drivers/net/vxlan.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * VXLAN: Virtual eXtensible Local Area Network
   4  *
   5  * Copyright (c) 2012-2013 Vyatta Inc.
   6  */
   7
   8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   9
  10 #include <linux/kernel.h>
  11 #include <linux/module.h>
  12 #include <linux/errno.h>
  13 #include <linux/slab.h>
  14 #include <linux/udp.h>
  15 #include <linux/igmp.h>
  16 #include <linux/if_ether.h>
  17 #include <linux/ethtool.h>
  18 #include <net/arp.h>
  19 #include <net/ndisc.h>
  20 #include <net/ipv6_stubs.h>
  21 #include <net/ip.h>
  22 #include <net/icmp.h>
  23 #include <net/rtnetlink.h>
  24 #include <net/inet_ecn.h>
  25 #include <net/net_namespace.h>
  26 #include <net/netns/generic.h>
  27 #include <net/tun_proto.h>
  28 #include <net/vxlan.h>
  29
  30 #if IS_ENABLED(CONFIG_IPV6)
  31 #include <net/ip6_tunnel.h>
  32 #include <net/ip6_checksum.h>
  33 #endif
  34
  35 #define VXLAN_VERSION   "0.1"
  36
  37 #define PORT_HASH_BITS  8
  38 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  39 #define FDB_AGE_DEFAULT 300 /* 5 min */
  40 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  41
  42 /* UDP port for VXLAN traffic.
  43  * The IANA assigned port is 4789, but the Linux default is 8472
  44  * for compatibility with early adopters.
  45  */
  46 static unsigned short vxlan_port __read_mostly = 8472;
  47 module_param_named(udp_port, vxlan_port, ushort, 0444);
  48 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  49
  50 static bool log_ecn_error = true;
  51 module_param(log_ecn_error, bool, 0644);
  52 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  53
  54 static unsigned int vxlan_net_id;
  55 static struct rtnl_link_ops vxlan_link_ops;
  56
  57 static const u8 all_zeros_mac[ETH_ALEN + 2];
  58
  59 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  60
  61 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan);
  62
  63 /* per-network namespace private data for this module */
  64 struct vxlan_net {
  65         struct list_head  vxlan_list;
  66         struct hlist_head sock_list[PORT_HASH_SIZE];
  67         spinlock_t        sock_lock;
  68 };
  69
  70 /* Forwarding table entry */
  71 struct vxlan_fdb {
  72         struct hlist_node hlist;        /* linked list of entries */
  73         struct rcu_head   rcu;
  74         unsigned long     updated;      /* jiffies */
  75         unsigned long     used;
  76         struct list_head  remotes;
  77         u8                eth_addr[ETH_ALEN];
  78         u16               state;        /* see ndm_state */
  79         __be32            vni;
  80         u16               flags;        /* see ndm_flags and below */
  81 };
  82
  83 #define NTF_VXLAN_ADDED_BY_USER 0x100
  84
  85 /* salt for hash table */
  86 static u32 vxlan_salt __read_mostly;
  87
  88 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  89 {
  90         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  91                ip_tunnel_collect_metadata();
  92 }
  93
  94 #if IS_ENABLED(CONFIG_IPV6)
  95 static inline
  96 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
  97 {
  98         if (a->sa.sa_family != b->sa.sa_family)
  99                 return false;
 100         if (a->sa.sa_family == AF_INET6)
 101                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 102         else
 103                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 104 }
 105
 106 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 107 {
 108         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 109                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 110                 ip->sa.sa_family = AF_INET6;
 111                 return 0;
 112         } else if (nla_len(nla) >= sizeof(__be32)) {
 113                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 114                 ip->sa.sa_family = AF_INET;
 115                 return 0;
 116         } else {
 117                 return -EAFNOSUPPORT;
 118         }
 119 }
 120
 121 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 122                               const union vxlan_addr *ip)
 123 {
 124         if (ip->sa.sa_family == AF_INET6)
 125                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 126         else
 127                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 128 }
 129
 130 #else /* !CONFIG_IPV6 */
 131
 132 static inline
 133 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 134 {
 135         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 136 }
 137
 138 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 139 {
 140         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 141                 return -EAFNOSUPPORT;
 142         } else if (nla_len(nla) >= sizeof(__be32)) {
 143                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 144                 ip->sa.sa_family = AF_INET;
 145                 return 0;
 146         } else {
 147                 return -EAFNOSUPPORT;
 148         }
 149 }
 150
 151 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 152                               const union vxlan_addr *ip)
 153 {
 154         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 155 }
 156 #endif
 157
 158 /* Virtual Network hash table head */
 159 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 160 {
 161         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 162 }
 163
 164 /* Socket hash table head */
 165 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 166 {
 167         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 168
 169         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 170 }
 171
 172 /* First remote destination for a forwarding entry.
 173  * Guaranteed to be non-NULL because remotes are never deleted.
 174  */
 175 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 176 {
 177         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 178 }
 179
 180 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 181 {
 182         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 183 }
 184
 185 /* Find VXLAN socket based on network namespace, address family and UDP port
 186  * and enabled unshareable flags.
 187  */
 188 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 189                                           __be16 port, u32 flags, int ifindex)
 190 {
 191         struct vxlan_sock *vs;
 192
 193         flags &= VXLAN_F_RCV_FLAGS;
 194
 195         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 196                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 197                     vxlan_get_sk_family(vs) == family &&
 198                     vs->flags == flags &&
 199                     vs->sock->sk->sk_bound_dev_if == ifindex)
 200                         return vs;
 201         }
 202         return NULL;
 203 }
 204
 205 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, int ifindex,
 206                                            __be32 vni)
 207 {
 208         struct vxlan_dev_node *node;
 209
 210         /* For flow based devices, map all packets to VNI 0 */
 211         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 212                 vni = 0;
 213
 214         hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) {
 215                 if (node->vxlan->default_dst.remote_vni != vni)
 216                         continue;
 217
 218                 if (IS_ENABLED(CONFIG_IPV6)) {
 219                         const struct vxlan_config *cfg = &node->vxlan->cfg;
 220
 221                         if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) &&
 222                             cfg->remote_ifindex != ifindex)
 223                                 continue;
 224                 }
 225
 226                 return node->vxlan;
 227         }
 228
 229         return NULL;
 230 }
 231
 232 /* Look up VNI in a per net namespace table */
 233 static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex,
 234                                         __be32 vni, sa_family_t family,
 235                                         __be16 port, u32 flags)
 236 {
 237         struct vxlan_sock *vs;
 238
 239         vs = vxlan_find_sock(net, family, port, flags, ifindex);
 240         if (!vs)
 241                 return NULL;
 242
 243         return vxlan_vs_find_vni(vs, ifindex, vni);
 244 }
 245
 246 /* Fill in neighbour message in skbuff. */
 247 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 248                           const struct vxlan_fdb *fdb,
 249                           u32 portid, u32 seq, int type, unsigned int flags,
 250                           const struct vxlan_rdst *rdst)
 251 {
 252         unsigned long now = jiffies;
 253         struct nda_cacheinfo ci;
 254         struct nlmsghdr *nlh;
 255         struct ndmsg *ndm;
 256         bool send_ip, send_eth;
 257
 258         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 259         if (nlh == NULL)
 260                 return -EMSGSIZE;
 261
 262         ndm = nlmsg_data(nlh);
 263         memset(ndm, 0, sizeof(*ndm));
 264
 265         send_eth = send_ip = true;
 266
 267         if (type == RTM_GETNEIGH) {
 268                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 269                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 270                 ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
 271         } else
 272                 ndm->ndm_family = AF_BRIDGE;
 273         ndm->ndm_state = fdb->state;
 274         ndm->ndm_ifindex = vxlan->dev->ifindex;
 275         ndm->ndm_flags = fdb->flags;
 276         if (rdst->offloaded)
 277                 ndm->ndm_flags |= NTF_OFFLOADED;
 278         ndm->ndm_type = RTN_UNICAST;
 279
 280         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 281             nla_put_s32(skb, NDA_LINK_NETNSID,
 282                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 283                 goto nla_put_failure;
 284
 285         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 286                 goto nla_put_failure;
 287
 288         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 289                 goto nla_put_failure;
 290
 291         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 292             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 293                 goto nla_put_failure;
 294         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 295             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 296                 goto nla_put_failure;
 297         if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
 298             nla_put_u32(skb, NDA_SRC_VNI,
 299                         be32_to_cpu(fdb->vni)))
 300                 goto nla_put_failure;
 301         if (rdst->remote_ifindex &&
 302             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 303                 goto nla_put_failure;
 304
 305         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 306         ci.ndm_confirmed = 0;
 307         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 308         ci.ndm_refcnt    = 0;
 309
 310         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 311                 goto nla_put_failure;
 312
 313         nlmsg_end(skb, nlh);
 314         return 0;
 315
 316 nla_put_failure:
 317         nlmsg_cancel(skb, nlh);
 318         return -EMSGSIZE;
 319 }
 320
 321 static inline size_t vxlan_nlmsg_size(void)
 322 {
 323         return NLMSG_ALIGN(sizeof(struct ndmsg))
 324                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 325                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 326                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 327                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 328                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 329                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 330                 + nla_total_size(sizeof(struct nda_cacheinfo));
 331 }
 332
 333 static void __vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 334                                struct vxlan_rdst *rd, int type)
 335 {
 336         struct net *net = dev_net(vxlan->dev);
 337         struct sk_buff *skb;
 338         int err = -ENOBUFS;
 339
 340         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 341         if (skb == NULL)
 342                 goto errout;
 343
 344         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 345         if (err < 0) {
 346                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 347                 WARN_ON(err == -EMSGSIZE);
 348                 kfree_skb(skb);
 349                 goto errout;
 350         }
 351
 352         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 353         return;
 354 errout:
 355         if (err < 0)
 356                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 357 }
 358
 359 static void vxlan_fdb_switchdev_notifier_info(const struct vxlan_dev *vxlan,
 360                             const struct vxlan_fdb *fdb,
 361                             const struct vxlan_rdst *rd,
 362                             struct netlink_ext_ack *extack,
 363                             struct switchdev_notifier_vxlan_fdb_info *fdb_info)
 364 {
 365         fdb_info->info.dev = vxlan->dev;
 366         fdb_info->info.extack = extack;
 367         fdb_info->remote_ip = rd->remote_ip;
 368         fdb_info->remote_port = rd->remote_port;
 369         fdb_info->remote_vni = rd->remote_vni;
 370         fdb_info->remote_ifindex = rd->remote_ifindex;
 371         memcpy(fdb_info->eth_addr, fdb->eth_addr, ETH_ALEN);
 372         fdb_info->vni = fdb->vni;
 373         fdb_info->offloaded = rd->offloaded;
 374         fdb_info->added_by_user = fdb->flags & NTF_VXLAN_ADDED_BY_USER;
 375 }
 376
 377 static int vxlan_fdb_switchdev_call_notifiers(struct vxlan_dev *vxlan,
 378                                               struct vxlan_fdb *fdb,
 379                                               struct vxlan_rdst *rd,
 380                                               bool adding,
 381                                               struct netlink_ext_ack *extack)
 382 {
 383         struct switchdev_notifier_vxlan_fdb_info info;
 384         enum switchdev_notifier_type notifier_type;
 385         int ret;
 386
 387         if (WARN_ON(!rd))
 388                 return 0;
 389
 390         notifier_type = adding ? SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE
 391                                : SWITCHDEV_VXLAN_FDB_DEL_TO_DEVICE;
 392         vxlan_fdb_switchdev_notifier_info(vxlan, fdb, rd, NULL, &info);
 393         ret = call_switchdev_notifiers(notifier_type, vxlan->dev,
 394                                        &info.info, extack);
 395         return notifier_to_errno(ret);
 396 }
 397
 398 static int vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 399                             struct vxlan_rdst *rd, int type, bool swdev_notify,
 400                             struct netlink_ext_ack *extack)
 401 {
 402         int err;
 403
 404         if (swdev_notify) {
 405                 switch (type) {
 406                 case RTM_NEWNEIGH:
 407                         err = vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
 408                                                                  true, extack);
 409                         if (err)
 410                                 return err;
 411                         break;
 412                 case RTM_DELNEIGH:
 413                         vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
 414                                                            false, extack);
 415                         break;
 416                 }
 417         }
 418
 419         __vxlan_fdb_notify(vxlan, fdb, rd, type);
 420         return 0;
 421 }
 422
 423 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 424 {
 425         struct vxlan_dev *vxlan = netdev_priv(dev);
 426         struct vxlan_fdb f = {
 427                 .state = NUD_STALE,
 428         };
 429         struct vxlan_rdst remote = {
 430                 .remote_ip = *ipa, /* goes to NDA_DST */
 431                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 432         };
 433
 434         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
 435 }
 436
 437 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 438 {
 439         struct vxlan_fdb f = {
 440                 .state = NUD_STALE,
 441         };
 442         struct vxlan_rdst remote = { };
 443
 444         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 445
 446         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
 447 }
 448
 449 /* Hash Ethernet address */
 450 static u32 eth_hash(const unsigned char *addr)
 451 {
 452         u64 value = get_unaligned((u64 *)addr);
 453
 454         /* only want 6 bytes */
 455 #ifdef __BIG_ENDIAN
 456         value >>= 16;
 457 #else
 458         value <<= 16;
 459 #endif
 460         return hash_64(value, FDB_HASH_BITS);
 461 }
 462
 463 static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
 464 {
 465         /* use 1 byte of OUI and 3 bytes of NIC */
 466         u32 key = get_unaligned((u32 *)(addr + 2));
 467
 468         return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
 469 }
 470
 471 /* Hash chain to use given mac address */
 472 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 473                                                 const u8 *mac, __be32 vni)
 474 {
 475         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)
 476                 return &vxlan->fdb_head[eth_vni_hash(mac, vni)];
 477         else
 478                 return &vxlan->fdb_head[eth_hash(mac)];
 479 }
 480
 481 /* Look up Ethernet address in forwarding table */
 482 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 483                                           const u8 *mac, __be32 vni)
 484 {
 485         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
 486         struct vxlan_fdb *f;
 487
 488         hlist_for_each_entry_rcu(f, head, hlist) {
 489                 if (ether_addr_equal(mac, f->eth_addr)) {
 490                         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
 491                                 if (vni == f->vni)
 492                                         return f;
 493                         } else {
 494                                 return f;
 495                         }
 496                 }
 497         }
 498
 499         return NULL;
 500 }
 501
 502 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 503                                         const u8 *mac, __be32 vni)
 504 {
 505         struct vxlan_fdb *f;
 506
 507         f = __vxlan_find_mac(vxlan, mac, vni);
 508         if (f && f->used != jiffies)
 509                 f->used = jiffies;
 510
 511         return f;
 512 }
 513
 514 /* caller should hold vxlan->hash_lock */
 515 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 516                                               union vxlan_addr *ip, __be16 port,
 517                                               __be32 vni, __u32 ifindex)
 518 {
 519         struct vxlan_rdst *rd;
 520
 521         list_for_each_entry(rd, &f->remotes, list) {
 522                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 523                     rd->remote_port == port &&
 524                     rd->remote_vni == vni &&
 525                     rd->remote_ifindex == ifindex)
 526                         return rd;
 527         }
 528
 529         return NULL;
 530 }
 531
 532 int vxlan_fdb_find_uc(struct net_device *dev, const u8 *mac, __be32 vni,
 533                       struct switchdev_notifier_vxlan_fdb_info *fdb_info)
 534 {
 535         struct vxlan_dev *vxlan = netdev_priv(dev);
 536         u8 eth_addr[ETH_ALEN + 2] = { 0 };
 537         struct vxlan_rdst *rdst;
 538         struct vxlan_fdb *f;
 539         int rc = 0;
 540
 541         if (is_multicast_ether_addr(mac) ||
 542             is_zero_ether_addr(mac))
 543                 return -EINVAL;
 544
 545         ether_addr_copy(eth_addr, mac);
 546
 547         rcu_read_lock();
 548
 549         f = __vxlan_find_mac(vxlan, eth_addr, vni);
 550         if (!f) {
 551                 rc = -ENOENT;
 552                 goto out;
 553         }
 554
 555         rdst = first_remote_rcu(f);
 556         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, NULL, fdb_info);
 557
 558 out:
 559         rcu_read_unlock();
 560         return rc;
 561 }
 562 EXPORT_SYMBOL_GPL(vxlan_fdb_find_uc);
 563
 564 static int vxlan_fdb_notify_one(struct notifier_block *nb,
 565                                 const struct vxlan_dev *vxlan,
 566                                 const struct vxlan_fdb *f,
 567                                 const struct vxlan_rdst *rdst,
 568                                 struct netlink_ext_ack *extack)
 569 {
 570         struct switchdev_notifier_vxlan_fdb_info fdb_info;
 571         int rc;
 572
 573         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, extack, &fdb_info);
 574         rc = nb->notifier_call(nb, SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE,
 575                                &fdb_info);
 576         return notifier_to_errno(rc);
 577 }
 578
 579 int vxlan_fdb_replay(const struct net_device *dev, __be32 vni,
 580                      struct notifier_block *nb,
 581                      struct netlink_ext_ack *extack)
 582 {
 583         struct vxlan_dev *vxlan;
 584         struct vxlan_rdst *rdst;
 585         struct vxlan_fdb *f;
 586         unsigned int h;
 587         int rc = 0;
 588
 589         if (!netif_is_vxlan(dev))
 590                 return -EINVAL;
 591         vxlan = netdev_priv(dev);
 592
 593         spin_lock_bh(&vxlan->hash_lock);
 594         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 595                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) {
 596                         if (f->vni == vni) {
 597                                 list_for_each_entry(rdst, &f->remotes, list) {
 598                                         rc = vxlan_fdb_notify_one(nb, vxlan,
 599                                                                   f, rdst,
 600                                                                   extack);
 601                                         if (rc)
 602                                                 goto out;
 603                                 }
 604                         }
 605                 }
 606         }
 607
 608 out:
 609         spin_unlock_bh(&vxlan->hash_lock);
 610         return rc;
 611 }
 612 EXPORT_SYMBOL_GPL(vxlan_fdb_replay);
 613
 614 void vxlan_fdb_clear_offload(const struct net_device *dev, __be32 vni)
 615 {
 616         struct vxlan_dev *vxlan;
 617         struct vxlan_rdst *rdst;
 618         struct vxlan_fdb *f;
 619         unsigned int h;
 620
 621         if (!netif_is_vxlan(dev))
 622                 return;
 623         vxlan = netdev_priv(dev);
 624
 625         spin_lock_bh(&vxlan->hash_lock);
 626         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 627                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist)
 628                         if (f->vni == vni)
 629                                 list_for_each_entry(rdst, &f->remotes, list)
 630                                         rdst->offloaded = false;
 631         }
 632         spin_unlock_bh(&vxlan->hash_lock);
 633 }
 634 EXPORT_SYMBOL_GPL(vxlan_fdb_clear_offload);
 635
 636 /* Replace destination of unicast mac */
 637 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 638                              union vxlan_addr *ip, __be16 port, __be32 vni,
 639                              __u32 ifindex, struct vxlan_rdst *oldrd)
 640 {
 641         struct vxlan_rdst *rd;
 642
 643         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 644         if (rd)
 645                 return 0;
 646
 647         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 648         if (!rd)
 649                 return 0;
 650
 651         *oldrd = *rd;
 652         dst_cache_reset(&rd->dst_cache);
 653         rd->remote_ip = *ip;
 654         rd->remote_port = port;
 655         rd->remote_vni = vni;
 656         rd->remote_ifindex = ifindex;
 657         rd->offloaded = false;
 658         return 1;
 659 }
 660
 661 /* Add/update destinations for multicast */
 662 static int vxlan_fdb_append(struct vxlan_fdb *f,
 663                             union vxlan_addr *ip, __be16 port, __be32 vni,
 664                             __u32 ifindex, struct vxlan_rdst **rdp)
 665 {
 666         struct vxlan_rdst *rd;
 667
 668         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 669         if (rd)
 670                 return 0;
 671
 672         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 673         if (rd == NULL)
 674                 return -ENOBUFS;
 675
 676         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 677                 kfree(rd);
 678                 return -ENOBUFS;
 679         }
 680
 681         rd->remote_ip = *ip;
 682         rd->remote_port = port;
 683         rd->offloaded = false;
 684         rd->remote_vni = vni;
 685         rd->remote_ifindex = ifindex;
 686
 687         list_add_tail_rcu(&rd->list, &f->remotes);
 688
 689         *rdp = rd;
 690         return 1;
 691 }
 692
 693 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 694                                           unsigned int off,
 695                                           struct vxlanhdr *vh, size_t hdrlen,
 696                                           __be32 vni_field,
 697                                           struct gro_remcsum *grc,
 698                                           bool nopartial)
 699 {
 700         size_t start, offset;
 701
 702         if (skb->remcsum_offload)
 703                 return vh;
 704
 705         if (!NAPI_GRO_CB(skb)->csum_valid)
 706                 return NULL;
 707
 708         start = vxlan_rco_start(vni_field);
 709         offset = start + vxlan_rco_offset(vni_field);
 710
 711         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 712                                      start, offset, grc, nopartial);
 713
 714         skb->remcsum_offload = 1;
 715
 716         return vh;
 717 }
 718
 719 static struct sk_buff *vxlan_gro_receive(struct sock *sk,
 720                                          struct list_head *head,
 721                                          struct sk_buff *skb)
 722 {
 723         struct sk_buff *pp = NULL;
 724         struct sk_buff *p;
 725         struct vxlanhdr *vh, *vh2;
 726         unsigned int hlen, off_vx;
 727         int flush = 1;
 728         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 729         __be32 flags;
 730         struct gro_remcsum grc;
 731
 732         skb_gro_remcsum_init(&grc);
 733
 734         off_vx = skb_gro_offset(skb);
 735         hlen = off_vx + sizeof(*vh);
 736         vh   = skb_gro_header_fast(skb, off_vx);
 737         if (skb_gro_header_hard(skb, hlen)) {
 738                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 739                 if (unlikely(!vh))
 740                         goto out;
 741         }
 742
 743         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 744
 745         flags = vh->vx_flags;
 746
 747         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 748                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 749                                        vh->vx_vni, &grc,
 750                                        !!(vs->flags &
 751                                           VXLAN_F_REMCSUM_NOPARTIAL));
 752
 753                 if (!vh)
 754                         goto out;
 755         }
 756
 757         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 758
 759         list_for_each_entry(p, head, list) {
 760                 if (!NAPI_GRO_CB(p)->same_flow)
 761                         continue;
 762
 763                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 764                 if (vh->vx_flags != vh2->vx_flags ||
 765                     vh->vx_vni != vh2->vx_vni) {
 766                         NAPI_GRO_CB(p)->same_flow = 0;
 767                         continue;
 768                 }
 769         }
 770
 771         pp = call_gro_receive(eth_gro_receive, head, skb);
 772         flush = 0;
 773
 774 out:
 775         skb_gro_flush_final_remcsum(skb, pp, flush, &grc);
 776
 777         return pp;
 778 }
 779
 780 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 781 {
 782         /* Sets 'skb->inner_mac_header' since we are always called with
 783          * 'skb->encapsulation' set.
 784          */
 785         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 786 }
 787
 788 static struct vxlan_fdb *vxlan_fdb_alloc(struct vxlan_dev *vxlan,
 789                                          const u8 *mac, __u16 state,
 790                                          __be32 src_vni, __u16 ndm_flags)
 791 {
 792         struct vxlan_fdb *f;
 793
 794         f = kmalloc(sizeof(*f), GFP_ATOMIC);
 795         if (!f)
 796                 return NULL;
 797         f->state = state;
 798         f->flags = ndm_flags;
 799         f->updated = f->used = jiffies;
 800         f->vni = src_vni;
 801         INIT_LIST_HEAD(&f->remotes);
 802         memcpy(f->eth_addr, mac, ETH_ALEN);
 803
 804         return f;
 805 }
 806
 807 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 808                             const u8 *mac, union vxlan_addr *ip,
 809                             __u16 state, __be16 port, __be32 src_vni,
 810                             __be32 vni, __u32 ifindex, __u16 ndm_flags,
 811                             struct vxlan_fdb **fdb)
 812 {
 813         struct vxlan_rdst *rd = NULL;
 814         struct vxlan_fdb *f;
 815         int rc;
 816
 817         if (vxlan->cfg.addrmax &&
 818             vxlan->addrcnt >= vxlan->cfg.addrmax)
 819                 return -ENOSPC;
 820
 821         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 822         f = vxlan_fdb_alloc(vxlan, mac, state, src_vni, ndm_flags);
 823         if (!f)
 824                 return -ENOMEM;
 825
 826         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 827         if (rc < 0) {
 828                 kfree(f);
 829                 return rc;
 830         }
 831
 832         ++vxlan->addrcnt;
 833         hlist_add_head_rcu(&f->hlist,
 834                            vxlan_fdb_head(vxlan, mac, src_vni));
 835
 836         *fdb = f;
 837
 838         return 0;
 839 }
 840
 841 static void vxlan_fdb_free(struct rcu_head *head)
 842 {
 843         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 844         struct vxlan_rdst *rd, *nd;
 845
 846         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 847                 dst_cache_destroy(&rd->dst_cache);
 848                 kfree(rd);
 849         }
 850         kfree(f);
 851 }
 852
 853 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 854                               bool do_notify, bool swdev_notify)
 855 {
 856         struct vxlan_rdst *rd;
 857
 858         netdev_dbg(vxlan->dev, "delete %pM\n", f->eth_addr);
 859
 860         --vxlan->addrcnt;
 861         if (do_notify)
 862                 list_for_each_entry(rd, &f->remotes, list)
 863                         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH,
 864                                          swdev_notify, NULL);
 865
 866         hlist_del_rcu(&f->hlist);
 867         call_rcu(&f->rcu, vxlan_fdb_free);
 868 }
 869
 870 static void vxlan_dst_free(struct rcu_head *head)
 871 {
 872         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 873
 874         dst_cache_destroy(&rd->dst_cache);
 875         kfree(rd);
 876 }
 877
 878 static int vxlan_fdb_update_existing(struct vxlan_dev *vxlan,
 879                                      union vxlan_addr *ip,
 880                                      __u16 state, __u16 flags,
 881                                      __be16 port, __be32 vni,
 882                                      __u32 ifindex, __u16 ndm_flags,
 883                                      struct vxlan_fdb *f,
 884                                      bool swdev_notify,
 885                                      struct netlink_ext_ack *extack)
 886 {
 887         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
 888         struct vxlan_rdst *rd = NULL;
 889         struct vxlan_rdst oldrd;
 890         int notify = 0;
 891         int rc = 0;
 892         int err;
 893
 894         /* Do not allow an externally learned entry to take over an entry added
 895          * by the user.
 896          */
 897         if (!(fdb_flags & NTF_EXT_LEARNED) ||
 898             !(f->flags & NTF_VXLAN_ADDED_BY_USER)) {
 899                 if (f->state != state) {
 900                         f->state = state;
 901                         f->updated = jiffies;
 902                         notify = 1;
 903                 }
 904                 if (f->flags != fdb_flags) {
 905                         f->flags = fdb_flags;
 906                         f->updated = jiffies;
 907                         notify = 1;
 908                 }
 909         }
 910
 911         if ((flags & NLM_F_REPLACE)) {
 912                 /* Only change unicasts */
 913                 if (!(is_multicast_ether_addr(f->eth_addr) ||
 914                       is_zero_ether_addr(f->eth_addr))) {
 915                         rc = vxlan_fdb_replace(f, ip, port, vni,
 916                                                ifindex, &oldrd);
 917                         notify |= rc;
 918                 } else {
 919                         return -EOPNOTSUPP;
 920                 }
 921         }
 922         if ((flags & NLM_F_APPEND) &&
 923             (is_multicast_ether_addr(f->eth_addr) ||
 924              is_zero_ether_addr(f->eth_addr))) {
 925                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 926
 927                 if (rc < 0)
 928                         return rc;
 929                 notify |= rc;
 930         }
 931
 932         if (ndm_flags & NTF_USE)
 933                 f->used = jiffies;
 934
 935         if (notify) {
 936                 if (rd == NULL)
 937                         rd = first_remote_rtnl(f);
 938
 939                 err = vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH,
 940                                        swdev_notify, extack);
 941                 if (err)
 942                         goto err_notify;
 943         }
 944
 945         return 0;
 946
 947 err_notify:
 948         if ((flags & NLM_F_REPLACE) && rc)
 949                 *rd = oldrd;
 950         else if ((flags & NLM_F_APPEND) && rc) {
 951                 list_del_rcu(&rd->list);
 952                 call_rcu(&rd->rcu, vxlan_dst_free);
 953         }
 954         return err;
 955 }
 956
 957 static int vxlan_fdb_update_create(struct vxlan_dev *vxlan,
 958                                    const u8 *mac, union vxlan_addr *ip,
 959                                    __u16 state, __u16 flags,
 960                                    __be16 port, __be32 src_vni, __be32 vni,
 961                                    __u32 ifindex, __u16 ndm_flags,
 962                                    bool swdev_notify,
 963                                    struct netlink_ext_ack *extack)
 964 {
 965         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
 966         struct vxlan_fdb *f;
 967         int rc;
 968
 969         /* Disallow replace to add a multicast entry */
 970         if ((flags & NLM_F_REPLACE) &&
 971             (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 972                 return -EOPNOTSUPP;
 973
 974         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 975         rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni,
 976                               vni, ifindex, fdb_flags, &f);
 977         if (rc < 0)
 978                 return rc;
 979
 980         rc = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH,
 981                               swdev_notify, extack);
 982         if (rc)
 983                 goto err_notify;
 984
 985         return 0;
 986
 987 err_notify:
 988         vxlan_fdb_destroy(vxlan, f, false, false);
 989         return rc;
 990 }
 991
 992 /* Add new entry to forwarding table -- assumes lock held */
 993 static int vxlan_fdb_update(struct vxlan_dev *vxlan,
 994                             const u8 *mac, union vxlan_addr *ip,
 995                             __u16 state, __u16 flags,
 996                             __be16 port, __be32 src_vni, __be32 vni,
 997                             __u32 ifindex, __u16 ndm_flags,
 998                             bool swdev_notify,
 999                             struct netlink_ext_ack *extack)
1000 {
1001         struct vxlan_fdb *f;
1002
1003         f = __vxlan_find_mac(vxlan, mac, src_vni);
1004         if (f) {
1005                 if (flags & NLM_F_EXCL) {
1006                         netdev_dbg(vxlan->dev,
1007                                    "lost race to create %pM\n", mac);
1008                         return -EEXIST;
1009                 }
1010
1011                 return vxlan_fdb_update_existing(vxlan, ip, state, flags, port,
1012                                                  vni, ifindex, ndm_flags, f,
1013                                                  swdev_notify, extack);
1014         } else {
1015                 if (!(flags & NLM_F_CREATE))
1016                         return -ENOENT;
1017
1018                 return vxlan_fdb_update_create(vxlan, mac, ip, state, flags,
1019                                                port, src_vni, vni, ifindex,
1020                                                ndm_flags, swdev_notify, extack);
1021         }
1022 }
1023
1024 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
1025                                   struct vxlan_rdst *rd, bool swdev_notify)
1026 {
1027         list_del_rcu(&rd->list);
1028         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, swdev_notify, NULL);
1029         call_rcu(&rd->rcu, vxlan_dst_free);
1030 }
1031
1032 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
1033                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
1034                            __be32 *vni, u32 *ifindex)
1035 {
1036         struct net *net = dev_net(vxlan->dev);
1037         int err;
1038
1039         if (tb[NDA_DST]) {
1040                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
1041                 if (err)
1042                         return err;
1043         } else {
1044                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
1045                 if (remote->sa.sa_family == AF_INET) {
1046                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
1047                         ip->sa.sa_family = AF_INET;
1048 #if IS_ENABLED(CONFIG_IPV6)
1049                 } else {
1050                         ip->sin6.sin6_addr = in6addr_any;
1051                         ip->sa.sa_family = AF_INET6;
1052 #endif
1053                 }
1054         }
1055
1056         if (tb[NDA_PORT]) {
1057                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
1058                         return -EINVAL;
1059                 *port = nla_get_be16(tb[NDA_PORT]);
1060         } else {
1061                 *port = vxlan->cfg.dst_port;
1062         }
1063
1064         if (tb[NDA_VNI]) {
1065                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
1066                         return -EINVAL;
1067                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1068         } else {
1069                 *vni = vxlan->default_dst.remote_vni;
1070         }
1071
1072         if (tb[NDA_SRC_VNI]) {
1073                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
1074                         return -EINVAL;
1075                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
1076         } else {
1077                 *src_vni = vxlan->default_dst.remote_vni;
1078         }
1079
1080         if (tb[NDA_IFINDEX]) {
1081                 struct net_device *tdev;
1082
1083                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
1084                         return -EINVAL;
1085                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
1086                 tdev = __dev_get_by_index(net, *ifindex);
1087                 if (!tdev)
1088                         return -EADDRNOTAVAIL;
1089         } else {
1090                 *ifindex = 0;
1091         }
1092
1093         return 0;
1094 }
1095
1096 /* Add static entry (via netlink) */
1097 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
1098                          struct net_device *dev,
1099                          const unsigned char *addr, u16 vid, u16 flags,
1100                          struct netlink_ext_ack *extack)
1101 {
1102         struct vxlan_dev *vxlan = netdev_priv(dev);
1103         /* struct net *net = dev_net(vxlan->dev); */
1104         union vxlan_addr ip;
1105         __be16 port;
1106         __be32 src_vni, vni;
1107         u32 ifindex;
1108         int err;
1109
1110         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
1111                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
1112                         ndm->ndm_state);
1113                 return -EINVAL;
1114         }
1115
1116         if (tb[NDA_DST] == NULL)
1117                 return -EINVAL;
1118
1119         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
1120         if (err)
1121                 return err;
1122
1123         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
1124                 return -EAFNOSUPPORT;
1125
1126         spin_lock_bh(&vxlan->hash_lock);
1127         err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags,
1128                                port, src_vni, vni, ifindex,
1129                                ndm->ndm_flags | NTF_VXLAN_ADDED_BY_USER,
1130                                true, extack);
1131         spin_unlock_bh(&vxlan->hash_lock);
1132
1133         return err;
1134 }
1135
1136 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
1137                               const unsigned char *addr, union vxlan_addr ip,
1138                               __be16 port, __be32 src_vni, __be32 vni,
1139                               u32 ifindex, bool swdev_notify)
1140 {
1141         struct vxlan_fdb *f;
1142         struct vxlan_rdst *rd = NULL;
1143         int err = -ENOENT;
1144
1145         f = vxlan_find_mac(vxlan, addr, src_vni);
1146         if (!f)
1147                 return err;
1148
1149         if (!vxlan_addr_any(&ip)) {
1150                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
1151                 if (!rd)
1152                         goto out;
1153         }
1154
1155         /* remove a destination if it's not the only one on the list,
1156          * otherwise destroy the fdb entry
1157          */
1158         if (rd && !list_is_singular(&f->remotes)) {
1159                 vxlan_fdb_dst_destroy(vxlan, f, rd, swdev_notify);
1160                 goto out;
1161         }
1162
1163         vxlan_fdb_destroy(vxlan, f, true, swdev_notify);
1164
1165 out:
1166         return 0;
1167 }
1168
1169 /* Delete entry (via netlink) */
1170 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
1171                             struct net_device *dev,
1172                             const unsigned char *addr, u16 vid)
1173 {
1174         struct vxlan_dev *vxlan = netdev_priv(dev);
1175         union vxlan_addr ip;
1176         __be32 src_vni, vni;
1177         __be16 port;
1178         u32 ifindex;
1179         int err;
1180
1181         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex);
1182         if (err)
1183                 return err;
1184
1185         spin_lock_bh(&vxlan->hash_lock);
1186         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
1187                                  true);
1188         spin_unlock_bh(&vxlan->hash_lock);
1189
1190         return err;
1191 }
1192
1193 /* Dump forwarding table */
1194 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1195                           struct net_device *dev,
1196                           struct net_device *filter_dev, int *idx)
1197 {
1198         struct vxlan_dev *vxlan = netdev_priv(dev);
1199         unsigned int h;
1200         int err = 0;
1201
1202         for (h = 0; h < FDB_HASH_SIZE; ++h) {
1203                 struct vxlan_fdb *f;
1204
1205                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
1206                         struct vxlan_rdst *rd;
1207
1208                         list_for_each_entry_rcu(rd, &f->remotes, list) {
1209                                 if (*idx < cb->args[2])
1210                                         goto skip;
1211
1212                                 err = vxlan_fdb_info(skb, vxlan, f,
1213                                                      NETLINK_CB(cb->skb).portid,
1214                                                      cb->nlh->nlmsg_seq,
1215                                                      RTM_NEWNEIGH,
1216                                                      NLM_F_MULTI, rd);
1217                                 if (err < 0)
1218                                         goto out;
1219 skip:
1220                                 *idx += 1;
1221                         }
1222                 }
1223         }
1224 out:
1225         return err;
1226 }
1227
1228 static int vxlan_fdb_get(struct sk_buff *skb,
1229                          struct nlattr *tb[],
1230                          struct net_device *dev,
1231                          const unsigned char *addr,
1232                          u16 vid, u32 portid, u32 seq,
1233                          struct netlink_ext_ack *extack)
1234 {
1235         struct vxlan_dev *vxlan = netdev_priv(dev);
1236         struct vxlan_fdb *f;
1237         __be32 vni;
1238         int err;
1239
1240         if (tb[NDA_VNI])
1241                 vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1242         else
1243                 vni = vxlan->default_dst.remote_vni;
1244
1245         rcu_read_lock();
1246
1247         f = __vxlan_find_mac(vxlan, addr, vni);
1248         if (!f) {
1249                 NL_SET_ERR_MSG(extack, "Fdb entry not found");
1250                 err = -ENOENT;
1251                 goto errout;
1252         }
1253
1254         err = vxlan_fdb_info(skb, vxlan, f, portid, seq,
1255                              RTM_NEWNEIGH, 0, first_remote_rcu(f));
1256 errout:
1257         rcu_read_unlock();
1258         return err;
1259 }
1260
1261 /* Watch incoming packets to learn mapping between Ethernet address
1262  * and Tunnel endpoint.
1263  * Return true if packet is bogus and should be dropped.
1264  */
1265 static bool vxlan_snoop(struct net_device *dev,
1266                         union vxlan_addr *src_ip, const u8 *src_mac,
1267                         u32 src_ifindex, __be32 vni)
1268 {
1269         struct vxlan_dev *vxlan = netdev_priv(dev);
1270         struct vxlan_fdb *f;
1271         u32 ifindex = 0;
1272
1273 #if IS_ENABLED(CONFIG_IPV6)
1274         if (src_ip->sa.sa_family == AF_INET6 &&
1275             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
1276                 ifindex = src_ifindex;
1277 #endif
1278
1279         f = vxlan_find_mac(vxlan, src_mac, vni);
1280         if (likely(f)) {
1281                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1282
1283                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
1284                            rdst->remote_ifindex == ifindex))
1285                         return false;
1286
1287                 /* Don't migrate static entries, drop packets */
1288                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
1289                         return true;
1290
1291                 if (net_ratelimit())
1292                         netdev_info(dev,
1293                                     "%pM migrated from %pIS to %pIS\n",
1294                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1295
1296                 rdst->remote_ip = *src_ip;
1297                 f->updated = jiffies;
1298                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH, true, NULL);
1299         } else {
1300                 /* learned new entry */
1301                 spin_lock(&vxlan->hash_lock);
1302
1303                 /* close off race between vxlan_flush and incoming packets */
1304                 if (netif_running(dev))
1305                         vxlan_fdb_update(vxlan, src_mac, src_ip,
1306                                          NUD_REACHABLE,
1307                                          NLM_F_EXCL|NLM_F_CREATE,
1308                                          vxlan->cfg.dst_port,
1309                                          vni,
1310                                          vxlan->default_dst.remote_vni,
1311                                          ifindex, NTF_SELF, true, NULL);
1312                 spin_unlock(&vxlan->hash_lock);
1313         }
1314
1315         return false;
1316 }
1317
1318 /* See if multicast group is already in use by other ID */
1319 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1320 {
1321         struct vxlan_dev *vxlan;
1322         struct vxlan_sock *sock4;
1323 #if IS_ENABLED(CONFIG_IPV6)
1324         struct vxlan_sock *sock6;
1325 #endif
1326         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1327
1328         sock4 = rtnl_dereference(dev->vn4_sock);
1329
1330         /* The vxlan_sock is only used by dev, leaving group has
1331          * no effect on other vxlan devices.
1332          */
1333         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1334                 return false;
1335 #if IS_ENABLED(CONFIG_IPV6)
1336         sock6 = rtnl_dereference(dev->vn6_sock);
1337         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1338                 return false;
1339 #endif
1340
1341         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1342                 if (!netif_running(vxlan->dev) || vxlan == dev)
1343                         continue;
1344
1345                 if (family == AF_INET &&
1346                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1347                         continue;
1348 #if IS_ENABLED(CONFIG_IPV6)
1349                 if (family == AF_INET6 &&
1350                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1351                         continue;
1352 #endif
1353
1354                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1355                                       &dev->default_dst.remote_ip))
1356                         continue;
1357
1358                 if (vxlan->default_dst.remote_ifindex !=
1359                     dev->default_dst.remote_ifindex)
1360                         continue;
1361
1362                 return true;
1363         }
1364
1365         return false;
1366 }
1367
1368 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1369 {
1370         struct vxlan_net *vn;
1371
1372         if (!vs)
1373                 return false;
1374         if (!refcount_dec_and_test(&vs->refcnt))
1375                 return false;
1376
1377         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1378         spin_lock(&vn->sock_lock);
1379         hlist_del_rcu(&vs->hlist);
1380         udp_tunnel_notify_del_rx_port(vs->sock,
1381                                       (vs->flags & VXLAN_F_GPE) ?
1382                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1383                                       UDP_TUNNEL_TYPE_VXLAN);
1384         spin_unlock(&vn->sock_lock);
1385
1386         return true;
1387 }
1388
1389 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1390 {
1391         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1392 #if IS_ENABLED(CONFIG_IPV6)
1393         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1394
1395         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1396 #endif
1397
1398         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1399         synchronize_net();
1400
1401         vxlan_vs_del_dev(vxlan);
1402
1403         if (__vxlan_sock_release_prep(sock4)) {
1404                 udp_tunnel_sock_release(sock4->sock);
1405                 kfree(sock4);
1406         }
1407
1408 #if IS_ENABLED(CONFIG_IPV6)
1409         if (__vxlan_sock_release_prep(sock6)) {
1410                 udp_tunnel_sock_release(sock6->sock);
1411                 kfree(sock6);
1412         }
1413 #endif
1414 }
1415
1416 /* Update multicast group membership when first VNI on
1417  * multicast address is brought up
1418  */
1419 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1420 {
1421         struct sock *sk;
1422         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1423         int ifindex = vxlan->default_dst.remote_ifindex;
1424         int ret = -EINVAL;
1425
1426         if (ip->sa.sa_family == AF_INET) {
1427                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1428                 struct ip_mreqn mreq = {
1429                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1430                         .imr_ifindex            = ifindex,
1431                 };
1432
1433                 sk = sock4->sock->sk;
1434                 lock_sock(sk);
1435                 ret = ip_mc_join_group(sk, &mreq);
1436                 release_sock(sk);
1437 #if IS_ENABLED(CONFIG_IPV6)
1438         } else {
1439                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1440
1441                 sk = sock6->sock->sk;
1442                 lock_sock(sk);
1443                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1444                                                    &ip->sin6.sin6_addr);
1445                 release_sock(sk);
1446 #endif
1447         }
1448
1449         return ret;
1450 }
1451
1452 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1453 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1454 {
1455         struct sock *sk;
1456         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1457         int ifindex = vxlan->default_dst.remote_ifindex;
1458         int ret = -EINVAL;
1459
1460         if (ip->sa.sa_family == AF_INET) {
1461                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1462                 struct ip_mreqn mreq = {
1463                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1464                         .imr_ifindex            = ifindex,
1465                 };
1466
1467                 sk = sock4->sock->sk;
1468                 lock_sock(sk);
1469                 ret = ip_mc_leave_group(sk, &mreq);
1470                 release_sock(sk);
1471 #if IS_ENABLED(CONFIG_IPV6)
1472         } else {
1473                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1474
1475                 sk = sock6->sock->sk;
1476                 lock_sock(sk);
1477                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1478                                                    &ip->sin6.sin6_addr);
1479                 release_sock(sk);
1480 #endif
1481         }
1482
1483         return ret;
1484 }
1485
1486 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1487                           struct sk_buff *skb, u32 vxflags)
1488 {
1489         size_t start, offset;
1490
1491         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1492                 goto out;
1493
1494         start = vxlan_rco_start(unparsed->vx_vni);
1495         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1496
1497         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1498                 return false;
1499
1500         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1501                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1502 out:
1503         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1504         unparsed->vx_vni &= VXLAN_VNI_MASK;
1505         return true;
1506 }
1507
1508 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1509                                 struct sk_buff *skb, u32 vxflags,
1510                                 struct vxlan_metadata *md)
1511 {
1512         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1513         struct metadata_dst *tun_dst;
1514
1515         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1516                 goto out;
1517
1518         md->gbp = ntohs(gbp->policy_id);
1519
1520         tun_dst = (struct metadata_dst *)skb_dst(skb);
1521         if (tun_dst) {
1522                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1523                 tun_dst->u.tun_info.options_len = sizeof(*md);
1524         }
1525         if (gbp->dont_learn)
1526                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1527
1528         if (gbp->policy_applied)
1529                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1530
1531         /* In flow-based mode, GBP is carried in dst_metadata */
1532         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1533                 skb->mark = md->gbp;
1534 out:
1535         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1536 }
1537
1538 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1539                                 __be16 *protocol,
1540                                 struct sk_buff *skb, u32 vxflags)
1541 {
1542         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1543
1544         /* Need to have Next Protocol set for interfaces in GPE mode. */
1545         if (!gpe->np_applied)
1546                 return false;
1547         /* "The initial version is 0. If a receiver does not support the
1548          * version indicated it MUST drop the packet.
1549          */
1550         if (gpe->version != 0)
1551                 return false;
1552         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1553          * processing MUST occur." However, we don't implement OAM
1554          * processing, thus drop the packet.
1555          */
1556         if (gpe->oam_flag)
1557                 return false;
1558
1559         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1560         if (!*protocol)
1561                 return false;
1562
1563         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1564         return true;
1565 }
1566
1567 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1568                           struct vxlan_sock *vs,
1569                           struct sk_buff *skb, __be32 vni)
1570 {
1571         union vxlan_addr saddr;
1572         u32 ifindex = skb->dev->ifindex;
1573
1574         skb_reset_mac_header(skb);
1575         skb->protocol = eth_type_trans(skb, vxlan->dev);
1576         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1577
1578         /* Ignore packet loops (and multicast echo) */
1579         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1580                 return false;
1581
1582         /* Get address from the outer IP header */
1583         if (vxlan_get_sk_family(vs) == AF_INET) {
1584                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1585                 saddr.sa.sa_family = AF_INET;
1586 #if IS_ENABLED(CONFIG_IPV6)
1587         } else {
1588                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1589                 saddr.sa.sa_family = AF_INET6;
1590 #endif
1591         }
1592
1593         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1594             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1595                 return false;
1596
1597         return true;
1598 }
1599
1600 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1601                                   struct sk_buff *skb)
1602 {
1603         int err = 0;
1604
1605         if (vxlan_get_sk_family(vs) == AF_INET)
1606                 err = IP_ECN_decapsulate(oiph, skb);
1607 #if IS_ENABLED(CONFIG_IPV6)
1608         else
1609                 err = IP6_ECN_decapsulate(oiph, skb);
1610 #endif
1611
1612         if (unlikely(err) && log_ecn_error) {
1613                 if (vxlan_get_sk_family(vs) == AF_INET)
1614                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1615                                              &((struct iphdr *)oiph)->saddr,
1616                                              ((struct iphdr *)oiph)->tos);
1617                 else
1618                         net_info_ratelimited("non-ECT from %pI6\n",
1619                                              &((struct ipv6hdr *)oiph)->saddr);
1620         }
1621         return err <= 1;
1622 }
1623
1624 /* Callback from net/ipv4/udp.c to receive packets */
1625 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1626 {
1627         struct pcpu_sw_netstats *stats;
1628         struct vxlan_dev *vxlan;
1629         struct vxlan_sock *vs;
1630         struct vxlanhdr unparsed;
1631         struct vxlan_metadata _md;
1632         struct vxlan_metadata *md = &_md;
1633         __be16 protocol = htons(ETH_P_TEB);
1634         bool raw_proto = false;
1635         void *oiph;
1636         __be32 vni = 0;
1637
1638         /* Need UDP and VXLAN header to be present */
1639         if (!pskb_may_pull(skb, VXLAN_HLEN))
1640                 goto drop;
1641
1642         unparsed = *vxlan_hdr(skb);
1643         /* VNI flag always required to be set */
1644         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1645                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1646                            ntohl(vxlan_hdr(skb)->vx_flags),
1647                            ntohl(vxlan_hdr(skb)->vx_vni));
1648                 /* Return non vxlan pkt */
1649                 goto drop;
1650         }
1651         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1652         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1653
1654         vs = rcu_dereference_sk_user_data(sk);
1655         if (!vs)
1656                 goto drop;
1657
1658         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1659
1660         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1661         if (!vxlan)
1662                 goto drop;
1663
1664         /* For backwards compatibility, only allow reserved fields to be
1665          * used by VXLAN extensions if explicitly requested.
1666          */
1667         if (vs->flags & VXLAN_F_GPE) {
1668                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1669                         goto drop;
1670                 raw_proto = true;
1671         }
1672
1673         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1674                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1675                         goto drop;
1676
1677         if (vxlan_collect_metadata(vs)) {
1678                 struct metadata_dst *tun_dst;
1679
1680                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1681                                          key32_to_tunnel_id(vni), sizeof(*md));
1682
1683                 if (!tun_dst)
1684                         goto drop;
1685
1686                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1687
1688                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1689         } else {
1690                 memset(md, 0, sizeof(*md));
1691         }
1692
1693         if (vs->flags & VXLAN_F_REMCSUM_RX)
1694                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1695                         goto drop;
1696         if (vs->flags & VXLAN_F_GBP)
1697                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1698         /* Note that GBP and GPE can never be active together. This is
1699          * ensured in vxlan_dev_configure.
1700          */
1701
1702         if (unparsed.vx_flags || unparsed.vx_vni) {
1703                 /* If there are any unprocessed flags remaining treat
1704                  * this as a malformed packet. This behavior diverges from
1705                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1706                  * in reserved fields are to be ignored. The approach here
1707                  * maintains compatibility with previous stack code, and also
1708                  * is more robust and provides a little more security in
1709                  * adding extensions to VXLAN.
1710                  */
1711                 goto drop;
1712         }
1713
1714         if (!raw_proto) {
1715                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1716                         goto drop;
1717         } else {
1718                 skb_reset_mac_header(skb);
1719                 skb->dev = vxlan->dev;
1720                 skb->pkt_type = PACKET_HOST;
1721         }
1722
1723         oiph = skb_network_header(skb);
1724         skb_reset_network_header(skb);
1725
1726         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1727                 ++vxlan->dev->stats.rx_frame_errors;
1728                 ++vxlan->dev->stats.rx_errors;
1729                 goto drop;
1730         }
1731
1732         rcu_read_lock();
1733
1734         if (unlikely(!(vxlan->dev->flags & IFF_UP))) {
1735                 rcu_read_unlock();
1736                 atomic_long_inc(&vxlan->dev->rx_dropped);
1737                 goto drop;
1738         }
1739
1740         stats = this_cpu_ptr(vxlan->dev->tstats);
1741         u64_stats_update_begin(&stats->syncp);
1742         stats->rx_packets++;
1743         stats->rx_bytes += skb->len;
1744         u64_stats_update_end(&stats->syncp);
1745
1746         gro_cells_receive(&vxlan->gro_cells, skb);
1747
1748         rcu_read_unlock();
1749
1750         return 0;
1751
1752 drop:
1753         /* Consume bad packet */
1754         kfree_skb(skb);
1755         return 0;
1756 }
1757
1758 /* Callback from net/ipv{4,6}/udp.c to check that we have a VNI for errors */
1759 static int vxlan_err_lookup(struct sock *sk, struct sk_buff *skb)
1760 {
1761         struct vxlan_dev *vxlan;
1762         struct vxlan_sock *vs;
1763         struct vxlanhdr *hdr;
1764         __be32 vni;
1765
1766         if (!pskb_may_pull(skb, skb_transport_offset(skb) + VXLAN_HLEN))
1767                 return -EINVAL;
1768
1769         hdr = vxlan_hdr(skb);
1770
1771         if (!(hdr->vx_flags & VXLAN_HF_VNI))
1772                 return -EINVAL;
1773
1774         vs = rcu_dereference_sk_user_data(sk);
1775         if (!vs)
1776                 return -ENOENT;
1777
1778         vni = vxlan_vni(hdr->vx_vni);
1779         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1780         if (!vxlan)
1781                 return -ENOENT;
1782
1783         return 0;
1784 }
1785
1786 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1787 {
1788         struct vxlan_dev *vxlan = netdev_priv(dev);
1789         struct arphdr *parp;
1790         u8 *arpptr, *sha;
1791         __be32 sip, tip;
1792         struct neighbour *n;
1793
1794         if (dev->flags & IFF_NOARP)
1795                 goto out;
1796
1797         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1798                 dev->stats.tx_dropped++;
1799                 goto out;
1800         }
1801         parp = arp_hdr(skb);
1802
1803         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1804              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1805             parp->ar_pro != htons(ETH_P_IP) ||
1806             parp->ar_op != htons(ARPOP_REQUEST) ||
1807             parp->ar_hln != dev->addr_len ||
1808             parp->ar_pln != 4)
1809                 goto out;
1810         arpptr = (u8 *)parp + sizeof(struct arphdr);
1811         sha = arpptr;
1812         arpptr += dev->addr_len;        /* sha */
1813         memcpy(&sip, arpptr, sizeof(sip));
1814         arpptr += sizeof(sip);
1815         arpptr += dev->addr_len;        /* tha */
1816         memcpy(&tip, arpptr, sizeof(tip));
1817
1818         if (ipv4_is_loopback(tip) ||
1819             ipv4_is_multicast(tip))
1820                 goto out;
1821
1822         n = neigh_lookup(&arp_tbl, &tip, dev);
1823
1824         if (n) {
1825                 struct vxlan_fdb *f;
1826                 struct sk_buff  *reply;
1827
1828                 if (!(n->nud_state & NUD_CONNECTED)) {
1829                         neigh_release(n);
1830                         goto out;
1831                 }
1832
1833                 f = vxlan_find_mac(vxlan, n->ha, vni);
1834                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1835                         /* bridge-local neighbor */
1836                         neigh_release(n);
1837                         goto out;
1838                 }
1839
1840                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1841                                 n->ha, sha);
1842
1843                 neigh_release(n);
1844
1845                 if (reply == NULL)
1846                         goto out;
1847
1848                 skb_reset_mac_header(reply);
1849                 __skb_pull(reply, skb_network_offset(reply));
1850                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1851                 reply->pkt_type = PACKET_HOST;
1852
1853                 if (netif_rx_ni(reply) == NET_RX_DROP)
1854                         dev->stats.rx_dropped++;
1855         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1856                 union vxlan_addr ipa = {
1857                         .sin.sin_addr.s_addr = tip,
1858                         .sin.sin_family = AF_INET,
1859                 };
1860
1861                 vxlan_ip_miss(dev, &ipa);
1862         }
1863 out:
1864         consume_skb(skb);
1865         return NETDEV_TX_OK;
1866 }
1867
1868 #if IS_ENABLED(CONFIG_IPV6)
1869 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1870         struct neighbour *n, bool isrouter)
1871 {
1872         struct net_device *dev = request->dev;
1873         struct sk_buff *reply;
1874         struct nd_msg *ns, *na;
1875         struct ipv6hdr *pip6;
1876         u8 *daddr;
1877         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1878         int ns_olen;
1879         int i, len;
1880
1881         if (dev == NULL || !pskb_may_pull(request, request->len))
1882                 return NULL;
1883
1884         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1885                 sizeof(*na) + na_olen + dev->needed_tailroom;
1886         reply = alloc_skb(len, GFP_ATOMIC);
1887         if (reply == NULL)
1888                 return NULL;
1889
1890         reply->protocol = htons(ETH_P_IPV6);
1891         reply->dev = dev;
1892         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1893         skb_push(reply, sizeof(struct ethhdr));
1894         skb_reset_mac_header(reply);
1895
1896         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
1897
1898         daddr = eth_hdr(request)->h_source;
1899         ns_olen = request->len - skb_network_offset(request) -
1900                 sizeof(struct ipv6hdr) - sizeof(*ns);
1901         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1902                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1903                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1904                         break;
1905                 }
1906         }
1907
1908         /* Ethernet header */
1909         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1910         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1911         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1912         reply->protocol = htons(ETH_P_IPV6);
1913
1914         skb_pull(reply, sizeof(struct ethhdr));
1915         skb_reset_network_header(reply);
1916         skb_put(reply, sizeof(struct ipv6hdr));
1917
1918         /* IPv6 header */
1919
1920         pip6 = ipv6_hdr(reply);
1921         memset(pip6, 0, sizeof(struct ipv6hdr));
1922         pip6->version = 6;
1923         pip6->priority = ipv6_hdr(request)->priority;
1924         pip6->nexthdr = IPPROTO_ICMPV6;
1925         pip6->hop_limit = 255;
1926         pip6->daddr = ipv6_hdr(request)->saddr;
1927         pip6->saddr = *(struct in6_addr *)n->primary_key;
1928
1929         skb_pull(reply, sizeof(struct ipv6hdr));
1930         skb_reset_transport_header(reply);
1931
1932         /* Neighbor Advertisement */
1933         na = skb_put_zero(reply, sizeof(*na) + na_olen);
1934         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1935         na->icmph.icmp6_router = isrouter;
1936         na->icmph.icmp6_override = 1;
1937         na->icmph.icmp6_solicited = 1;
1938         na->target = ns->target;
1939         ether_addr_copy(&na->opt[2], n->ha);
1940         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1941         na->opt[1] = na_olen >> 3;
1942
1943         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1944                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1945                 csum_partial(na, sizeof(*na)+na_olen, 0));
1946
1947         pip6->payload_len = htons(sizeof(*na)+na_olen);
1948
1949         skb_push(reply, sizeof(struct ipv6hdr));
1950
1951         reply->ip_summed = CHECKSUM_UNNECESSARY;
1952
1953         return reply;
1954 }
1955
1956 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1957 {
1958         struct vxlan_dev *vxlan = netdev_priv(dev);
1959         const struct in6_addr *daddr;
1960         const struct ipv6hdr *iphdr;
1961         struct inet6_dev *in6_dev;
1962         struct neighbour *n;
1963         struct nd_msg *msg;
1964
1965         in6_dev = __in6_dev_get(dev);
1966         if (!in6_dev)
1967                 goto out;
1968
1969         iphdr = ipv6_hdr(skb);
1970         daddr = &iphdr->daddr;
1971         msg = (struct nd_msg *)(iphdr + 1);
1972
1973         if (ipv6_addr_loopback(daddr) ||
1974             ipv6_addr_is_multicast(&msg->target))
1975                 goto out;
1976
1977         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1978
1979         if (n) {
1980                 struct vxlan_fdb *f;
1981                 struct sk_buff *reply;
1982
1983                 if (!(n->nud_state & NUD_CONNECTED)) {
1984                         neigh_release(n);
1985                         goto out;
1986                 }
1987
1988                 f = vxlan_find_mac(vxlan, n->ha, vni);
1989                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1990                         /* bridge-local neighbor */
1991                         neigh_release(n);
1992                         goto out;
1993                 }
1994
1995                 reply = vxlan_na_create(skb, n,
1996                                         !!(f ? f->flags & NTF_ROUTER : 0));
1997
1998                 neigh_release(n);
1999
2000                 if (reply == NULL)
2001                         goto out;
2002
2003                 if (netif_rx_ni(reply) == NET_RX_DROP)
2004                         dev->stats.rx_dropped++;
2005
2006         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2007                 union vxlan_addr ipa = {
2008                         .sin6.sin6_addr = msg->target,
2009                         .sin6.sin6_family = AF_INET6,
2010                 };
2011
2012                 vxlan_ip_miss(dev, &ipa);
2013         }
2014
2015 out:
2016         consume_skb(skb);
2017         return NETDEV_TX_OK;
2018 }
2019 #endif
2020
2021 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
2022 {
2023         struct vxlan_dev *vxlan = netdev_priv(dev);
2024         struct neighbour *n;
2025
2026         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
2027                 return false;
2028
2029         n = NULL;
2030         switch (ntohs(eth_hdr(skb)->h_proto)) {
2031         case ETH_P_IP:
2032         {
2033                 struct iphdr *pip;
2034
2035                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
2036                         return false;
2037                 pip = ip_hdr(skb);
2038                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
2039                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2040                         union vxlan_addr ipa = {
2041                                 .sin.sin_addr.s_addr = pip->daddr,
2042                                 .sin.sin_family = AF_INET,
2043                         };
2044
2045                         vxlan_ip_miss(dev, &ipa);
2046                         return false;
2047                 }
2048
2049                 break;
2050         }
2051 #if IS_ENABLED(CONFIG_IPV6)
2052         case ETH_P_IPV6:
2053         {
2054                 struct ipv6hdr *pip6;
2055
2056                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
2057                         return false;
2058                 pip6 = ipv6_hdr(skb);
2059                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
2060                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2061                         union vxlan_addr ipa = {
2062                                 .sin6.sin6_addr = pip6->daddr,
2063                                 .sin6.sin6_family = AF_INET6,
2064                         };
2065
2066                         vxlan_ip_miss(dev, &ipa);
2067                         return false;
2068                 }
2069
2070                 break;
2071         }
2072 #endif
2073         default:
2074                 return false;
2075         }
2076
2077         if (n) {
2078                 bool diff;
2079
2080                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
2081                 if (diff) {
2082                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
2083                                 dev->addr_len);
2084                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
2085                 }
2086                 neigh_release(n);
2087                 return diff;
2088         }
2089
2090         return false;
2091 }
2092
2093 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
2094                                 struct vxlan_metadata *md)
2095 {
2096         struct vxlanhdr_gbp *gbp;
2097
2098         if (!md->gbp)
2099                 return;
2100
2101         gbp = (struct vxlanhdr_gbp *)vxh;
2102         vxh->vx_flags |= VXLAN_HF_GBP;
2103
2104         if (md->gbp & VXLAN_GBP_DONT_LEARN)
2105                 gbp->dont_learn = 1;
2106
2107         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
2108                 gbp->policy_applied = 1;
2109
2110         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
2111 }
2112
2113 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
2114                                __be16 protocol)
2115 {
2116         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
2117
2118         gpe->np_applied = 1;
2119         gpe->next_protocol = tun_p_from_eth_p(protocol);
2120         if (!gpe->next_protocol)
2121                 return -EPFNOSUPPORT;
2122         return 0;
2123 }
2124
2125 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
2126                            int iphdr_len, __be32 vni,
2127                            struct vxlan_metadata *md, u32 vxflags,
2128                            bool udp_sum)
2129 {
2130         struct vxlanhdr *vxh;
2131         int min_headroom;
2132         int err;
2133         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
2134         __be16 inner_protocol = htons(ETH_P_TEB);
2135
2136         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
2137             skb->ip_summed == CHECKSUM_PARTIAL) {
2138                 int csum_start = skb_checksum_start_offset(skb);
2139
2140                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
2141                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
2142                     (skb->csum_offset == offsetof(struct udphdr, check) ||
2143                      skb->csum_offset == offsetof(struct tcphdr, check)))
2144                         type |= SKB_GSO_TUNNEL_REMCSUM;
2145         }
2146
2147         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
2148                         + VXLAN_HLEN + iphdr_len;
2149
2150         /* Need space for new headers (invalidates iph ptr) */
2151         err = skb_cow_head(skb, min_headroom);
2152         if (unlikely(err))
2153                 return err;
2154
2155         err = iptunnel_handle_offloads(skb, type);
2156         if (err)
2157                 return err;
2158
2159         vxh = __skb_push(skb, sizeof(*vxh));
2160         vxh->vx_flags = VXLAN_HF_VNI;
2161         vxh->vx_vni = vxlan_vni_field(vni);
2162
2163         if (type & SKB_GSO_TUNNEL_REMCSUM) {
2164                 unsigned int start;
2165
2166                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
2167                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
2168                 vxh->vx_flags |= VXLAN_HF_RCO;
2169
2170                 if (!skb_is_gso(skb)) {
2171                         skb->ip_summed = CHECKSUM_NONE;
2172                         skb->encapsulation = 0;
2173                 }
2174         }
2175
2176         if (vxflags & VXLAN_F_GBP)
2177                 vxlan_build_gbp_hdr(vxh, vxflags, md);
2178         if (vxflags & VXLAN_F_GPE) {
2179                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
2180                 if (err < 0)
2181                         return err;
2182                 inner_protocol = skb->protocol;
2183         }
2184
2185         skb_set_inner_protocol(skb, inner_protocol);
2186         return 0;
2187 }
2188
2189 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
2190                                       struct vxlan_sock *sock4,
2191                                       struct sk_buff *skb, int oif, u8 tos,
2192                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
2193                                       struct dst_cache *dst_cache,
2194                                       const struct ip_tunnel_info *info)
2195 {
2196         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2197         struct rtable *rt = NULL;
2198         struct flowi4 fl4;
2199
2200         if (!sock4)
2201                 return ERR_PTR(-EIO);
2202
2203         if (tos && !info)
2204                 use_cache = false;
2205         if (use_cache) {
2206                 rt = dst_cache_get_ip4(dst_cache, saddr);
2207                 if (rt)
2208                         return rt;
2209         }
2210
2211         memset(&fl4, 0, sizeof(fl4));
2212         fl4.flowi4_oif = oif;
2213         fl4.flowi4_tos = RT_TOS(tos);
2214         fl4.flowi4_mark = skb->mark;
2215         fl4.flowi4_proto = IPPROTO_UDP;
2216         fl4.daddr = daddr;
2217         fl4.saddr = *saddr;
2218         fl4.fl4_dport = dport;
2219         fl4.fl4_sport = sport;
2220
2221         rt = ip_route_output_key(vxlan->net, &fl4);
2222         if (likely(!IS_ERR(rt))) {
2223                 if (rt->dst.dev == dev) {
2224                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
2225                         ip_rt_put(rt);
2226                         return ERR_PTR(-ELOOP);
2227                 }
2228
2229                 *saddr = fl4.saddr;
2230                 if (use_cache)
2231                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2232         } else {
2233                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
2234                 return ERR_PTR(-ENETUNREACH);
2235         }
2236         return rt;
2237 }
2238
2239 #if IS_ENABLED(CONFIG_IPV6)
2240 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
2241                                           struct net_device *dev,
2242                                           struct vxlan_sock *sock6,
2243                                           struct sk_buff *skb, int oif, u8 tos,
2244                                           __be32 label,
2245                                           const struct in6_addr *daddr,
2246                                           struct in6_addr *saddr,
2247                                           __be16 dport, __be16 sport,
2248                                           struct dst_cache *dst_cache,
2249                                           const struct ip_tunnel_info *info)
2250 {
2251         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2252         struct dst_entry *ndst;
2253         struct flowi6 fl6;
2254         int err;
2255
2256         if (!sock6)
2257                 return ERR_PTR(-EIO);
2258
2259         if (tos && !info)
2260                 use_cache = false;
2261         if (use_cache) {
2262                 ndst = dst_cache_get_ip6(dst_cache, saddr);
2263                 if (ndst)
2264                         return ndst;
2265         }
2266
2267         memset(&fl6, 0, sizeof(fl6));
2268         fl6.flowi6_oif = oif;
2269         fl6.daddr = *daddr;
2270         fl6.saddr = *saddr;
2271         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
2272         fl6.flowi6_mark = skb->mark;
2273         fl6.flowi6_proto = IPPROTO_UDP;
2274         fl6.fl6_dport = dport;
2275         fl6.fl6_sport = sport;
2276
2277         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
2278                                          sock6->sock->sk,
2279                                          &ndst, &fl6);
2280         if (unlikely(err < 0)) {
2281                 netdev_dbg(dev, "no route to %pI6\n", daddr);
2282                 return ERR_PTR(-ENETUNREACH);
2283         }
2284
2285         if (unlikely(ndst->dev == dev)) {
2286                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
2287                 dst_release(ndst);
2288                 return ERR_PTR(-ELOOP);
2289         }
2290
2291         *saddr = fl6.saddr;
2292         if (use_cache)
2293                 dst_cache_set_ip6(dst_cache, ndst, saddr);
2294         return ndst;
2295 }
2296 #endif
2297
2298 /* Bypass encapsulation if the destination is local */
2299 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2300                                struct vxlan_dev *dst_vxlan, __be32 vni)
2301 {
2302         struct pcpu_sw_netstats *tx_stats, *rx_stats;
2303         union vxlan_addr loopback;
2304         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2305         struct net_device *dev;
2306         int len = skb->len;
2307
2308         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2309         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2310         skb->pkt_type = PACKET_HOST;
2311         skb->encapsulation = 0;
2312         skb->dev = dst_vxlan->dev;
2313         __skb_pull(skb, skb_network_offset(skb));
2314
2315         if (remote_ip->sa.sa_family == AF_INET) {
2316                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2317                 loopback.sa.sa_family =  AF_INET;
2318 #if IS_ENABLED(CONFIG_IPV6)
2319         } else {
2320                 loopback.sin6.sin6_addr = in6addr_loopback;
2321                 loopback.sa.sa_family =  AF_INET6;
2322 #endif
2323         }
2324
2325         rcu_read_lock();
2326         dev = skb->dev;
2327         if (unlikely(!(dev->flags & IFF_UP))) {
2328                 kfree_skb(skb);
2329                 goto drop;
2330         }
2331
2332         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
2333                 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni);
2334
2335         u64_stats_update_begin(&tx_stats->syncp);
2336         tx_stats->tx_packets++;
2337         tx_stats->tx_bytes += len;
2338         u64_stats_update_end(&tx_stats->syncp);
2339
2340         if (netif_rx(skb) == NET_RX_SUCCESS) {
2341                 u64_stats_update_begin(&rx_stats->syncp);
2342                 rx_stats->rx_packets++;
2343                 rx_stats->rx_bytes += len;
2344                 u64_stats_update_end(&rx_stats->syncp);
2345         } else {
2346 drop:
2347                 dev->stats.rx_dropped++;
2348         }
2349         rcu_read_unlock();
2350 }
2351
2352 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2353                                  struct vxlan_dev *vxlan,
2354                                  union vxlan_addr *daddr,
2355                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2356                                  struct dst_entry *dst,
2357                                  u32 rt_flags)
2358 {
2359 #if IS_ENABLED(CONFIG_IPV6)
2360         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2361          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2362          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2363          */
2364         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2365 #endif
2366         /* Bypass encapsulation if the destination is local */
2367         if (rt_flags & RTCF_LOCAL &&
2368             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2369                 struct vxlan_dev *dst_vxlan;
2370
2371                 dst_release(dst);
2372                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2373                                            daddr->sa.sa_family, dst_port,
2374                                            vxlan->cfg.flags);
2375                 if (!dst_vxlan) {
2376                         dev->stats.tx_errors++;
2377                         kfree_skb(skb);
2378
2379                         return -ENOENT;
2380                 }
2381                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2382                 return 1;
2383         }
2384
2385         return 0;
2386 }
2387
2388 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2389                            __be32 default_vni, struct vxlan_rdst *rdst,
2390                            bool did_rsc)
2391 {
2392         struct dst_cache *dst_cache;
2393         struct ip_tunnel_info *info;
2394         struct vxlan_dev *vxlan = netdev_priv(dev);
2395         const struct iphdr *old_iph = ip_hdr(skb);
2396         union vxlan_addr *dst;
2397         union vxlan_addr remote_ip, local_ip;
2398         struct vxlan_metadata _md;
2399         struct vxlan_metadata *md = &_md;
2400         __be16 src_port = 0, dst_port;
2401         struct dst_entry *ndst = NULL;
2402         __be32 vni, label;
2403         __u8 tos, ttl;
2404         int ifindex;
2405         int err;
2406         u32 flags = vxlan->cfg.flags;
2407         bool udp_sum = false;
2408         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2409
2410         info = skb_tunnel_info(skb);
2411
2412         if (rdst) {
2413                 dst = &rdst->remote_ip;
2414                 if (vxlan_addr_any(dst)) {
2415                         if (did_rsc) {
2416                                 /* short-circuited back to local bridge */
2417                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2418                                 return;
2419                         }
2420                         goto drop;
2421                 }
2422
2423                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2424                 vni = (rdst->remote_vni) ? : default_vni;
2425                 ifindex = rdst->remote_ifindex;
2426                 local_ip = vxlan->cfg.saddr;
2427                 dst_cache = &rdst->dst_cache;
2428                 md->gbp = skb->mark;
2429                 if (flags & VXLAN_F_TTL_INHERIT) {
2430                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2431                 } else {
2432                         ttl = vxlan->cfg.ttl;
2433                         if (!ttl && vxlan_addr_multicast(dst))
2434                                 ttl = 1;
2435                 }
2436
2437                 tos = vxlan->cfg.tos;
2438                 if (tos == 1)
2439                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2440
2441                 if (dst->sa.sa_family == AF_INET)
2442                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2443                 else
2444                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2445                 label = vxlan->cfg.label;
2446         } else {
2447                 if (!info) {
2448                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2449                                   dev->name);
2450                         goto drop;
2451                 }
2452                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2453                 if (remote_ip.sa.sa_family == AF_INET) {
2454                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2455                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2456                 } else {
2457                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2458                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2459                 }
2460                 dst = &remote_ip;
2461                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2462                 vni = tunnel_id_to_key32(info->key.tun_id);
2463                 ifindex = 0;
2464                 dst_cache = &info->dst_cache;
2465                 if (info->options_len &&
2466                     info->key.tun_flags & TUNNEL_VXLAN_OPT)
2467                         md = ip_tunnel_info_opts(info);
2468                 ttl = info->key.ttl;
2469                 tos = info->key.tos;
2470                 label = info->key.label;
2471                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2472         }
2473         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2474                                      vxlan->cfg.port_max, true);
2475
2476         rcu_read_lock();
2477         if (dst->sa.sa_family == AF_INET) {
2478                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2479                 struct rtable *rt;
2480                 __be16 df = 0;
2481
2482                 if (!ifindex)
2483                         ifindex = sock4->sock->sk->sk_bound_dev_if;
2484
2485                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2486                                      dst->sin.sin_addr.s_addr,
2487                                      &local_ip.sin.sin_addr.s_addr,
2488                                      dst_port, src_port,
2489                                      dst_cache, info);
2490                 if (IS_ERR(rt)) {
2491                         err = PTR_ERR(rt);
2492                         goto tx_error;
2493                 }
2494
2495                 if (!info) {
2496                         /* Bypass encapsulation if the destination is local */
2497                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2498                                                     dst_port, ifindex, vni,
2499                                                     &rt->dst, rt->rt_flags);
2500                         if (err)
2501                                 goto out_unlock;
2502
2503                         if (vxlan->cfg.df == VXLAN_DF_SET) {
2504                                 df = htons(IP_DF);
2505                         } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) {
2506                                 struct ethhdr *eth = eth_hdr(skb);
2507
2508                                 if (ntohs(eth->h_proto) == ETH_P_IPV6 ||
2509                                     (ntohs(eth->h_proto) == ETH_P_IP &&
2510                                      old_iph->frag_off & htons(IP_DF)))
2511                                         df = htons(IP_DF);
2512                         }
2513                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2514                         df = htons(IP_DF);
2515                 }
2516
2517                 ndst = &rt->dst;
2518                 skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM);
2519
2520                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2521                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2522                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2523                                       vni, md, flags, udp_sum);
2524                 if (err < 0)
2525                         goto tx_error;
2526
2527                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2528                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2529                                     src_port, dst_port, xnet, !udp_sum);
2530 #if IS_ENABLED(CONFIG_IPV6)
2531         } else {
2532                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2533
2534                 if (!ifindex)
2535                         ifindex = sock6->sock->sk->sk_bound_dev_if;
2536
2537                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2538                                         label, &dst->sin6.sin6_addr,
2539                                         &local_ip.sin6.sin6_addr,
2540                                         dst_port, src_port,
2541                                         dst_cache, info);
2542                 if (IS_ERR(ndst)) {
2543                         err = PTR_ERR(ndst);
2544                         ndst = NULL;
2545                         goto tx_error;
2546                 }
2547
2548                 if (!info) {
2549                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2550
2551                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2552                                                     dst_port, ifindex, vni,
2553                                                     ndst, rt6i_flags);
2554                         if (err)
2555                                 goto out_unlock;
2556                 }
2557
2558                 skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM);
2559
2560                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2561                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2562                 skb_scrub_packet(skb, xnet);
2563                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2564                                       vni, md, flags, udp_sum);
2565                 if (err < 0)
2566                         goto tx_error;
2567
2568                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2569                                      &local_ip.sin6.sin6_addr,
2570                                      &dst->sin6.sin6_addr, tos, ttl,
2571                                      label, src_port, dst_port, !udp_sum);
2572 #endif
2573         }
2574 out_unlock:
2575         rcu_read_unlock();
2576         return;
2577
2578 drop:
2579         dev->stats.tx_dropped++;
2580         dev_kfree_skb(skb);
2581         return;
2582
2583 tx_error:
2584         rcu_read_unlock();
2585         if (err == -ELOOP)
2586                 dev->stats.collisions++;
2587         else if (err == -ENETUNREACH)
2588                 dev->stats.tx_carrier_errors++;
2589         dst_release(ndst);
2590         dev->stats.tx_errors++;
2591         kfree_skb(skb);
2592 }
2593
2594 /* Transmit local packets over Vxlan
2595  *
2596  * Outer IP header inherits ECN and DF from inner header.
2597  * Outer UDP destination is the VXLAN assigned port.
2598  *           source port is based on hash of flow
2599  */
2600 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2601 {
2602         struct vxlan_dev *vxlan = netdev_priv(dev);
2603         struct vxlan_rdst *rdst, *fdst = NULL;
2604         const struct ip_tunnel_info *info;
2605         bool did_rsc = false;
2606         struct vxlan_fdb *f;
2607         struct ethhdr *eth;
2608         __be32 vni = 0;
2609
2610         info = skb_tunnel_info(skb);
2611
2612         skb_reset_mac_header(skb);
2613
2614         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2615                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2616                     info->mode & IP_TUNNEL_INFO_TX) {
2617                         vni = tunnel_id_to_key32(info->key.tun_id);
2618                 } else {
2619                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2620                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2621                         else
2622                                 kfree_skb(skb);
2623                         return NETDEV_TX_OK;
2624                 }
2625         }
2626
2627         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2628                 eth = eth_hdr(skb);
2629                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2630                         return arp_reduce(dev, skb, vni);
2631 #if IS_ENABLED(CONFIG_IPV6)
2632                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2633                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2634                                             sizeof(struct nd_msg)) &&
2635                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2636                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2637
2638                         if (m->icmph.icmp6_code == 0 &&
2639                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2640                                 return neigh_reduce(dev, skb, vni);
2641                 }
2642 #endif
2643         }
2644
2645         eth = eth_hdr(skb);
2646         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2647         did_rsc = false;
2648
2649         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2650             (ntohs(eth->h_proto) == ETH_P_IP ||
2651              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2652                 did_rsc = route_shortcircuit(dev, skb);
2653                 if (did_rsc)
2654                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2655         }
2656
2657         if (f == NULL) {
2658                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2659                 if (f == NULL) {
2660                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2661                             !is_multicast_ether_addr(eth->h_dest))
2662                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2663
2664                         dev->stats.tx_dropped++;
2665                         kfree_skb(skb);
2666                         return NETDEV_TX_OK;
2667                 }
2668         }
2669
2670         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2671                 struct sk_buff *skb1;
2672
2673                 if (!fdst) {
2674                         fdst = rdst;
2675                         continue;
2676                 }
2677                 skb1 = skb_clone(skb, GFP_ATOMIC);
2678                 if (skb1)
2679                         vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2680         }
2681
2682         if (fdst)
2683                 vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2684         else
2685                 kfree_skb(skb);
2686         return NETDEV_TX_OK;
2687 }
2688
2689 /* Walk the forwarding table and purge stale entries */
2690 static void vxlan_cleanup(struct timer_list *t)
2691 {
2692         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2693         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2694         unsigned int h;
2695
2696         if (!netif_running(vxlan->dev))
2697                 return;
2698
2699         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2700                 struct hlist_node *p, *n;
2701
2702                 spin_lock(&vxlan->hash_lock);
2703                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2704                         struct vxlan_fdb *f
2705                                 = container_of(p, struct vxlan_fdb, hlist);
2706                         unsigned long timeout;
2707
2708                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2709                                 continue;
2710
2711                         if (f->flags & NTF_EXT_LEARNED)
2712                                 continue;
2713
2714                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2715                         if (time_before_eq(timeout, jiffies)) {
2716                                 netdev_dbg(vxlan->dev,
2717                                            "garbage collect %pM\n",
2718                                            f->eth_addr);
2719                                 f->state = NUD_STALE;
2720                                 vxlan_fdb_destroy(vxlan, f, true, true);
2721                         } else if (time_before(timeout, next_timer))
2722                                 next_timer = timeout;
2723                 }
2724                 spin_unlock(&vxlan->hash_lock);
2725         }
2726
2727         mod_timer(&vxlan->age_timer, next_timer);
2728 }
2729
2730 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2731 {
2732         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2733
2734         spin_lock(&vn->sock_lock);
2735         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2736 #if IS_ENABLED(CONFIG_IPV6)
2737         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2738 #endif
2739         spin_unlock(&vn->sock_lock);
2740 }
2741
2742 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2743                              struct vxlan_dev_node *node)
2744 {
2745         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2746         __be32 vni = vxlan->default_dst.remote_vni;
2747
2748         node->vxlan = vxlan;
2749         spin_lock(&vn->sock_lock);
2750         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2751         spin_unlock(&vn->sock_lock);
2752 }
2753
2754 /* Setup stats when device is created */
2755 static int vxlan_init(struct net_device *dev)
2756 {
2757         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2758         if (!dev->tstats)
2759                 return -ENOMEM;
2760
2761         return 0;
2762 }
2763
2764 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2765 {
2766         struct vxlan_fdb *f;
2767
2768         spin_lock_bh(&vxlan->hash_lock);
2769         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2770         if (f)
2771                 vxlan_fdb_destroy(vxlan, f, true, true);
2772         spin_unlock_bh(&vxlan->hash_lock);
2773 }
2774
2775 static void vxlan_uninit(struct net_device *dev)
2776 {
2777         struct vxlan_dev *vxlan = netdev_priv(dev);
2778
2779         gro_cells_destroy(&vxlan->gro_cells);
2780
2781         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2782
2783         free_percpu(dev->tstats);
2784 }
2785
2786 /* Start ageing timer and join group when device is brought up */
2787 static int vxlan_open(struct net_device *dev)
2788 {
2789         struct vxlan_dev *vxlan = netdev_priv(dev);
2790         int ret;
2791
2792         ret = vxlan_sock_add(vxlan);
2793         if (ret < 0)
2794                 return ret;
2795
2796         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2797                 ret = vxlan_igmp_join(vxlan);
2798                 if (ret == -EADDRINUSE)
2799                         ret = 0;
2800                 if (ret) {
2801                         vxlan_sock_release(vxlan);
2802                         return ret;
2803                 }
2804         }
2805
2806         if (vxlan->cfg.age_interval)
2807                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2808
2809         return ret;
2810 }
2811
2812 /* Purge the forwarding table */
2813 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2814 {
2815         unsigned int h;
2816
2817         spin_lock_bh(&vxlan->hash_lock);
2818         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2819                 struct hlist_node *p, *n;
2820                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2821                         struct vxlan_fdb *f
2822                                 = container_of(p, struct vxlan_fdb, hlist);
2823                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
2824                                 continue;
2825                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2826                         if (!is_zero_ether_addr(f->eth_addr))
2827                                 vxlan_fdb_destroy(vxlan, f, true, true);
2828                 }
2829         }
2830         spin_unlock_bh(&vxlan->hash_lock);
2831 }
2832
2833 /* Cleanup timer and forwarding table on shutdown */
2834 static int vxlan_stop(struct net_device *dev)
2835 {
2836         struct vxlan_dev *vxlan = netdev_priv(dev);
2837         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2838         int ret = 0;
2839
2840         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2841             !vxlan_group_used(vn, vxlan))
2842                 ret = vxlan_igmp_leave(vxlan);
2843
2844         del_timer_sync(&vxlan->age_timer);
2845
2846         vxlan_flush(vxlan, false);
2847         vxlan_sock_release(vxlan);
2848
2849         return ret;
2850 }
2851
2852 /* Stub, nothing needs to be done. */
2853 static void vxlan_set_multicast_list(struct net_device *dev)
2854 {
2855 }
2856
2857 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2858 {
2859         struct vxlan_dev *vxlan = netdev_priv(dev);
2860         struct vxlan_rdst *dst = &vxlan->default_dst;
2861         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2862                                                          dst->remote_ifindex);
2863         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
2864
2865         /* This check is different than dev->max_mtu, because it looks at
2866          * the lowerdev->mtu, rather than the static dev->max_mtu
2867          */
2868         if (lowerdev) {
2869                 int max_mtu = lowerdev->mtu -
2870                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2871                 if (new_mtu > max_mtu)
2872                         return -EINVAL;
2873         }
2874
2875         dev->mtu = new_mtu;
2876         return 0;
2877 }
2878
2879 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2880 {
2881         struct vxlan_dev *vxlan = netdev_priv(dev);
2882         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2883         __be16 sport, dport;
2884
2885         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2886                                   vxlan->cfg.port_max, true);
2887         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2888
2889         if (ip_tunnel_info_af(info) == AF_INET) {
2890                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2891                 struct rtable *rt;
2892
2893                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2894                                      info->key.u.ipv4.dst,
2895                                      &info->key.u.ipv4.src, dport, sport,
2896                                      &info->dst_cache, info);
2897                 if (IS_ERR(rt))
2898                         return PTR_ERR(rt);
2899                 ip_rt_put(rt);
2900         } else {
2901 #if IS_ENABLED(CONFIG_IPV6)
2902                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2903                 struct dst_entry *ndst;
2904
2905                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2906                                         info->key.label, &info->key.u.ipv6.dst,
2907                                         &info->key.u.ipv6.src, dport, sport,
2908                                         &info->dst_cache, info);
2909                 if (IS_ERR(ndst))
2910                         return PTR_ERR(ndst);
2911                 dst_release(ndst);
2912 #else /* !CONFIG_IPV6 */
2913                 return -EPFNOSUPPORT;
2914 #endif
2915         }
2916         info->key.tp_src = sport;
2917         info->key.tp_dst = dport;
2918         return 0;
2919 }
2920
2921 static const struct net_device_ops vxlan_netdev_ether_ops = {
2922         .ndo_init               = vxlan_init,
2923         .ndo_uninit             = vxlan_uninit,
2924         .ndo_open               = vxlan_open,
2925         .ndo_stop               = vxlan_stop,
2926         .ndo_start_xmit         = vxlan_xmit,
2927         .ndo_get_stats64        = ip_tunnel_get_stats64,
2928         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2929         .ndo_change_mtu         = vxlan_change_mtu,
2930         .ndo_validate_addr      = eth_validate_addr,
2931         .ndo_set_mac_address    = eth_mac_addr,
2932         .ndo_fdb_add            = vxlan_fdb_add,
2933         .ndo_fdb_del            = vxlan_fdb_delete,
2934         .ndo_fdb_dump           = vxlan_fdb_dump,
2935         .ndo_fdb_get            = vxlan_fdb_get,
2936         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2937         .ndo_change_proto_down  = dev_change_proto_down_generic,
2938 };
2939
2940 static const struct net_device_ops vxlan_netdev_raw_ops = {
2941         .ndo_init               = vxlan_init,
2942         .ndo_uninit             = vxlan_uninit,
2943         .ndo_open               = vxlan_open,
2944         .ndo_stop               = vxlan_stop,
2945         .ndo_start_xmit         = vxlan_xmit,
2946         .ndo_get_stats64        = ip_tunnel_get_stats64,
2947         .ndo_change_mtu         = vxlan_change_mtu,
2948         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2949 };
2950
2951 /* Info for udev, that this is a virtual tunnel endpoint */
2952 static struct device_type vxlan_type = {
2953         .name = "vxlan",
2954 };
2955
2956 /* Calls the ndo_udp_tunnel_add of the caller in order to
2957  * supply the listening VXLAN udp ports. Callers are expected
2958  * to implement the ndo_udp_tunnel_add.
2959  */
2960 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
2961 {
2962         struct vxlan_sock *vs;
2963         struct net *net = dev_net(dev);
2964         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2965         unsigned int i;
2966
2967         spin_lock(&vn->sock_lock);
2968         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2969                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
2970                         unsigned short type;
2971
2972                         if (vs->flags & VXLAN_F_GPE)
2973                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
2974                         else
2975                                 type = UDP_TUNNEL_TYPE_VXLAN;
2976
2977                         if (push)
2978                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
2979                         else
2980                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
2981                 }
2982         }
2983         spin_unlock(&vn->sock_lock);
2984 }
2985
2986 /* Initialize the device structure. */
2987 static void vxlan_setup(struct net_device *dev)
2988 {
2989         struct vxlan_dev *vxlan = netdev_priv(dev);
2990         unsigned int h;
2991
2992         eth_hw_addr_random(dev);
2993         ether_setup(dev);
2994
2995         dev->needs_free_netdev = true;
2996         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2997
2998         dev->features   |= NETIF_F_LLTX;
2999         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
3000         dev->features   |= NETIF_F_RXCSUM;
3001         dev->features   |= NETIF_F_GSO_SOFTWARE;
3002
3003         dev->vlan_features = dev->features;
3004         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
3005         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
3006         netif_keep_dst(dev);
3007         dev->priv_flags |= IFF_NO_QUEUE;
3008
3009         /* MTU range: 68 - 65535 */
3010         dev->min_mtu = ETH_MIN_MTU;
3011         dev->max_mtu = ETH_MAX_MTU;
3012
3013         INIT_LIST_HEAD(&vxlan->next);
3014         spin_lock_init(&vxlan->hash_lock);
3015
3016         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
3017
3018         vxlan->dev = dev;
3019
3020         gro_cells_init(&vxlan->gro_cells, dev);
3021
3022         for (h = 0; h < FDB_HASH_SIZE; ++h)
3023                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
3024 }
3025
3026 static void vxlan_ether_setup(struct net_device *dev)
3027 {
3028         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
3029         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3030         dev->netdev_ops = &vxlan_netdev_ether_ops;
3031 }
3032
3033 static void vxlan_raw_setup(struct net_device *dev)
3034 {
3035         dev->header_ops = NULL;
3036         dev->type = ARPHRD_NONE;
3037         dev->hard_header_len = 0;
3038         dev->addr_len = 0;
3039         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
3040         dev->netdev_ops = &vxlan_netdev_raw_ops;
3041 }
3042
3043 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
3044         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
3045         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
3046         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
3047         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
3048         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
3049         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
3050         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
3051         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
3052         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
3053         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
3054         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
3055         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
3056         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
3057         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
3058         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
3059         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
3060         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
3061         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
3062         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
3063         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
3064         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
3065         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
3066         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
3067         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
3068         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
3069         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
3070         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
3071         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
3072         [IFLA_VXLAN_DF]         = { .type = NLA_U8 },
3073 };
3074
3075 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
3076                           struct netlink_ext_ack *extack)
3077 {
3078         if (tb[IFLA_ADDRESS]) {
3079                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
3080                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3081                                             "Provided link layer address is not Ethernet");
3082                         return -EINVAL;
3083                 }
3084
3085                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
3086                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3087                                             "Provided Ethernet address is not unicast");
3088                         return -EADDRNOTAVAIL;
3089                 }
3090         }
3091
3092         if (tb[IFLA_MTU]) {
3093                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
3094
3095                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
3096                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3097                                             "MTU must be between 68 and 65535");
3098                         return -EINVAL;
3099                 }
3100         }
3101
3102         if (!data) {
3103                 NL_SET_ERR_MSG(extack,
3104                                "Required attributes not provided to perform the operation");
3105                 return -EINVAL;
3106         }
3107
3108         if (data[IFLA_VXLAN_ID]) {
3109                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
3110
3111                 if (id >= VXLAN_N_VID) {
3112                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID],
3113                                             "VXLAN ID must be lower than 16777216");
3114                         return -ERANGE;
3115                 }
3116         }
3117
3118         if (data[IFLA_VXLAN_PORT_RANGE]) {
3119                 const struct ifla_vxlan_port_range *p
3120                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3121
3122                 if (ntohs(p->high) < ntohs(p->low)) {
3123                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
3124                                             "Invalid source port range");
3125                         return -EINVAL;
3126                 }
3127         }
3128
3129         if (data[IFLA_VXLAN_DF]) {
3130                 enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]);
3131
3132                 if (df < 0 || df > VXLAN_DF_MAX) {
3133                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_DF],
3134                                             "Invalid DF attribute");
3135                         return -EINVAL;
3136                 }
3137         }
3138
3139         return 0;
3140 }
3141
3142 static void vxlan_get_drvinfo(struct net_device *netdev,
3143                               struct ethtool_drvinfo *drvinfo)
3144 {
3145         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
3146         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
3147 }
3148
3149 static const struct ethtool_ops vxlan_ethtool_ops = {
3150         .get_drvinfo    = vxlan_get_drvinfo,
3151         .get_link       = ethtool_op_get_link,
3152 };
3153
3154 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
3155                                         __be16 port, u32 flags, int ifindex)
3156 {
3157         struct socket *sock;
3158         struct udp_port_cfg udp_conf;
3159         int err;
3160
3161         memset(&udp_conf, 0, sizeof(udp_conf));
3162
3163         if (ipv6) {
3164                 udp_conf.family = AF_INET6;
3165                 udp_conf.use_udp6_rx_checksums =
3166                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
3167                 udp_conf.ipv6_v6only = 1;
3168         } else {
3169                 udp_conf.family = AF_INET;
3170         }
3171
3172         udp_conf.local_udp_port = port;
3173         udp_conf.bind_ifindex = ifindex;
3174
3175         /* Open UDP socket */
3176         err = udp_sock_create(net, &udp_conf, &sock);
3177         if (err < 0)
3178                 return ERR_PTR(err);
3179
3180         return sock;
3181 }
3182
3183 /* Create new listen socket if needed */
3184 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
3185                                               __be16 port, u32 flags,
3186                                               int ifindex)
3187 {
3188         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3189         struct vxlan_sock *vs;
3190         struct socket *sock;
3191         unsigned int h;
3192         struct udp_tunnel_sock_cfg tunnel_cfg;
3193
3194         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
3195         if (!vs)
3196                 return ERR_PTR(-ENOMEM);
3197
3198         for (h = 0; h < VNI_HASH_SIZE; ++h)
3199                 INIT_HLIST_HEAD(&vs->vni_list[h]);
3200
3201         sock = vxlan_create_sock(net, ipv6, port, flags, ifindex);
3202         if (IS_ERR(sock)) {
3203                 kfree(vs);
3204                 return ERR_CAST(sock);
3205         }
3206
3207         vs->sock = sock;
3208         refcount_set(&vs->refcnt, 1);
3209         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
3210
3211         spin_lock(&vn->sock_lock);
3212         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
3213         udp_tunnel_notify_add_rx_port(sock,
3214                                       (vs->flags & VXLAN_F_GPE) ?
3215                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
3216                                       UDP_TUNNEL_TYPE_VXLAN);
3217         spin_unlock(&vn->sock_lock);
3218
3219         /* Mark socket as an encapsulation socket. */
3220         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
3221         tunnel_cfg.sk_user_data = vs;
3222         tunnel_cfg.encap_type = 1;
3223         tunnel_cfg.encap_rcv = vxlan_rcv;
3224         tunnel_cfg.encap_err_lookup = vxlan_err_lookup;
3225         tunnel_cfg.encap_destroy = NULL;
3226         tunnel_cfg.gro_receive = vxlan_gro_receive;
3227         tunnel_cfg.gro_complete = vxlan_gro_complete;
3228
3229         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
3230
3231         return vs;
3232 }
3233
3234 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
3235 {
3236         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3237         struct vxlan_sock *vs = NULL;
3238         struct vxlan_dev_node *node;
3239         int l3mdev_index = 0;
3240
3241         if (vxlan->cfg.remote_ifindex)
3242                 l3mdev_index = l3mdev_master_upper_ifindex_by_index(
3243                         vxlan->net, vxlan->cfg.remote_ifindex);
3244
3245         if (!vxlan->cfg.no_share) {
3246                 spin_lock(&vn->sock_lock);
3247                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
3248                                      vxlan->cfg.dst_port, vxlan->cfg.flags,
3249                                      l3mdev_index);
3250                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
3251                         spin_unlock(&vn->sock_lock);
3252                         return -EBUSY;
3253                 }
3254                 spin_unlock(&vn->sock_lock);
3255         }
3256         if (!vs)
3257                 vs = vxlan_socket_create(vxlan->net, ipv6,
3258                                          vxlan->cfg.dst_port, vxlan->cfg.flags,
3259                                          l3mdev_index);
3260         if (IS_ERR(vs))
3261                 return PTR_ERR(vs);
3262 #if IS_ENABLED(CONFIG_IPV6)
3263         if (ipv6) {
3264                 rcu_assign_pointer(vxlan->vn6_sock, vs);
3265                 node = &vxlan->hlist6;
3266         } else
3267 #endif
3268         {
3269                 rcu_assign_pointer(vxlan->vn4_sock, vs);
3270                 node = &vxlan->hlist4;
3271         }
3272         vxlan_vs_add_dev(vs, vxlan, node);
3273         return 0;
3274 }
3275
3276 static int vxlan_sock_add(struct vxlan_dev *vxlan)
3277 {
3278         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
3279         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
3280         bool ipv4 = !ipv6 || metadata;
3281         int ret = 0;
3282
3283         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
3284 #if IS_ENABLED(CONFIG_IPV6)
3285         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
3286         if (ipv6) {
3287                 ret = __vxlan_sock_add(vxlan, true);
3288                 if (ret < 0 && ret != -EAFNOSUPPORT)
3289                         ipv4 = false;
3290         }
3291 #endif
3292         if (ipv4)
3293                 ret = __vxlan_sock_add(vxlan, false);
3294         if (ret < 0)
3295                 vxlan_sock_release(vxlan);
3296         return ret;
3297 }
3298
3299 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
3300                                  struct net_device **lower,
3301                                  struct vxlan_dev *old,
3302                                  struct netlink_ext_ack *extack)
3303 {
3304         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
3305         struct vxlan_dev *tmp;
3306         bool use_ipv6 = false;
3307
3308         if (conf->flags & VXLAN_F_GPE) {
3309                 /* For now, allow GPE only together with
3310                  * COLLECT_METADATA. This can be relaxed later; in such
3311                  * case, the other side of the PtP link will have to be
3312                  * provided.
3313                  */
3314                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
3315                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3316                         NL_SET_ERR_MSG(extack,
3317                                        "VXLAN GPE does not support this combination of attributes");
3318                         return -EINVAL;
3319                 }
3320         }
3321
3322         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
3323                 /* Unless IPv6 is explicitly requested, assume IPv4 */
3324                 conf->remote_ip.sa.sa_family = AF_INET;
3325                 conf->saddr.sa.sa_family = AF_INET;
3326         } else if (!conf->remote_ip.sa.sa_family) {
3327                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
3328         } else if (!conf->saddr.sa.sa_family) {
3329                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
3330         }
3331
3332         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
3333                 NL_SET_ERR_MSG(extack,
3334                                "Local and remote address must be from the same family");
3335                 return -EINVAL;
3336         }
3337
3338         if (vxlan_addr_multicast(&conf->saddr)) {
3339                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
3340                 return -EINVAL;
3341         }
3342
3343         if (conf->saddr.sa.sa_family == AF_INET6) {
3344                 if (!IS_ENABLED(CONFIG_IPV6)) {
3345                         NL_SET_ERR_MSG(extack,
3346                                        "IPv6 support not enabled in the kernel");
3347                         return -EPFNOSUPPORT;
3348                 }
3349                 use_ipv6 = true;
3350                 conf->flags |= VXLAN_F_IPV6;
3351
3352                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3353                         int local_type =
3354                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3355                         int remote_type =
3356                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3357
3358                         if (local_type & IPV6_ADDR_LINKLOCAL) {
3359                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3360                                     (remote_type != IPV6_ADDR_ANY)) {
3361                                         NL_SET_ERR_MSG(extack,
3362                                                        "Invalid combination of local and remote address scopes");
3363                                         return -EINVAL;
3364                                 }
3365
3366                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3367                         } else {
3368                                 if (remote_type ==
3369                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3370                                         NL_SET_ERR_MSG(extack,
3371                                                        "Invalid combination of local and remote address scopes");
3372                                         return -EINVAL;
3373                                 }
3374
3375                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3376                         }
3377                 }
3378         }
3379
3380         if (conf->label && !use_ipv6) {
3381                 NL_SET_ERR_MSG(extack,
3382                                "Label attribute only applies to IPv6 VXLAN devices");
3383                 return -EINVAL;
3384         }
3385
3386         if (conf->remote_ifindex) {
3387                 struct net_device *lowerdev;
3388
3389                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3390                 if (!lowerdev) {
3391                         NL_SET_ERR_MSG(extack,
3392                                        "Invalid local interface, device not found");
3393                         return -ENODEV;
3394                 }
3395
3396 #if IS_ENABLED(CONFIG_IPV6)
3397                 if (use_ipv6) {
3398                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3399                         if (idev && idev->cnf.disable_ipv6) {
3400                                 NL_SET_ERR_MSG(extack,
3401                                                "IPv6 support disabled by administrator");
3402                                 return -EPERM;
3403                         }
3404                 }
3405 #endif
3406
3407                 *lower = lowerdev;
3408         } else {
3409                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3410                         NL_SET_ERR_MSG(extack,
3411                                        "Local interface required for multicast remote destination");
3412
3413                         return -EINVAL;
3414                 }
3415
3416 #if IS_ENABLED(CONFIG_IPV6)
3417                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3418                         NL_SET_ERR_MSG(extack,
3419                                        "Local interface required for link-local local/remote addresses");
3420                         return -EINVAL;
3421                 }
3422 #endif
3423
3424                 *lower = NULL;
3425         }
3426
3427         if (!conf->dst_port) {
3428                 if (conf->flags & VXLAN_F_GPE)
3429                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3430                 else
3431                         conf->dst_port = htons(vxlan_port);
3432         }
3433
3434         if (!conf->age_interval)
3435                 conf->age_interval = FDB_AGE_DEFAULT;
3436
3437         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3438                 if (tmp == old)
3439                         continue;
3440
3441                 if (tmp->cfg.vni != conf->vni)
3442                         continue;
3443                 if (tmp->cfg.dst_port != conf->dst_port)
3444                         continue;
3445                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3446                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3447                         continue;
3448
3449                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3450                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3451                         continue;
3452
3453                 NL_SET_ERR_MSG(extack,
3454                                "A VXLAN device with the specified VNI already exists");
3455                 return -EEXIST;
3456         }
3457
3458         return 0;
3459 }
3460
3461 static void vxlan_config_apply(struct net_device *dev,
3462                                struct vxlan_config *conf,
3463                                struct net_device *lowerdev,
3464                                struct net *src_net,
3465                                bool changelink)
3466 {
3467         struct vxlan_dev *vxlan = netdev_priv(dev);
3468         struct vxlan_rdst *dst = &vxlan->default_dst;
3469         unsigned short needed_headroom = ETH_HLEN;
3470         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3471         int max_mtu = ETH_MAX_MTU;
3472
3473         if (!changelink) {
3474                 if (conf->flags & VXLAN_F_GPE)
3475                         vxlan_raw_setup(dev);
3476                 else
3477                         vxlan_ether_setup(dev);
3478
3479                 if (conf->mtu)
3480                         dev->mtu = conf->mtu;
3481
3482                 vxlan->net = src_net;
3483         }
3484
3485         dst->remote_vni = conf->vni;
3486
3487         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3488
3489         if (lowerdev) {
3490                 dst->remote_ifindex = conf->remote_ifindex;
3491
3492                 dev->gso_max_size = lowerdev->gso_max_size;
3493                 dev->gso_max_segs = lowerdev->gso_max_segs;
3494
3495                 needed_headroom = lowerdev->hard_header_len;
3496
3497                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3498                                            VXLAN_HEADROOM);
3499                 if (max_mtu < ETH_MIN_MTU)
3500                         max_mtu = ETH_MIN_MTU;
3501
3502                 if (!changelink && !conf->mtu)
3503                         dev->mtu = max_mtu;
3504         }
3505
3506         if (dev->mtu > max_mtu)
3507                 dev->mtu = max_mtu;
3508
3509         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3510                 needed_headroom += VXLAN6_HEADROOM;
3511         else
3512                 needed_headroom += VXLAN_HEADROOM;
3513         dev->needed_headroom = needed_headroom;
3514
3515         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3516 }
3517
3518 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3519                                struct vxlan_config *conf, bool changelink,
3520                                struct netlink_ext_ack *extack)
3521 {
3522         struct vxlan_dev *vxlan = netdev_priv(dev);
3523         struct net_device *lowerdev;
3524         int ret;
3525
3526         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3527         if (ret)
3528                 return ret;
3529
3530         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3531
3532         return 0;
3533 }
3534
3535 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3536                               struct vxlan_config *conf,
3537                               struct netlink_ext_ack *extack)
3538 {
3539         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3540         struct vxlan_dev *vxlan = netdev_priv(dev);
3541         struct vxlan_fdb *f = NULL;
3542         bool unregister = false;
3543         int err;
3544
3545         err = vxlan_dev_configure(net, dev, conf, false, extack);
3546         if (err)
3547                 return err;
3548
3549         dev->ethtool_ops = &vxlan_ethtool_ops;
3550
3551         /* create an fdb entry for a valid default destination */
3552         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
3553                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3554                                        &vxlan->default_dst.remote_ip,
3555                                        NUD_REACHABLE | NUD_PERMANENT,
3556                                        vxlan->cfg.dst_port,
3557                                        vxlan->default_dst.remote_vni,
3558                                        vxlan->default_dst.remote_vni,
3559                                        vxlan->default_dst.remote_ifindex,
3560                                        NTF_SELF, &f);
3561                 if (err)
3562                         return err;
3563         }
3564
3565         err = register_netdevice(dev);
3566         if (err)
3567                 goto errout;
3568         unregister = true;
3569
3570         err = rtnl_configure_link(dev, NULL);
3571         if (err)
3572                 goto errout;
3573
3574         /* notify default fdb entry */
3575         if (f) {
3576                 err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f),
3577                                        RTM_NEWNEIGH, true, extack);
3578                 if (err)
3579                         goto errout;
3580         }
3581
3582         list_add(&vxlan->next, &vn->vxlan_list);
3583         return 0;
3584
3585 errout:
3586         /* unregister_netdevice() destroys the default FDB entry with deletion
3587          * notification. But the addition notification was not sent yet, so
3588          * destroy the entry by hand here.
3589          */
3590         if (f)
3591                 vxlan_fdb_destroy(vxlan, f, false, false);
3592         if (unregister)
3593                 unregister_netdevice(dev);
3594         return err;
3595 }
3596
3597 /* Set/clear flags based on attribute */
3598 static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[],
3599                           int attrtype, unsigned long mask, bool changelink,
3600                           bool changelink_supported,
3601                           struct netlink_ext_ack *extack)
3602 {
3603         unsigned long flags;
3604
3605         if (!tb[attrtype])
3606                 return 0;
3607
3608         if (changelink && !changelink_supported) {
3609                 vxlan_flag_attr_error(attrtype, extack);
3610                 return -EOPNOTSUPP;
3611         }
3612
3613         if (vxlan_policy[attrtype].type == NLA_FLAG)
3614                 flags = conf->flags | mask;
3615         else if (nla_get_u8(tb[attrtype]))
3616                 flags = conf->flags | mask;
3617         else
3618                 flags = conf->flags & ~mask;
3619
3620         conf->flags = flags;
3621
3622         return 0;
3623 }
3624
3625 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3626                          struct net_device *dev, struct vxlan_config *conf,
3627                          bool changelink, struct netlink_ext_ack *extack)
3628 {
3629         struct vxlan_dev *vxlan = netdev_priv(dev);
3630         int err = 0;
3631
3632         memset(conf, 0, sizeof(*conf));
3633
3634         /* if changelink operation, start with old existing cfg */
3635         if (changelink)
3636                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3637
3638         if (data[IFLA_VXLAN_ID]) {
3639                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3640
3641                 if (changelink && (vni != conf->vni)) {
3642                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI");
3643                         return -EOPNOTSUPP;
3644                 }
3645                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3646         }
3647
3648         if (data[IFLA_VXLAN_GROUP]) {
3649                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) {
3650                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group");
3651                         return -EOPNOTSUPP;
3652                 }
3653
3654                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3655                 conf->remote_ip.sa.sa_family = AF_INET;
3656         } else if (data[IFLA_VXLAN_GROUP6]) {
3657                 if (!IS_ENABLED(CONFIG_IPV6)) {
3658                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel");
3659                         return -EPFNOSUPPORT;
3660                 }
3661
3662                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) {
3663                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group");
3664                         return -EOPNOTSUPP;
3665                 }
3666
3667                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3668                 conf->remote_ip.sa.sa_family = AF_INET6;
3669         }
3670
3671         if (data[IFLA_VXLAN_LOCAL]) {
3672                 if (changelink && (conf->saddr.sa.sa_family != AF_INET)) {
3673                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old");
3674                         return -EOPNOTSUPP;
3675                 }
3676
3677                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3678                 conf->saddr.sa.sa_family = AF_INET;
3679         } else if (data[IFLA_VXLAN_LOCAL6]) {
3680                 if (!IS_ENABLED(CONFIG_IPV6)) {
3681                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel");
3682                         return -EPFNOSUPPORT;
3683                 }
3684
3685                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) {
3686                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old");
3687                         return -EOPNOTSUPP;
3688                 }
3689
3690                 /* TODO: respect scope id */
3691                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3692                 conf->saddr.sa.sa_family = AF_INET6;
3693         }
3694
3695         if (data[IFLA_VXLAN_LINK])
3696                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3697
3698         if (data[IFLA_VXLAN_TOS])
3699                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3700
3701         if (data[IFLA_VXLAN_TTL])
3702                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3703
3704         if (data[IFLA_VXLAN_TTL_INHERIT]) {
3705                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT,
3706                                     VXLAN_F_TTL_INHERIT, changelink, false,
3707                                     extack);
3708                 if (err)
3709                         return err;
3710
3711         }
3712
3713         if (data[IFLA_VXLAN_LABEL])
3714                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3715                              IPV6_FLOWLABEL_MASK;
3716
3717         if (data[IFLA_VXLAN_LEARNING]) {
3718                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING,
3719                                     VXLAN_F_LEARN, changelink, true,
3720                                     extack);
3721                 if (err)
3722                         return err;
3723         } else if (!changelink) {
3724                 /* default to learn on a new device */
3725                 conf->flags |= VXLAN_F_LEARN;
3726         }
3727
3728         if (data[IFLA_VXLAN_AGEING])
3729                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3730
3731         if (data[IFLA_VXLAN_PROXY]) {
3732                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY,
3733                                     VXLAN_F_PROXY, changelink, false,
3734                                     extack);
3735                 if (err)
3736                         return err;
3737         }
3738
3739         if (data[IFLA_VXLAN_RSC]) {
3740                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC,
3741                                     VXLAN_F_RSC, changelink, false,
3742                                     extack);
3743                 if (err)
3744                         return err;
3745         }
3746
3747         if (data[IFLA_VXLAN_L2MISS]) {
3748                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS,
3749                                     VXLAN_F_L2MISS, changelink, false,
3750                                     extack);
3751                 if (err)
3752                         return err;
3753         }
3754
3755         if (data[IFLA_VXLAN_L3MISS]) {
3756                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS,
3757                                     VXLAN_F_L3MISS, changelink, false,
3758                                     extack);
3759                 if (err)
3760                         return err;
3761         }
3762
3763         if (data[IFLA_VXLAN_LIMIT]) {
3764                 if (changelink) {
3765                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT],
3766                                             "Cannot change limit");
3767                         return -EOPNOTSUPP;
3768                 }
3769                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
3770         }
3771
3772         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
3773                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA,
3774                                     VXLAN_F_COLLECT_METADATA, changelink, false,
3775                                     extack);
3776                 if (err)
3777                         return err;
3778         }
3779
3780         if (data[IFLA_VXLAN_PORT_RANGE]) {
3781                 if (!changelink) {
3782                         const struct ifla_vxlan_port_range *p
3783                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3784                         conf->port_min = ntohs(p->low);
3785                         conf->port_max = ntohs(p->high);
3786                 } else {
3787                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
3788                                             "Cannot change port range");
3789                         return -EOPNOTSUPP;
3790                 }
3791         }
3792
3793         if (data[IFLA_VXLAN_PORT]) {
3794                 if (changelink) {
3795                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT],
3796                                             "Cannot change port");
3797                         return -EOPNOTSUPP;
3798                 }
3799                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3800         }
3801
3802         if (data[IFLA_VXLAN_UDP_CSUM]) {
3803                 if (changelink) {
3804                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM],
3805                                             "Cannot change UDP_CSUM flag");
3806                         return -EOPNOTSUPP;
3807                 }
3808                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3809                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3810         }
3811
3812         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
3813                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3814                                     VXLAN_F_UDP_ZERO_CSUM6_TX, changelink,
3815                                     false, extack);
3816                 if (err)
3817                         return err;
3818         }
3819
3820         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
3821                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3822                                     VXLAN_F_UDP_ZERO_CSUM6_RX, changelink,
3823                                     false, extack);
3824                 if (err)
3825                         return err;
3826         }
3827
3828         if (data[IFLA_VXLAN_REMCSUM_TX]) {
3829                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX,
3830                                     VXLAN_F_REMCSUM_TX, changelink, false,
3831                                     extack);
3832                 if (err)
3833                         return err;
3834         }
3835
3836         if (data[IFLA_VXLAN_REMCSUM_RX]) {
3837                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX,
3838                                     VXLAN_F_REMCSUM_RX, changelink, false,
3839                                     extack);
3840                 if (err)
3841                         return err;
3842         }
3843
3844         if (data[IFLA_VXLAN_GBP]) {
3845                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP,
3846                                     VXLAN_F_GBP, changelink, false, extack);
3847                 if (err)
3848                         return err;
3849         }
3850
3851         if (data[IFLA_VXLAN_GPE]) {
3852                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE,
3853                                     VXLAN_F_GPE, changelink, false,
3854                                     extack);
3855                 if (err)
3856                         return err;
3857         }
3858
3859         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
3860                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL,
3861                                     VXLAN_F_REMCSUM_NOPARTIAL, changelink,
3862                                     false, extack);
3863                 if (err)
3864                         return err;
3865         }
3866
3867         if (tb[IFLA_MTU]) {
3868                 if (changelink) {
3869                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3870                                             "Cannot change mtu");
3871                         return -EOPNOTSUPP;
3872                 }
3873                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
3874         }
3875
3876         if (data[IFLA_VXLAN_DF])
3877                 conf->df = nla_get_u8(data[IFLA_VXLAN_DF]);
3878
3879         return 0;
3880 }
3881
3882 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
3883                          struct nlattr *tb[], struct nlattr *data[],
3884                          struct netlink_ext_ack *extack)
3885 {
3886         struct vxlan_config conf;
3887         int err;
3888
3889         err = vxlan_nl2conf(tb, data, dev, &conf, false, extack);
3890         if (err)
3891                 return err;
3892
3893         return __vxlan_dev_create(src_net, dev, &conf, extack);
3894 }
3895
3896 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
3897                             struct nlattr *data[],
3898                             struct netlink_ext_ack *extack)
3899 {
3900         struct vxlan_dev *vxlan = netdev_priv(dev);
3901         struct vxlan_rdst *dst = &vxlan->default_dst;
3902         struct net_device *lowerdev;
3903         struct vxlan_config conf;
3904         int err;
3905
3906         err = vxlan_nl2conf(tb, data, dev, &conf, true, extack);
3907         if (err)
3908                 return err;
3909
3910         err = vxlan_config_validate(vxlan->net, &conf, &lowerdev,
3911                                     vxlan, extack);
3912         if (err)
3913                 return err;
3914
3915         /* handle default dst entry */
3916         if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) {
3917                 spin_lock_bh(&vxlan->hash_lock);
3918                 if (!vxlan_addr_any(&conf.remote_ip)) {
3919                         err = vxlan_fdb_update(vxlan, all_zeros_mac,
3920                                                &conf.remote_ip,
3921                                                NUD_REACHABLE | NUD_PERMANENT,
3922                                                NLM_F_APPEND | NLM_F_CREATE,
3923                                                vxlan->cfg.dst_port,
3924                                                conf.vni, conf.vni,
3925                                                conf.remote_ifindex,
3926                                                NTF_SELF, true, extack);
3927                         if (err) {
3928                                 spin_unlock_bh(&vxlan->hash_lock);
3929                                 return err;
3930                         }
3931                 }
3932                 if (!vxlan_addr_any(&dst->remote_ip))
3933                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
3934                                            dst->remote_ip,
3935                                            vxlan->cfg.dst_port,
3936                                            dst->remote_vni,
3937                                            dst->remote_vni,
3938                                            dst->remote_ifindex,
3939                                            true);
3940                 spin_unlock_bh(&vxlan->hash_lock);
3941         }
3942
3943         if (conf.age_interval != vxlan->cfg.age_interval)
3944                 mod_timer(&vxlan->age_timer, jiffies);
3945
3946         vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true);
3947         return 0;
3948 }
3949
3950 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3951 {
3952         struct vxlan_dev *vxlan = netdev_priv(dev);
3953
3954         vxlan_flush(vxlan, true);
3955
3956         list_del(&vxlan->next);
3957         unregister_netdevice_queue(dev, head);
3958 }
3959
3960 static size_t vxlan_get_size(const struct net_device *dev)
3961 {
3962
3963         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3964                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3965                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3966                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3967                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3968                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL_INHERIT */
3969                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3970                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_DF */
3971                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3972                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3973                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3974                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3975                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3976                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3977                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3978                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3979                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3980                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3981                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3982                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3983                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3984                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3985                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3986                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3987                 0;
3988 }
3989
3990 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3991 {
3992         const struct vxlan_dev *vxlan = netdev_priv(dev);
3993         const struct vxlan_rdst *dst = &vxlan->default_dst;
3994         struct ifla_vxlan_port_range ports = {
3995                 .low =  htons(vxlan->cfg.port_min),
3996                 .high = htons(vxlan->cfg.port_max),
3997         };
3998
3999         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
4000                 goto nla_put_failure;
4001
4002         if (!vxlan_addr_any(&dst->remote_ip)) {
4003                 if (dst->remote_ip.sa.sa_family == AF_INET) {
4004                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
4005                                             dst->remote_ip.sin.sin_addr.s_addr))
4006                                 goto nla_put_failure;
4007 #if IS_ENABLED(CONFIG_IPV6)
4008                 } else {
4009                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
4010                                              &dst->remote_ip.sin6.sin6_addr))
4011                                 goto nla_put_failure;
4012 #endif
4013                 }
4014         }
4015
4016         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
4017                 goto nla_put_failure;
4018
4019         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
4020                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
4021                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
4022                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
4023                                 goto nla_put_failure;
4024 #if IS_ENABLED(CONFIG_IPV6)
4025                 } else {
4026                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
4027                                              &vxlan->cfg.saddr.sin6.sin6_addr))
4028                                 goto nla_put_failure;
4029 #endif
4030                 }
4031         }
4032
4033         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
4034             nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT,
4035                        !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) ||
4036             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
4037             nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) ||
4038             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
4039             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
4040                         !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
4041             nla_put_u8(skb, IFLA_VXLAN_PROXY,
4042                         !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
4043             nla_put_u8(skb, IFLA_VXLAN_RSC,
4044                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
4045             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
4046                         !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
4047             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
4048                         !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
4049             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
4050                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
4051             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
4052             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
4053             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
4054             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
4055                         !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
4056             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4057                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
4058             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4059                         !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
4060             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
4061                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
4062             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
4063                         !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
4064                 goto nla_put_failure;
4065
4066         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
4067                 goto nla_put_failure;
4068
4069         if (vxlan->cfg.flags & VXLAN_F_GBP &&
4070             nla_put_flag(skb, IFLA_VXLAN_GBP))
4071                 goto nla_put_failure;
4072
4073         if (vxlan->cfg.flags & VXLAN_F_GPE &&
4074             nla_put_flag(skb, IFLA_VXLAN_GPE))
4075                 goto nla_put_failure;
4076
4077         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
4078             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
4079                 goto nla_put_failure;
4080
4081         return 0;
4082
4083 nla_put_failure:
4084         return -EMSGSIZE;
4085 }
4086
4087 static struct net *vxlan_get_link_net(const struct net_device *dev)
4088 {
4089         struct vxlan_dev *vxlan = netdev_priv(dev);
4090
4091         return vxlan->net;
4092 }
4093
4094 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
4095         .kind           = "vxlan",
4096         .maxtype        = IFLA_VXLAN_MAX,
4097         .policy         = vxlan_policy,
4098         .priv_size      = sizeof(struct vxlan_dev),
4099         .setup          = vxlan_setup,
4100         .validate       = vxlan_validate,
4101         .newlink        = vxlan_newlink,
4102         .changelink     = vxlan_changelink,
4103         .dellink        = vxlan_dellink,
4104         .get_size       = vxlan_get_size,
4105         .fill_info      = vxlan_fill_info,
4106         .get_link_net   = vxlan_get_link_net,
4107 };
4108
4109 struct net_device *vxlan_dev_create(struct net *net, const char *name,
4110                                     u8 name_assign_type,
4111                                     struct vxlan_config *conf)
4112 {
4113         struct nlattr *tb[IFLA_MAX + 1];
4114         struct net_device *dev;
4115         int err;
4116
4117         memset(&tb, 0, sizeof(tb));
4118
4119         dev = rtnl_create_link(net, name, name_assign_type,
4120                                &vxlan_link_ops, tb, NULL);
4121         if (IS_ERR(dev))
4122                 return dev;
4123
4124         err = __vxlan_dev_create(net, dev, conf, NULL);
4125         if (err < 0) {
4126                 free_netdev(dev);
4127                 return ERR_PTR(err);
4128         }
4129
4130         err = rtnl_configure_link(dev, NULL);
4131         if (err < 0) {
4132                 LIST_HEAD(list_kill);
4133
4134                 vxlan_dellink(dev, &list_kill);
4135                 unregister_netdevice_many(&list_kill);
4136                 return ERR_PTR(err);
4137         }
4138
4139         return dev;
4140 }
4141 EXPORT_SYMBOL_GPL(vxlan_dev_create);
4142
4143 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
4144                                              struct net_device *dev)
4145 {
4146         struct vxlan_dev *vxlan, *next;
4147         LIST_HEAD(list_kill);
4148
4149         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4150                 struct vxlan_rdst *dst = &vxlan->default_dst;
4151
4152                 /* In case we created vxlan device with carrier
4153                  * and we loose the carrier due to module unload
4154                  * we also need to remove vxlan device. In other
4155                  * cases, it's not necessary and remote_ifindex
4156                  * is 0 here, so no matches.
4157                  */
4158                 if (dst->remote_ifindex == dev->ifindex)
4159                         vxlan_dellink(vxlan->dev, &list_kill);
4160         }
4161
4162         unregister_netdevice_many(&list_kill);
4163 }
4164
4165 static int vxlan_netdevice_event(struct notifier_block *unused,
4166                                  unsigned long event, void *ptr)
4167 {
4168         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4169         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
4170
4171         if (event == NETDEV_UNREGISTER) {
4172                 vxlan_offload_rx_ports(dev, false);
4173                 vxlan_handle_lowerdev_unregister(vn, dev);
4174         } else if (event == NETDEV_REGISTER) {
4175                 vxlan_offload_rx_ports(dev, true);
4176         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
4177                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
4178                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
4179         }
4180
4181         return NOTIFY_DONE;
4182 }
4183
4184 static struct notifier_block vxlan_notifier_block __read_mostly = {
4185         .notifier_call = vxlan_netdevice_event,
4186 };
4187
4188 static void
4189 vxlan_fdb_offloaded_set(struct net_device *dev,
4190                         struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4191 {
4192         struct vxlan_dev *vxlan = netdev_priv(dev);
4193         struct vxlan_rdst *rdst;
4194         struct vxlan_fdb *f;
4195
4196         spin_lock_bh(&vxlan->hash_lock);
4197
4198         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4199         if (!f)
4200                 goto out;
4201
4202         rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip,
4203                                    fdb_info->remote_port,
4204                                    fdb_info->remote_vni,
4205                                    fdb_info->remote_ifindex);
4206         if (!rdst)
4207                 goto out;
4208
4209         rdst->offloaded = fdb_info->offloaded;
4210
4211 out:
4212         spin_unlock_bh(&vxlan->hash_lock);
4213 }
4214
4215 static int
4216 vxlan_fdb_external_learn_add(struct net_device *dev,
4217                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4218 {
4219         struct vxlan_dev *vxlan = netdev_priv(dev);
4220         struct netlink_ext_ack *extack;
4221         int err;
4222
4223         extack = switchdev_notifier_info_to_extack(&fdb_info->info);
4224
4225         spin_lock_bh(&vxlan->hash_lock);
4226         err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip,
4227                                NUD_REACHABLE,
4228                                NLM_F_CREATE | NLM_F_REPLACE,
4229                                fdb_info->remote_port,
4230                                fdb_info->vni,
4231                                fdb_info->remote_vni,
4232                                fdb_info->remote_ifindex,
4233                                NTF_USE | NTF_SELF | NTF_EXT_LEARNED,
4234                                false, extack);
4235         spin_unlock_bh(&vxlan->hash_lock);
4236
4237         return err;
4238 }
4239
4240 static int
4241 vxlan_fdb_external_learn_del(struct net_device *dev,
4242                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4243 {
4244         struct vxlan_dev *vxlan = netdev_priv(dev);
4245         struct vxlan_fdb *f;
4246         int err = 0;
4247
4248         spin_lock_bh(&vxlan->hash_lock);
4249
4250         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4251         if (!f)
4252                 err = -ENOENT;
4253         else if (f->flags & NTF_EXT_LEARNED)
4254                 err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr,
4255                                          fdb_info->remote_ip,
4256                                          fdb_info->remote_port,
4257                                          fdb_info->vni,
4258                                          fdb_info->remote_vni,
4259                                          fdb_info->remote_ifindex,
4260                                          false);
4261
4262         spin_unlock_bh(&vxlan->hash_lock);
4263
4264         return err;
4265 }
4266
4267 static int vxlan_switchdev_event(struct notifier_block *unused,
4268                                  unsigned long event, void *ptr)
4269 {
4270         struct net_device *dev = switchdev_notifier_info_to_dev(ptr);
4271         struct switchdev_notifier_vxlan_fdb_info *fdb_info;
4272         int err = 0;
4273
4274         switch (event) {
4275         case SWITCHDEV_VXLAN_FDB_OFFLOADED:
4276                 vxlan_fdb_offloaded_set(dev, ptr);
4277                 break;
4278         case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE:
4279                 fdb_info = ptr;
4280                 err = vxlan_fdb_external_learn_add(dev, fdb_info);
4281                 if (err) {
4282                         err = notifier_from_errno(err);
4283                         break;
4284                 }
4285                 fdb_info->offloaded = true;
4286                 vxlan_fdb_offloaded_set(dev, fdb_info);
4287                 break;
4288         case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE:
4289                 fdb_info = ptr;
4290                 err = vxlan_fdb_external_learn_del(dev, fdb_info);
4291                 if (err) {
4292                         err = notifier_from_errno(err);
4293                         break;
4294                 }
4295                 fdb_info->offloaded = false;
4296                 vxlan_fdb_offloaded_set(dev, fdb_info);
4297                 break;
4298         }
4299
4300         return err;
4301 }
4302
4303 static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = {
4304         .notifier_call = vxlan_switchdev_event,
4305 };
4306
4307 static __net_init int vxlan_init_net(struct net *net)
4308 {
4309         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4310         unsigned int h;
4311
4312         INIT_LIST_HEAD(&vn->vxlan_list);
4313         spin_lock_init(&vn->sock_lock);
4314
4315         for (h = 0; h < PORT_HASH_SIZE; ++h)
4316                 INIT_HLIST_HEAD(&vn->sock_list[h]);
4317
4318         return 0;
4319 }
4320
4321 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
4322 {
4323         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4324         struct vxlan_dev *vxlan, *next;
4325         struct net_device *dev, *aux;
4326         unsigned int h;
4327
4328         for_each_netdev_safe(net, dev, aux)
4329                 if (dev->rtnl_link_ops == &vxlan_link_ops)
4330                         unregister_netdevice_queue(dev, head);
4331
4332         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4333                 /* If vxlan->dev is in the same netns, it has already been added
4334                  * to the list by the previous loop.
4335                  */
4336                 if (!net_eq(dev_net(vxlan->dev), net))
4337                         unregister_netdevice_queue(vxlan->dev, head);
4338         }
4339
4340         for (h = 0; h < PORT_HASH_SIZE; ++h)
4341                 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
4342 }
4343
4344 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
4345 {
4346         struct net *net;
4347         LIST_HEAD(list);
4348
4349         rtnl_lock();
4350         list_for_each_entry(net, net_list, exit_list)
4351                 vxlan_destroy_tunnels(net, &list);
4352
4353         unregister_netdevice_many(&list);
4354         rtnl_unlock();
4355 }
4356
4357 static struct pernet_operations vxlan_net_ops = {
4358         .init = vxlan_init_net,
4359         .exit_batch = vxlan_exit_batch_net,
4360         .id   = &vxlan_net_id,
4361         .size = sizeof(struct vxlan_net),
4362 };
4363
4364 static int __init vxlan_init_module(void)
4365 {
4366         int rc;
4367
4368         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
4369
4370         rc = register_pernet_subsys(&vxlan_net_ops);
4371         if (rc)
4372                 goto out1;
4373
4374         rc = register_netdevice_notifier(&vxlan_notifier_block);
4375         if (rc)
4376                 goto out2;
4377
4378         rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block);
4379         if (rc)
4380                 goto out3;
4381
4382         rc = rtnl_link_register(&vxlan_link_ops);
4383         if (rc)
4384                 goto out4;
4385
4386         return 0;
4387 out4:
4388         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4389 out3:
4390         unregister_netdevice_notifier(&vxlan_notifier_block);
4391 out2:
4392         unregister_pernet_subsys(&vxlan_net_ops);
4393 out1:
4394         return rc;
4395 }
4396 late_initcall(vxlan_init_module);
4397
4398 static void __exit vxlan_cleanup_module(void)
4399 {
4400         rtnl_link_unregister(&vxlan_link_ops);
4401         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4402         unregister_netdevice_notifier(&vxlan_notifier_block);
4403         unregister_pernet_subsys(&vxlan_net_ops);
4404         /* rcu_barrier() is called by netns */
4405 }
4406 module_exit(vxlan_cleanup_module);
4407
4408 MODULE_LICENSE("GPL");
4409 MODULE_VERSION(VXLAN_VERSION);
4410 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
4411 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
4412 MODULE_ALIAS_RTNL_LINK("vxlan");