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(struct vxlan_dev *vxlan,
 797                                          const u8 *mac, __u16 state,
 798                                          __be32 src_vni, __u16 ndm_flags)
 799 {
 800         struct vxlan_fdb *f;
 801
 802         f = kmalloc(sizeof(*f), GFP_ATOMIC);
 803         if (!f)
 804                 return NULL;
 805         f->state = state;
 806         f->flags = ndm_flags;
 807         f->updated = f->used = jiffies;
 808         f->vni = src_vni;
 809         INIT_LIST_HEAD(&f->remotes);
 810         memcpy(f->eth_addr, mac, ETH_ALEN);
 811
 812         return f;
 813 }
 814
 815 static void vxlan_fdb_insert(struct vxlan_dev *vxlan, const u8 *mac,
 816                              __be32 src_vni, struct vxlan_fdb *f)
 817 {
 818         ++vxlan->addrcnt;
 819         hlist_add_head_rcu(&f->hlist,
 820                            vxlan_fdb_head(vxlan, mac, src_vni));
 821 }
 822
 823 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 824                             const u8 *mac, union vxlan_addr *ip,
 825                             __u16 state, __be16 port, __be32 src_vni,
 826                             __be32 vni, __u32 ifindex, __u16 ndm_flags,
 827                             struct vxlan_fdb **fdb)
 828 {
 829         struct vxlan_rdst *rd = NULL;
 830         struct vxlan_fdb *f;
 831         int rc;
 832
 833         if (vxlan->cfg.addrmax &&
 834             vxlan->addrcnt >= vxlan->cfg.addrmax)
 835                 return -ENOSPC;
 836
 837         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 838         f = vxlan_fdb_alloc(vxlan, mac, state, src_vni, ndm_flags);
 839         if (!f)
 840                 return -ENOMEM;
 841
 842         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 843         if (rc < 0) {
 844                 kfree(f);
 845                 return rc;
 846         }
 847
 848         *fdb = f;
 849
 850         return 0;
 851 }
 852
 853 static void __vxlan_fdb_free(struct vxlan_fdb *f)
 854 {
 855         struct vxlan_rdst *rd, *nd;
 856
 857         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 858                 dst_cache_destroy(&rd->dst_cache);
 859                 kfree(rd);
 860         }
 861         kfree(f);
 862 }
 863
 864 static void vxlan_fdb_free(struct rcu_head *head)
 865 {
 866         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 867
 868         __vxlan_fdb_free(f);
 869 }
 870
 871 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 872                               bool do_notify, bool swdev_notify)
 873 {
 874         struct vxlan_rdst *rd;
 875
 876         netdev_dbg(vxlan->dev, "delete %pM\n", f->eth_addr);
 877
 878         --vxlan->addrcnt;
 879         if (do_notify)
 880                 list_for_each_entry(rd, &f->remotes, list)
 881                         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH,
 882                                          swdev_notify, NULL);
 883
 884         hlist_del_rcu(&f->hlist);
 885         call_rcu(&f->rcu, vxlan_fdb_free);
 886 }
 887
 888 static void vxlan_dst_free(struct rcu_head *head)
 889 {
 890         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 891
 892         dst_cache_destroy(&rd->dst_cache);
 893         kfree(rd);
 894 }
 895
 896 static int vxlan_fdb_update_existing(struct vxlan_dev *vxlan,
 897                                      union vxlan_addr *ip,
 898                                      __u16 state, __u16 flags,
 899                                      __be16 port, __be32 vni,
 900                                      __u32 ifindex, __u16 ndm_flags,
 901                                      struct vxlan_fdb *f,
 902                                      bool swdev_notify,
 903                                      struct netlink_ext_ack *extack)
 904 {
 905         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
 906         struct vxlan_rdst *rd = NULL;
 907         struct vxlan_rdst oldrd;
 908         int notify = 0;
 909         int rc = 0;
 910         int err;
 911
 912         /* Do not allow an externally learned entry to take over an entry added
 913          * by the user.
 914          */
 915         if (!(fdb_flags & NTF_EXT_LEARNED) ||
 916             !(f->flags & NTF_VXLAN_ADDED_BY_USER)) {
 917                 if (f->state != state) {
 918                         f->state = state;
 919                         f->updated = jiffies;
 920                         notify = 1;
 921                 }
 922                 if (f->flags != fdb_flags) {
 923                         f->flags = fdb_flags;
 924                         f->updated = jiffies;
 925                         notify = 1;
 926                 }
 927         }
 928
 929         if ((flags & NLM_F_REPLACE)) {
 930                 /* Only change unicasts */
 931                 if (!(is_multicast_ether_addr(f->eth_addr) ||
 932                       is_zero_ether_addr(f->eth_addr))) {
 933                         rc = vxlan_fdb_replace(f, ip, port, vni,
 934                                                ifindex, &oldrd);
 935                         notify |= rc;
 936                 } else {
 937                         return -EOPNOTSUPP;
 938                 }
 939         }
 940         if ((flags & NLM_F_APPEND) &&
 941             (is_multicast_ether_addr(f->eth_addr) ||
 942              is_zero_ether_addr(f->eth_addr))) {
 943                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 944
 945                 if (rc < 0)
 946                         return rc;
 947                 notify |= rc;
 948         }
 949
 950         if (ndm_flags & NTF_USE)
 951                 f->used = jiffies;
 952
 953         if (notify) {
 954                 if (rd == NULL)
 955                         rd = first_remote_rtnl(f);
 956
 957                 err = vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH,
 958                                        swdev_notify, extack);
 959                 if (err)
 960                         goto err_notify;
 961         }
 962
 963         return 0;
 964
 965 err_notify:
 966         if ((flags & NLM_F_REPLACE) && rc)
 967                 *rd = oldrd;
 968         else if ((flags & NLM_F_APPEND) && rc) {
 969                 list_del_rcu(&rd->list);
 970                 call_rcu(&rd->rcu, vxlan_dst_free);
 971         }
 972         return err;
 973 }
 974
 975 static int vxlan_fdb_update_create(struct vxlan_dev *vxlan,
 976                                    const u8 *mac, union vxlan_addr *ip,
 977                                    __u16 state, __u16 flags,
 978                                    __be16 port, __be32 src_vni, __be32 vni,
 979                                    __u32 ifindex, __u16 ndm_flags,
 980                                    bool swdev_notify,
 981                                    struct netlink_ext_ack *extack)
 982 {
 983         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
 984         struct vxlan_fdb *f;
 985         int rc;
 986
 987         /* Disallow replace to add a multicast entry */
 988         if ((flags & NLM_F_REPLACE) &&
 989             (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 990                 return -EOPNOTSUPP;
 991
 992         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 993         rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni,
 994                               vni, ifindex, fdb_flags, &f);
 995         if (rc < 0)
 996                 return rc;
 997
 998         vxlan_fdb_insert(vxlan, mac, src_vni, f);
 999         rc = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH,
1000                               swdev_notify, extack);
1001         if (rc)
1002                 goto err_notify;
1003
1004         return 0;
1005
1006 err_notify:
1007         vxlan_fdb_destroy(vxlan, f, false, false);
1008         return rc;
1009 }
1010
1011 /* Add new entry to forwarding table -- assumes lock held */
1012 static int vxlan_fdb_update(struct vxlan_dev *vxlan,
1013                             const u8 *mac, union vxlan_addr *ip,
1014                             __u16 state, __u16 flags,
1015                             __be16 port, __be32 src_vni, __be32 vni,
1016                             __u32 ifindex, __u16 ndm_flags,
1017                             bool swdev_notify,
1018                             struct netlink_ext_ack *extack)
1019 {
1020         struct vxlan_fdb *f;
1021
1022         f = __vxlan_find_mac(vxlan, mac, src_vni);
1023         if (f) {
1024                 if (flags & NLM_F_EXCL) {
1025                         netdev_dbg(vxlan->dev,
1026                                    "lost race to create %pM\n", mac);
1027                         return -EEXIST;
1028                 }
1029
1030                 return vxlan_fdb_update_existing(vxlan, ip, state, flags, port,
1031                                                  vni, ifindex, ndm_flags, f,
1032                                                  swdev_notify, extack);
1033         } else {
1034                 if (!(flags & NLM_F_CREATE))
1035                         return -ENOENT;
1036
1037                 return vxlan_fdb_update_create(vxlan, mac, ip, state, flags,
1038                                                port, src_vni, vni, ifindex,
1039                                                ndm_flags, swdev_notify, extack);
1040         }
1041 }
1042
1043 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
1044                                   struct vxlan_rdst *rd, bool swdev_notify)
1045 {
1046         list_del_rcu(&rd->list);
1047         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, swdev_notify, NULL);
1048         call_rcu(&rd->rcu, vxlan_dst_free);
1049 }
1050
1051 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
1052                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
1053                            __be32 *vni, u32 *ifindex)
1054 {
1055         struct net *net = dev_net(vxlan->dev);
1056         int err;
1057
1058         if (tb[NDA_DST]) {
1059                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
1060                 if (err)
1061                         return err;
1062         } else {
1063                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
1064                 if (remote->sa.sa_family == AF_INET) {
1065                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
1066                         ip->sa.sa_family = AF_INET;
1067 #if IS_ENABLED(CONFIG_IPV6)
1068                 } else {
1069                         ip->sin6.sin6_addr = in6addr_any;
1070                         ip->sa.sa_family = AF_INET6;
1071 #endif
1072                 }
1073         }
1074
1075         if (tb[NDA_PORT]) {
1076                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
1077                         return -EINVAL;
1078                 *port = nla_get_be16(tb[NDA_PORT]);
1079         } else {
1080                 *port = vxlan->cfg.dst_port;
1081         }
1082
1083         if (tb[NDA_VNI]) {
1084                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
1085                         return -EINVAL;
1086                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1087         } else {
1088                 *vni = vxlan->default_dst.remote_vni;
1089         }
1090
1091         if (tb[NDA_SRC_VNI]) {
1092                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
1093                         return -EINVAL;
1094                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
1095         } else {
1096                 *src_vni = vxlan->default_dst.remote_vni;
1097         }
1098
1099         if (tb[NDA_IFINDEX]) {
1100                 struct net_device *tdev;
1101
1102                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
1103                         return -EINVAL;
1104                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
1105                 tdev = __dev_get_by_index(net, *ifindex);
1106                 if (!tdev)
1107                         return -EADDRNOTAVAIL;
1108         } else {
1109                 *ifindex = 0;
1110         }
1111
1112         return 0;
1113 }
1114
1115 /* Add static entry (via netlink) */
1116 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
1117                          struct net_device *dev,
1118                          const unsigned char *addr, u16 vid, u16 flags,
1119                          struct netlink_ext_ack *extack)
1120 {
1121         struct vxlan_dev *vxlan = netdev_priv(dev);
1122         /* struct net *net = dev_net(vxlan->dev); */
1123         union vxlan_addr ip;
1124         __be16 port;
1125         __be32 src_vni, vni;
1126         u32 ifindex;
1127         u32 hash_index;
1128         int err;
1129
1130         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
1131                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
1132                         ndm->ndm_state);
1133                 return -EINVAL;
1134         }
1135
1136         if (tb[NDA_DST] == NULL)
1137                 return -EINVAL;
1138
1139         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
1140         if (err)
1141                 return err;
1142
1143         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
1144                 return -EAFNOSUPPORT;
1145
1146         hash_index = fdb_head_index(vxlan, addr, src_vni);
1147         spin_lock_bh(&vxlan->hash_lock[hash_index]);
1148         err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags,
1149                                port, src_vni, vni, ifindex,
1150                                ndm->ndm_flags | NTF_VXLAN_ADDED_BY_USER,
1151                                true, extack);
1152         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1153
1154         return err;
1155 }
1156
1157 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
1158                               const unsigned char *addr, union vxlan_addr ip,
1159                               __be16 port, __be32 src_vni, __be32 vni,
1160                               u32 ifindex, bool swdev_notify)
1161 {
1162         struct vxlan_fdb *f;
1163         struct vxlan_rdst *rd = NULL;
1164         int err = -ENOENT;
1165
1166         f = vxlan_find_mac(vxlan, addr, src_vni);
1167         if (!f)
1168                 return err;
1169
1170         if (!vxlan_addr_any(&ip)) {
1171                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
1172                 if (!rd)
1173                         goto out;
1174         }
1175
1176         /* remove a destination if it's not the only one on the list,
1177          * otherwise destroy the fdb entry
1178          */
1179         if (rd && !list_is_singular(&f->remotes)) {
1180                 vxlan_fdb_dst_destroy(vxlan, f, rd, swdev_notify);
1181                 goto out;
1182         }
1183
1184         vxlan_fdb_destroy(vxlan, f, true, swdev_notify);
1185
1186 out:
1187         return 0;
1188 }
1189
1190 /* Delete entry (via netlink) */
1191 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
1192                             struct net_device *dev,
1193                             const unsigned char *addr, u16 vid)
1194 {
1195         struct vxlan_dev *vxlan = netdev_priv(dev);
1196         union vxlan_addr ip;
1197         __be32 src_vni, vni;
1198         __be16 port;
1199         u32 ifindex;
1200         u32 hash_index;
1201         int err;
1202
1203         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
1204         if (err)
1205                 return err;
1206
1207         hash_index = fdb_head_index(vxlan, addr, src_vni);
1208         spin_lock_bh(&vxlan->hash_lock[hash_index]);
1209         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
1210                                  true);
1211         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1212
1213         return err;
1214 }
1215
1216 /* Dump forwarding table */
1217 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1218                           struct net_device *dev,
1219                           struct net_device *filter_dev, int *idx)
1220 {
1221         struct vxlan_dev *vxlan = netdev_priv(dev);
1222         unsigned int h;
1223         int err = 0;
1224
1225         for (h = 0; h < FDB_HASH_SIZE; ++h) {
1226                 struct vxlan_fdb *f;
1227
1228                 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 + 1]) {
1928                         kfree_skb(reply);
1929                         return NULL;
1930                 }
1931                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1932                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1933                         break;
1934                 }
1935         }
1936
1937         /* Ethernet header */
1938         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1939         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1940         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1941         reply->protocol = htons(ETH_P_IPV6);
1942
1943         skb_pull(reply, sizeof(struct ethhdr));
1944         skb_reset_network_header(reply);
1945         skb_put(reply, sizeof(struct ipv6hdr));
1946
1947         /* IPv6 header */
1948
1949         pip6 = ipv6_hdr(reply);
1950         memset(pip6, 0, sizeof(struct ipv6hdr));
1951         pip6->version = 6;
1952         pip6->priority = ipv6_hdr(request)->priority;
1953         pip6->nexthdr = IPPROTO_ICMPV6;
1954         pip6->hop_limit = 255;
1955         pip6->daddr = ipv6_hdr(request)->saddr;
1956         pip6->saddr = *(struct in6_addr *)n->primary_key;
1957
1958         skb_pull(reply, sizeof(struct ipv6hdr));
1959         skb_reset_transport_header(reply);
1960
1961         /* Neighbor Advertisement */
1962         na = skb_put_zero(reply, sizeof(*na) + na_olen);
1963         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1964         na->icmph.icmp6_router = isrouter;
1965         na->icmph.icmp6_override = 1;
1966         na->icmph.icmp6_solicited = 1;
1967         na->target = ns->target;
1968         ether_addr_copy(&na->opt[2], n->ha);
1969         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1970         na->opt[1] = na_olen >> 3;
1971
1972         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1973                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1974                 csum_partial(na, sizeof(*na)+na_olen, 0));
1975
1976         pip6->payload_len = htons(sizeof(*na)+na_olen);
1977
1978         skb_push(reply, sizeof(struct ipv6hdr));
1979
1980         reply->ip_summed = CHECKSUM_UNNECESSARY;
1981
1982         return reply;
1983 }
1984
1985 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1986 {
1987         struct vxlan_dev *vxlan = netdev_priv(dev);
1988         const struct in6_addr *daddr;
1989         const struct ipv6hdr *iphdr;
1990         struct inet6_dev *in6_dev;
1991         struct neighbour *n;
1992         struct nd_msg *msg;
1993
1994         in6_dev = __in6_dev_get(dev);
1995         if (!in6_dev)
1996                 goto out;
1997
1998         iphdr = ipv6_hdr(skb);
1999         daddr = &iphdr->daddr;
2000         msg = (struct nd_msg *)(iphdr + 1);
2001
2002         if (ipv6_addr_loopback(daddr) ||
2003             ipv6_addr_is_multicast(&msg->target))
2004                 goto out;
2005
2006         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
2007
2008         if (n) {
2009                 struct vxlan_fdb *f;
2010                 struct sk_buff *reply;
2011
2012                 if (!(n->nud_state & NUD_CONNECTED)) {
2013                         neigh_release(n);
2014                         goto out;
2015                 }
2016
2017                 f = vxlan_find_mac(vxlan, n->ha, vni);
2018                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2019                         /* bridge-local neighbor */
2020                         neigh_release(n);
2021                         goto out;
2022                 }
2023
2024                 reply = vxlan_na_create(skb, n,
2025                                         !!(f ? f->flags & NTF_ROUTER : 0));
2026
2027                 neigh_release(n);
2028
2029                 if (reply == NULL)
2030                         goto out;
2031
2032                 if (netif_rx_ni(reply) == NET_RX_DROP)
2033                         dev->stats.rx_dropped++;
2034
2035         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2036                 union vxlan_addr ipa = {
2037                         .sin6.sin6_addr = msg->target,
2038                         .sin6.sin6_family = AF_INET6,
2039                 };
2040
2041                 vxlan_ip_miss(dev, &ipa);
2042         }
2043
2044 out:
2045         consume_skb(skb);
2046         return NETDEV_TX_OK;
2047 }
2048 #endif
2049
2050 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
2051 {
2052         struct vxlan_dev *vxlan = netdev_priv(dev);
2053         struct neighbour *n;
2054
2055         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
2056                 return false;
2057
2058         n = NULL;
2059         switch (ntohs(eth_hdr(skb)->h_proto)) {
2060         case ETH_P_IP:
2061         {
2062                 struct iphdr *pip;
2063
2064                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
2065                         return false;
2066                 pip = ip_hdr(skb);
2067                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
2068                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2069                         union vxlan_addr ipa = {
2070                                 .sin.sin_addr.s_addr = pip->daddr,
2071                                 .sin.sin_family = AF_INET,
2072                         };
2073
2074                         vxlan_ip_miss(dev, &ipa);
2075                         return false;
2076                 }
2077
2078                 break;
2079         }
2080 #if IS_ENABLED(CONFIG_IPV6)
2081         case ETH_P_IPV6:
2082         {
2083                 struct ipv6hdr *pip6;
2084
2085                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
2086                         return false;
2087                 pip6 = ipv6_hdr(skb);
2088                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
2089                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2090                         union vxlan_addr ipa = {
2091                                 .sin6.sin6_addr = pip6->daddr,
2092                                 .sin6.sin6_family = AF_INET6,
2093                         };
2094
2095                         vxlan_ip_miss(dev, &ipa);
2096                         return false;
2097                 }
2098
2099                 break;
2100         }
2101 #endif
2102         default:
2103                 return false;
2104         }
2105
2106         if (n) {
2107                 bool diff;
2108
2109                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
2110                 if (diff) {
2111                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
2112                                 dev->addr_len);
2113                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
2114                 }
2115                 neigh_release(n);
2116                 return diff;
2117         }
2118
2119         return false;
2120 }
2121
2122 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
2123                                 struct vxlan_metadata *md)
2124 {
2125         struct vxlanhdr_gbp *gbp;
2126
2127         if (!md->gbp)
2128                 return;
2129
2130         gbp = (struct vxlanhdr_gbp *)vxh;
2131         vxh->vx_flags |= VXLAN_HF_GBP;
2132
2133         if (md->gbp & VXLAN_GBP_DONT_LEARN)
2134                 gbp->dont_learn = 1;
2135
2136         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
2137                 gbp->policy_applied = 1;
2138
2139         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
2140 }
2141
2142 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
2143                                __be16 protocol)
2144 {
2145         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
2146
2147         gpe->np_applied = 1;
2148         gpe->next_protocol = tun_p_from_eth_p(protocol);
2149         if (!gpe->next_protocol)
2150                 return -EPFNOSUPPORT;
2151         return 0;
2152 }
2153
2154 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
2155                            int iphdr_len, __be32 vni,
2156                            struct vxlan_metadata *md, u32 vxflags,
2157                            bool udp_sum)
2158 {
2159         struct vxlanhdr *vxh;
2160         int min_headroom;
2161         int err;
2162         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
2163         __be16 inner_protocol = htons(ETH_P_TEB);
2164
2165         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
2166             skb->ip_summed == CHECKSUM_PARTIAL) {
2167                 int csum_start = skb_checksum_start_offset(skb);
2168
2169                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
2170                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
2171                     (skb->csum_offset == offsetof(struct udphdr, check) ||
2172                      skb->csum_offset == offsetof(struct tcphdr, check)))
2173                         type |= SKB_GSO_TUNNEL_REMCSUM;
2174         }
2175
2176         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
2177                         + VXLAN_HLEN + iphdr_len;
2178
2179         /* Need space for new headers (invalidates iph ptr) */
2180         err = skb_cow_head(skb, min_headroom);
2181         if (unlikely(err))
2182                 return err;
2183
2184         err = iptunnel_handle_offloads(skb, type);
2185         if (err)
2186                 return err;
2187
2188         vxh = __skb_push(skb, sizeof(*vxh));
2189         vxh->vx_flags = VXLAN_HF_VNI;
2190         vxh->vx_vni = vxlan_vni_field(vni);
2191
2192         if (type & SKB_GSO_TUNNEL_REMCSUM) {
2193                 unsigned int start;
2194
2195                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
2196                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
2197                 vxh->vx_flags |= VXLAN_HF_RCO;
2198
2199                 if (!skb_is_gso(skb)) {
2200                         skb->ip_summed = CHECKSUM_NONE;
2201                         skb->encapsulation = 0;
2202                 }
2203         }
2204
2205         if (vxflags & VXLAN_F_GBP)
2206                 vxlan_build_gbp_hdr(vxh, vxflags, md);
2207         if (vxflags & VXLAN_F_GPE) {
2208                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
2209                 if (err < 0)
2210                         return err;
2211                 inner_protocol = skb->protocol;
2212         }
2213
2214         skb_set_inner_protocol(skb, inner_protocol);
2215         return 0;
2216 }
2217
2218 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
2219                                       struct vxlan_sock *sock4,
2220                                       struct sk_buff *skb, int oif, u8 tos,
2221                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
2222                                       struct dst_cache *dst_cache,
2223                                       const struct ip_tunnel_info *info)
2224 {
2225         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2226         struct rtable *rt = NULL;
2227         struct flowi4 fl4;
2228
2229         if (!sock4)
2230                 return ERR_PTR(-EIO);
2231
2232         if (tos && !info)
2233                 use_cache = false;
2234         if (use_cache) {
2235                 rt = dst_cache_get_ip4(dst_cache, saddr);
2236                 if (rt)
2237                         return rt;
2238         }
2239
2240         memset(&fl4, 0, sizeof(fl4));
2241         fl4.flowi4_oif = oif;
2242         fl4.flowi4_tos = RT_TOS(tos);
2243         fl4.flowi4_mark = skb->mark;
2244         fl4.flowi4_proto = IPPROTO_UDP;
2245         fl4.daddr = daddr;
2246         fl4.saddr = *saddr;
2247         fl4.fl4_dport = dport;
2248         fl4.fl4_sport = sport;
2249
2250         rt = ip_route_output_key(vxlan->net, &fl4);
2251         if (!IS_ERR(rt)) {
2252                 if (rt->dst.dev == dev) {
2253                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
2254                         ip_rt_put(rt);
2255                         return ERR_PTR(-ELOOP);
2256                 }
2257
2258                 *saddr = fl4.saddr;
2259                 if (use_cache)
2260                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2261         } else {
2262                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
2263                 return ERR_PTR(-ENETUNREACH);
2264         }
2265         return rt;
2266 }
2267
2268 #if IS_ENABLED(CONFIG_IPV6)
2269 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
2270                                           struct net_device *dev,
2271                                           struct vxlan_sock *sock6,
2272                                           struct sk_buff *skb, int oif, u8 tos,
2273                                           __be32 label,
2274                                           const struct in6_addr *daddr,
2275                                           struct in6_addr *saddr,
2276                                           __be16 dport, __be16 sport,
2277                                           struct dst_cache *dst_cache,
2278                                           const struct ip_tunnel_info *info)
2279 {
2280         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2281         struct dst_entry *ndst;
2282         struct flowi6 fl6;
2283
2284         if (!sock6)
2285                 return ERR_PTR(-EIO);
2286
2287         if (tos && !info)
2288                 use_cache = false;
2289         if (use_cache) {
2290                 ndst = dst_cache_get_ip6(dst_cache, saddr);
2291                 if (ndst)
2292                         return ndst;
2293         }
2294
2295         memset(&fl6, 0, sizeof(fl6));
2296         fl6.flowi6_oif = oif;
2297         fl6.daddr = *daddr;
2298         fl6.saddr = *saddr;
2299         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
2300         fl6.flowi6_mark = skb->mark;
2301         fl6.flowi6_proto = IPPROTO_UDP;
2302         fl6.fl6_dport = dport;
2303         fl6.fl6_sport = sport;
2304
2305         ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk,
2306                                                &fl6, NULL);
2307         if (unlikely(IS_ERR(ndst))) {
2308                 netdev_dbg(dev, "no route to %pI6\n", daddr);
2309                 return ERR_PTR(-ENETUNREACH);
2310         }
2311
2312         if (unlikely(ndst->dev == dev)) {
2313                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
2314                 dst_release(ndst);
2315                 return ERR_PTR(-ELOOP);
2316         }
2317
2318         *saddr = fl6.saddr;
2319         if (use_cache)
2320                 dst_cache_set_ip6(dst_cache, ndst, saddr);
2321         return ndst;
2322 }
2323 #endif
2324
2325 /* Bypass encapsulation if the destination is local */
2326 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2327                                struct vxlan_dev *dst_vxlan, __be32 vni)
2328 {
2329         struct pcpu_sw_netstats *tx_stats, *rx_stats;
2330         union vxlan_addr loopback;
2331         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2332         struct net_device *dev;
2333         int len = skb->len;
2334
2335         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2336         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2337         skb->pkt_type = PACKET_HOST;
2338         skb->encapsulation = 0;
2339         skb->dev = dst_vxlan->dev;
2340         __skb_pull(skb, skb_network_offset(skb));
2341
2342         if (remote_ip->sa.sa_family == AF_INET) {
2343                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2344                 loopback.sa.sa_family =  AF_INET;
2345 #if IS_ENABLED(CONFIG_IPV6)
2346         } else {
2347                 loopback.sin6.sin6_addr = in6addr_loopback;
2348                 loopback.sa.sa_family =  AF_INET6;
2349 #endif
2350         }
2351
2352         rcu_read_lock();
2353         dev = skb->dev;
2354         if (unlikely(!(dev->flags & IFF_UP))) {
2355                 kfree_skb(skb);
2356                 goto drop;
2357         }
2358
2359         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
2360                 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni);
2361
2362         u64_stats_update_begin(&tx_stats->syncp);
2363         tx_stats->tx_packets++;
2364         tx_stats->tx_bytes += len;
2365         u64_stats_update_end(&tx_stats->syncp);
2366
2367         if (netif_rx(skb) == NET_RX_SUCCESS) {
2368                 u64_stats_update_begin(&rx_stats->syncp);
2369                 rx_stats->rx_packets++;
2370                 rx_stats->rx_bytes += len;
2371                 u64_stats_update_end(&rx_stats->syncp);
2372         } else {
2373 drop:
2374                 dev->stats.rx_dropped++;
2375         }
2376         rcu_read_unlock();
2377 }
2378
2379 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2380                                  struct vxlan_dev *vxlan,
2381                                  union vxlan_addr *daddr,
2382                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2383                                  struct dst_entry *dst,
2384                                  u32 rt_flags)
2385 {
2386 #if IS_ENABLED(CONFIG_IPV6)
2387         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2388          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2389          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2390          */
2391         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2392 #endif
2393         /* Bypass encapsulation if the destination is local */
2394         if (rt_flags & RTCF_LOCAL &&
2395             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2396                 struct vxlan_dev *dst_vxlan;
2397
2398                 dst_release(dst);
2399                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2400                                            daddr->sa.sa_family, dst_port,
2401                                            vxlan->cfg.flags);
2402                 if (!dst_vxlan) {
2403                         dev->stats.tx_errors++;
2404                         kfree_skb(skb);
2405
2406                         return -ENOENT;
2407                 }
2408                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2409                 return 1;
2410         }
2411
2412         return 0;
2413 }
2414
2415 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2416                            __be32 default_vni, struct vxlan_rdst *rdst,
2417                            bool did_rsc)
2418 {
2419         struct dst_cache *dst_cache;
2420         struct ip_tunnel_info *info;
2421         struct vxlan_dev *vxlan = netdev_priv(dev);
2422         const struct iphdr *old_iph = ip_hdr(skb);
2423         union vxlan_addr *dst;
2424         union vxlan_addr remote_ip, local_ip;
2425         struct vxlan_metadata _md;
2426         struct vxlan_metadata *md = &_md;
2427         __be16 src_port = 0, dst_port;
2428         struct dst_entry *ndst = NULL;
2429         __be32 vni, label;
2430         __u8 tos, ttl;
2431         int ifindex;
2432         int err;
2433         u32 flags = vxlan->cfg.flags;
2434         bool udp_sum = false;
2435         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2436
2437         info = skb_tunnel_info(skb);
2438
2439         if (rdst) {
2440                 dst = &rdst->remote_ip;
2441                 if (vxlan_addr_any(dst)) {
2442                         if (did_rsc) {
2443                                 /* short-circuited back to local bridge */
2444                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2445                                 return;
2446                         }
2447                         goto drop;
2448                 }
2449
2450                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2451                 vni = (rdst->remote_vni) ? : default_vni;
2452                 ifindex = rdst->remote_ifindex;
2453                 local_ip = vxlan->cfg.saddr;
2454                 dst_cache = &rdst->dst_cache;
2455                 md->gbp = skb->mark;
2456                 if (flags & VXLAN_F_TTL_INHERIT) {
2457                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2458                 } else {
2459                         ttl = vxlan->cfg.ttl;
2460                         if (!ttl && vxlan_addr_multicast(dst))
2461                                 ttl = 1;
2462                 }
2463
2464                 tos = vxlan->cfg.tos;
2465                 if (tos == 1)
2466                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2467
2468                 if (dst->sa.sa_family == AF_INET)
2469                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2470                 else
2471                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2472                 label = vxlan->cfg.label;
2473         } else {
2474                 if (!info) {
2475                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2476                                   dev->name);
2477                         goto drop;
2478                 }
2479                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2480                 if (remote_ip.sa.sa_family == AF_INET) {
2481                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2482                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2483                 } else {
2484                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2485                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2486                 }
2487                 dst = &remote_ip;
2488                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2489                 vni = tunnel_id_to_key32(info->key.tun_id);
2490                 ifindex = 0;
2491                 dst_cache = &info->dst_cache;
2492                 if (info->key.tun_flags & TUNNEL_VXLAN_OPT) {
2493                         if (info->options_len < sizeof(*md))
2494                                 goto drop;
2495                         md = ip_tunnel_info_opts(info);
2496                 }
2497                 ttl = info->key.ttl;
2498                 tos = info->key.tos;
2499                 label = info->key.label;
2500                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2501         }
2502         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2503                                      vxlan->cfg.port_max, true);
2504
2505         rcu_read_lock();
2506         if (dst->sa.sa_family == AF_INET) {
2507                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2508                 struct rtable *rt;
2509                 __be16 df = 0;
2510
2511                 if (!ifindex)
2512                         ifindex = sock4->sock->sk->sk_bound_dev_if;
2513
2514                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2515                                      dst->sin.sin_addr.s_addr,
2516                                      &local_ip.sin.sin_addr.s_addr,
2517                                      dst_port, src_port,
2518                                      dst_cache, info);
2519                 if (IS_ERR(rt)) {
2520                         err = PTR_ERR(rt);
2521                         goto tx_error;
2522                 }
2523
2524                 if (!info) {
2525                         /* Bypass encapsulation if the destination is local */
2526                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2527                                                     dst_port, ifindex, vni,
2528                                                     &rt->dst, rt->rt_flags);
2529                         if (err)
2530                                 goto out_unlock;
2531
2532                         if (vxlan->cfg.df == VXLAN_DF_SET) {
2533                                 df = htons(IP_DF);
2534                         } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) {
2535                                 struct ethhdr *eth = eth_hdr(skb);
2536
2537                                 if (ntohs(eth->h_proto) == ETH_P_IPV6 ||
2538                                     (ntohs(eth->h_proto) == ETH_P_IP &&
2539                                      old_iph->frag_off & htons(IP_DF)))
2540                                         df = htons(IP_DF);
2541                         }
2542                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2543                         df = htons(IP_DF);
2544                 }
2545
2546                 ndst = &rt->dst;
2547                 skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM);
2548
2549                 tos = ip_tunnel_ecn_encap(RT_TOS(tos), old_iph, skb);
2550                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2551                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2552                                       vni, md, flags, udp_sum);
2553                 if (err < 0)
2554                         goto tx_error;
2555
2556                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2557                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2558                                     src_port, dst_port, xnet, !udp_sum);
2559 #if IS_ENABLED(CONFIG_IPV6)
2560         } else {
2561                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2562
2563                 if (!ifindex)
2564                         ifindex = sock6->sock->sk->sk_bound_dev_if;
2565
2566                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2567                                         label, &dst->sin6.sin6_addr,
2568                                         &local_ip.sin6.sin6_addr,
2569                                         dst_port, src_port,
2570                                         dst_cache, info);
2571                 if (IS_ERR(ndst)) {
2572                         err = PTR_ERR(ndst);
2573                         ndst = NULL;
2574                         goto tx_error;
2575                 }
2576
2577                 if (!info) {
2578                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2579
2580                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2581                                                     dst_port, ifindex, vni,
2582                                                     ndst, rt6i_flags);
2583                         if (err)
2584                                 goto out_unlock;
2585                 }
2586
2587                 skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM);
2588
2589                 tos = ip_tunnel_ecn_encap(RT_TOS(tos), old_iph, skb);
2590                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2591                 skb_scrub_packet(skb, xnet);
2592                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2593                                       vni, md, flags, udp_sum);
2594                 if (err < 0)
2595                         goto tx_error;
2596
2597                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2598                                      &local_ip.sin6.sin6_addr,
2599                                      &dst->sin6.sin6_addr, tos, ttl,
2600                                      label, src_port, dst_port, !udp_sum);
2601 #endif
2602         }
2603 out_unlock:
2604         rcu_read_unlock();
2605         return;
2606
2607 drop:
2608         dev->stats.tx_dropped++;
2609         dev_kfree_skb(skb);
2610         return;
2611
2612 tx_error:
2613         rcu_read_unlock();
2614         if (err == -ELOOP)
2615                 dev->stats.collisions++;
2616         else if (err == -ENETUNREACH)
2617                 dev->stats.tx_carrier_errors++;
2618         dst_release(ndst);
2619         dev->stats.tx_errors++;
2620         kfree_skb(skb);
2621 }
2622
2623 /* Transmit local packets over Vxlan
2624  *
2625  * Outer IP header inherits ECN and DF from inner header.
2626  * Outer UDP destination is the VXLAN assigned port.
2627  *           source port is based on hash of flow
2628  */
2629 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2630 {
2631         struct vxlan_dev *vxlan = netdev_priv(dev);
2632         struct vxlan_rdst *rdst, *fdst = NULL;
2633         const struct ip_tunnel_info *info;
2634         bool did_rsc = false;
2635         struct vxlan_fdb *f;
2636         struct ethhdr *eth;
2637         __be32 vni = 0;
2638
2639         info = skb_tunnel_info(skb);
2640
2641         skb_reset_mac_header(skb);
2642
2643         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2644                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2645                     info->mode & IP_TUNNEL_INFO_TX) {
2646                         vni = tunnel_id_to_key32(info->key.tun_id);
2647                 } else {
2648                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2649                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2650                         else
2651                                 kfree_skb(skb);
2652                         return NETDEV_TX_OK;
2653                 }
2654         }
2655
2656         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2657                 eth = eth_hdr(skb);
2658                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2659                         return arp_reduce(dev, skb, vni);
2660 #if IS_ENABLED(CONFIG_IPV6)
2661                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2662                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2663                                             sizeof(struct nd_msg)) &&
2664                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2665                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2666
2667                         if (m->icmph.icmp6_code == 0 &&
2668                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2669                                 return neigh_reduce(dev, skb, vni);
2670                 }
2671 #endif
2672         }
2673
2674         eth = eth_hdr(skb);
2675         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2676         did_rsc = false;
2677
2678         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2679             (ntohs(eth->h_proto) == ETH_P_IP ||
2680              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2681                 did_rsc = route_shortcircuit(dev, skb);
2682                 if (did_rsc)
2683                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2684         }
2685
2686         if (f == NULL) {
2687                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2688                 if (f == NULL) {
2689                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2690                             !is_multicast_ether_addr(eth->h_dest))
2691                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2692
2693                         dev->stats.tx_dropped++;
2694                         kfree_skb(skb);
2695                         return NETDEV_TX_OK;
2696                 }
2697         }
2698
2699         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2700                 struct sk_buff *skb1;
2701
2702                 if (!fdst) {
2703                         fdst = rdst;
2704                         continue;
2705                 }
2706                 skb1 = skb_clone(skb, GFP_ATOMIC);
2707                 if (skb1)
2708                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2709         }
2710
2711         if (fdst)
2712                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2713         else
2714                 kfree_skb(skb);
2715         return NETDEV_TX_OK;
2716 }
2717
2718 /* Walk the forwarding table and purge stale entries */
2719 static void vxlan_cleanup(struct timer_list *t)
2720 {
2721         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2722         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2723         unsigned int h;
2724
2725         if (!netif_running(vxlan->dev))
2726                 return;
2727
2728         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2729                 struct hlist_node *p, *n;
2730
2731                 spin_lock(&vxlan->hash_lock[h]);
2732                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2733                         struct vxlan_fdb *f
2734                                 = container_of(p, struct vxlan_fdb, hlist);
2735                         unsigned long timeout;
2736
2737                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2738                                 continue;
2739
2740                         if (f->flags & NTF_EXT_LEARNED)
2741                                 continue;
2742
2743                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2744                         if (time_before_eq(timeout, jiffies)) {
2745                                 netdev_dbg(vxlan->dev,
2746                                            "garbage collect %pM\n",
2747                                            f->eth_addr);
2748                                 f->state = NUD_STALE;
2749                                 vxlan_fdb_destroy(vxlan, f, true, true);
2750                         } else if (time_before(timeout, next_timer))
2751                                 next_timer = timeout;
2752                 }
2753                 spin_unlock(&vxlan->hash_lock[h]);
2754         }
2755
2756         mod_timer(&vxlan->age_timer, next_timer);
2757 }
2758
2759 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2760 {
2761         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2762
2763         spin_lock(&vn->sock_lock);
2764         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2765 #if IS_ENABLED(CONFIG_IPV6)
2766         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2767 #endif
2768         spin_unlock(&vn->sock_lock);
2769 }
2770
2771 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2772                              struct vxlan_dev_node *node)
2773 {
2774         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2775         __be32 vni = vxlan->default_dst.remote_vni;
2776
2777         node->vxlan = vxlan;
2778         spin_lock(&vn->sock_lock);
2779         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2780         spin_unlock(&vn->sock_lock);
2781 }
2782
2783 /* Setup stats when device is created */
2784 static int vxlan_init(struct net_device *dev)
2785 {
2786         struct vxlan_dev *vxlan = netdev_priv(dev);
2787         int err;
2788
2789         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2790         if (!dev->tstats)
2791                 return -ENOMEM;
2792
2793         err = gro_cells_init(&vxlan->gro_cells, dev);
2794         if (err) {
2795                 free_percpu(dev->tstats);
2796                 return err;
2797         }
2798
2799         return 0;
2800 }
2801
2802 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2803 {
2804         struct vxlan_fdb *f;
2805         u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, vni);
2806
2807         spin_lock_bh(&vxlan->hash_lock[hash_index]);
2808         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2809         if (f)
2810                 vxlan_fdb_destroy(vxlan, f, true, true);
2811         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
2812 }
2813
2814 static void vxlan_uninit(struct net_device *dev)
2815 {
2816         struct vxlan_dev *vxlan = netdev_priv(dev);
2817
2818         gro_cells_destroy(&vxlan->gro_cells);
2819
2820         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2821
2822         free_percpu(dev->tstats);
2823 }
2824
2825 /* Start ageing timer and join group when device is brought up */
2826 static int vxlan_open(struct net_device *dev)
2827 {
2828         struct vxlan_dev *vxlan = netdev_priv(dev);
2829         int ret;
2830
2831         ret = vxlan_sock_add(vxlan);
2832         if (ret < 0)
2833                 return ret;
2834
2835         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2836                 ret = vxlan_igmp_join(vxlan);
2837                 if (ret == -EADDRINUSE)
2838                         ret = 0;
2839                 if (ret) {
2840                         vxlan_sock_release(vxlan);
2841                         return ret;
2842                 }
2843         }
2844
2845         if (vxlan->cfg.age_interval)
2846                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2847
2848         return ret;
2849 }
2850
2851 /* Purge the forwarding table */
2852 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2853 {
2854         unsigned int h;
2855
2856         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2857                 struct hlist_node *p, *n;
2858
2859                 spin_lock_bh(&vxlan->hash_lock[h]);
2860                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2861                         struct vxlan_fdb *f
2862                                 = container_of(p, struct vxlan_fdb, hlist);
2863                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2864                                 continue;
2865                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2866                         if (!is_zero_ether_addr(f->eth_addr))
2867                                 vxlan_fdb_destroy(vxlan, f, true, true);
2868                 }
2869                 spin_unlock_bh(&vxlan->hash_lock[h]);
2870         }
2871 }
2872
2873 /* Cleanup timer and forwarding table on shutdown */
2874 static int vxlan_stop(struct net_device *dev)
2875 {
2876         struct vxlan_dev *vxlan = netdev_priv(dev);
2877         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2878         int ret = 0;
2879
2880         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2881             !vxlan_group_used(vn, vxlan))
2882                 ret = vxlan_igmp_leave(vxlan);
2883
2884         del_timer_sync(&vxlan->age_timer);
2885
2886         vxlan_flush(vxlan, false);
2887         vxlan_sock_release(vxlan);
2888
2889         return ret;
2890 }
2891
2892 /* Stub, nothing needs to be done. */
2893 static void vxlan_set_multicast_list(struct net_device *dev)
2894 {
2895 }
2896
2897 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2898 {
2899         struct vxlan_dev *vxlan = netdev_priv(dev);
2900         struct vxlan_rdst *dst = &vxlan->default_dst;
2901         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2902                                                          dst->remote_ifindex);
2903         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
2904
2905         /* This check is different than dev->max_mtu, because it looks at
2906          * the lowerdev->mtu, rather than the static dev->max_mtu
2907          */
2908         if (lowerdev) {
2909                 int max_mtu = lowerdev->mtu -
2910                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2911                 if (new_mtu > max_mtu)
2912                         return -EINVAL;
2913         }
2914
2915         dev->mtu = new_mtu;
2916         return 0;
2917 }
2918
2919 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2920 {
2921         struct vxlan_dev *vxlan = netdev_priv(dev);
2922         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2923         __be16 sport, dport;
2924
2925         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2926                                   vxlan->cfg.port_max, true);
2927         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2928
2929         if (ip_tunnel_info_af(info) == AF_INET) {
2930                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2931                 struct rtable *rt;
2932
2933                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2934                                      info->key.u.ipv4.dst,
2935                                      &info->key.u.ipv4.src, dport, sport,
2936                                      &info->dst_cache, info);
2937                 if (IS_ERR(rt))
2938                         return PTR_ERR(rt);
2939                 ip_rt_put(rt);
2940         } else {
2941 #if IS_ENABLED(CONFIG_IPV6)
2942                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2943                 struct dst_entry *ndst;
2944
2945                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2946                                         info->key.label, &info->key.u.ipv6.dst,
2947                                         &info->key.u.ipv6.src, dport, sport,
2948                                         &info->dst_cache, info);
2949                 if (IS_ERR(ndst))
2950                         return PTR_ERR(ndst);
2951                 dst_release(ndst);
2952 #else /* !CONFIG_IPV6 */
2953                 return -EPFNOSUPPORT;
2954 #endif
2955         }
2956         info->key.tp_src = sport;
2957         info->key.tp_dst = dport;
2958         return 0;
2959 }
2960
2961 static const struct net_device_ops vxlan_netdev_ether_ops = {
2962         .ndo_init               = vxlan_init,
2963         .ndo_uninit             = vxlan_uninit,
2964         .ndo_open               = vxlan_open,
2965         .ndo_stop               = vxlan_stop,
2966         .ndo_start_xmit         = vxlan_xmit,
2967         .ndo_get_stats64        = ip_tunnel_get_stats64,
2968         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2969         .ndo_change_mtu         = vxlan_change_mtu,
2970         .ndo_validate_addr      = eth_validate_addr,
2971         .ndo_set_mac_address    = eth_mac_addr,
2972         .ndo_fdb_add            = vxlan_fdb_add,
2973         .ndo_fdb_del            = vxlan_fdb_delete,
2974         .ndo_fdb_dump           = vxlan_fdb_dump,
2975         .ndo_fdb_get            = vxlan_fdb_get,
2976         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2977         .ndo_change_proto_down  = dev_change_proto_down_generic,
2978 };
2979
2980 static const struct net_device_ops vxlan_netdev_raw_ops = {
2981         .ndo_init               = vxlan_init,
2982         .ndo_uninit             = vxlan_uninit,
2983         .ndo_open               = vxlan_open,
2984         .ndo_stop               = vxlan_stop,
2985         .ndo_start_xmit         = vxlan_xmit,
2986         .ndo_get_stats64        = ip_tunnel_get_stats64,
2987         .ndo_change_mtu         = vxlan_change_mtu,
2988         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2989 };
2990
2991 /* Info for udev, that this is a virtual tunnel endpoint */
2992 static struct device_type vxlan_type = {
2993         .name = "vxlan",
2994 };
2995
2996 /* Calls the ndo_udp_tunnel_add of the caller in order to
2997  * supply the listening VXLAN udp ports. Callers are expected
2998  * to implement the ndo_udp_tunnel_add.
2999  */
3000 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
3001 {
3002         struct vxlan_sock *vs;
3003         struct net *net = dev_net(dev);
3004         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3005         unsigned int i;
3006
3007         spin_lock(&vn->sock_lock);
3008         for (i = 0; i < PORT_HASH_SIZE; ++i) {
3009                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
3010                         unsigned short type;
3011
3012                         if (vs->flags & VXLAN_F_GPE)
3013                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
3014                         else
3015                                 type = UDP_TUNNEL_TYPE_VXLAN;
3016
3017                         if (push)
3018                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
3019                         else
3020                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
3021                 }
3022         }
3023         spin_unlock(&vn->sock_lock);
3024 }
3025
3026 /* Initialize the device structure. */
3027 static void vxlan_setup(struct net_device *dev)
3028 {
3029         struct vxlan_dev *vxlan = netdev_priv(dev);
3030         unsigned int h;
3031
3032         eth_hw_addr_random(dev);
3033         ether_setup(dev);
3034
3035         dev->needs_free_netdev = true;
3036         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
3037
3038         dev->features   |= NETIF_F_LLTX;
3039         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
3040         dev->features   |= NETIF_F_RXCSUM;
3041         dev->features   |= NETIF_F_GSO_SOFTWARE;
3042
3043         dev->vlan_features = dev->features;
3044         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
3045         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
3046         netif_keep_dst(dev);
3047         dev->priv_flags |= IFF_NO_QUEUE;
3048
3049         /* MTU range: 68 - 65535 */
3050         dev->min_mtu = ETH_MIN_MTU;
3051         dev->max_mtu = ETH_MAX_MTU;
3052
3053         INIT_LIST_HEAD(&vxlan->next);
3054
3055         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
3056
3057         vxlan->dev = dev;
3058
3059         for (h = 0; h < FDB_HASH_SIZE; ++h) {
3060                 spin_lock_init(&vxlan->hash_lock[h]);
3061                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
3062         }
3063 }
3064
3065 static void vxlan_ether_setup(struct net_device *dev)
3066 {
3067         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
3068         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3069         dev->netdev_ops = &vxlan_netdev_ether_ops;
3070 }
3071
3072 static void vxlan_raw_setup(struct net_device *dev)
3073 {
3074         dev->header_ops = NULL;
3075         dev->type = ARPHRD_NONE;
3076         dev->hard_header_len = 0;
3077         dev->addr_len = 0;
3078         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
3079         dev->netdev_ops = &vxlan_netdev_raw_ops;
3080 }
3081
3082 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
3083         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
3084         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
3085         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
3086         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
3087         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
3088         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
3089         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
3090         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
3091         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
3092         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
3093         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
3094         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
3095         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
3096         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
3097         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
3098         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
3099         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
3100         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
3101         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
3102         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
3103         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
3104         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
3105         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
3106         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
3107         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
3108         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
3109         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
3110         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
3111         [IFLA_VXLAN_DF]         = { .type = NLA_U8 },
3112 };
3113
3114 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
3115                           struct netlink_ext_ack *extack)
3116 {
3117         if (tb[IFLA_ADDRESS]) {
3118                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
3119                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3120                                             "Provided link layer address is not Ethernet");
3121                         return -EINVAL;
3122                 }
3123
3124                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
3125                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3126                                             "Provided Ethernet address is not unicast");
3127                         return -EADDRNOTAVAIL;
3128                 }
3129         }
3130
3131         if (tb[IFLA_MTU]) {
3132                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
3133
3134                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
3135                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3136                                             "MTU must be between 68 and 65535");
3137                         return -EINVAL;
3138                 }
3139         }
3140
3141         if (!data) {
3142                 NL_SET_ERR_MSG(extack,
3143                                "Required attributes not provided to perform the operation");
3144                 return -EINVAL;
3145         }
3146
3147         if (data[IFLA_VXLAN_ID]) {
3148                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
3149
3150                 if (id >= VXLAN_N_VID) {
3151                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_ID],
3152                                             "VXLAN ID must be lower than 16777216");
3153                         return -ERANGE;
3154                 }
3155         }
3156
3157         if (data[IFLA_VXLAN_PORT_RANGE]) {
3158                 const struct ifla_vxlan_port_range *p
3159                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3160
3161                 if (ntohs(p->high) < ntohs(p->low)) {
3162                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_PORT_RANGE],
3163                                             "Invalid source port range");
3164                         return -EINVAL;
3165                 }
3166         }
3167
3168         if (data[IFLA_VXLAN_DF]) {
3169                 enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]);
3170
3171                 if (df < 0 || df > VXLAN_DF_MAX) {
3172                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_DF],
3173                                             "Invalid DF attribute");
3174                         return -EINVAL;
3175                 }
3176         }
3177
3178         return 0;
3179 }
3180
3181 static void vxlan_get_drvinfo(struct net_device *netdev,
3182                               struct ethtool_drvinfo *drvinfo)
3183 {
3184         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
3185         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
3186 }
3187
3188 static const struct ethtool_ops vxlan_ethtool_ops = {
3189         .get_drvinfo    = vxlan_get_drvinfo,
3190         .get_link       = ethtool_op_get_link,
3191 };
3192
3193 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
3194                                         __be16 port, u32 flags, int ifindex)
3195 {
3196         struct socket *sock;
3197         struct udp_port_cfg udp_conf;
3198         int err;
3199
3200         memset(&udp_conf, 0, sizeof(udp_conf));
3201
3202         if (ipv6) {
3203                 udp_conf.family = AF_INET6;
3204                 udp_conf.use_udp6_rx_checksums =
3205                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
3206                 udp_conf.ipv6_v6only = 1;
3207         } else {
3208                 udp_conf.family = AF_INET;
3209         }
3210
3211         udp_conf.local_udp_port = port;
3212         udp_conf.bind_ifindex = ifindex;
3213
3214         /* Open UDP socket */
3215         err = udp_sock_create(net, &udp_conf, &sock);
3216         if (err < 0)
3217                 return ERR_PTR(err);
3218
3219         return sock;
3220 }
3221
3222 /* Create new listen socket if needed */
3223 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
3224                                               __be16 port, u32 flags,
3225                                               int ifindex)
3226 {
3227         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3228         struct vxlan_sock *vs;
3229         struct socket *sock;
3230         unsigned int h;
3231         struct udp_tunnel_sock_cfg tunnel_cfg;
3232
3233         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
3234         if (!vs)
3235                 return ERR_PTR(-ENOMEM);
3236
3237         for (h = 0; h < VNI_HASH_SIZE; ++h)
3238                 INIT_HLIST_HEAD(&vs->vni_list[h]);
3239
3240         sock = vxlan_create_sock(net, ipv6, port, flags, ifindex);
3241         if (IS_ERR(sock)) {
3242                 kfree(vs);
3243                 return ERR_CAST(sock);
3244         }
3245
3246         vs->sock = sock;
3247         refcount_set(&vs->refcnt, 1);
3248         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
3249
3250         spin_lock(&vn->sock_lock);
3251         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
3252         udp_tunnel_notify_add_rx_port(sock,
3253                                       (vs->flags & VXLAN_F_GPE) ?
3254                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
3255                                       UDP_TUNNEL_TYPE_VXLAN);
3256         spin_unlock(&vn->sock_lock);
3257
3258         /* Mark socket as an encapsulation socket. */
3259         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
3260         tunnel_cfg.sk_user_data = vs;
3261         tunnel_cfg.encap_type = 1;
3262         tunnel_cfg.encap_rcv = vxlan_rcv;
3263         tunnel_cfg.encap_err_lookup = vxlan_err_lookup;
3264         tunnel_cfg.encap_destroy = NULL;
3265         tunnel_cfg.gro_receive = vxlan_gro_receive;
3266         tunnel_cfg.gro_complete = vxlan_gro_complete;
3267
3268         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
3269
3270         return vs;
3271 }
3272
3273 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
3274 {
3275         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3276         struct vxlan_sock *vs = NULL;
3277         struct vxlan_dev_node *node;
3278         int l3mdev_index = 0;
3279
3280         if (vxlan->cfg.remote_ifindex)
3281                 l3mdev_index = l3mdev_master_upper_ifindex_by_index(
3282                         vxlan->net, vxlan->cfg.remote_ifindex);
3283
3284         if (!vxlan->cfg.no_share) {
3285                 spin_lock(&vn->sock_lock);
3286                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
3287                                      vxlan->cfg.dst_port, vxlan->cfg.flags,
3288                                      l3mdev_index);
3289                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
3290                         spin_unlock(&vn->sock_lock);
3291                         return -EBUSY;
3292                 }
3293                 spin_unlock(&vn->sock_lock);
3294         }
3295         if (!vs)
3296                 vs = vxlan_socket_create(vxlan->net, ipv6,
3297                                          vxlan->cfg.dst_port, vxlan->cfg.flags,
3298                                          l3mdev_index);
3299         if (IS_ERR(vs))
3300                 return PTR_ERR(vs);
3301 #if IS_ENABLED(CONFIG_IPV6)
3302         if (ipv6) {
3303                 rcu_assign_pointer(vxlan->vn6_sock, vs);
3304                 node = &vxlan->hlist6;
3305         } else
3306 #endif
3307         {
3308                 rcu_assign_pointer(vxlan->vn4_sock, vs);
3309                 node = &vxlan->hlist4;
3310         }
3311         vxlan_vs_add_dev(vs, vxlan, node);
3312         return 0;
3313 }
3314
3315 static int vxlan_sock_add(struct vxlan_dev *vxlan)
3316 {
3317         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
3318         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
3319         bool ipv4 = !ipv6 || metadata;
3320         int ret = 0;
3321
3322         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
3323 #if IS_ENABLED(CONFIG_IPV6)
3324         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
3325         if (ipv6) {
3326                 ret = __vxlan_sock_add(vxlan, true);
3327                 if (ret < 0 && ret != -EAFNOSUPPORT)
3328                         ipv4 = false;
3329         }
3330 #endif
3331         if (ipv4)
3332                 ret = __vxlan_sock_add(vxlan, false);
3333         if (ret < 0)
3334                 vxlan_sock_release(vxlan);
3335         return ret;
3336 }
3337
3338 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
3339                                  struct net_device **lower,
3340                                  struct vxlan_dev *old,
3341                                  struct netlink_ext_ack *extack)
3342 {
3343         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
3344         struct vxlan_dev *tmp;
3345         bool use_ipv6 = false;
3346
3347         if (conf->flags & VXLAN_F_GPE) {
3348                 /* For now, allow GPE only together with
3349                  * COLLECT_METADATA. This can be relaxed later; in such
3350                  * case, the other side of the PtP link will have to be
3351                  * provided.
3352                  */
3353                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
3354                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3355                         NL_SET_ERR_MSG(extack,
3356                                        "VXLAN GPE does not support this combination of attributes");
3357                         return -EINVAL;
3358                 }
3359         }
3360
3361         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
3362                 /* Unless IPv6 is explicitly requested, assume IPv4 */
3363                 conf->remote_ip.sa.sa_family = AF_INET;
3364                 conf->saddr.sa.sa_family = AF_INET;
3365         } else if (!conf->remote_ip.sa.sa_family) {
3366                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
3367         } else if (!conf->saddr.sa.sa_family) {
3368                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
3369         }
3370
3371         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
3372                 NL_SET_ERR_MSG(extack,
3373                                "Local and remote address must be from the same family");
3374                 return -EINVAL;
3375         }
3376
3377         if (vxlan_addr_multicast(&conf->saddr)) {
3378                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
3379                 return -EINVAL;
3380         }
3381
3382         if (conf->saddr.sa.sa_family == AF_INET6) {
3383                 if (!IS_ENABLED(CONFIG_IPV6)) {
3384                         NL_SET_ERR_MSG(extack,
3385                                        "IPv6 support not enabled in the kernel");
3386                         return -EPFNOSUPPORT;
3387                 }
3388                 use_ipv6 = true;
3389                 conf->flags |= VXLAN_F_IPV6;
3390
3391                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3392                         int local_type =
3393                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3394                         int remote_type =
3395                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3396
3397                         if (local_type & IPV6_ADDR_LINKLOCAL) {
3398                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3399                                     (remote_type != IPV6_ADDR_ANY)) {
3400                                         NL_SET_ERR_MSG(extack,
3401                                                        "Invalid combination of local and remote address scopes");
3402                                         return -EINVAL;
3403                                 }
3404
3405                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3406                         } else {
3407                                 if (remote_type ==
3408                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3409                                         NL_SET_ERR_MSG(extack,
3410                                                        "Invalid combination of local and remote address scopes");
3411                                         return -EINVAL;
3412                                 }
3413
3414                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3415                         }
3416                 }
3417         }
3418
3419         if (conf->label && !use_ipv6) {
3420                 NL_SET_ERR_MSG(extack,
3421                                "Label attribute only applies to IPv6 VXLAN devices");
3422                 return -EINVAL;
3423         }
3424
3425         if (conf->remote_ifindex) {
3426                 struct net_device *lowerdev;
3427
3428                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3429                 if (!lowerdev) {
3430                         NL_SET_ERR_MSG(extack,
3431                                        "Invalid local interface, device not found");
3432                         return -ENODEV;
3433                 }
3434
3435 #if IS_ENABLED(CONFIG_IPV6)
3436                 if (use_ipv6) {
3437                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3438                         if (idev && idev->cnf.disable_ipv6) {
3439                                 NL_SET_ERR_MSG(extack,
3440                                                "IPv6 support disabled by administrator");
3441                                 return -EPERM;
3442                         }
3443                 }
3444 #endif
3445
3446                 *lower = lowerdev;
3447         } else {
3448                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3449                         NL_SET_ERR_MSG(extack,
3450                                        "Local interface required for multicast remote destination");
3451
3452                         return -EINVAL;
3453                 }
3454
3455 #if IS_ENABLED(CONFIG_IPV6)
3456                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3457                         NL_SET_ERR_MSG(extack,
3458                                        "Local interface required for link-local local/remote addresses");
3459                         return -EINVAL;
3460                 }
3461 #endif
3462
3463                 *lower = NULL;
3464         }
3465
3466         if (!conf->dst_port) {
3467                 if (conf->flags & VXLAN_F_GPE)
3468                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3469                 else
3470                         conf->dst_port = htons(vxlan_port);
3471         }
3472
3473         if (!conf->age_interval)
3474                 conf->age_interval = FDB_AGE_DEFAULT;
3475
3476         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3477                 if (tmp == old)
3478                         continue;
3479
3480                 if (tmp->cfg.vni != conf->vni)
3481                         continue;
3482                 if (tmp->cfg.dst_port != conf->dst_port)
3483                         continue;
3484                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3485                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3486                         continue;
3487
3488                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3489                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3490                         continue;
3491
3492                 NL_SET_ERR_MSG(extack,
3493                                "A VXLAN device with the specified VNI already exists");
3494                 return -EEXIST;
3495         }
3496
3497         return 0;
3498 }
3499
3500 static void vxlan_config_apply(struct net_device *dev,
3501                                struct vxlan_config *conf,
3502                                struct net_device *lowerdev,
3503                                struct net *src_net,
3504                                bool changelink)
3505 {
3506         struct vxlan_dev *vxlan = netdev_priv(dev);
3507         struct vxlan_rdst *dst = &vxlan->default_dst;
3508         unsigned short needed_headroom = ETH_HLEN;
3509         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3510         int max_mtu = ETH_MAX_MTU;
3511
3512         if (!changelink) {
3513                 if (conf->flags & VXLAN_F_GPE)
3514                         vxlan_raw_setup(dev);
3515                 else
3516                         vxlan_ether_setup(dev);
3517
3518                 if (conf->mtu)
3519                         dev->mtu = conf->mtu;
3520
3521                 vxlan->net = src_net;
3522         }
3523
3524         dst->remote_vni = conf->vni;
3525
3526         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3527
3528         if (lowerdev) {
3529                 dst->remote_ifindex = conf->remote_ifindex;
3530
3531                 dev->gso_max_size = lowerdev->gso_max_size;
3532                 dev->gso_max_segs = lowerdev->gso_max_segs;
3533
3534                 needed_headroom = lowerdev->hard_header_len;
3535
3536                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3537                                            VXLAN_HEADROOM);
3538                 if (max_mtu < ETH_MIN_MTU)
3539                         max_mtu = ETH_MIN_MTU;
3540
3541                 if (!changelink && !conf->mtu)
3542                         dev->mtu = max_mtu;
3543         }
3544
3545         if (dev->mtu > max_mtu)
3546                 dev->mtu = max_mtu;
3547
3548         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3549                 needed_headroom += VXLAN6_HEADROOM;
3550         else
3551                 needed_headroom += VXLAN_HEADROOM;
3552         dev->needed_headroom = needed_headroom;
3553
3554         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3555 }
3556
3557 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3558                                struct vxlan_config *conf, bool changelink,
3559                                struct netlink_ext_ack *extack)
3560 {
3561         struct vxlan_dev *vxlan = netdev_priv(dev);
3562         struct net_device *lowerdev;
3563         int ret;
3564
3565         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3566         if (ret)
3567                 return ret;
3568
3569         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3570
3571         return 0;
3572 }
3573
3574 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3575                               struct vxlan_config *conf,
3576                               struct netlink_ext_ack *extack)
3577 {
3578         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3579         struct vxlan_dev *vxlan = netdev_priv(dev);
3580         struct net_device *remote_dev = NULL;
3581         struct vxlan_fdb *f = NULL;
3582         bool unregister = false;
3583         struct vxlan_rdst *dst;
3584         int err;
3585
3586         dst = &vxlan->default_dst;
3587         err = vxlan_dev_configure(net, dev, conf, false, extack);
3588         if (err)
3589                 return err;
3590
3591         dev->ethtool_ops = &vxlan_ethtool_ops;
3592
3593         /* create an fdb entry for a valid default destination */
3594         if (!vxlan_addr_any(&dst->remote_ip)) {
3595                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3596                                        &dst->remote_ip,
3597                                        NUD_REACHABLE | NUD_PERMANENT,
3598                                        vxlan->cfg.dst_port,
3599                                        dst->remote_vni,
3600                                        dst->remote_vni,
3601                                        dst->remote_ifindex,
3602                                        NTF_SELF, &f);
3603                 if (err)
3604                         return err;
3605         }
3606
3607         err = register_netdevice(dev);
3608         if (err)
3609                 goto errout;
3610         unregister = true;
3611
3612         if (dst->remote_ifindex) {
3613                 remote_dev = __dev_get_by_index(net, dst->remote_ifindex);
3614                 if (!remote_dev)
3615                         goto errout;
3616
3617                 err = netdev_upper_dev_link(remote_dev, dev, extack);
3618                 if (err)
3619                         goto errout;
3620         }
3621
3622         err = rtnl_configure_link(dev, NULL);
3623         if (err)
3624                 goto unlink;
3625
3626         if (f) {
3627                 vxlan_fdb_insert(vxlan, all_zeros_mac, dst->remote_vni, f);
3628
3629                 /* notify default fdb entry */
3630                 err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f),
3631                                        RTM_NEWNEIGH, true, extack);
3632                 if (err) {
3633                         vxlan_fdb_destroy(vxlan, f, false, false);
3634                         if (remote_dev)
3635                                 netdev_upper_dev_unlink(remote_dev, dev);
3636                         goto unregister;
3637                 }
3638         }
3639
3640         list_add(&vxlan->next, &vn->vxlan_list);
3641         if (remote_dev)
3642                 dst->remote_dev = remote_dev;
3643         return 0;
3644 unlink:
3645         if (remote_dev)
3646                 netdev_upper_dev_unlink(remote_dev, dev);
3647 errout:
3648         /* unregister_netdevice() destroys the default FDB entry with deletion
3649          * notification. But the addition notification was not sent yet, so
3650          * destroy the entry by hand here.
3651          */
3652         if (f)
3653                 __vxlan_fdb_free(f);
3654 unregister:
3655         if (unregister)
3656                 unregister_netdevice(dev);
3657         return err;
3658 }
3659
3660 /* Set/clear flags based on attribute */
3661 static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[],
3662                           int attrtype, unsigned long mask, bool changelink,
3663                           bool changelink_supported,
3664                           struct netlink_ext_ack *extack)
3665 {
3666         unsigned long flags;
3667
3668         if (!tb[attrtype])
3669                 return 0;
3670
3671         if (changelink && !changelink_supported) {
3672                 vxlan_flag_attr_error(attrtype, extack);
3673                 return -EOPNOTSUPP;
3674         }
3675
3676         if (vxlan_policy[attrtype].type == NLA_FLAG)
3677                 flags = conf->flags | mask;
3678         else if (nla_get_u8(tb[attrtype]))
3679                 flags = conf->flags | mask;
3680         else
3681                 flags = conf->flags & ~mask;
3682
3683         conf->flags = flags;
3684
3685         return 0;
3686 }
3687
3688 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3689                          struct net_device *dev, struct vxlan_config *conf,
3690                          bool changelink, struct netlink_ext_ack *extack)
3691 {
3692         struct vxlan_dev *vxlan = netdev_priv(dev);
3693         int err = 0;
3694
3695         memset(conf, 0, sizeof(*conf));
3696
3697         /* if changelink operation, start with old existing cfg */
3698         if (changelink)
3699                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3700
3701         if (data[IFLA_VXLAN_ID]) {
3702                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3703
3704                 if (changelink && (vni != conf->vni)) {
3705                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI");
3706                         return -EOPNOTSUPP;
3707                 }
3708                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3709         }
3710
3711         if (data[IFLA_VXLAN_GROUP]) {
3712                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) {
3713                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group");
3714                         return -EOPNOTSUPP;
3715                 }
3716
3717                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3718                 conf->remote_ip.sa.sa_family = AF_INET;
3719         } else if (data[IFLA_VXLAN_GROUP6]) {
3720                 if (!IS_ENABLED(CONFIG_IPV6)) {
3721                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel");
3722                         return -EPFNOSUPPORT;
3723                 }
3724
3725                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) {
3726                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group");
3727                         return -EOPNOTSUPP;
3728                 }
3729
3730                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3731                 conf->remote_ip.sa.sa_family = AF_INET6;
3732         }
3733
3734         if (data[IFLA_VXLAN_LOCAL]) {
3735                 if (changelink && (conf->saddr.sa.sa_family != AF_INET)) {
3736                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old");
3737                         return -EOPNOTSUPP;
3738                 }
3739
3740                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3741                 conf->saddr.sa.sa_family = AF_INET;
3742         } else if (data[IFLA_VXLAN_LOCAL6]) {
3743                 if (!IS_ENABLED(CONFIG_IPV6)) {
3744                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel");
3745                         return -EPFNOSUPPORT;
3746                 }
3747
3748                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) {
3749                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old");
3750                         return -EOPNOTSUPP;
3751                 }
3752
3753                 /* TODO: respect scope id */
3754                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3755                 conf->saddr.sa.sa_family = AF_INET6;
3756         }
3757
3758         if (data[IFLA_VXLAN_LINK])
3759                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3760
3761         if (data[IFLA_VXLAN_TOS])
3762                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3763
3764         if (data[IFLA_VXLAN_TTL])
3765                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3766
3767         if (data[IFLA_VXLAN_TTL_INHERIT]) {
3768                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT,
3769                                     VXLAN_F_TTL_INHERIT, changelink, false,
3770                                     extack);
3771                 if (err)
3772                         return err;
3773
3774         }
3775
3776         if (data[IFLA_VXLAN_LABEL])
3777                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3778                              IPV6_FLOWLABEL_MASK;
3779
3780         if (data[IFLA_VXLAN_LEARNING]) {
3781                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING,
3782                                     VXLAN_F_LEARN, changelink, true,
3783                                     extack);
3784                 if (err)
3785                         return err;
3786         } else if (!changelink) {
3787                 /* default to learn on a new device */
3788                 conf->flags |= VXLAN_F_LEARN;
3789         }
3790
3791         if (data[IFLA_VXLAN_AGEING])
3792                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3793
3794         if (data[IFLA_VXLAN_PROXY]) {
3795                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY,
3796                                     VXLAN_F_PROXY, changelink, false,
3797                                     extack);
3798                 if (err)
3799                         return err;
3800         }
3801
3802         if (data[IFLA_VXLAN_RSC]) {
3803                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC,
3804                                     VXLAN_F_RSC, changelink, false,
3805                                     extack);
3806                 if (err)
3807                         return err;
3808         }
3809
3810         if (data[IFLA_VXLAN_L2MISS]) {
3811                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS,
3812                                     VXLAN_F_L2MISS, changelink, false,
3813                                     extack);
3814                 if (err)
3815                         return err;
3816         }
3817
3818         if (data[IFLA_VXLAN_L3MISS]) {
3819                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS,
3820                                     VXLAN_F_L3MISS, changelink, false,
3821                                     extack);
3822                 if (err)
3823                         return err;
3824         }
3825
3826         if (data[IFLA_VXLAN_LIMIT]) {
3827                 if (changelink) {
3828                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT],
3829                                             "Cannot change limit");
3830                         return -EOPNOTSUPP;
3831                 }
3832                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3833         }
3834
3835         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3836                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA,
3837                                     VXLAN_F_COLLECT_METADATA, changelink, false,
3838                                     extack);
3839                 if (err)
3840                         return err;
3841         }
3842
3843         if (data[IFLA_VXLAN_PORT_RANGE]) {
3844                 if (!changelink) {
3845                         const struct ifla_vxlan_port_range *p
3846                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3847                         conf->port_min = ntohs(p->low);
3848                         conf->port_max = ntohs(p->high);
3849                 } else {
3850                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
3851                                             "Cannot change port range");
3852                         return -EOPNOTSUPP;
3853                 }
3854         }
3855
3856         if (data[IFLA_VXLAN_PORT]) {
3857                 if (changelink) {
3858                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT],
3859                                             "Cannot change port");
3860                         return -EOPNOTSUPP;
3861                 }
3862                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3863         }
3864
3865         if (data[IFLA_VXLAN_UDP_CSUM]) {
3866                 if (changelink) {
3867                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM],
3868                                             "Cannot change UDP_CSUM flag");
3869                         return -EOPNOTSUPP;
3870                 }
3871                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3872                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3873         }
3874
3875         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3876                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3877                                     VXLAN_F_UDP_ZERO_CSUM6_TX, changelink,
3878                                     false, extack);
3879                 if (err)
3880                         return err;
3881         }
3882
3883         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3884                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3885                                     VXLAN_F_UDP_ZERO_CSUM6_RX, changelink,
3886                                     false, extack);
3887                 if (err)
3888                         return err;
3889         }
3890
3891         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3892                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX,
3893                                     VXLAN_F_REMCSUM_TX, changelink, false,
3894                                     extack);
3895                 if (err)
3896                         return err;
3897         }
3898
3899         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3900                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX,
3901                                     VXLAN_F_REMCSUM_RX, changelink, false,
3902                                     extack);
3903                 if (err)
3904                         return err;
3905         }
3906
3907         if (data[IFLA_VXLAN_GBP]) {
3908                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP,
3909                                     VXLAN_F_GBP, changelink, false, extack);
3910                 if (err)
3911                         return err;
3912         }
3913
3914         if (data[IFLA_VXLAN_GPE]) {
3915                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE,
3916                                     VXLAN_F_GPE, changelink, false,
3917                                     extack);
3918                 if (err)
3919                         return err;
3920         }
3921
3922         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3923                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL,
3924                                     VXLAN_F_REMCSUM_NOPARTIAL, changelink,
3925                                     false, extack);
3926                 if (err)
3927                         return err;
3928         }
3929
3930         if (tb[IFLA_MTU]) {
3931                 if (changelink) {
3932                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3933                                             "Cannot change mtu");
3934                         return -EOPNOTSUPP;
3935                 }
3936                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3937         }
3938
3939         if (data[IFLA_VXLAN_DF])
3940                 conf->df = nla_get_u8(data[IFLA_VXLAN_DF]);
3941
3942         return 0;
3943 }
3944
3945 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3946                          struct nlattr *tb[], struct nlattr *data[],
3947                          struct netlink_ext_ack *extack)
3948 {
3949         struct vxlan_config conf;
3950         int err;
3951
3952         err = vxlan_nl2conf(tb, data, dev, &conf, false, extack);
3953         if (err)
3954                 return err;
3955
3956         return __vxlan_dev_create(src_net, dev, &conf, extack);
3957 }
3958
3959 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3960                             struct nlattr *data[],
3961                             struct netlink_ext_ack *extack)
3962 {
3963         struct vxlan_dev *vxlan = netdev_priv(dev);
3964         struct net_device *lowerdev;
3965         struct vxlan_config conf;
3966         struct vxlan_rdst *dst;
3967         int err;
3968
3969         dst = &vxlan->default_dst;
3970         err = vxlan_nl2conf(tb, data, dev, &conf, true, extack);
3971         if (err)
3972                 return err;
3973
3974         err = vxlan_config_validate(vxlan->net, &conf, &lowerdev,
3975                                     vxlan, extack);
3976         if (err)
3977                 return err;
3978
3979         if (dst->remote_dev == lowerdev)
3980                 lowerdev = NULL;
3981
3982         err = netdev_adjacent_change_prepare(dst->remote_dev, lowerdev, dev,
3983                                              extack);
3984         if (err)
3985                 return err;
3986
3987         /* handle default dst entry */
3988         if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) {
3989                 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, conf.vni);
3990
3991                 spin_lock_bh(&vxlan->hash_lock[hash_index]);
3992                 if (!vxlan_addr_any(&conf.remote_ip)) {
3993                         err = vxlan_fdb_update(vxlan, all_zeros_mac,
3994                                                &conf.remote_ip,
3995                                                NUD_REACHABLE | NUD_PERMANENT,
3996                                                NLM_F_APPEND | NLM_F_CREATE,
3997                                                vxlan->cfg.dst_port,
3998                                                conf.vni, conf.vni,
3999                                                conf.remote_ifindex,
4000                                                NTF_SELF, true, extack);
4001                         if (err) {
4002                                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4003                                 netdev_adjacent_change_abort(dst->remote_dev,
4004                                                              lowerdev, dev);
4005                                 return err;
4006                         }
4007                 }
4008                 if (!vxlan_addr_any(&dst->remote_ip))
4009                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
4010                                            dst->remote_ip,
4011                                            vxlan->cfg.dst_port,
4012                                            dst->remote_vni,
4013                                            dst->remote_vni,
4014                                            dst->remote_ifindex,
4015                                            true);
4016                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4017         }
4018
4019         if (conf.age_interval != vxlan->cfg.age_interval)
4020                 mod_timer(&vxlan->age_timer, jiffies);
4021
4022         netdev_adjacent_change_commit(dst->remote_dev, lowerdev, dev);
4023         if (lowerdev && lowerdev != dst->remote_dev) {
4024                 dst->remote_dev = lowerdev;
4025                 netdev_update_lockdep_key(lowerdev);
4026         }
4027         vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true);
4028         return 0;
4029 }
4030
4031 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
4032 {
4033         struct vxlan_dev *vxlan = netdev_priv(dev);
4034
4035         vxlan_flush(vxlan, true);
4036
4037         list_del(&vxlan->next);
4038         unregister_netdevice_queue(dev, head);
4039         if (vxlan->default_dst.remote_dev)
4040                 netdev_upper_dev_unlink(vxlan->default_dst.remote_dev, dev);
4041 }
4042
4043 static size_t vxlan_get_size(const struct net_device *dev)
4044 {
4045
4046         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
4047                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
4048                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
4049                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
4050                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
4051                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL_INHERIT */
4052                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
4053                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_DF */
4054                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
4055                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
4056                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
4057                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
4058                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
4059                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
4060                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
4061                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
4062                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
4063                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
4064                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
4065                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
4066                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
4067                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
4068                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
4069                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
4070                 0;
4071 }
4072
4073 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
4074 {
4075         const struct vxlan_dev *vxlan = netdev_priv(dev);
4076         const struct vxlan_rdst *dst = &vxlan->default_dst;
4077         struct ifla_vxlan_port_range ports = {
4078                 .low =  htons(vxlan->cfg.port_min),
4079                 .high = htons(vxlan->cfg.port_max),
4080         };
4081
4082         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
4083                 goto nla_put_failure;
4084
4085         if (!vxlan_addr_any(&dst->remote_ip)) {
4086                 if (dst->remote_ip.sa.sa_family == AF_INET) {
4087                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
4088                                             dst->remote_ip.sin.sin_addr.s_addr))
4089                                 goto nla_put_failure;
4090 #if IS_ENABLED(CONFIG_IPV6)
4091                 } else {
4092                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
4093                                              &dst->remote_ip.sin6.sin6_addr))
4094                                 goto nla_put_failure;
4095 #endif
4096                 }
4097         }
4098
4099         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
4100                 goto nla_put_failure;
4101
4102         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
4103                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
4104                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
4105                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
4106                                 goto nla_put_failure;
4107 #if IS_ENABLED(CONFIG_IPV6)
4108                 } else {
4109                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
4110                                              &vxlan->cfg.saddr.sin6.sin6_addr))
4111                                 goto nla_put_failure;
4112 #endif
4113                 }
4114         }
4115
4116         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
4117             nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT,
4118                        !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) ||
4119             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
4120             nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) ||
4121             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
4122             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
4123                         !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
4124             nla_put_u8(skb, IFLA_VXLAN_PROXY,
4125                         !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
4126             nla_put_u8(skb, IFLA_VXLAN_RSC,
4127                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
4128             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
4129                         !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
4130             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
4131                         !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
4132             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
4133                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
4134             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
4135             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
4136             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
4137             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
4138                         !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
4139             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4140                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
4141             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4142                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
4143             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
4144                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
4145             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
4146                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
4147                 goto nla_put_failure;
4148
4149         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
4150                 goto nla_put_failure;
4151
4152         if (vxlan->cfg.flags & VXLAN_F_GBP &&
4153             nla_put_flag(skb, IFLA_VXLAN_GBP))
4154                 goto nla_put_failure;
4155
4156         if (vxlan->cfg.flags & VXLAN_F_GPE &&
4157             nla_put_flag(skb, IFLA_VXLAN_GPE))
4158                 goto nla_put_failure;
4159
4160         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
4161             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
4162                 goto nla_put_failure;
4163
4164         return 0;
4165
4166 nla_put_failure:
4167         return -EMSGSIZE;
4168 }
4169
4170 static struct net *vxlan_get_link_net(const struct net_device *dev)
4171 {
4172         struct vxlan_dev *vxlan = netdev_priv(dev);
4173
4174         return vxlan->net;
4175 }
4176
4177 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
4178         .kind           = "vxlan",
4179         .maxtype        = IFLA_VXLAN_MAX,
4180         .policy         = vxlan_policy,
4181         .priv_size      = sizeof(struct vxlan_dev),
4182         .setup          = vxlan_setup,
4183         .validate       = vxlan_validate,
4184         .newlink        = vxlan_newlink,
4185         .changelink     = vxlan_changelink,
4186         .dellink        = vxlan_dellink,
4187         .get_size       = vxlan_get_size,
4188         .fill_info      = vxlan_fill_info,
4189         .get_link_net   = vxlan_get_link_net,
4190 };
4191
4192 struct net_device *vxlan_dev_create(struct net *net, const char *name,
4193                                     u8 name_assign_type,
4194                                     struct vxlan_config *conf)
4195 {
4196         struct nlattr *tb[IFLA_MAX + 1];
4197         struct net_device *dev;
4198         int err;
4199
4200         memset(&tb, 0, sizeof(tb));
4201
4202         dev = rtnl_create_link(net, name, name_assign_type,
4203                                &vxlan_link_ops, tb, NULL);
4204         if (IS_ERR(dev))
4205                 return dev;
4206
4207         err = __vxlan_dev_create(net, dev, conf, NULL);
4208         if (err < 0) {
4209                 free_netdev(dev);
4210                 return ERR_PTR(err);
4211         }
4212
4213         err = rtnl_configure_link(dev, NULL);
4214         if (err < 0) {
4215                 LIST_HEAD(list_kill);
4216
4217                 vxlan_dellink(dev, &list_kill);
4218                 unregister_netdevice_many(&list_kill);
4219                 return ERR_PTR(err);
4220         }
4221
4222         return dev;
4223 }
4224 EXPORT_SYMBOL_GPL(vxlan_dev_create);
4225
4226 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
4227                                              struct net_device *dev)
4228 {
4229         struct vxlan_dev *vxlan, *next;
4230         LIST_HEAD(list_kill);
4231
4232         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4233                 struct vxlan_rdst *dst = &vxlan->default_dst;
4234
4235                 /* In case we created vxlan device with carrier
4236                  * and we loose the carrier due to module unload
4237                  * we also need to remove vxlan device. In other
4238                  * cases, it's not necessary and remote_ifindex
4239                  * is 0 here, so no matches.
4240                  */
4241                 if (dst->remote_ifindex == dev->ifindex)
4242                         vxlan_dellink(vxlan->dev, &list_kill);
4243         }
4244
4245         unregister_netdevice_many(&list_kill);
4246 }
4247
4248 static int vxlan_netdevice_event(struct notifier_block *unused,
4249                                  unsigned long event, void *ptr)
4250 {
4251         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4252         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
4253
4254         if (event == NETDEV_UNREGISTER) {
4255                 vxlan_offload_rx_ports(dev, false);
4256                 vxlan_handle_lowerdev_unregister(vn, dev);
4257         } else if (event == NETDEV_REGISTER) {
4258                 vxlan_offload_rx_ports(dev, true);
4259         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
4260                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
4261                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
4262         }
4263
4264         return NOTIFY_DONE;
4265 }
4266
4267 static struct notifier_block vxlan_notifier_block __read_mostly = {
4268         .notifier_call = vxlan_netdevice_event,
4269 };
4270
4271 static void
4272 vxlan_fdb_offloaded_set(struct net_device *dev,
4273                         struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4274 {
4275         struct vxlan_dev *vxlan = netdev_priv(dev);
4276         struct vxlan_rdst *rdst;
4277         struct vxlan_fdb *f;
4278         u32 hash_index;
4279
4280         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4281
4282         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4283
4284         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4285         if (!f)
4286                 goto out;
4287
4288         rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip,
4289                                    fdb_info->remote_port,
4290                                    fdb_info->remote_vni,
4291                                    fdb_info->remote_ifindex);
4292         if (!rdst)
4293                 goto out;
4294
4295         rdst->offloaded = fdb_info->offloaded;
4296
4297 out:
4298         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4299 }
4300
4301 static int
4302 vxlan_fdb_external_learn_add(struct net_device *dev,
4303                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4304 {
4305         struct vxlan_dev *vxlan = netdev_priv(dev);
4306         struct netlink_ext_ack *extack;
4307         u32 hash_index;
4308         int err;
4309
4310         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4311         extack = switchdev_notifier_info_to_extack(&fdb_info->info);
4312
4313         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4314         err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip,
4315                                NUD_REACHABLE,
4316                                NLM_F_CREATE | NLM_F_REPLACE,
4317                                fdb_info->remote_port,
4318                                fdb_info->vni,
4319                                fdb_info->remote_vni,
4320                                fdb_info->remote_ifindex,
4321                                NTF_USE | NTF_SELF | NTF_EXT_LEARNED,
4322                                false, extack);
4323         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4324
4325         return err;
4326 }
4327
4328 static int
4329 vxlan_fdb_external_learn_del(struct net_device *dev,
4330                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4331 {
4332         struct vxlan_dev *vxlan = netdev_priv(dev);
4333         struct vxlan_fdb *f;
4334         u32 hash_index;
4335         int err = 0;
4336
4337         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4338         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4339
4340         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4341         if (!f)
4342                 err = -ENOENT;
4343         else if (f->flags & NTF_EXT_LEARNED)
4344                 err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr,
4345                                          fdb_info->remote_ip,
4346                                          fdb_info->remote_port,
4347                                          fdb_info->vni,
4348                                          fdb_info->remote_vni,
4349                                          fdb_info->remote_ifindex,
4350                                          false);
4351
4352         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4353
4354         return err;
4355 }
4356
4357 static int vxlan_switchdev_event(struct notifier_block *unused,
4358                                  unsigned long event, void *ptr)
4359 {
4360         struct net_device *dev = switchdev_notifier_info_to_dev(ptr);
4361         struct switchdev_notifier_vxlan_fdb_info *fdb_info;
4362         int err = 0;
4363
4364         switch (event) {
4365         case SWITCHDEV_VXLAN_FDB_OFFLOADED:
4366                 vxlan_fdb_offloaded_set(dev, ptr);
4367                 break;
4368         case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE:
4369                 fdb_info = ptr;
4370                 err = vxlan_fdb_external_learn_add(dev, fdb_info);
4371                 if (err) {
4372                         err = notifier_from_errno(err);
4373                         break;
4374                 }
4375                 fdb_info->offloaded = true;
4376                 vxlan_fdb_offloaded_set(dev, fdb_info);
4377                 break;
4378         case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE:
4379                 fdb_info = ptr;
4380                 err = vxlan_fdb_external_learn_del(dev, fdb_info);
4381                 if (err) {
4382                         err = notifier_from_errno(err);
4383                         break;
4384                 }
4385                 fdb_info->offloaded = false;
4386                 vxlan_fdb_offloaded_set(dev, fdb_info);
4387                 break;
4388         }
4389
4390         return err;
4391 }
4392
4393 static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = {
4394         .notifier_call = vxlan_switchdev_event,
4395 };
4396
4397 static __net_init int vxlan_init_net(struct net *net)
4398 {
4399         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4400         unsigned int h;
4401
4402         INIT_LIST_HEAD(&vn->vxlan_list);
4403         spin_lock_init(&vn->sock_lock);
4404
4405         for (h = 0; h < PORT_HASH_SIZE; ++h)
4406                 INIT_HLIST_HEAD(&vn->sock_list[h]);
4407
4408         return 0;
4409 }
4410
4411 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
4412 {
4413         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4414         struct vxlan_dev *vxlan, *next;
4415         struct net_device *dev, *aux;
4416         unsigned int h;
4417
4418         for_each_netdev_safe(net, dev, aux)
4419                 if (dev->rtnl_link_ops == &vxlan_link_ops)
4420                         unregister_netdevice_queue(dev, head);
4421
4422         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4423                 /* If vxlan->dev is in the same netns, it has already been added
4424                  * to the list by the previous loop.
4425                  */
4426                 if (!net_eq(dev_net(vxlan->dev), net))
4427                         unregister_netdevice_queue(vxlan->dev, head);
4428         }
4429
4430         for (h = 0; h < PORT_HASH_SIZE; ++h)
4431                 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
4432 }
4433
4434 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
4435 {
4436         struct net *net;
4437         LIST_HEAD(list);
4438
4439         rtnl_lock();
4440         list_for_each_entry(net, net_list, exit_list)
4441                 vxlan_destroy_tunnels(net, &list);
4442
4443         unregister_netdevice_many(&list);
4444         rtnl_unlock();
4445 }
4446
4447 static struct pernet_operations vxlan_net_ops = {
4448         .init = vxlan_init_net,
4449         .exit_batch = vxlan_exit_batch_net,
4450         .id   = &vxlan_net_id,
4451         .size = sizeof(struct vxlan_net),
4452 };
4453
4454 static int __init vxlan_init_module(void)
4455 {
4456         int rc;
4457
4458         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
4459
4460         rc = register_pernet_subsys(&vxlan_net_ops);
4461         if (rc)
4462                 goto out1;
4463
4464         rc = register_netdevice_notifier(&vxlan_notifier_block);
4465         if (rc)
4466                 goto out2;
4467
4468         rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block);
4469         if (rc)
4470                 goto out3;
4471
4472         rc = rtnl_link_register(&vxlan_link_ops);
4473         if (rc)
4474                 goto out4;
4475
4476         return 0;
4477 out4:
4478         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4479 out3:
4480         unregister_netdevice_notifier(&vxlan_notifier_block);
4481 out2:
4482         unregister_pernet_subsys(&vxlan_net_ops);
4483 out1:
4484         return rc;
4485 }
4486 late_initcall(vxlan_init_module);
4487
4488 static void __exit vxlan_cleanup_module(void)
4489 {
4490         rtnl_link_unregister(&vxlan_link_ops);
4491         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4492         unregister_netdevice_notifier(&vxlan_notifier_block);
4493         unregister_pernet_subsys(&vxlan_net_ops);
4494         /* rcu_barrier() is called by netns */
4495 }
4496 module_exit(vxlan_cleanup_module);
4497
4498 MODULE_LICENSE("GPL");
4499 MODULE_VERSION(VXLAN_VERSION);
4500 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
4501 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
4502 MODULE_ALIAS_RTNL_LINK("vxlan");