlan_raw_setup(struct net_device *dev)
3069 {
3070         dev->header_ops = NULL;
3071         dev->type = ARPHRD_NONE;
3072         dev->hard_header_len = 0;
3073         dev->addr_len = 0;
3074         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
3075         dev->netdev_ops = &vxlan_netdev_raw_ops;
3076 }
3077
3078 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
3079         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
3080         [IFLA_VXLAN_GROUP]      = { .len = sizeof_field(struct iphdr, daddr) },
3081         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
3082         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
3083         [IFLA_VXLAN_LOCAL]      = { .len = sizeof_field(struct iphdr, saddr) },
3084         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
3085         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
3086         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
3087         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
3088         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
3089         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
3090         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
3091         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
3092         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
3093         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
3094         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
3095         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
3096         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
3097         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
3098         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
3099         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
3100         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
3101         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
3102         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
3103         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
3104         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
3105         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
3106         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
3107         [IFLA_VXLAN_DF]         = { .type = NLA_U8 },
3108 };
3109
3110 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
3111                           struct netlink_ext_ack *extack)
3112 {
3113         if (tb[IFLA_ADDRESS]) {
3114                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
3115                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3116                                             "Provided link layer address is not Ethernet");
3117                         return -EINVAL;
3118                 }
3119
3120                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
3121                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3122                                             "Provided Ethernet address is not unicast");
3123                         return -EADDRNOTAVAIL;
3124                 }
3125         }
3126
3127         if (tb[IFLA_MTU]) {
3128                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
3129
3130                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
3131                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3132                                             "MTU must be between 68 and 65535");
3133                         return -EINVAL;
3134                 }
3135         }
3136
3137         if (!data) {
3138                 NL_SET_ERR_MSG(extack,
3139                                "Required attributes not provided to perform the operation");
3140                 return -EINVAL;
3141         }
3142
3143         if (data[IFLA_VXLAN_ID]) {
3144                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
3145
3146                 if (id >= VXLAN_N_VID) {
3147                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_ID],
3148                                             "VXLAN ID must be lower than 16777216");
3149                         return -ERANGE;
3150                 }
3151         }
3152
3153         if (data[IFLA_VXLAN_PORT_RANGE]) {
3154                 const struct ifla_vxlan_port_range *p
3155                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3156
3157                 if (ntohs(p->high) < ntohs(p->low)) {
3158                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_PORT_RANGE],
3159                                             "Invalid source port range");
3160                         return -EINVAL;
3161                 }
3162         }
3163
3164         if (data[IFLA_VXLAN_DF]) {
3165                 enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]);
3166
3167                 if (df < 0 || df > VXLAN_DF_MAX) {
3168                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_DF],
3169                                             "Invalid DF attribute");
3170                         return -EINVAL;
3171                 }
3172         }
3173
3174         return 0;
3175 }
3176
3177 static void vxlan_get_drvinfo(struct net_device *netdev,
3178                               struct ethtool_drvinfo *drvinfo)
3179 {
3180         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
3181         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
3182 }
3183
3184 static int vxlan_get_link_ksettings(struct net_device *dev,
3185                                     struct ethtool_link_ksettings *cmd)
3186 {
3187         struct vxlan_dev *vxlan = netdev_priv(dev);
3188         struct vxlan_rdst *dst = &vxlan->default_dst;
3189         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
3190                                                          dst->remote_ifindex);
3191
3192         if (!lowerdev) {
3193                 cmd->base.duplex = DUPLEX_UNKNOWN;
3194                 cmd->base.port = PORT_OTHER;
3195                 cmd->base.speed = SPEED_UNKNOWN;
3196
3197                 return 0;
3198         }
3199
3200         return __ethtool_get_link_ksettings(lowerdev, cmd);
3201 }
3202
3203 static const struct ethtool_ops vxlan_ethtool_ops = {
3204         .get_drvinfo            = vxlan_get_drvinfo,
3205         .get_link               = ethtool_op_get_link,
3206         .get_link_ksettings     = vxlan_get_link_ksettings,
3207 };
3208
3209 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
3210                                         __be16 port, u32 flags, int ifindex)
3211 {
3212         struct socket *sock;
3213         struct udp_port_cfg udp_conf;
3214         int err;
3215
3216         memset(&udp_conf, 0, sizeof(udp_conf));
3217
3218         if (ipv6) {
3219                 udp_conf.family = AF_INET6;
3220                 udp_conf.use_udp6_rx_checksums =
3221                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
3222                 udp_conf.ipv6_v6only = 1;
3223         } else {
3224                 udp_conf.family = AF_INET;
3225         }
3226
3227         udp_conf.local_udp_port = port;
3228         udp_conf.bind_ifindex = ifindex;
3229
3230         /* Open UDP socket */
3231         err = udp_sock_create(net, &udp_conf, &sock);
3232         if (err < 0)
3233                 return ERR_PTR(err);
3234
3235         return sock;
3236 }
3237
3238 /* Create new listen socket if needed */
3239 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
3240                                               __be16 port, u32 flags,
3241                                               int ifindex)
3242 {
3243         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3244         struct vxlan_sock *vs;
3245         struct socket *sock;
3246         unsigned int h;
3247         struct udp_tunnel_sock_cfg tunnel_cfg;
3248
3249         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
3250         if (!vs)
3251                 return ERR_PTR(-ENOMEM);
3252
3253         for (h = 0; h < VNI_HASH_SIZE; ++h)
3254                 INIT_HLIST_HEAD(&vs->vni_list[h]);
3255
3256         sock = vxlan_create_sock(net, ipv6, port, flags, ifindex);
3257         if (IS_ERR(sock)) {
3258                 kfree(vs);
3259                 return ERR_CAST(sock);
3260         }
3261
3262         vs->sock = sock;
3263         refcount_set(&vs->refcnt, 1);
3264         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
3265
3266         spin_lock(&vn->sock_lock);
3267         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
3268         udp_tunnel_notify_add_rx_port(sock,
3269                                       (vs->flags & VXLAN_F_GPE) ?
3270                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
3271                                       UDP_TUNNEL_TYPE_VXLAN);
3272         spin_unlock(&vn->sock_lock);
3273
3274         /* Mark socket as an encapsulation socket. */
3275         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
3276         tunnel_cfg.sk_user_data = vs;
3277         tunnel_cfg.encap_type = 1;
3278         tunnel_cfg.encap_rcv = vxlan_rcv;
3279         tunnel_cfg.encap_err_lookup = vxlan_err_lookup;
3280         tunnel_cfg.encap_destroy = NULL;
3281         tunnel_cfg.gro_receive = vxlan_gro_receive;
3282         tunnel_cfg.gro_complete = vxlan_gro_complete;
3283
3284         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
3285
3286         return vs;
3287 }
3288
3289 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
3290 {
3291         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3292         struct vxlan_sock *vs = NULL;
3293         struct vxlan_dev_node *node;
3294         int l3mdev_index = 0;
3295
3296         if (vxlan->cfg.remote_ifindex)
3297                 l3mdev_index = l3mdev_master_upper_ifindex_by_index(
3298                         vxlan->net, vxlan->cfg.remote_ifindex);
3299
3300         if (!vxlan->cfg.no_share) {
3301                 spin_lock(&vn->sock_lock);
3302                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
3303                                      vxlan->cfg.dst_port, vxlan->cfg.flags,
3304                                      l3mdev_index);
3305                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
3306                         spin_unlock(&vn->sock_lock);
3307                         return -EBUSY;
3308                 }
3309                 spin_unlock(&vn->sock_lock);
3310         }
3311         if (!vs)
3312                 vs = vxlan_socket_create(vxlan->net, ipv6,
3313                                          vxlan->cfg.dst_port, vxlan->cfg.flags,
3314                                          l3mdev_index);
3315         if (IS_ERR(vs))
3316                 return PTR_ERR(vs);
3317 #if IS_ENABLED(CONFIG_IPV6)
3318         if (ipv6) {
3319                 rcu_assign_pointer(vxlan->vn6_sock, vs);
3320                 node = &vxlan->hlist6;
3321         } else
3322 #endif
3323         {
3324                 rcu_assign_pointer(vxlan->vn4_sock, vs);
3325                 node = &vxlan->hlist4;
3326         }
3327         vxlan_vs_add_dev(vs, vxlan, node);
3328         return 0;
3329 }
3330
3331 static int vxlan_sock_add(struct vxlan_dev *vxlan)
3332 {
3333         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
3334         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
3335         bool ipv4 = !ipv6 || metadata;
3336         int ret = 0;
3337
3338         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
3339 #if IS_ENABLED(CONFIG_IPV6)
3340         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
3341         if (ipv6) {
3342                 ret = __vxlan_sock_add(vxlan, true);
3343                 if (ret < 0 && ret != -EAFNOSUPPORT)
3344                         ipv4 = false;
3345         }
3346 #endif
3347         if (ipv4)
3348                 ret = __vxlan_sock_add(vxlan, false);
3349         if (ret < 0)
3350                 vxlan_sock_release(vxlan);
3351         return ret;
3352 }
3353
3354 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
3355                                  struct net_device **lower,
3356                                  struct vxlan_dev *old,
3357                                  struct netlink_ext_ack *extack)
3358 {
3359         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
3360         struct vxlan_dev *tmp;
3361         bool use_ipv6 = false;
3362
3363         if (conf->flags & VXLAN_F_GPE) {
3364                 /* For now, allow GPE only together with
3365                  * COLLECT_METADATA. This can be relaxed later; in such
3366                  * case, the other side of the PtP link will have to be
3367                  * provided.
3368                  */
3369                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
3370                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3371                         NL_SET_ERR_MSG(extack,
3372                                        "VXLAN GPE does not support this combination of attributes");
3373                         return -EINVAL;
3374                 }
3375         }
3376
3377         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
3378                 /* Unless IPv6 is explicitly requested, assume IPv4 */
3379                 conf->remote_ip.sa.sa_family = AF_INET;
3380                 conf->saddr.sa.sa_family = AF_INET;
3381         } else if (!conf->remote_ip.sa.sa_family) {
3382                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
3383         } else if (!conf->saddr.sa.sa_family) {
3384                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
3385         }
3386
3387         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
3388                 NL_SET_ERR_MSG(extack,
3389                                "Local and remote address must be from the same family");
3390                 return -EINVAL;
3391         }
3392
3393         if (vxlan_addr_multicast(&conf->saddr)) {
3394                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
3395                 return -EINVAL;
3396         }
3397
3398         if (conf->saddr.sa.sa_family == AF_INET6) {
3399                 if (!IS_ENABLED(CONFIG_IPV6)) {
3400                         NL_SET_ERR_MSG(extack,
3401                                        "IPv6 support not enabled in the kernel");
3402                         return -EPFNOSUPPORT;
3403                 }
3404                 use_ipv6 = true;
3405                 conf->flags |= VXLAN_F_IPV6;
3406
3407                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3408                         int local_type =
3409                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3410                         int remote_type =
3411                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3412
3413                         if (local_type & IPV6_ADDR_LINKLOCAL) {
3414                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3415                                     (remote_type != IPV6_ADDR_ANY)) {
3416                                         NL_SET_ERR_MSG(extack,
3417                                                        "Invalid combination of local and remote address scopes");
3418                                         return -EINVAL;
3419                                 }
3420
3421                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3422                         } else {
3423                                 if (remote_type ==
3424                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3425                                         NL_SET_ERR_MSG(extack,
3426                                                        "Invalid combination of local and remote address scopes");
3427                                         return -EINVAL;
3428                                 }
3429
3430                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3431                         }
3432                 }
3433         }
3434
3435         if (conf->label && !use_ipv6) {
3436                 NL_SET_ERR_MSG(extack,
3437                                "Label attribute only applies to IPv6 VXLAN devices");
3438                 return -EINVAL;
3439         }
3440
3441         if (conf->remote_ifindex) {
3442                 struct net_device *lowerdev;
3443
3444                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3445                 if (!lowerdev) {
3446                         NL_SET_ERR_MSG(extack,
3447                                        "Invalid local interface, device not found");
3448                         return -ENODEV;
3449                 }
3450
3451 #if IS_ENABLED(CONFIG_IPV6)
3452                 if (use_ipv6) {
3453                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3454                         if (idev && idev->cnf.disable_ipv6) {
3455                                 NL_SET_ERR_MSG(extack,
3456                                                "IPv6 support disabled by administrator");
3457                                 return -EPERM;
3458                         }
3459                 }
3460 #endif
3461
3462                 *lower = lowerdev;
3463         } else {
3464                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3465                         NL_SET_ERR_MSG(extack,
3466                                        "Local interface required for multicast remote destination");
3467
3468                         return -EINVAL;
3469                 }
3470
3471 #if IS_ENABLED(CONFIG_IPV6)
3472                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3473                         NL_SET_ERR_MSG(extack,
3474                                        "Local interface required for link-local local/remote addresses");
3475                         return -EINVAL;
3476                 }
3477 #endif
3478
3479                 *lower = NULL;
3480         }
3481
3482         if (!conf->dst_port) {
3483                 if (conf->flags & VXLAN_F_GPE)
3484                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3485                 else
3486                         conf->dst_port = htons(vxlan_port);
3487         }
3488
3489         if (!conf->age_interval)
3490                 conf->age_interval = FDB_AGE_DEFAULT;
3491
3492         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3493                 if (tmp == old)
3494                         continue;
3495
3496                 if (tmp->cfg.vni != conf->vni)
3497                         continue;
3498                 if (tmp->cfg.dst_port != conf->dst_port)
3499                         continue;
3500                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3501                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3502                         continue;
3503
3504                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3505                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3506                         continue;
3507
3508                 NL_SET_ERR_MSG(extack,
3509                                "A VXLAN device with the specified VNI already exists");
3510                 return -EEXIST;
3511         }
3512
3513         return 0;
3514 }
3515
3516 static void vxlan_config_apply(struct net_device *dev,
3517                                struct vxlan_config *conf,
3518                                struct net_device *lowerdev,
3519                                struct net *src_net,
3520                                bool changelink)
3521 {
3522         struct vxlan_dev *vxlan = netdev_priv(dev);
3523         struct vxlan_rdst *dst = &vxlan->default_dst;
3524         unsigned short needed_headroom = ETH_HLEN;
3525         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3526         int max_mtu = ETH_MAX_MTU;
3527
3528         if (!changelink) {
3529                 if (conf->flags & VXLAN_F_GPE)
3530                         vxlan_raw_setup(dev);
3531                 else
3532                         vxlan_ether_setup(dev);
3533
3534                 if (conf->mtu)
3535                         dev->mtu = conf->mtu;
3536
3537                 vxlan->net = src_net;
3538         }
3539
3540         dst->remote_vni = conf->vni;
3541
3542         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3543
3544         if (lowerdev) {
3545                 dst->remote_ifindex = conf->remote_ifindex;
3546
3547                 dev->gso_max_size = lowerdev->gso_max_size;
3548                 dev->gso_max_segs = lowerdev->gso_max_segs;
3549
3550                 needed_headroom = lowerdev->hard_header_len;
3551
3552                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3553                                            VXLAN_HEADROOM);
3554                 if (max_mtu < ETH_MIN_MTU)
3555                         max_mtu = ETH_MIN_MTU;
3556
3557                 if (!changelink && !conf->mtu)
3558                         dev->mtu = max_mtu;
3559         }
3560
3561         if (dev->mtu > max_mtu)
3562                 dev->mtu = max_mtu;
3563
3564         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3565                 needed_headroom += VXLAN6_HEADROOM;
3566         else
3567                 needed_headroom += VXLAN_HEADROOM;
3568         dev->needed_headroom = needed_headroom;
3569
3570         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3571 }
3572
3573 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3574                                struct vxlan_config *conf, bool changelink,
3575                                struct netlink_ext_ack *extack)
3576 {
3577         struct vxlan_dev *vxlan = netdev_priv(dev);
3578         struct net_device *lowerdev;
3579         int ret;
3580
3581         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3582         if (ret)
3583                 return ret;
3584
3585         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3586
3587         return 0;
3588 }
3589
3590 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3591                               struct vxlan_config *conf,
3592                               struct netlink_ext_ack *extack)
3593 {
3594         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3595         struct vxlan_dev *vxlan = netdev_priv(dev);
3596         struct net_device *remote_dev = NULL;
3597         struct vxlan_fdb *f = NULL;
3598         bool unregister = false;
3599         struct vxlan_rdst *dst;
3600         int err;
3601
3602         dst = &vxlan->default_dst;
3603         err = vxlan_dev_configure(net, dev, conf, false, extack);
3604         if (err)
3605                 return err;
3606
3607         dev->ethtool_ops = &vxlan_ethtool_ops;
3608
3609         /* create an fdb entry for a valid default destination */
3610         if (!vxlan_addr_any(&dst->remote_ip)) {
3611                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3612                                        &dst->remote_ip,
3613                                        NUD_REACHABLE | NUD_PERMANENT,
3614                                        vxlan->cfg.dst_port,
3615                                        dst->remote_vni,
3616                                        dst->remote_vni,
3617                                        dst->remote_ifindex,
3618                                        NTF_SELF, &f);
3619                 if (err)
3620                         return err;
3621         }
3622
3623         err = register_netdevice(dev);
3624         if (err)
3625                 goto errout;
3626         unregister = true;
3627
3628         if (dst->remote_ifindex) {
3629                 remote_dev = __dev_get_by_index(net, dst->remote_ifindex);
3630                 if (!remote_dev)
3631                         goto errout;
3632
3633                 err = netdev_upper_dev_link(remote_dev, dev, extack);
3634                 if (err)
3635                         goto errout;
3636         }
3637
3638         err = rtnl_configure_link(dev, NULL);
3639         if (err)
3640                 goto unlink;
3641
3642         if (f) {
3643                 vxlan_fdb_insert(vxlan, all_zeros_mac, dst->remote_vni, f);
3644
3645                 /* notify default fdb entry */
3646                 err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f),
3647                                        RTM_NEWNEIGH, true, extack);
3648                 if (err) {
3649                         vxlan_fdb_destroy(vxlan, f, false, false);
3650                         if (remote_dev)
3651                                 netdev_upper_dev_unlink(remote_dev, dev);
3652                         goto unregister;
3653                 }
3654         }
3655
3656         list_add(&vxlan->next, &vn->vxlan_list);
3657         if (remote_dev)
3658                 dst->remote_dev = remote_dev;
3659         return 0;
3660 unlink:
3661         if (remote_dev)
3662                 netdev_upper_dev_unlink(remote_dev, dev);
3663 errout:
3664         /* unregister_netdevice() destroys the default FDB entry with deletion
3665          * notification. But the addition notification was not sent yet, so
3666          * destroy the entry by hand here.
3667          */
3668         if (f)
3669                 __vxlan_fdb_free(f);
3670 unregister:
3671         if (unregister)
3672                 unregister_netdevice(dev);
3673         return err;
3674 }
3675
3676 /* Set/clear flags based on attribute */
3677 static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[],
3678                           int attrtype, unsigned long mask, bool changelink,
3679                           bool changelink_supported,
3680                           struct netlink_ext_ack *extack)
3681 {
3682         unsigned long flags;
3683
3684         if (!tb[attrtype])
3685                 return 0;
3686
3687         if (changelink && !changelink_supported) {
3688                 vxlan_flag_attr_error(attrtype, extack);
3689                 return -EOPNOTSUPP;
3690         }
3691
3692         if (vxlan_policy[attrtype].type == NLA_FLAG)
3693                 flags = conf->flags | mask;
3694         else if (nla_get_u8(tb[attrtype]))
3695                 flags = conf->flags | mask;
3696         else
3697                 flags = conf->flags & ~mask;
3698
3699         conf->flags = flags;
3700
3701         return 0;
3702 }
3703
3704 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3705                          struct net_device *dev, struct vxlan_config *conf,
3706                          bool changelink, struct netlink_ext_ack *extack)
3707 {
3708         struct vxlan_dev *vxlan = netdev_priv(dev);
3709         int err = 0;
3710
3711         memset(conf, 0, sizeof(*conf));
3712
3713         /* if changelink operation, start with old existing cfg */
3714         if (changelink)
3715                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3716
3717         if (data[IFLA_VXLAN_ID]) {
3718                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3719
3720                 if (changelink && (vni != conf->vni)) {
3721                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI");
3722                         return -EOPNOTSUPP;
3723                 }
3724                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3725         }
3726
3727         if (data[IFLA_VXLAN_GROUP]) {
3728                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) {
3729                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group");
3730                         return -EOPNOTSUPP;
3731                 }
3732
3733                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3734                 conf->remote_ip.sa.sa_family = AF_INET;
3735         } else if (data[IFLA_VXLAN_GROUP6]) {
3736                 if (!IS_ENABLED(CONFIG_IPV6)) {
3737                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel");
3738                         return -EPFNOSUPPORT;
3739                 }
3740
3741                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) {
3742                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group");
3743                         return -EOPNOTSUPP;
3744                 }
3745
3746                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3747                 conf->remote_ip.sa.sa_family = AF_INET6;
3748         }
3749
3750         if (data[IFLA_VXLAN_LOCAL]) {
3751                 if (changelink && (conf->saddr.sa.sa_family != AF_INET)) {
3752                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old");
3753                         return -EOPNOTSUPP;
3754                 }
3755
3756                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3757                 conf->saddr.sa.sa_family = AF_INET;
3758         } else if (data[IFLA_VXLAN_LOCAL6]) {
3759                 if (!IS_ENABLED(CONFIG_IPV6)) {
3760                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel");
3761                         return -EPFNOSUPPORT;
3762                 }
3763
3764                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) {
3765                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old");
3766                         return -EOPNOTSUPP;
3767                 }
3768
3769                 /* TODO: respect scope id */
3770                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3771                 conf->saddr.sa.sa_family = AF_INET6;
3772         }
3773
3774         if (data[IFLA_VXLAN_LINK])
3775                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3776
3777         if (data[IFLA_VXLAN_TOS])
3778                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3779
3780         if (data[IFLA_VXLAN_TTL])
3781                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3782
3783         if (data[IFLA_VXLAN_TTL_INHERIT]) {
3784                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT,
3785                                     VXLAN_F_TTL_INHERIT, changelink, false,
3786                                     extack);
3787                 if (err)
3788                         return err;
3789
3790         }
3791
3792         if (data[IFLA_VXLAN_LABEL])
3793                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3794                              IPV6_FLOWLABEL_MASK;
3795
3796         if (data[IFLA_VXLAN_LEARNING]) {
3797                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING,
3798                                     VXLAN_F_LEARN, changelink, true,
3799                                     extack);
3800                 if (err)
3801                         return err;
3802         } else if (!changelink) {
3803                 /* default to learn on a new device */
3804                 conf->flags |= VXLAN_F_LEARN;
3805         }
3806
3807         if (data[IFLA_VXLAN_AGEING])
3808                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3809
3810         if (data[IFLA_VXLAN_PROXY]) {
3811                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY,
3812                                     VXLAN_F_PROXY, changelink, false,
3813                                     extack);
3814                 if (err)
3815                         return err;
3816         }
3817
3818         if (data[IFLA_VXLAN_RSC]) {
3819                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC,
3820                                     VXLAN_F_RSC, changelink, false,
3821                                     extack);
3822                 if (err)
3823                         return err;
3824         }
3825
3826         if (data[IFLA_VXLAN_L2MISS]) {
3827                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS,
3828                                     VXLAN_F_L2MISS, changelink, false,
3829                                     extack);
3830                 if (err)
3831                         return err;
3832         }
3833
3834         if (data[IFLA_VXLAN_L3MISS]) {
3835                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS,
3836                                     VXLAN_F_L3MISS, changelink, false,
3837                                     extack);
3838                 if (err)
3839                         return err;
3840         }
3841
3842         if (data[IFLA_VXLAN_LIMIT]) {
3843                 if (changelink) {
3844                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT],
3845                                             "Cannot change limit");
3846                         return -EOPNOTSUPP;
3847                 }
3848                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3849         }
3850
3851         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3852                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA,
3853                                     VXLAN_F_COLLECT_METADATA, changelink, false,
3854                                     extack);
3855                 if (err)
3856                         return err;
3857         }
3858
3859         if (data[IFLA_VXLAN_PORT_RANGE]) {
3860                 if (!changelink) {
3861                         const struct ifla_vxlan_port_range *p
3862                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3863                         conf->port_min = ntohs(p->low);
3864                         conf->port_max = ntohs(p->high);
3865                 } else {
3866                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
3867                                             "Cannot change port range");
3868                         return -EOPNOTSUPP;
3869                 }
3870         }
3871
3872         if (data[IFLA_VXLAN_PORT]) {
3873                 if (changelink) {
3874                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT],
3875                                             "Cannot change port");
3876                         return -EOPNOTSUPP;
3877                 }
3878                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3879         }
3880
3881         if (data[IFLA_VXLAN_UDP_CSUM]) {
3882                 if (changelink) {
3883                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM],
3884                                             "Cannot change UDP_CSUM flag");
3885                         return -EOPNOTSUPP;
3886                 }
3887                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3888                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3889         }
3890
3891         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3892                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3893                                     VXLAN_F_UDP_ZERO_CSUM6_TX, changelink,
3894                                     false, extack);
3895                 if (err)
3896                         return err;
3897         }
3898
3899         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3900                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3901                                     VXLAN_F_UDP_ZERO_CSUM6_RX, changelink,
3902                                     false, extack);
3903                 if (err)
3904                         return err;
3905         }
3906
3907         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3908                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX,
3909                                     VXLAN_F_REMCSUM_TX, changelink, false,
3910                                     extack);
3911                 if (err)
3912                         return err;
3913         }
3914
3915         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3916                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX,
3917                                     VXLAN_F_REMCSUM_RX, changelink, false,
3918                                     extack);
3919                 if (err)
3920                         return err;
3921         }
3922
3923         if (data[IFLA_VXLAN_GBP]) {
3924                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP,
3925                                     VXLAN_F_GBP, changelink, false, extack);
3926                 if (err)
3927                         return err;
3928         }
3929
3930         if (data[IFLA_VXLAN_GPE]) {
3931                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE,
3932                                     VXLAN_F_GPE, changelink, false,
3933                                     extack);
3934                 if (err)
3935                         return err;
3936         }
3937
3938         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3939                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL,
3940                                     VXLAN_F_REMCSUM_NOPARTIAL, changelink,
3941                                     false, extack);
3942                 if (err)
3943                         return err;
3944         }
3945
3946         if (tb[IFLA_MTU]) {
3947                 if (changelink) {
3948                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3949                                             "Cannot change mtu");
3950                         return -EOPNOTSUPP;
3951                 }
3952                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3953         }
3954
3955         if (data[IFLA_VXLAN_DF])
3956                 conf->df = nla_get_u8(data[IFLA_VXLAN_DF]);
3957
3958         return 0;
3959 }
3960
3961 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3962                          struct nlattr *tb[], struct nlattr *data[],
3963                          struct netlink_ext_ack *extack)
3964 {
3965         struct vxlan_config conf;
3966         int err;
3967
3968         err = vxlan_nl2conf(tb, data, dev, &conf, false, extack);
3969         if (err)
3970                 return err;
3971
3972         return __vxlan_dev_create(src_net, dev, &conf, extack);
3973 }
3974
3975 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3976                             struct nlattr *data[],
3977                             struct netlink_ext_ack *extack)
3978 {
3979         struct vxlan_dev *vxlan = netdev_priv(dev);
3980         struct net_device *lowerdev;
3981         struct vxlan_config conf;
3982         struct vxlan_rdst *dst;
3983         int err;
3984
3985         dst = &vxlan->default_dst;
3986         err = vxlan_nl2conf(tb, data, dev, &conf, true, extack);
3987         if (err)
3988                 return err;
3989
3990         err = vxlan_config_validate(vxlan->net, &conf, &lowerdev,
3991                                     vxlan, extack);
3992         if (err)
3993                 return err;
3994
3995         if (dst->remote_dev == lowerdev)
3996                 lowerdev = NULL;
3997
3998         err = netdev_adjacent_change_prepare(dst->remote_dev, lowerdev, dev,
3999                                              extack);
4000         if (err)
4001                 return err;
4002
4003         /* handle default dst entry */
4004         if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) {
4005                 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, conf.vni);
4006
4007                 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4008                 if (!vxlan_addr_any(&conf.remote_ip)) {
4009                         err = vxlan_fdb_update(vxlan, all_zeros_mac,
4010                                                &conf.remote_ip,
4011                                                NUD_REACHABLE | NUD_PERMANENT,
4012                                                NLM_F_APPEND | NLM_F_CREATE,
4013                                                vxlan->cfg.dst_port,
4014                                                conf.vni, conf.vni,
4015                                                conf.remote_ifindex,
4016                                                NTF_SELF, true, extack);
4017                         if (err) {
4018                                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4019                                 netdev_adjacent_change_abort(dst->remote_dev,
4020                                                              lowerdev, dev);
4021                                 return err;
4022                         }
4023                 }
4024                 if (!vxlan_addr_any(&dst->remote_ip))
4025                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
4026                                            dst->remote_ip,
4027                                            vxlan->cfg.dst_port,
4028                                            dst->remote_vni,
4029                                            dst->remote_vni,
4030                                            dst->remote_ifindex,
4031                                            true);
4032                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4033         }
4034
4035         if (conf.age_interval != vxlan->cfg.age_interval)
4036                 mod_timer(&vxlan->age_timer, jiffies);
4037
4038         netdev_adjacent_change_commit(dst->remote_dev, lowerdev, dev);
4039         if (lowerdev && lowerdev != dst->remote_dev) {
4040                 dst->remote_dev = lowerdev;
4041                 netdev_update_lockdep_key(lowerdev);
4042         }
4043         vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true);
4044         return 0;
4045 }
4046
4047 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
4048 {
4049         struct vxlan_dev *vxlan = netdev_priv(dev);
4050
4051         vxlan_flush(vxlan, true);
4052
4053         list_del(&vxlan->next);
4054         unregister_netdevice_queue(dev, head);
4055         if (vxlan->default_dst.remote_dev)
4056                 netdev_upper_dev_unlink(vxlan->default_dst.remote_dev, dev);
4057 }
4058
4059 static size_t vxlan_get_size(const struct net_device *dev)
4060 {
4061
4062         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
4063                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
4064                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
4065                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
4066                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
4067                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL_INHERIT */
4068                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
4069                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_DF */
4070                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
4071                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
4072                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
4073                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
4074                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
4075                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
4076                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
4077                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
4078                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
4079                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
4080                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
4081                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
4082                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
4083                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
4084                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
4085                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
4086                 0;
4087 }
4088
4089 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
4090 {
4091         const struct vxlan_dev *vxlan = netdev_priv(dev);
4092         const struct vxlan_rdst *dst = &vxlan->default_dst;
4093         struct ifla_vxlan_port_range ports = {
4094                 .low =  htons(vxlan->cfg.port_min),
4095                 .high = htons(vxlan->cfg.port_max),
4096         };
4097
4098         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
4099                 goto nla_put_failure;
4100
4101         if (!vxlan_addr_any(&dst->remote_ip)) {
4102                 if (dst->remote_ip.sa.sa_family == AF_INET) {
4103                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
4104                                             dst->remote_ip.sin.sin_addr.s_addr))
4105                                 goto nla_put_failure;
4106 #if IS_ENABLED(CONFIG_IPV6)
4107                 } else {
4108                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
4109                                              &dst->remote_ip.sin6.sin6_addr))
4110                                 goto nla_put_failure;
4111 #endif
4112                 }
4113         }
4114
4115         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
4116                 goto nla_put_failure;
4117
4118         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
4119                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
4120                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
4121                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
4122                                 goto nla_put_failure;
4123 #if IS_ENABLED(CONFIG_IPV6)
4124                 } else {
4125                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
4126                                              &vxlan->cfg.saddr.sin6.sin6_addr))
4127                                 goto nla_put_failure;
4128 #endif
4129                 }
4130         }
4131
4132         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
4133             nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT,
4134                        !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) ||
4135             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
4136             nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) ||
4137             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
4138             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
4139                        !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
4140             nla_put_u8(skb, IFLA_VXLAN_PROXY,
4141                        !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
4142             nla_put_u8(skb, IFLA_VXLAN_RSC,
4143                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
4144             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
4145                        !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
4146             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
4147                        !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
4148             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
4149                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
4150             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
4151             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
4152             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
4153             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
4154                        !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
4155             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4156                        !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
4157             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4158                        !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
4159             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
4160                        !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
4161             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
4162                        !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
4163                 goto nla_put_failure;
4164
4165         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
4166                 goto nla_put_failure;
4167
4168         if (vxlan->cfg.flags & VXLAN_F_GBP &&
4169             nla_put_flag(skb, IFLA_VXLAN_GBP))
4170                 goto nla_put_failure;
4171
4172         if (vxlan->cfg.flags & VXLAN_F_GPE &&
4173             nla_put_flag(skb, IFLA_VXLAN_GPE))
4174                 goto nla_put_failure;
4175
4176         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
4177             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
4178                 goto nla_put_failure;
4179
4180         return 0;
4181
4182 nla_put_failure:
4183         return -EMSGSIZE;
4184 }
4185
4186 static struct net *vxlan_get_link_net(const struct net_device *dev)
4187 {
4188         struct vxlan_dev *vxlan = netdev_priv(dev);
4189
4190         return vxlan->net;
4191 }
4192
4193 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
4194         .kind           = "vxlan",
4195         .maxtype        = IFLA_VXLAN_MAX,
4196         .policy         = vxlan_policy,
4197         .priv_size      = sizeof(struct vxlan_dev),
4198         .setup          = vxlan_setup,
4199         .validate       = vxlan_validate,
4200         .newlink        = vxlan_newlink,
4201         .changelink     = vxlan_changelink,
4202         .dellink        = vxlan_dellink,
4203         .get_size       = vxlan_get_size,
4204         .fill_info      = vxlan_fill_info,
4205         .get_link_net   = vxlan_get_link_net,
4206 };
4207
4208 struct net_device *vxlan_dev_create(struct net *net, const char *name,
4209                                     u8 name_assign_type,
4210                                     struct vxlan_config *conf)
4211 {
4212         struct nlattr *tb[IFLA_MAX + 1];
4213         struct net_device *dev;
4214         int err;
4215
4216         memset(&tb, 0, sizeof(tb));
4217
4218         dev = rtnl_create_link(net, name, name_assign_type,
4219                                &vxlan_link_ops, tb, NULL);
4220         if (IS_ERR(dev))
4221                 return dev;
4222
4223         err = __vxlan_dev_create(net, dev, conf, NULL);
4224         if (err < 0) {
4225                 free_netdev(dev);
4226                 return ERR_PTR(err);
4227         }
4228
4229         err = rtnl_configure_link(dev, NULL);
4230         if (err < 0) {
4231                 LIST_HEAD(list_kill);
4232
4233                 vxlan_dellink(dev, &list_kill);
4234                 unregister_netdevice_many(&list_kill);
4235                 return ERR_PTR(err);
4236         }
4237
4238         return dev;
4239 }
4240 EXPORT_SYMBOL_GPL(vxlan_dev_create);
4241
4242 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
4243                                              struct net_device *dev)
4244 {
4245         struct vxlan_dev *vxlan, *next;
4246         LIST_HEAD(list_kill);
4247
4248         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4249                 struct vxlan_rdst *dst = &vxlan->default_dst;
4250
4251                 /* In case we created vxlan device with carrier
4252                  * and we loose the carrier due to module unload
4253                  * we also need to remove vxlan device. In other
4254                  * cases, it's not necessary and remote_ifindex
4255                  * is 0 here, so no matches.
4256                  */
4257                 if (dst->remote_ifindex == dev->ifindex)
4258                         vxlan_dellink(vxlan->dev, &list_kill);
4259         }
4260
4261         unregister_netdevice_many(&list_kill);
4262 }
4263
4264 static int vxlan_netdevice_event(struct notifier_block *unused,
4265                                  unsigned long event, void *ptr)
4266 {
4267         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4268         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
4269
4270         if (event == NETDEV_UNREGISTER) {
4271                 vxlan_offload_rx_ports(dev, false);
4272                 vxlan_handle_lowerdev_unregister(vn, dev);
4273         } else if (event == NETDEV_REGISTER) {
4274                 vxlan_offload_rx_ports(dev, true);
4275         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
4276                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
4277                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
4278         }
4279
4280         return NOTIFY_DONE;
4281 }
4282
4283 static struct notifier_block vxlan_notifier_block __read_mostly = {
4284         .notifier_call = vxlan_netdevice_event,
4285 };
4286
4287 static void
4288 vxlan_fdb_offloaded_set(struct net_device *dev,
4289                         struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4290 {
4291         struct vxlan_dev *vxlan = netdev_priv(dev);
4292         struct vxlan_rdst *rdst;
4293         struct vxlan_fdb *f;
4294         u32 hash_index;
4295
4296         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4297
4298         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4299
4300         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4301         if (!f)
4302                 goto out;
4303
4304         rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip,
4305                                    fdb_info->remote_port,
4306                                    fdb_info->remote_vni,
4307                                    fdb_info->remote_ifindex);
4308         if (!rdst)
4309                 goto out;
4310
4311         rdst->offloaded = fdb_info->offloaded;
4312
4313 out:
4314         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4315 }
4316
4317 static int
4318 vxlan_fdb_external_learn_add(struct net_device *dev,
4319                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4320 {
4321         struct vxlan_dev *vxlan = netdev_priv(dev);
4322         struct netlink_ext_ack *extack;
4323         u32 hash_index;
4324         int err;
4325
4326         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4327         extack = switchdev_notifier_info_to_extack(&fdb_info->info);
4328
4329         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4330         err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip,
4331                                NUD_REACHABLE,
4332                                NLM_F_CREATE | NLM_F_REPLACE,
4333                                fdb_info->remote_port,
4334                                fdb_info->vni,
4335                                fdb_info->remote_vni,
4336                                fdb_info->remote_ifindex,
4337                                NTF_USE | NTF_SELF | NTF_EXT_LEARNED,
4338                                false, extack);
4339         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4340
4341         return err;
4342 }
4343
4344 static int
4345 vxlan_fdb_external_learn_del(struct net_device *dev,
4346                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4347 {
4348         struct vxlan_dev *vxlan = netdev_priv(dev);
4349         struct vxlan_fdb *f;
4350         u32 hash_index;
4351         int err = 0;
4352
4353         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4354         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4355
4356         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4357         if (!f)
4358                 err = -ENOENT;
4359         else if (f->flags & NTF_EXT_LEARNED)
4360                 err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr,
4361                                          fdb_info->remote_ip,
4362                                          fdb_info->remote_port,
4363                                          fdb_info->vni,
4364                                          fdb_info->remote_vni,
4365                                          fdb_info->remote_ifindex,
4366                                          false);
4367
4368         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4369
4370         return err;
4371 }
4372
4373 static int vxlan_switchdev_event(struct notifier_block *unused,
4374                                  unsigned long event, void *ptr)
4375 {
4376         struct net_device *dev = switchdev_notifier_info_to_dev(ptr);
4377         struct switchdev_notifier_vxlan_fdb_info *fdb_info;
4378         int err = 0;
4379
4380         switch (event) {
4381         case SWITCHDEV_VXLAN_FDB_OFFLOADED:
4382                 vxlan_fdb_offloaded_set(dev, ptr);
4383                 break;
4384         case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE:
4385                 fdb_info = ptr;
4386                 err = vxlan_fdb_external_learn_add(dev, fdb_info);
4387                 if (err) {
4388                         err = notifier_from_errno(err);
4389                         break;
4390                 }
4391                 fdb_info->offloaded = true;
4392                 vxlan_fdb_offloaded_set(dev, fdb_info);
4393                 break;
4394         case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE:
4395                 fdb_info = ptr;
4396                 err = vxlan_fdb_external_learn_del(dev, fdb_info);
4397                 if (err) {
4398                         err = notifier_from_errno(err);
4399                         break;
4400                 }
4401                 fdb_info->offloaded = false;
4402                 vxlan_fdb_offloaded_set(dev, fdb_info);
4403                 break;
4404         }
4405
4406         return err;
4407 }
4408
4409 static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = {
4410         .notifier_call = vxlan_switchdev_event,
4411 };
4412
4413 static __net_init int vxlan_init_net(struct net *net)
4414 {
4415         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4416         unsigned int h;
4417
4418         INIT_LIST_HEAD(&vn->vxlan_list);
4419         spin_lock_init(&vn->sock_lock);
4420
4421         for (h = 0; h < PORT_HASH_SIZE; ++h)
4422                 INIT_HLIST_HEAD(&vn->sock_list[h]);
4423
4424         return 0;
4425 }
4426
4427 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
4428 {
4429         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4430         struct vxlan_dev *vxlan, *next;
4431         struct net_device *dev, *aux;
4432         unsigned int h;
4433
4434         for_each_netdev_safe(net, dev, aux)
4435                 if (dev->rtnl_link_ops == &vxlan_link_ops)
4436                         unregister_netdevice_queue(dev, head);
4437
4438         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4439                 /* If vxlan->dev is in the same netns, it has already been added
4440                  * to the list by the previous loop.
4441                  */
4442                 if (!net_eq(dev_net(vxlan->dev), net))
4443                         unregister_netdevice_queue(vxlan->dev, head);
4444         }
4445
4446         for (h = 0; h < PORT_HASH_SIZE; ++h)
4447                 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
4448 }
4449
4450 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
4451 {
4452         struct net *net;
4453         LIST_HEAD(list);
4454
4455         rtnl_lock();
4456         list_for_each_entry(net, net_list, exit_list)
4457                 vxlan_destroy_tunnels(net, &list);
4458
4459         unregister_netdevice_many(&list);
4460         rtnl_unlock();
4461 }
4462
4463 static struct pernet_operations vxlan_net_ops = {
4464         .init = vxlan_init_net,
4465         .exit_batch = vxlan_exit_batch_net,
4466         .id   = &vxlan_net_id,
4467         .size = sizeof(struct vxlan_net),
4468 };
4469
4470 static int __init vxlan_init_module(void)
4471 {
4472         int rc;
4473
4474         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
4475
4476         rc = register_pernet_subsys(&vxlan_net_ops);
4477         if (rc)
4478                 goto out1;
4479
4480         rc = register_netdevice_notifier(&vxlan_notifier_block);
4481         if (rc)
4482                 goto out2;
4483
4484         rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block);
4485         if (rc)
4486                 goto out3;
4487
4488         rc = rtnl_link_register(&vxlan_link_ops);
4489         if (rc)
4490                 goto out4;
4491
4492         return 0;
4493 out4:
4494         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4495 out3:
4496         unregister_netdevice_notifier(&vxlan_notifier_block);
4497 out2:
4498         unregister_pernet_subsys(&vxlan_net_ops);
4499 out1:
4500         return rc;
4501 }
4502 late_initcall(vxlan_init_module);
4503
4504 static void __exit vxlan_cleanup_module(void)
4505 {
4506         rtnl_link_unregister(&vxlan_link_ops);
4507         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4508         unregister_netdevice_notifier(&vxlan_notifier_block);
4509         unregister_pernet_subsys(&vxlan_net_ops);
4510         /* rcu_barrier() is called by netns */
4511 }
4512 module_exit(vxlan_cleanup_module);
4513
4514 MODULE_LICENSE("GPL");
4515 MODULE_VERSION(VXLAN_VERSION);
4516 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
4517 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
4518 MODULE_ALIAS_RTNL_LINK("vxlan");