drivers/net/vxlan/vxlan_core.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/gro.h>
  21 #include <net/ipv6_stubs.h>
  22 #include <net/ip.h>
  23 #include <net/icmp.h>
  24 #include <net/rtnetlink.h>
  25 #include <net/inet_ecn.h>
  26 #include <net/net_namespace.h>
  27 #include <net/netns/generic.h>
  28 #include <net/tun_proto.h>
  29 #include <net/vxlan.h>
  30 #include <net/nexthop.h>
  31
  32 #if IS_ENABLED(CONFIG_IPV6)
  33 #include <net/ip6_tunnel.h>
  34 #include <net/ip6_checksum.h>
  35 #endif
  36
  37 #include "vxlan_private.h"
  38
  39 #define VXLAN_VERSION   "0.1"
  40
  41 #define FDB_AGE_DEFAULT 300 /* 5 min */
  42 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  43
  44 /* UDP port for VXLAN traffic.
  45  * The IANA assigned port is 4789, but the Linux default is 8472
  46  * for compatibility with early adopters.
  47  */
  48 static unsigned short vxlan_port __read_mostly = 8472;
  49 module_param_named(udp_port, vxlan_port, ushort, 0444);
  50 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  51
  52 static bool log_ecn_error = true;
  53 module_param(log_ecn_error, bool, 0644);
  54 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  55
  56 unsigned int vxlan_net_id;
  57
  58 const u8 all_zeros_mac[ETH_ALEN + 2];
  59 static struct rtnl_link_ops vxlan_link_ops;
  60
  61 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  62
  63 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan);
  64
  65 /* salt for hash table */
  66 static u32 vxlan_salt __read_mostly;
  67
  68 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  69 {
  70         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  71                ip_tunnel_collect_metadata();
  72 }
  73
  74 /* Find VXLAN socket based on network namespace, address family, UDP port,
  75  * enabled unshareable flags and socket device binding (see l3mdev with
  76  * non-default VRF).
  77  */
  78 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
  79                                           __be16 port, u32 flags, int ifindex)
  80 {
  81         struct vxlan_sock *vs;
  82
  83         flags &= VXLAN_F_RCV_FLAGS;
  84
  85         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
  86                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
  87                     vxlan_get_sk_family(vs) == family &&
  88                     vs->flags == flags &&
  89                     vs->sock->sk->sk_bound_dev_if == ifindex)
  90                         return vs;
  91         }
  92         return NULL;
  93 }
  94
  95 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs,
  96                                            int ifindex, __be32 vni,
  97                                            struct vxlan_vni_node **vninode)
  98 {
  99         struct vxlan_vni_node *vnode;
 100         struct vxlan_dev_node *node;
 101
 102         /* For flow based devices, map all packets to VNI 0 */
 103         if (vs->flags & VXLAN_F_COLLECT_METADATA &&
 104             !(vs->flags & VXLAN_F_VNIFILTER))
 105                 vni = 0;
 106
 107         hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) {
 108                 if (!node->vxlan)
 109                         continue;
 110                 vnode = NULL;
 111                 if (node->vxlan->cfg.flags & VXLAN_F_VNIFILTER) {
 112                         vnode = vxlan_vnifilter_lookup(node->vxlan, vni);
 113                         if (!vnode)
 114                                 continue;
 115                 } else if (node->vxlan->default_dst.remote_vni != vni) {
 116                         continue;
 117                 }
 118
 119                 if (IS_ENABLED(CONFIG_IPV6)) {
 120                         const struct vxlan_config *cfg = &node->vxlan->cfg;
 121
 122                         if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) &&
 123                             cfg->remote_ifindex != ifindex)
 124                                 continue;
 125                 }
 126
 127                 if (vninode)
 128                         *vninode = vnode;
 129                 return node->vxlan;
 130         }
 131
 132         return NULL;
 133 }
 134
 135 /* Look up VNI in a per net namespace table */
 136 static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex,
 137                                         __be32 vni, sa_family_t family,
 138                                         __be16 port, u32 flags)
 139 {
 140         struct vxlan_sock *vs;
 141
 142         vs = vxlan_find_sock(net, family, port, flags, ifindex);
 143         if (!vs)
 144                 return NULL;
 145
 146         return vxlan_vs_find_vni(vs, ifindex, vni, NULL);
 147 }
 148
 149 /* Fill in neighbour message in skbuff. */
 150 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 151                           const struct vxlan_fdb *fdb,
 152                           u32 portid, u32 seq, int type, unsigned int flags,
 153                           const struct vxlan_rdst *rdst)
 154 {
 155         unsigned long now = jiffies;
 156         struct nda_cacheinfo ci;
 157         bool send_ip, send_eth;
 158         struct nlmsghdr *nlh;
 159         struct nexthop *nh;
 160         struct ndmsg *ndm;
 161         int nh_family;
 162         u32 nh_id;
 163
 164         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 165         if (nlh == NULL)
 166                 return -EMSGSIZE;
 167
 168         ndm = nlmsg_data(nlh);
 169         memset(ndm, 0, sizeof(*ndm));
 170
 171         send_eth = send_ip = true;
 172
 173         rcu_read_lock();
 174         nh = rcu_dereference(fdb->nh);
 175         if (nh) {
 176                 nh_family = nexthop_get_family(nh);
 177                 nh_id = nh->id;
 178         }
 179         rcu_read_unlock();
 180
 181         if (type == RTM_GETNEIGH) {
 182                 if (rdst) {
 183                         send_ip = !vxlan_addr_any(&rdst->remote_ip);
 184                         ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
 185                 } else if (nh) {
 186                         ndm->ndm_family = nh_family;
 187                 }
 188                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 189         } else
 190                 ndm->ndm_family = AF_BRIDGE;
 191         ndm->ndm_state = fdb->state;
 192         ndm->ndm_ifindex = vxlan->dev->ifindex;
 193         ndm->ndm_flags = fdb->flags;
 194         if (rdst && rdst->offloaded)
 195                 ndm->ndm_flags |= NTF_OFFLOADED;
 196         ndm->ndm_type = RTN_UNICAST;
 197
 198         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 199             nla_put_s32(skb, NDA_LINK_NETNSID,
 200                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 201                 goto nla_put_failure;
 202
 203         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 204                 goto nla_put_failure;
 205         if (nh) {
 206                 if (nla_put_u32(skb, NDA_NH_ID, nh_id))
 207                         goto nla_put_failure;
 208         } else if (rdst) {
 209                 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST,
 210                                                   &rdst->remote_ip))
 211                         goto nla_put_failure;
 212
 213                 if (rdst->remote_port &&
 214                     rdst->remote_port != vxlan->cfg.dst_port &&
 215                     nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 216                         goto nla_put_failure;
 217                 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 218                     nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 219                         goto nla_put_failure;
 220                 if (rdst->remote_ifindex &&
 221                     nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 222                         goto nla_put_failure;
 223         }
 224
 225         if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
 226             nla_put_u32(skb, NDA_SRC_VNI,
 227                         be32_to_cpu(fdb->vni)))
 228                 goto nla_put_failure;
 229
 230         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 231         ci.ndm_confirmed = 0;
 232         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 233         ci.ndm_refcnt    = 0;
 234
 235         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 236                 goto nla_put_failure;
 237
 238         nlmsg_end(skb, nlh);
 239         return 0;
 240
 241 nla_put_failure:
 242         nlmsg_cancel(skb, nlh);
 243         return -EMSGSIZE;
 244 }
 245
 246 static inline size_t vxlan_nlmsg_size(void)
 247 {
 248         return NLMSG_ALIGN(sizeof(struct ndmsg))
 249                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 250                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 251                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 252                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 253                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 254                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 255                 + nla_total_size(sizeof(struct nda_cacheinfo));
 256 }
 257
 258 static void __vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 259                                struct vxlan_rdst *rd, int type)
 260 {
 261         struct net *net = dev_net(vxlan->dev);
 262         struct sk_buff *skb;
 263         int err = -ENOBUFS;
 264
 265         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 266         if (skb == NULL)
 267                 goto errout;
 268
 269         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 270         if (err < 0) {
 271                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 272                 WARN_ON(err == -EMSGSIZE);
 273                 kfree_skb(skb);
 274                 goto errout;
 275         }
 276
 277         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 278         return;
 279 errout:
 280         if (err < 0)
 281                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 282 }
 283
 284 static void vxlan_fdb_switchdev_notifier_info(const struct vxlan_dev *vxlan,
 285                             const struct vxlan_fdb *fdb,
 286                             const struct vxlan_rdst *rd,
 287                             struct netlink_ext_ack *extack,
 288                             struct switchdev_notifier_vxlan_fdb_info *fdb_info)
 289 {
 290         fdb_info->info.dev = vxlan->dev;
 291         fdb_info->info.extack = extack;
 292         fdb_info->remote_ip = rd->remote_ip;
 293         fdb_info->remote_port = rd->remote_port;
 294         fdb_info->remote_vni = rd->remote_vni;
 295         fdb_info->remote_ifindex = rd->remote_ifindex;
 296         memcpy(fdb_info->eth_addr, fdb->eth_addr, ETH_ALEN);
 297         fdb_info->vni = fdb->vni;
 298         fdb_info->offloaded = rd->offloaded;
 299         fdb_info->added_by_user = fdb->flags & NTF_VXLAN_ADDED_BY_USER;
 300 }
 301
 302 static int vxlan_fdb_switchdev_call_notifiers(struct vxlan_dev *vxlan,
 303                                               struct vxlan_fdb *fdb,
 304                                               struct vxlan_rdst *rd,
 305                                               bool adding,
 306                                               struct netlink_ext_ack *extack)
 307 {
 308         struct switchdev_notifier_vxlan_fdb_info info;
 309         enum switchdev_notifier_type notifier_type;
 310         int ret;
 311
 312         if (WARN_ON(!rd))
 313                 return 0;
 314
 315         notifier_type = adding ? SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE
 316                                : SWITCHDEV_VXLAN_FDB_DEL_TO_DEVICE;
 317         vxlan_fdb_switchdev_notifier_info(vxlan, fdb, rd, NULL, &info);
 318         ret = call_switchdev_notifiers(notifier_type, vxlan->dev,
 319                                        &info.info, extack);
 320         return notifier_to_errno(ret);
 321 }
 322
 323 static int vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 324                             struct vxlan_rdst *rd, int type, bool swdev_notify,
 325                             struct netlink_ext_ack *extack)
 326 {
 327         int err;
 328
 329         if (swdev_notify && rd) {
 330                 switch (type) {
 331                 case RTM_NEWNEIGH:
 332                         err = vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
 333                                                                  true, extack);
 334                         if (err)
 335                                 return err;
 336                         break;
 337                 case RTM_DELNEIGH:
 338                         vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
 339                                                            false, extack);
 340                         break;
 341                 }
 342         }
 343
 344         __vxlan_fdb_notify(vxlan, fdb, rd, type);
 345         return 0;
 346 }
 347
 348 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 349 {
 350         struct vxlan_dev *vxlan = netdev_priv(dev);
 351         struct vxlan_fdb f = {
 352                 .state = NUD_STALE,
 353         };
 354         struct vxlan_rdst remote = {
 355                 .remote_ip = *ipa, /* goes to NDA_DST */
 356                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 357         };
 358
 359         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
 360 }
 361
 362 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 363 {
 364         struct vxlan_fdb f = {
 365                 .state = NUD_STALE,
 366         };
 367         struct vxlan_rdst remote = { };
 368
 369         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 370
 371         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
 372 }
 373
 374 /* Hash Ethernet address */
 375 static u32 eth_hash(const unsigned char *addr)
 376 {
 377         u64 value = get_unaligned((u64 *)addr);
 378
 379         /* only want 6 bytes */
 380 #ifdef __BIG_ENDIAN
 381         value >>= 16;
 382 #else
 383         value <<= 16;
 384 #endif
 385         return hash_64(value, FDB_HASH_BITS);
 386 }
 387
 388 u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
 389 {
 390         /* use 1 byte of OUI and 3 bytes of NIC */
 391         u32 key = get_unaligned((u32 *)(addr + 2));
 392
 393         return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
 394 }
 395
 396 u32 fdb_head_index(struct vxlan_dev *vxlan, const u8 *mac, __be32 vni)
 397 {
 398         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)
 399                 return eth_vni_hash(mac, vni);
 400         else
 401                 return eth_hash(mac);
 402 }
 403
 404 /* Hash chain to use given mac address */
 405 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 406                                                 const u8 *mac, __be32 vni)
 407 {
 408         return &vxlan->fdb_head[fdb_head_index(vxlan, mac, vni)];
 409 }
 410
 411 /* Look up Ethernet address in forwarding table */
 412 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 413                                           const u8 *mac, __be32 vni)
 414 {
 415         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
 416         struct vxlan_fdb *f;
 417
 418         hlist_for_each_entry_rcu(f, head, hlist) {
 419                 if (ether_addr_equal(mac, f->eth_addr)) {
 420                         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
 421                                 if (vni == f->vni)
 422                                         return f;
 423                         } else {
 424                                 return f;
 425                         }
 426                 }
 427         }
 428
 429         return NULL;
 430 }
 431
 432 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 433                                         const u8 *mac, __be32 vni)
 434 {
 435         struct vxlan_fdb *f;
 436
 437         f = __vxlan_find_mac(vxlan, mac, vni);
 438         if (f && f->used != jiffies)
 439                 f->used = jiffies;
 440
 441         return f;
 442 }
 443
 444 /* caller should hold vxlan->hash_lock */
 445 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 446                                               union vxlan_addr *ip, __be16 port,
 447                                               __be32 vni, __u32 ifindex)
 448 {
 449         struct vxlan_rdst *rd;
 450
 451         list_for_each_entry(rd, &f->remotes, list) {
 452                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 453                     rd->remote_port == port &&
 454                     rd->remote_vni == vni &&
 455                     rd->remote_ifindex == ifindex)
 456                         return rd;
 457         }
 458
 459         return NULL;
 460 }
 461
 462 int vxlan_fdb_find_uc(struct net_device *dev, const u8 *mac, __be32 vni,
 463                       struct switchdev_notifier_vxlan_fdb_info *fdb_info)
 464 {
 465         struct vxlan_dev *vxlan = netdev_priv(dev);
 466         u8 eth_addr[ETH_ALEN + 2] = { 0 };
 467         struct vxlan_rdst *rdst;
 468         struct vxlan_fdb *f;
 469         int rc = 0;
 470
 471         if (is_multicast_ether_addr(mac) ||
 472             is_zero_ether_addr(mac))
 473                 return -EINVAL;
 474
 475         ether_addr_copy(eth_addr, mac);
 476
 477         rcu_read_lock();
 478
 479         f = __vxlan_find_mac(vxlan, eth_addr, vni);
 480         if (!f) {
 481                 rc = -ENOENT;
 482                 goto out;
 483         }
 484
 485         rdst = first_remote_rcu(f);
 486         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, NULL, fdb_info);
 487
 488 out:
 489         rcu_read_unlock();
 490         return rc;
 491 }
 492 EXPORT_SYMBOL_GPL(vxlan_fdb_find_uc);
 493
 494 static int vxlan_fdb_notify_one(struct notifier_block *nb,
 495                                 const struct vxlan_dev *vxlan,
 496                                 const struct vxlan_fdb *f,
 497                                 const struct vxlan_rdst *rdst,
 498                                 struct netlink_ext_ack *extack)
 499 {
 500         struct switchdev_notifier_vxlan_fdb_info fdb_info;
 501         int rc;
 502
 503         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, extack, &fdb_info);
 504         rc = nb->notifier_call(nb, SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE,
 505                                &fdb_info);
 506         return notifier_to_errno(rc);
 507 }
 508
 509 int vxlan_fdb_replay(const struct net_device *dev, __be32 vni,
 510                      struct notifier_block *nb,
 511                      struct netlink_ext_ack *extack)
 512 {
 513         struct vxlan_dev *vxlan;
 514         struct vxlan_rdst *rdst;
 515         struct vxlan_fdb *f;
 516         unsigned int h;
 517         int rc = 0;
 518
 519         if (!netif_is_vxlan(dev))
 520                 return -EINVAL;
 521         vxlan = netdev_priv(dev);
 522
 523         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 524                 spin_lock_bh(&vxlan->hash_lock[h]);
 525                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) {
 526                         if (f->vni == vni) {
 527                                 list_for_each_entry(rdst, &f->remotes, list) {
 528                                         rc = vxlan_fdb_notify_one(nb, vxlan,
 529                                                                   f, rdst,
 530                                                                   extack);
 531                                         if (rc)
 532                                                 goto unlock;
 533                                 }
 534                         }
 535                 }
 536                 spin_unlock_bh(&vxlan->hash_lock[h]);
 537         }
 538         return 0;
 539
 540 unlock:
 541         spin_unlock_bh(&vxlan->hash_lock[h]);
 542         return rc;
 543 }
 544 EXPORT_SYMBOL_GPL(vxlan_fdb_replay);
 545
 546 void vxlan_fdb_clear_offload(const struct net_device *dev, __be32 vni)
 547 {
 548         struct vxlan_dev *vxlan;
 549         struct vxlan_rdst *rdst;
 550         struct vxlan_fdb *f;
 551         unsigned int h;
 552
 553         if (!netif_is_vxlan(dev))
 554                 return;
 555         vxlan = netdev_priv(dev);
 556
 557         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 558                 spin_lock_bh(&vxlan->hash_lock[h]);
 559                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist)
 560                         if (f->vni == vni)
 561                                 list_for_each_entry(rdst, &f->remotes, list)
 562                                         rdst->offloaded = false;
 563                 spin_unlock_bh(&vxlan->hash_lock[h]);
 564         }
 565
 566 }
 567 EXPORT_SYMBOL_GPL(vxlan_fdb_clear_offload);
 568
 569 /* Replace destination of unicast mac */
 570 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 571                              union vxlan_addr *ip, __be16 port, __be32 vni,
 572                              __u32 ifindex, struct vxlan_rdst *oldrd)
 573 {
 574         struct vxlan_rdst *rd;
 575
 576         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 577         if (rd)
 578                 return 0;
 579
 580         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 581         if (!rd)
 582                 return 0;
 583
 584         *oldrd = *rd;
 585         dst_cache_reset(&rd->dst_cache);
 586         rd->remote_ip = *ip;
 587         rd->remote_port = port;
 588         rd->remote_vni = vni;
 589         rd->remote_ifindex = ifindex;
 590         rd->offloaded = false;
 591         return 1;
 592 }
 593
 594 /* Add/update destinations for multicast */
 595 static int vxlan_fdb_append(struct vxlan_fdb *f,
 596                             union vxlan_addr *ip, __be16 port, __be32 vni,
 597                             __u32 ifindex, struct vxlan_rdst **rdp)
 598 {
 599         struct vxlan_rdst *rd;
 600
 601         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 602         if (rd)
 603                 return 0;
 604
 605         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 606         if (rd == NULL)
 607                 return -ENOMEM;
 608
 609         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 610                 kfree(rd);
 611                 return -ENOMEM;
 612         }
 613
 614         rd->remote_ip = *ip;
 615         rd->remote_port = port;
 616         rd->offloaded = false;
 617         rd->remote_vni = vni;
 618         rd->remote_ifindex = ifindex;
 619
 620         list_add_tail_rcu(&rd->list, &f->remotes);
 621
 622         *rdp = rd;
 623         return 1;
 624 }
 625
 626 static bool vxlan_parse_gpe_proto(struct vxlanhdr *hdr, __be16 *protocol)
 627 {
 628         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)hdr;
 629
 630         /* Need to have Next Protocol set for interfaces in GPE mode. */
 631         if (!gpe->np_applied)
 632                 return false;
 633         /* "The initial version is 0. If a receiver does not support the
 634          * version indicated it MUST drop the packet.
 635          */
 636         if (gpe->version != 0)
 637                 return false;
 638         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
 639          * processing MUST occur." However, we don't implement OAM
 640          * processing, thus drop the packet.
 641          */
 642         if (gpe->oam_flag)
 643                 return false;
 644
 645         *protocol = tun_p_to_eth_p(gpe->next_protocol);
 646         if (!*protocol)
 647                 return false;
 648
 649         return true;
 650 }
 651
 652 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 653                                           unsigned int off,
 654                                           struct vxlanhdr *vh, size_t hdrlen,
 655                                           __be32 vni_field,
 656                                           struct gro_remcsum *grc,
 657                                           bool nopartial)
 658 {
 659         size_t start, offset;
 660
 661         if (skb->remcsum_offload)
 662                 return vh;
 663
 664         if (!NAPI_GRO_CB(skb)->csum_valid)
 665                 return NULL;
 666
 667         start = vxlan_rco_start(vni_field);
 668         offset = start + vxlan_rco_offset(vni_field);
 669
 670         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 671                                      start, offset, grc, nopartial);
 672
 673         skb->remcsum_offload = 1;
 674
 675         return vh;
 676 }
 677
 678 static struct vxlanhdr *vxlan_gro_prepare_receive(struct sock *sk,
 679                                                   struct list_head *head,
 680                                                   struct sk_buff *skb,
 681                                                   struct gro_remcsum *grc)
 682 {
 683         struct sk_buff *p;
 684         struct vxlanhdr *vh, *vh2;
 685         unsigned int hlen, off_vx;
 686         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 687         __be32 flags;
 688
 689         skb_gro_remcsum_init(grc);
 690
 691         off_vx = skb_gro_offset(skb);
 692         hlen = off_vx + sizeof(*vh);
 693         vh = skb_gro_header(skb, hlen, off_vx);
 694         if (unlikely(!vh))
 695                 return NULL;
 696
 697         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 698
 699         flags = vh->vx_flags;
 700
 701         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 702                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 703                                        vh->vx_vni, grc,
 704                                        !!(vs->flags &
 705                                           VXLAN_F_REMCSUM_NOPARTIAL));
 706
 707                 if (!vh)
 708                         return NULL;
 709         }
 710
 711         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 712
 713         list_for_each_entry(p, head, list) {
 714                 if (!NAPI_GRO_CB(p)->same_flow)
 715                         continue;
 716
 717                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 718                 if (vh->vx_flags != vh2->vx_flags ||
 719                     vh->vx_vni != vh2->vx_vni) {
 720                         NAPI_GRO_CB(p)->same_flow = 0;
 721                         continue;
 722                 }
 723         }
 724
 725         return vh;
 726 }
 727
 728 static struct sk_buff *vxlan_gro_receive(struct sock *sk,
 729                                          struct list_head *head,
 730                                          struct sk_buff *skb)
 731 {
 732         struct sk_buff *pp = NULL;
 733         struct gro_remcsum grc;
 734         int flush = 1;
 735
 736         if (vxlan_gro_prepare_receive(sk, head, skb, &grc)) {
 737                 pp = call_gro_receive(eth_gro_receive, head, skb);
 738                 flush = 0;
 739         }
 740         skb_gro_flush_final_remcsum(skb, pp, flush, &grc);
 741         return pp;
 742 }
 743
 744 static struct sk_buff *vxlan_gpe_gro_receive(struct sock *sk,
 745                                              struct list_head *head,
 746                                              struct sk_buff *skb)
 747 {
 748         const struct packet_offload *ptype;
 749         struct sk_buff *pp = NULL;
 750         struct gro_remcsum grc;
 751         struct vxlanhdr *vh;
 752         __be16 protocol;
 753         int flush = 1;
 754
 755         vh = vxlan_gro_prepare_receive(sk, head, skb, &grc);
 756         if (vh) {
 757                 if (!vxlan_parse_gpe_proto(vh, &protocol))
 758                         goto out;
 759                 ptype = gro_find_receive_by_type(protocol);
 760                 if (!ptype)
 761                         goto out;
 762                 pp = call_gro_receive(ptype->callbacks.gro_receive, head, skb);
 763                 flush = 0;
 764         }
 765 out:
 766         skb_gro_flush_final_remcsum(skb, pp, flush, &grc);
 767         return pp;
 768 }
 769
 770 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 771 {
 772         /* Sets 'skb->inner_mac_header' since we are always called with
 773          * 'skb->encapsulation' set.
 774          */
 775         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 776 }
 777
 778 static int vxlan_gpe_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 779 {
 780         struct vxlanhdr *vh = (struct vxlanhdr *)(skb->data + nhoff);
 781         const struct packet_offload *ptype;
 782         int err = -ENOSYS;
 783         __be16 protocol;
 784
 785         if (!vxlan_parse_gpe_proto(vh, &protocol))
 786                 return err;
 787         ptype = gro_find_complete_by_type(protocol);
 788         if (ptype)
 789                 err = ptype->callbacks.gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 790         return err;
 791 }
 792
 793 static struct vxlan_fdb *vxlan_fdb_alloc(struct vxlan_dev *vxlan, const u8 *mac,
 794                                          __u16 state, __be32 src_vni,
 795                                          __u16 ndm_flags)
 796 {
 797         struct vxlan_fdb *f;
 798
 799         f = kmalloc(sizeof(*f), GFP_ATOMIC);
 800         if (!f)
 801                 return NULL;
 802         f->state = state;
 803         f->flags = ndm_flags;
 804         f->updated = f->used = jiffies;
 805         f->vni = src_vni;
 806         f->nh = NULL;
 807         RCU_INIT_POINTER(f->vdev, vxlan);
 808         INIT_LIST_HEAD(&f->nh_list);
 809         INIT_LIST_HEAD(&f->remotes);
 810         memcpy(f->eth_addr, mac, ETH_ALEN);
 811
 812         return f;
 813 }
 814
 815 static void vxlan_fdb_insert(struct vxlan_dev *vxlan, const u8 *mac,
 816                              __be32 src_vni, struct vxlan_fdb *f)
 817 {
 818         ++vxlan->addrcnt;
 819         hlist_add_head_rcu(&f->hlist,
 820                            vxlan_fdb_head(vxlan, mac, src_vni));
 821 }
 822
 823 static int vxlan_fdb_nh_update(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 824                                u32 nhid, struct netlink_ext_ack *extack)
 825 {
 826         struct nexthop *old_nh = rtnl_dereference(fdb->nh);
 827         struct nexthop *nh;
 828         int err = -EINVAL;
 829
 830         if (old_nh && old_nh->id == nhid)
 831                 return 0;
 832
 833         nh = nexthop_find_by_id(vxlan->net, nhid);
 834         if (!nh) {
 835                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
 836                 goto err_inval;
 837         }
 838
 839         if (!nexthop_get(nh)) {
 840                 NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
 841                 nh = NULL;
 842                 goto err_inval;
 843         }
 844         if (!nexthop_is_fdb(nh)) {
 845                 NL_SET_ERR_MSG(extack, "Nexthop is not a fdb nexthop");
 846                 goto err_inval;
 847         }
 848
 849         if (!nexthop_is_multipath(nh)) {
 850                 NL_SET_ERR_MSG(extack, "Nexthop is not a multipath group");
 851                 goto err_inval;
 852         }
 853
 854         /* check nexthop group family */
 855         switch (vxlan->default_dst.remote_ip.sa.sa_family) {
 856         case AF_INET:
 857                 if (!nexthop_has_v4(nh)) {
 858                         err = -EAFNOSUPPORT;
 859                         NL_SET_ERR_MSG(extack, "Nexthop group family not supported");
 860                         goto err_inval;
 861                 }
 862                 break;
 863         case AF_INET6:
 864                 if (nexthop_has_v4(nh)) {
 865                         err = -EAFNOSUPPORT;
 866                         NL_SET_ERR_MSG(extack, "Nexthop group family not supported");
 867                         goto err_inval;
 868                 }
 869         }
 870
 871         if (old_nh) {
 872                 list_del_rcu(&fdb->nh_list);
 873                 nexthop_put(old_nh);
 874         }
 875         rcu_assign_pointer(fdb->nh, nh);
 876         list_add_tail_rcu(&fdb->nh_list, &nh->fdb_list);
 877         return 1;
 878
 879 err_inval:
 880         if (nh)
 881                 nexthop_put(nh);
 882         return err;
 883 }
 884
 885 int vxlan_fdb_create(struct vxlan_dev *vxlan,
 886                      const u8 *mac, union vxlan_addr *ip,
 887                      __u16 state, __be16 port, __be32 src_vni,
 888                      __be32 vni, __u32 ifindex, __u16 ndm_flags,
 889                      u32 nhid, struct vxlan_fdb **fdb,
 890                      struct netlink_ext_ack *extack)
 891 {
 892         struct vxlan_rdst *rd = NULL;
 893         struct vxlan_fdb *f;
 894         int rc;
 895
 896         if (vxlan->cfg.addrmax &&
 897             vxlan->addrcnt >= vxlan->cfg.addrmax)
 898                 return -ENOSPC;
 899
 900         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 901         f = vxlan_fdb_alloc(vxlan, mac, state, src_vni, ndm_flags);
 902         if (!f)
 903                 return -ENOMEM;
 904
 905         if (nhid)
 906                 rc = vxlan_fdb_nh_update(vxlan, f, nhid, extack);
 907         else
 908                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 909         if (rc < 0)
 910                 goto errout;
 911
 912         *fdb = f;
 913
 914         return 0;
 915
 916 errout:
 917         kfree(f);
 918         return rc;
 919 }
 920
 921 static void __vxlan_fdb_free(struct vxlan_fdb *f)
 922 {
 923         struct vxlan_rdst *rd, *nd;
 924         struct nexthop *nh;
 925
 926         nh = rcu_dereference_raw(f->nh);
 927         if (nh) {
 928                 rcu_assign_pointer(f->nh, NULL);
 929                 rcu_assign_pointer(f->vdev, NULL);
 930                 nexthop_put(nh);
 931         }
 932
 933         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 934                 dst_cache_destroy(&rd->dst_cache);
 935                 kfree(rd);
 936         }
 937         kfree(f);
 938 }
 939
 940 static void vxlan_fdb_free(struct rcu_head *head)
 941 {
 942         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 943
 944         __vxlan_fdb_free(f);
 945 }
 946
 947 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 948                               bool do_notify, bool swdev_notify)
 949 {
 950         struct vxlan_rdst *rd;
 951
 952         netdev_dbg(vxlan->dev, "delete %pM\n", f->eth_addr);
 953
 954         --vxlan->addrcnt;
 955         if (do_notify) {
 956                 if (rcu_access_pointer(f->nh))
 957                         vxlan_fdb_notify(vxlan, f, NULL, RTM_DELNEIGH,
 958                                          swdev_notify, NULL);
 959                 else
 960                         list_for_each_entry(rd, &f->remotes, list)
 961                                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH,
 962                                                  swdev_notify, NULL);
 963         }
 964
 965         hlist_del_rcu(&f->hlist);
 966         list_del_rcu(&f->nh_list);
 967         call_rcu(&f->rcu, vxlan_fdb_free);
 968 }
 969
 970 static void vxlan_dst_free(struct rcu_head *head)
 971 {
 972         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
 973
 974         dst_cache_destroy(&rd->dst_cache);
 975         kfree(rd);
 976 }
 977
 978 static int vxlan_fdb_update_existing(struct vxlan_dev *vxlan,
 979                                      union vxlan_addr *ip,
 980                                      __u16 state, __u16 flags,
 981                                      __be16 port, __be32 vni,
 982                                      __u32 ifindex, __u16 ndm_flags,
 983                                      struct vxlan_fdb *f, u32 nhid,
 984                                      bool swdev_notify,
 985                                      struct netlink_ext_ack *extack)
 986 {
 987         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
 988         struct vxlan_rdst *rd = NULL;
 989         struct vxlan_rdst oldrd;
 990         int notify = 0;
 991         int rc = 0;
 992         int err;
 993
 994         if (nhid && !rcu_access_pointer(f->nh)) {
 995                 NL_SET_ERR_MSG(extack,
 996                                "Cannot replace an existing non nexthop fdb with a nexthop");
 997                 return -EOPNOTSUPP;
 998         }
 999
1000         if (nhid && (flags & NLM_F_APPEND)) {
1001                 NL_SET_ERR_MSG(extack,
1002                                "Cannot append to a nexthop fdb");
1003                 return -EOPNOTSUPP;
1004         }
1005
1006         /* Do not allow an externally learned entry to take over an entry added
1007          * by the user.
1008          */
1009         if (!(fdb_flags & NTF_EXT_LEARNED) ||
1010             !(f->flags & NTF_VXLAN_ADDED_BY_USER)) {
1011                 if (f->state != state) {
1012                         f->state = state;
1013                         f->updated = jiffies;
1014                         notify = 1;
1015                 }
1016                 if (f->flags != fdb_flags) {
1017                         f->flags = fdb_flags;
1018                         f->updated = jiffies;
1019                         notify = 1;
1020                 }
1021         }
1022
1023         if ((flags & NLM_F_REPLACE)) {
1024                 /* Only change unicasts */
1025                 if (!(is_multicast_ether_addr(f->eth_addr) ||
1026                       is_zero_ether_addr(f->eth_addr))) {
1027                         if (nhid) {
1028                                 rc = vxlan_fdb_nh_update(vxlan, f, nhid, extack);
1029                                 if (rc < 0)
1030                                         return rc;
1031                         } else {
1032                                 rc = vxlan_fdb_replace(f, ip, port, vni,
1033                                                        ifindex, &oldrd);
1034                         }
1035                         notify |= rc;
1036                 } else {
1037                         NL_SET_ERR_MSG(extack, "Cannot replace non-unicast fdb entries");
1038                         return -EOPNOTSUPP;
1039                 }
1040         }
1041         if ((flags & NLM_F_APPEND) &&
1042             (is_multicast_ether_addr(f->eth_addr) ||
1043              is_zero_ether_addr(f->eth_addr))) {
1044                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
1045
1046                 if (rc < 0)
1047                         return rc;
1048                 notify |= rc;
1049         }
1050
1051         if (ndm_flags & NTF_USE)
1052                 f->used = jiffies;
1053
1054         if (notify) {
1055                 if (rd == NULL)
1056                         rd = first_remote_rtnl(f);
1057
1058                 err = vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH,
1059                                        swdev_notify, extack);
1060                 if (err)
1061                         goto err_notify;
1062         }
1063
1064         return 0;
1065
1066 err_notify:
1067         if (nhid)
1068                 return err;
1069         if ((flags & NLM_F_REPLACE) && rc)
1070                 *rd = oldrd;
1071         else if ((flags & NLM_F_APPEND) && rc) {
1072                 list_del_rcu(&rd->list);
1073                 call_rcu(&rd->rcu, vxlan_dst_free);
1074         }
1075         return err;
1076 }
1077
1078 static int vxlan_fdb_update_create(struct vxlan_dev *vxlan,
1079                                    const u8 *mac, union vxlan_addr *ip,
1080                                    __u16 state, __u16 flags,
1081                                    __be16 port, __be32 src_vni, __be32 vni,
1082                                    __u32 ifindex, __u16 ndm_flags, u32 nhid,
1083                                    bool swdev_notify,
1084                                    struct netlink_ext_ack *extack)
1085 {
1086         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
1087         struct vxlan_fdb *f;
1088         int rc;
1089
1090         /* Disallow replace to add a multicast entry */
1091         if ((flags & NLM_F_REPLACE) &&
1092             (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
1093                 return -EOPNOTSUPP;
1094
1095         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
1096         rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni,
1097                               vni, ifindex, fdb_flags, nhid, &f, extack);
1098         if (rc < 0)
1099                 return rc;
1100
1101         vxlan_fdb_insert(vxlan, mac, src_vni, f);
1102         rc = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH,
1103                               swdev_notify, extack);
1104         if (rc)
1105                 goto err_notify;
1106
1107         return 0;
1108
1109 err_notify:
1110         vxlan_fdb_destroy(vxlan, f, false, false);
1111         return rc;
1112 }
1113
1114 /* Add new entry to forwarding table -- assumes lock held */
1115 int vxlan_fdb_update(struct vxlan_dev *vxlan,
1116                      const u8 *mac, union vxlan_addr *ip,
1117                      __u16 state, __u16 flags,
1118                      __be16 port, __be32 src_vni, __be32 vni,
1119                      __u32 ifindex, __u16 ndm_flags, u32 nhid,
1120                      bool swdev_notify,
1121                      struct netlink_ext_ack *extack)
1122 {
1123         struct vxlan_fdb *f;
1124
1125         f = __vxlan_find_mac(vxlan, mac, src_vni);
1126         if (f) {
1127                 if (flags & NLM_F_EXCL) {
1128                         netdev_dbg(vxlan->dev,
1129                                    "lost race to create %pM\n", mac);
1130                         return -EEXIST;
1131                 }
1132
1133                 return vxlan_fdb_update_existing(vxlan, ip, state, flags, port,
1134                                                  vni, ifindex, ndm_flags, f,
1135                                                  nhid, swdev_notify, extack);
1136         } else {
1137                 if (!(flags & NLM_F_CREATE))
1138                         return -ENOENT;
1139
1140                 return vxlan_fdb_update_create(vxlan, mac, ip, state, flags,
1141                                                port, src_vni, vni, ifindex,
1142                                                ndm_flags, nhid, swdev_notify,
1143                                                extack);
1144         }
1145 }
1146
1147 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
1148                                   struct vxlan_rdst *rd, bool swdev_notify)
1149 {
1150         list_del_rcu(&rd->list);
1151         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, swdev_notify, NULL);
1152         call_rcu(&rd->rcu, vxlan_dst_free);
1153 }
1154
1155 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
1156                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
1157                            __be32 *vni, u32 *ifindex, u32 *nhid,
1158                            struct netlink_ext_ack *extack)
1159 {
1160         struct net *net = dev_net(vxlan->dev);
1161         int err;
1162
1163         if (tb[NDA_NH_ID] &&
1164             (tb[NDA_DST] || tb[NDA_VNI] || tb[NDA_IFINDEX] || tb[NDA_PORT])) {
1165                 NL_SET_ERR_MSG(extack, "DST, VNI, ifindex and port are mutually exclusive with NH_ID");
1166                 return -EINVAL;
1167         }
1168
1169         if (tb[NDA_DST]) {
1170                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
1171                 if (err) {
1172                         NL_SET_ERR_MSG(extack, "Unsupported address family");
1173                         return err;
1174                 }
1175         } else {
1176                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
1177
1178                 if (remote->sa.sa_family == AF_INET) {
1179                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
1180                         ip->sa.sa_family = AF_INET;
1181 #if IS_ENABLED(CONFIG_IPV6)
1182                 } else {
1183                         ip->sin6.sin6_addr = in6addr_any;
1184                         ip->sa.sa_family = AF_INET6;
1185 #endif
1186                 }
1187         }
1188
1189         if (tb[NDA_PORT]) {
1190                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16)) {
1191                         NL_SET_ERR_MSG(extack, "Invalid vxlan port");
1192                         return -EINVAL;
1193                 }
1194                 *port = nla_get_be16(tb[NDA_PORT]);
1195         } else {
1196                 *port = vxlan->cfg.dst_port;
1197         }
1198
1199         if (tb[NDA_VNI]) {
1200                 if (nla_len(tb[NDA_VNI]) != sizeof(u32)) {
1201                         NL_SET_ERR_MSG(extack, "Invalid vni");
1202                         return -EINVAL;
1203                 }
1204                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1205         } else {
1206                 *vni = vxlan->default_dst.remote_vni;
1207         }
1208
1209         if (tb[NDA_SRC_VNI]) {
1210                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32)) {
1211                         NL_SET_ERR_MSG(extack, "Invalid src vni");
1212                         return -EINVAL;
1213                 }
1214                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
1215         } else {
1216                 *src_vni = vxlan->default_dst.remote_vni;
1217         }
1218
1219         if (tb[NDA_IFINDEX]) {
1220                 struct net_device *tdev;
1221
1222                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32)) {
1223                         NL_SET_ERR_MSG(extack, "Invalid ifindex");
1224                         return -EINVAL;
1225                 }
1226                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
1227                 tdev = __dev_get_by_index(net, *ifindex);
1228                 if (!tdev) {
1229                         NL_SET_ERR_MSG(extack, "Device not found");
1230                         return -EADDRNOTAVAIL;
1231                 }
1232         } else {
1233                 *ifindex = 0;
1234         }
1235
1236         if (tb[NDA_NH_ID])
1237                 *nhid = nla_get_u32(tb[NDA_NH_ID]);
1238         else
1239                 *nhid = 0;
1240
1241         return 0;
1242 }
1243
1244 /* Add static entry (via netlink) */
1245 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
1246                          struct net_device *dev,
1247                          const unsigned char *addr, u16 vid, u16 flags,
1248                          struct netlink_ext_ack *extack)
1249 {
1250         struct vxlan_dev *vxlan = netdev_priv(dev);
1251         /* struct net *net = dev_net(vxlan->dev); */
1252         union vxlan_addr ip;
1253         __be16 port;
1254         __be32 src_vni, vni;
1255         u32 ifindex, nhid;
1256         u32 hash_index;
1257         int err;
1258
1259         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
1260                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
1261                         ndm->ndm_state);
1262                 return -EINVAL;
1263         }
1264
1265         if (!tb || (!tb[NDA_DST] && !tb[NDA_NH_ID]))
1266                 return -EINVAL;
1267
1268         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex,
1269                               &nhid, extack);
1270         if (err)
1271                 return err;
1272
1273         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
1274                 return -EAFNOSUPPORT;
1275
1276         hash_index = fdb_head_index(vxlan, addr, src_vni);
1277         spin_lock_bh(&vxlan->hash_lock[hash_index]);
1278         err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags,
1279                                port, src_vni, vni, ifindex,
1280                                ndm->ndm_flags | NTF_VXLAN_ADDED_BY_USER,
1281                                nhid, true, extack);
1282         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1283
1284         return err;
1285 }
1286
1287 int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
1288                        const unsigned char *addr, union vxlan_addr ip,
1289                        __be16 port, __be32 src_vni, __be32 vni,
1290                        u32 ifindex, bool swdev_notify)
1291 {
1292         struct vxlan_rdst *rd = NULL;
1293         struct vxlan_fdb *f;
1294         int err = -ENOENT;
1295
1296         f = vxlan_find_mac(vxlan, addr, src_vni);
1297         if (!f)
1298                 return err;
1299
1300         if (!vxlan_addr_any(&ip)) {
1301                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
1302                 if (!rd)
1303                         goto out;
1304         }
1305
1306         /* remove a destination if it's not the only one on the list,
1307          * otherwise destroy the fdb entry
1308          */
1309         if (rd && !list_is_singular(&f->remotes)) {
1310                 vxlan_fdb_dst_destroy(vxlan, f, rd, swdev_notify);
1311                 goto out;
1312         }
1313
1314         vxlan_fdb_destroy(vxlan, f, true, swdev_notify);
1315
1316 out:
1317         return 0;
1318 }
1319
1320 /* Delete entry (via netlink) */
1321 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
1322                             struct net_device *dev,
1323                             const unsigned char *addr, u16 vid,
1324                             struct netlink_ext_ack *extack)
1325 {
1326         struct vxlan_dev *vxlan = netdev_priv(dev);
1327         union vxlan_addr ip;
1328         __be32 src_vni, vni;
1329         u32 ifindex, nhid;
1330         u32 hash_index;
1331         __be16 port;
1332         int err;
1333
1334         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex,
1335                               &nhid, extack);
1336         if (err)
1337                 return err;
1338
1339         hash_index = fdb_head_index(vxlan, addr, src_vni);
1340         spin_lock_bh(&vxlan->hash_lock[hash_index]);
1341         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
1342                                  true);
1343         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1344
1345         return err;
1346 }
1347
1348 /* Dump forwarding table */
1349 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1350                           struct net_device *dev,
1351                           struct net_device *filter_dev, int *idx)
1352 {
1353         struct vxlan_dev *vxlan = netdev_priv(dev);
1354         unsigned int h;
1355         int err = 0;
1356
1357         for (h = 0; h < FDB_HASH_SIZE; ++h) {
1358                 struct vxlan_fdb *f;
1359
1360                 rcu_read_lock();
1361                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
1362                         struct vxlan_rdst *rd;
1363
1364                         if (rcu_access_pointer(f->nh)) {
1365                                 if (*idx < cb->args[2])
1366                                         goto skip_nh;
1367                                 err = vxlan_fdb_info(skb, vxlan, f,
1368                                                      NETLINK_CB(cb->skb).portid,
1369                                                      cb->nlh->nlmsg_seq,
1370                                                      RTM_NEWNEIGH,
1371                                                      NLM_F_MULTI, NULL);
1372                                 if (err < 0) {
1373                                         rcu_read_unlock();
1374                                         goto out;
1375                                 }
1376 skip_nh:
1377                                 *idx += 1;
1378                                 continue;
1379                         }
1380
1381                         list_for_each_entry_rcu(rd, &f->remotes, list) {
1382                                 if (*idx < cb->args[2])
1383                                         goto skip;
1384
1385                                 err = vxlan_fdb_info(skb, vxlan, f,
1386                                                      NETLINK_CB(cb->skb).portid,
1387                                                      cb->nlh->nlmsg_seq,
1388                                                      RTM_NEWNEIGH,
1389                                                      NLM_F_MULTI, rd);
1390                                 if (err < 0) {
1391                                         rcu_read_unlock();
1392                                         goto out;
1393                                 }
1394 skip:
1395                                 *idx += 1;
1396                         }
1397                 }
1398                 rcu_read_unlock();
1399         }
1400 out:
1401         return err;
1402 }
1403
1404 static int vxlan_fdb_get(struct sk_buff *skb,
1405                          struct nlattr *tb[],
1406                          struct net_device *dev,
1407                          const unsigned char *addr,
1408                          u16 vid, u32 portid, u32 seq,
1409                          struct netlink_ext_ack *extack)
1410 {
1411         struct vxlan_dev *vxlan = netdev_priv(dev);
1412         struct vxlan_fdb *f;
1413         __be32 vni;
1414         int err;
1415
1416         if (tb[NDA_VNI])
1417                 vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1418         else
1419                 vni = vxlan->default_dst.remote_vni;
1420
1421         rcu_read_lock();
1422
1423         f = __vxlan_find_mac(vxlan, addr, vni);
1424         if (!f) {
1425                 NL_SET_ERR_MSG(extack, "Fdb entry not found");
1426                 err = -ENOENT;
1427                 goto errout;
1428         }
1429
1430         err = vxlan_fdb_info(skb, vxlan, f, portid, seq,
1431                              RTM_NEWNEIGH, 0, first_remote_rcu(f));
1432 errout:
1433         rcu_read_unlock();
1434         return err;
1435 }
1436
1437 /* Watch incoming packets to learn mapping between Ethernet address
1438  * and Tunnel endpoint.
1439  * Return true if packet is bogus and should be dropped.
1440  */
1441 static bool vxlan_snoop(struct net_device *dev,
1442                         union vxlan_addr *src_ip, const u8 *src_mac,
1443                         u32 src_ifindex, __be32 vni)
1444 {
1445         struct vxlan_dev *vxlan = netdev_priv(dev);
1446         struct vxlan_fdb *f;
1447         u32 ifindex = 0;
1448
1449 #if IS_ENABLED(CONFIG_IPV6)
1450         if (src_ip->sa.sa_family == AF_INET6 &&
1451             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
1452                 ifindex = src_ifindex;
1453 #endif
1454
1455         f = vxlan_find_mac(vxlan, src_mac, vni);
1456         if (likely(f)) {
1457                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1458
1459                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
1460                            rdst->remote_ifindex == ifindex))
1461                         return false;
1462
1463                 /* Don't migrate static entries, drop packets */
1464                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
1465                         return true;
1466
1467                 /* Don't override an fdb with nexthop with a learnt entry */
1468                 if (rcu_access_pointer(f->nh))
1469                         return true;
1470
1471                 if (net_ratelimit())
1472                         netdev_info(dev,
1473                                     "%pM migrated from %pIS to %pIS\n",
1474                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1475
1476                 rdst->remote_ip = *src_ip;
1477                 f->updated = jiffies;
1478                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH, true, NULL);
1479         } else {
1480                 u32 hash_index = fdb_head_index(vxlan, src_mac, vni);
1481
1482                 /* learned new entry */
1483                 spin_lock(&vxlan->hash_lock[hash_index]);
1484
1485                 /* close off race between vxlan_flush and incoming packets */
1486                 if (netif_running(dev))
1487                         vxlan_fdb_update(vxlan, src_mac, src_ip,
1488                                          NUD_REACHABLE,
1489                                          NLM_F_EXCL|NLM_F_CREATE,
1490                                          vxlan->cfg.dst_port,
1491                                          vni,
1492                                          vxlan->default_dst.remote_vni,
1493                                          ifindex, NTF_SELF, 0, true, NULL);
1494                 spin_unlock(&vxlan->hash_lock[hash_index]);
1495         }
1496
1497         return false;
1498 }
1499
1500 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1501 {
1502         struct vxlan_net *vn;
1503
1504         if (!vs)
1505                 return false;
1506         if (!refcount_dec_and_test(&vs->refcnt))
1507                 return false;
1508
1509         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1510         spin_lock(&vn->sock_lock);
1511         hlist_del_rcu(&vs->hlist);
1512         udp_tunnel_notify_del_rx_port(vs->sock,
1513                                       (vs->flags & VXLAN_F_GPE) ?
1514                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1515                                       UDP_TUNNEL_TYPE_VXLAN);
1516         spin_unlock(&vn->sock_lock);
1517
1518         return true;
1519 }
1520
1521 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1522 {
1523         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1524 #if IS_ENABLED(CONFIG_IPV6)
1525         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1526
1527         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1528 #endif
1529
1530         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1531         synchronize_net();
1532
1533         if (vxlan->cfg.flags & VXLAN_F_VNIFILTER)
1534                 vxlan_vs_del_vnigrp(vxlan);
1535         else
1536                 vxlan_vs_del_dev(vxlan);
1537
1538         if (__vxlan_sock_release_prep(sock4)) {
1539                 udp_tunnel_sock_release(sock4->sock);
1540                 kfree(sock4);
1541         }
1542
1543 #if IS_ENABLED(CONFIG_IPV6)
1544         if (__vxlan_sock_release_prep(sock6)) {
1545                 udp_tunnel_sock_release(sock6->sock);
1546                 kfree(sock6);
1547         }
1548 #endif
1549 }
1550
1551 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1552                           struct sk_buff *skb, u32 vxflags)
1553 {
1554         size_t start, offset;
1555
1556         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1557                 goto out;
1558
1559         start = vxlan_rco_start(unparsed->vx_vni);
1560         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1561
1562         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1563                 return false;
1564
1565         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1566                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1567 out:
1568         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1569         unparsed->vx_vni &= VXLAN_VNI_MASK;
1570         return true;
1571 }
1572
1573 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1574                                 struct sk_buff *skb, u32 vxflags,
1575                                 struct vxlan_metadata *md)
1576 {
1577         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1578         struct metadata_dst *tun_dst;
1579
1580         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1581                 goto out;
1582
1583         md->gbp = ntohs(gbp->policy_id);
1584
1585         tun_dst = (struct metadata_dst *)skb_dst(skb);
1586         if (tun_dst) {
1587                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1588                 tun_dst->u.tun_info.options_len = sizeof(*md);
1589         }
1590         if (gbp->dont_learn)
1591                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1592
1593         if (gbp->policy_applied)
1594                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1595
1596         /* In flow-based mode, GBP is carried in dst_metadata */
1597         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1598                 skb->mark = md->gbp;
1599 out:
1600         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1601 }
1602
1603 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1604                           struct vxlan_sock *vs,
1605                           struct sk_buff *skb, __be32 vni)
1606 {
1607         union vxlan_addr saddr;
1608         u32 ifindex = skb->dev->ifindex;
1609
1610         skb_reset_mac_header(skb);
1611         skb->protocol = eth_type_trans(skb, vxlan->dev);
1612         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1613
1614         /* Ignore packet loops (and multicast echo) */
1615         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1616                 return false;
1617
1618         /* Get address from the outer IP header */
1619         if (vxlan_get_sk_family(vs) == AF_INET) {
1620                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1621                 saddr.sa.sa_family = AF_INET;
1622 #if IS_ENABLED(CONFIG_IPV6)
1623         } else {
1624                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1625                 saddr.sa.sa_family = AF_INET6;
1626 #endif
1627         }
1628
1629         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1630             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1631                 return false;
1632
1633         return true;
1634 }
1635
1636 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1637                                   struct sk_buff *skb)
1638 {
1639         int err = 0;
1640
1641         if (vxlan_get_sk_family(vs) == AF_INET)
1642                 err = IP_ECN_decapsulate(oiph, skb);
1643 #if IS_ENABLED(CONFIG_IPV6)
1644         else
1645                 err = IP6_ECN_decapsulate(oiph, skb);
1646 #endif
1647
1648         if (unlikely(err) && log_ecn_error) {
1649                 if (vxlan_get_sk_family(vs) == AF_INET)
1650                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1651                                              &((struct iphdr *)oiph)->saddr,
1652                                              ((struct iphdr *)oiph)->tos);
1653                 else
1654                         net_info_ratelimited("non-ECT from %pI6\n",
1655                                              &((struct ipv6hdr *)oiph)->saddr);
1656         }
1657         return err <= 1;
1658 }
1659
1660 /* Callback from net/ipv4/udp.c to receive packets */
1661 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1662 {
1663         struct vxlan_vni_node *vninode = NULL;
1664         struct vxlan_dev *vxlan;
1665         struct vxlan_sock *vs;
1666         struct vxlanhdr unparsed;
1667         struct vxlan_metadata _md;
1668         struct vxlan_metadata *md = &_md;
1669         __be16 protocol = htons(ETH_P_TEB);
1670         bool raw_proto = false;
1671         void *oiph;
1672         __be32 vni = 0;
1673
1674         /* Need UDP and VXLAN header to be present */
1675         if (!pskb_may_pull(skb, VXLAN_HLEN))
1676                 goto drop;
1677
1678         unparsed = *vxlan_hdr(skb);
1679         /* VNI flag always required to be set */
1680         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1681                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1682                            ntohl(vxlan_hdr(skb)->vx_flags),
1683                            ntohl(vxlan_hdr(skb)->vx_vni));
1684                 /* Return non vxlan pkt */
1685                 goto drop;
1686         }
1687         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1688         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1689
1690         vs = rcu_dereference_sk_user_data(sk);
1691         if (!vs)
1692                 goto drop;
1693
1694         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1695
1696         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni, &vninode);
1697         if (!vxlan)
1698                 goto drop;
1699
1700         /* For backwards compatibility, only allow reserved fields to be
1701          * used by VXLAN extensions if explicitly requested.
1702          */
1703         if (vs->flags & VXLAN_F_GPE) {
1704                 if (!vxlan_parse_gpe_proto(&unparsed, &protocol))
1705                         goto drop;
1706                 unparsed.vx_flags &= ~VXLAN_GPE_USED_BITS;
1707                 raw_proto = true;
1708         }
1709
1710         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1711                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1712                 goto drop;
1713
1714         if (vs->flags & VXLAN_F_REMCSUM_RX)
1715                 if (unlikely(!vxlan_remcsum(&unparsed, skb, vs->flags)))
1716                         goto drop;
1717
1718         if (vxlan_collect_metadata(vs)) {
1719                 struct metadata_dst *tun_dst;
1720
1721                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1722                                          key32_to_tunnel_id(vni), sizeof(*md));
1723
1724                 if (!tun_dst)
1725                         goto drop;
1726
1727                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1728
1729                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1730         } else {
1731                 memset(md, 0, sizeof(*md));
1732         }
1733
1734         if (vs->flags & VXLAN_F_GBP)
1735                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1736         /* Note that GBP and GPE can never be active together. This is
1737          * ensured in vxlan_dev_configure.
1738          */
1739
1740         if (unparsed.vx_flags || unparsed.vx_vni) {
1741                 /* If there are any unprocessed flags remaining treat
1742                  * this as a malformed packet. This behavior diverges from
1743                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1744                  * in reserved fields are to be ignored. The approach here
1745                  * maintains compatibility with previous stack code, and also
1746                  * is more robust and provides a little more security in
1747                  * adding extensions to VXLAN.
1748                  */
1749                 goto drop;
1750         }
1751
1752         if (!raw_proto) {
1753                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1754                         goto drop;
1755         } else {
1756                 skb_reset_mac_header(skb);
1757                 skb->dev = vxlan->dev;
1758                 skb->pkt_type = PACKET_HOST;
1759         }
1760
1761         oiph = skb_network_header(skb);
1762         skb_reset_network_header(skb);
1763
1764         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1765                 ++vxlan->dev->stats.rx_frame_errors;
1766                 ++vxlan->dev->stats.rx_errors;
1767                 vxlan_vnifilter_count(vxlan, vni, vninode,
1768                                       VXLAN_VNI_STATS_RX_ERRORS, 0);
1769                 goto drop;
1770         }
1771
1772         rcu_read_lock();
1773
1774         if (unlikely(!(vxlan->dev->flags & IFF_UP))) {
1775                 rcu_read_unlock();
1776                 dev_core_stats_rx_dropped_inc(vxlan->dev);
1777                 vxlan_vnifilter_count(vxlan, vni, vninode,
1778                                       VXLAN_VNI_STATS_RX_DROPS, 0);
1779                 goto drop;
1780         }
1781
1782         dev_sw_netstats_rx_add(vxlan->dev, skb->len);
1783         vxlan_vnifilter_count(vxlan, vni, vninode, VXLAN_VNI_STATS_RX, skb->len);
1784         gro_cells_receive(&vxlan->gro_cells, skb);
1785
1786         rcu_read_unlock();
1787
1788         return 0;
1789
1790 drop:
1791         /* Consume bad packet */
1792         kfree_skb(skb);
1793         return 0;
1794 }
1795
1796 /* Callback from net/ipv{4,6}/udp.c to check that we have a VNI for errors */
1797 static int vxlan_err_lookup(struct sock *sk, struct sk_buff *skb)
1798 {
1799         struct vxlan_dev *vxlan;
1800         struct vxlan_sock *vs;
1801         struct vxlanhdr *hdr;
1802         __be32 vni;
1803
1804         if (!pskb_may_pull(skb, skb_transport_offset(skb) + VXLAN_HLEN))
1805                 return -EINVAL;
1806
1807         hdr = vxlan_hdr(skb);
1808
1809         if (!(hdr->vx_flags & VXLAN_HF_VNI))
1810                 return -EINVAL;
1811
1812         vs = rcu_dereference_sk_user_data(sk);
1813         if (!vs)
1814                 return -ENOENT;
1815
1816         vni = vxlan_vni(hdr->vx_vni);
1817         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni, NULL);
1818         if (!vxlan)
1819                 return -ENOENT;
1820
1821         return 0;
1822 }
1823
1824 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1825 {
1826         struct vxlan_dev *vxlan = netdev_priv(dev);
1827         struct arphdr *parp;
1828         u8 *arpptr, *sha;
1829         __be32 sip, tip;
1830         struct neighbour *n;
1831
1832         if (dev->flags & IFF_NOARP)
1833                 goto out;
1834
1835         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1836                 dev->stats.tx_dropped++;
1837                 goto out;
1838         }
1839         parp = arp_hdr(skb);
1840
1841         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1842              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1843             parp->ar_pro != htons(ETH_P_IP) ||
1844             parp->ar_op != htons(ARPOP_REQUEST) ||
1845             parp->ar_hln != dev->addr_len ||
1846             parp->ar_pln != 4)
1847                 goto out;
1848         arpptr = (u8 *)parp + sizeof(struct arphdr);
1849         sha = arpptr;
1850         arpptr += dev->addr_len;        /* sha */
1851         memcpy(&sip, arpptr, sizeof(sip));
1852         arpptr += sizeof(sip);
1853         arpptr += dev->addr_len;        /* tha */
1854         memcpy(&tip, arpptr, sizeof(tip));
1855
1856         if (ipv4_is_loopback(tip) ||
1857             ipv4_is_multicast(tip))
1858                 goto out;
1859
1860         n = neigh_lookup(&arp_tbl, &tip, dev);
1861
1862         if (n) {
1863                 struct vxlan_fdb *f;
1864                 struct sk_buff  *reply;
1865
1866                 if (!(READ_ONCE(n->nud_state) & NUD_CONNECTED)) {
1867                         neigh_release(n);
1868                         goto out;
1869                 }
1870
1871                 f = vxlan_find_mac(vxlan, n->ha, vni);
1872                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1873                         /* bridge-local neighbor */
1874                         neigh_release(n);
1875                         goto out;
1876                 }
1877
1878                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1879                                 n->ha, sha);
1880
1881                 neigh_release(n);
1882
1883                 if (reply == NULL)
1884                         goto out;
1885
1886                 skb_reset_mac_header(reply);
1887                 __skb_pull(reply, skb_network_offset(reply));
1888                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1889                 reply->pkt_type = PACKET_HOST;
1890
1891                 if (netif_rx(reply) == NET_RX_DROP) {
1892                         dev->stats.rx_dropped++;
1893                         vxlan_vnifilter_count(vxlan, vni, NULL,
1894                                               VXLAN_VNI_STATS_RX_DROPS, 0);
1895                 }
1896
1897         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
1898                 union vxlan_addr ipa = {
1899                         .sin.sin_addr.s_addr = tip,
1900                         .sin.sin_family = AF_INET,
1901                 };
1902
1903                 vxlan_ip_miss(dev, &ipa);
1904         }
1905 out:
1906         consume_skb(skb);
1907         return NETDEV_TX_OK;
1908 }
1909
1910 #if IS_ENABLED(CONFIG_IPV6)
1911 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1912         struct neighbour *n, bool isrouter)
1913 {
1914         struct net_device *dev = request->dev;
1915         struct sk_buff *reply;
1916         struct nd_msg *ns, *na;
1917         struct ipv6hdr *pip6;
1918         u8 *daddr;
1919         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1920         int ns_olen;
1921         int i, len;
1922
1923         if (dev == NULL || !pskb_may_pull(request, request->len))
1924                 return NULL;
1925
1926         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1927                 sizeof(*na) + na_olen + dev->needed_tailroom;
1928         reply = alloc_skb(len, GFP_ATOMIC);
1929         if (reply == NULL)
1930                 return NULL;
1931
1932         reply->protocol = htons(ETH_P_IPV6);
1933         reply->dev = dev;
1934         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1935         skb_push(reply, sizeof(struct ethhdr));
1936         skb_reset_mac_header(reply);
1937
1938         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
1939
1940         daddr = eth_hdr(request)->h_source;
1941         ns_olen = request->len - skb_network_offset(request) -
1942                 sizeof(struct ipv6hdr) - sizeof(*ns);
1943         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1944                 if (!ns->opt[i + 1]) {
1945                         kfree_skb(reply);
1946                         return NULL;
1947                 }
1948                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1949                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1950                         break;
1951                 }
1952         }
1953
1954         /* Ethernet header */
1955         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1956         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1957         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1958         reply->protocol = htons(ETH_P_IPV6);
1959
1960         skb_pull(reply, sizeof(struct ethhdr));
1961         skb_reset_network_header(reply);
1962         skb_put(reply, sizeof(struct ipv6hdr));
1963
1964         /* IPv6 header */
1965
1966         pip6 = ipv6_hdr(reply);
1967         memset(pip6, 0, sizeof(struct ipv6hdr));
1968         pip6->version = 6;
1969         pip6->priority = ipv6_hdr(request)->priority;
1970         pip6->nexthdr = IPPROTO_ICMPV6;
1971         pip6->hop_limit = 255;
1972         pip6->daddr = ipv6_hdr(request)->saddr;
1973         pip6->saddr = *(struct in6_addr *)n->primary_key;
1974
1975         skb_pull(reply, sizeof(struct ipv6hdr));
1976         skb_reset_transport_header(reply);
1977
1978         /* Neighbor Advertisement */
1979         na = skb_put_zero(reply, sizeof(*na) + na_olen);
1980         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1981         na->icmph.icmp6_router = isrouter;
1982         na->icmph.icmp6_override = 1;
1983         na->icmph.icmp6_solicited = 1;
1984         na->target = ns->target;
1985         ether_addr_copy(&na->opt[2], n->ha);
1986         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1987         na->opt[1] = na_olen >> 3;
1988
1989         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1990                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1991                 csum_partial(na, sizeof(*na)+na_olen, 0));
1992
1993         pip6->payload_len = htons(sizeof(*na)+na_olen);
1994
1995         skb_push(reply, sizeof(struct ipv6hdr));
1996
1997         reply->ip_summed = CHECKSUM_UNNECESSARY;
1998
1999         return reply;
2000 }
2001
2002 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
2003 {
2004         struct vxlan_dev *vxlan = netdev_priv(dev);
2005         const struct in6_addr *daddr;
2006         const struct ipv6hdr *iphdr;
2007         struct inet6_dev *in6_dev;
2008         struct neighbour *n;
2009         struct nd_msg *msg;
2010
2011         rcu_read_lock();
2012         in6_dev = __in6_dev_get(dev);
2013         if (!in6_dev)
2014                 goto out;
2015
2016         iphdr = ipv6_hdr(skb);
2017         daddr = &iphdr->daddr;
2018         msg = (struct nd_msg *)(iphdr + 1);
2019
2020         if (ipv6_addr_loopback(daddr) ||
2021             ipv6_addr_is_multicast(&msg->target))
2022                 goto out;
2023
2024         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
2025
2026         if (n) {
2027                 struct vxlan_fdb *f;
2028                 struct sk_buff *reply;
2029
2030                 if (!(READ_ONCE(n->nud_state) & NUD_CONNECTED)) {
2031                         neigh_release(n);
2032                         goto out;
2033                 }
2034
2035                 f = vxlan_find_mac(vxlan, n->ha, vni);
2036                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2037                         /* bridge-local neighbor */
2038                         neigh_release(n);
2039                         goto out;
2040                 }
2041
2042                 reply = vxlan_na_create(skb, n,
2043                                         !!(f ? f->flags & NTF_ROUTER : 0));
2044
2045                 neigh_release(n);
2046
2047                 if (reply == NULL)
2048                         goto out;
2049
2050                 if (netif_rx(reply) == NET_RX_DROP) {
2051                         dev->stats.rx_dropped++;
2052                         vxlan_vnifilter_count(vxlan, vni, NULL,
2053                                               VXLAN_VNI_STATS_RX_DROPS, 0);
2054                 }
2055         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2056                 union vxlan_addr ipa = {
2057                         .sin6.sin6_addr = msg->target,
2058                         .sin6.sin6_family = AF_INET6,
2059                 };
2060
2061                 vxlan_ip_miss(dev, &ipa);
2062         }
2063
2064 out:
2065         rcu_read_unlock();
2066         consume_skb(skb);
2067         return NETDEV_TX_OK;
2068 }
2069 #endif
2070
2071 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
2072 {
2073         struct vxlan_dev *vxlan = netdev_priv(dev);
2074         struct neighbour *n;
2075
2076         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
2077                 return false;
2078
2079         n = NULL;
2080         switch (ntohs(eth_hdr(skb)->h_proto)) {
2081         case ETH_P_IP:
2082         {
2083                 struct iphdr *pip;
2084
2085                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
2086                         return false;
2087                 pip = ip_hdr(skb);
2088                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
2089                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2090                         union vxlan_addr ipa = {
2091                                 .sin.sin_addr.s_addr = pip->daddr,
2092                                 .sin.sin_family = AF_INET,
2093                         };
2094
2095                         vxlan_ip_miss(dev, &ipa);
2096                         return false;
2097                 }
2098
2099                 break;
2100         }
2101 #if IS_ENABLED(CONFIG_IPV6)
2102         case ETH_P_IPV6:
2103         {
2104                 struct ipv6hdr *pip6;
2105
2106                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
2107                         return false;
2108                 pip6 = ipv6_hdr(skb);
2109                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
2110                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2111                         union vxlan_addr ipa = {
2112                                 .sin6.sin6_addr = pip6->daddr,
2113                                 .sin6.sin6_family = AF_INET6,
2114                         };
2115
2116                         vxlan_ip_miss(dev, &ipa);
2117                         return false;
2118                 }
2119
2120                 break;
2121         }
2122 #endif
2123         default:
2124                 return false;
2125         }
2126
2127         if (n) {
2128                 bool diff;
2129
2130                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
2131                 if (diff) {
2132                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
2133                                 dev->addr_len);
2134                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
2135                 }
2136                 neigh_release(n);
2137                 return diff;
2138         }
2139
2140         return false;
2141 }
2142
2143 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, __be16 protocol)
2144 {
2145         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
2146
2147         gpe->np_applied = 1;
2148         gpe->next_protocol = tun_p_from_eth_p(protocol);
2149         if (!gpe->next_protocol)
2150                 return -EPFNOSUPPORT;
2151         return 0;
2152 }
2153
2154 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
2155                            int iphdr_len, __be32 vni,
2156                            struct vxlan_metadata *md, u32 vxflags,
2157                            bool udp_sum)
2158 {
2159         struct vxlanhdr *vxh;
2160         int min_headroom;
2161         int err;
2162         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
2163         __be16 inner_protocol = htons(ETH_P_TEB);
2164
2165         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
2166             skb->ip_summed == CHECKSUM_PARTIAL) {
2167                 int csum_start = skb_checksum_start_offset(skb);
2168
2169                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
2170                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
2171                     (skb->csum_offset == offsetof(struct udphdr, check) ||
2172                      skb->csum_offset == offsetof(struct tcphdr, check)))
2173                         type |= SKB_GSO_TUNNEL_REMCSUM;
2174         }
2175
2176         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
2177                         + VXLAN_HLEN + iphdr_len;
2178
2179         /* Need space for new headers (invalidates iph ptr) */
2180         err = skb_cow_head(skb, min_headroom);
2181         if (unlikely(err))
2182                 return err;
2183
2184         err = iptunnel_handle_offloads(skb, type);
2185         if (err)
2186                 return err;
2187
2188         vxh = __skb_push(skb, sizeof(*vxh));
2189         vxh->vx_flags = VXLAN_HF_VNI;
2190         vxh->vx_vni = vxlan_vni_field(vni);
2191
2192         if (type & SKB_GSO_TUNNEL_REMCSUM) {
2193                 unsigned int start;
2194
2195                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
2196                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
2197                 vxh->vx_flags |= VXLAN_HF_RCO;
2198
2199                 if (!skb_is_gso(skb)) {
2200                         skb->ip_summed = CHECKSUM_NONE;
2201                         skb->encapsulation = 0;
2202                 }
2203         }
2204
2205         if (vxflags & VXLAN_F_GBP)
2206                 vxlan_build_gbp_hdr(vxh, md);
2207         if (vxflags & VXLAN_F_GPE) {
2208                 err = vxlan_build_gpe_hdr(vxh, skb->protocol);
2209                 if (err < 0)
2210                         return err;
2211                 inner_protocol = skb->protocol;
2212         }
2213
2214         skb_set_inner_protocol(skb, inner_protocol);
2215         return 0;
2216 }
2217
2218 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
2219                                       struct vxlan_sock *sock4,
2220                                       struct sk_buff *skb, int oif, u8 tos,
2221                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
2222                                       __u8 flow_flags, struct dst_cache *dst_cache,
2223                                       const struct ip_tunnel_info *info)
2224 {
2225         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2226         struct rtable *rt = NULL;
2227         struct flowi4 fl4;
2228
2229         if (!sock4)
2230                 return ERR_PTR(-EIO);
2231
2232         if (tos && !info)
2233                 use_cache = false;
2234         if (use_cache) {
2235                 rt = dst_cache_get_ip4(dst_cache, saddr);
2236                 if (rt)
2237                         return rt;
2238         }
2239
2240         memset(&fl4, 0, sizeof(fl4));
2241         fl4.flowi4_oif = oif;
2242         fl4.flowi4_tos = RT_TOS(tos);
2243         fl4.flowi4_mark = skb->mark;
2244         fl4.flowi4_proto = IPPROTO_UDP;
2245         fl4.daddr = daddr;
2246         fl4.saddr = *saddr;
2247         fl4.fl4_dport = dport;
2248         fl4.fl4_sport = sport;
2249         fl4.flowi4_flags = flow_flags;
2250
2251         rt = ip_route_output_key(vxlan->net, &fl4);
2252         if (!IS_ERR(rt)) {
2253                 if (rt->dst.dev == dev) {
2254                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
2255                         ip_rt_put(rt);
2256                         return ERR_PTR(-ELOOP);
2257                 }
2258
2259                 *saddr = fl4.saddr;
2260                 if (use_cache)
2261                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2262         } else {
2263                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
2264                 return ERR_PTR(-ENETUNREACH);
2265         }
2266         return rt;
2267 }
2268
2269 #if IS_ENABLED(CONFIG_IPV6)
2270 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
2271                                           struct net_device *dev,
2272                                           struct vxlan_sock *sock6,
2273                                           struct sk_buff *skb, int oif, u8 tos,
2274                                           __be32 label,
2275                                           const struct in6_addr *daddr,
2276                                           struct in6_addr *saddr,
2277                                           __be16 dport, __be16 sport,
2278                                           struct dst_cache *dst_cache,
2279                                           const struct ip_tunnel_info *info)
2280 {
2281         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2282         struct dst_entry *ndst;
2283         struct flowi6 fl6;
2284
2285         if (!sock6)
2286                 return ERR_PTR(-EIO);
2287
2288         if (tos && !info)
2289                 use_cache = false;
2290         if (use_cache) {
2291                 ndst = dst_cache_get_ip6(dst_cache, saddr);
2292                 if (ndst)
2293                         return ndst;
2294         }
2295
2296         memset(&fl6, 0, sizeof(fl6));
2297         fl6.flowi6_oif = oif;
2298         fl6.daddr = *daddr;
2299         fl6.saddr = *saddr;
2300         fl6.flowlabel = ip6_make_flowinfo(tos, label);
2301         fl6.flowi6_mark = skb->mark;
2302         fl6.flowi6_proto = IPPROTO_UDP;
2303         fl6.fl6_dport = dport;
2304         fl6.fl6_sport = sport;
2305
2306         ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk,
2307                                                &fl6, NULL);
2308         if (IS_ERR(ndst)) {
2309                 netdev_dbg(dev, "no route to %pI6\n", daddr);
2310                 return ERR_PTR(-ENETUNREACH);
2311         }
2312
2313         if (unlikely(ndst->dev == dev)) {
2314                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
2315                 dst_release(ndst);
2316                 return ERR_PTR(-ELOOP);
2317         }
2318
2319         *saddr = fl6.saddr;
2320         if (use_cache)
2321                 dst_cache_set_ip6(dst_cache, ndst, saddr);
2322         return ndst;
2323 }
2324 #endif
2325
2326 /* Bypass encapsulation if the destination is local */
2327 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2328                                struct vxlan_dev *dst_vxlan, __be32 vni,
2329                                bool snoop)
2330 {
2331         struct pcpu_sw_netstats *tx_stats, *rx_stats;
2332         union vxlan_addr loopback;
2333         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2334         struct net_device *dev;
2335         int len = skb->len;
2336
2337         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2338         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2339         skb->pkt_type = PACKET_HOST;
2340         skb->encapsulation = 0;
2341         skb->dev = dst_vxlan->dev;
2342         __skb_pull(skb, skb_network_offset(skb));
2343
2344         if (remote_ip->sa.sa_family == AF_INET) {
2345                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2346                 loopback.sa.sa_family =  AF_INET;
2347 #if IS_ENABLED(CONFIG_IPV6)
2348         } else {
2349                 loopback.sin6.sin6_addr = in6addr_loopback;
2350                 loopback.sa.sa_family =  AF_INET6;
2351 #endif
2352         }
2353
2354         rcu_read_lock();
2355         dev = skb->dev;
2356         if (unlikely(!(dev->flags & IFF_UP))) {
2357                 kfree_skb(skb);
2358                 goto drop;
2359         }
2360
2361         if ((dst_vxlan->cfg.flags & VXLAN_F_LEARN) && snoop)
2362                 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni);
2363
2364         u64_stats_update_begin(&tx_stats->syncp);
2365         u64_stats_inc(&tx_stats->tx_packets);
2366         u64_stats_add(&tx_stats->tx_bytes, len);
2367         u64_stats_update_end(&tx_stats->syncp);
2368         vxlan_vnifilter_count(src_vxlan, vni, NULL, VXLAN_VNI_STATS_TX, len);
2369
2370         if (__netif_rx(skb) == NET_RX_SUCCESS) {
2371                 u64_stats_update_begin(&rx_stats->syncp);
2372                 u64_stats_inc(&rx_stats->rx_packets);
2373                 u64_stats_add(&rx_stats->rx_bytes, len);
2374                 u64_stats_update_end(&rx_stats->syncp);
2375                 vxlan_vnifilter_count(dst_vxlan, vni, NULL, VXLAN_VNI_STATS_RX,
2376                                       len);
2377         } else {
2378 drop:
2379                 dev->stats.rx_dropped++;
2380                 vxlan_vnifilter_count(dst_vxlan, vni, NULL,
2381                                       VXLAN_VNI_STATS_RX_DROPS, 0);
2382         }
2383         rcu_read_unlock();
2384 }
2385
2386 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2387                                  struct vxlan_dev *vxlan,
2388                                  union vxlan_addr *daddr,
2389                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2390                                  struct dst_entry *dst,
2391                                  u32 rt_flags)
2392 {
2393 #if IS_ENABLED(CONFIG_IPV6)
2394         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2395          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2396          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2397          */
2398         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2399 #endif
2400         /* Bypass encapsulation if the destination is local */
2401         if (rt_flags & RTCF_LOCAL &&
2402             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) &&
2403             vxlan->cfg.flags & VXLAN_F_LOCALBYPASS) {
2404                 struct vxlan_dev *dst_vxlan;
2405
2406                 dst_release(dst);
2407                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2408                                            daddr->sa.sa_family, dst_port,
2409                                            vxlan->cfg.flags);
2410                 if (!dst_vxlan) {
2411                         dev->stats.tx_errors++;
2412                         vxlan_vnifilter_count(vxlan, vni, NULL,
2413                                               VXLAN_VNI_STATS_TX_ERRORS, 0);
2414                         kfree_skb(skb);
2415
2416                         return -ENOENT;
2417                 }
2418                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni, true);
2419                 return 1;
2420         }
2421
2422         return 0;
2423 }
2424
2425 void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2426                     __be32 default_vni, struct vxlan_rdst *rdst, bool did_rsc)
2427 {
2428         struct dst_cache *dst_cache;
2429         struct ip_tunnel_info *info;
2430         struct vxlan_dev *vxlan = netdev_priv(dev);
2431         const struct iphdr *old_iph = ip_hdr(skb);
2432         union vxlan_addr *dst;
2433         union vxlan_addr remote_ip, local_ip;
2434         struct vxlan_metadata _md;
2435         struct vxlan_metadata *md = &_md;
2436         unsigned int pkt_len = skb->len;
2437         __be16 src_port = 0, dst_port;
2438         struct dst_entry *ndst = NULL;
2439         __u8 tos, ttl, flow_flags = 0;
2440         int ifindex;
2441         int err;
2442         u32 flags = vxlan->cfg.flags;
2443         bool udp_sum = false;
2444         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2445         __be32 vni = 0;
2446 #if IS_ENABLED(CONFIG_IPV6)
2447         __be32 label;
2448 #endif
2449
2450         info = skb_tunnel_info(skb);
2451
2452         if (rdst) {
2453                 dst = &rdst->remote_ip;
2454                 if (vxlan_addr_any(dst)) {
2455                         if (did_rsc) {
2456                                 /* short-circuited back to local bridge */
2457                                 vxlan_encap_bypass(skb, vxlan, vxlan,
2458                                                    default_vni, true);
2459                                 return;
2460                         }
2461                         goto drop;
2462                 }
2463
2464                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2465                 vni = (rdst->remote_vni) ? : default_vni;
2466                 ifindex = rdst->remote_ifindex;
2467                 local_ip = vxlan->cfg.saddr;
2468                 dst_cache = &rdst->dst_cache;
2469                 md->gbp = skb->mark;
2470                 if (flags & VXLAN_F_TTL_INHERIT) {
2471                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2472                 } else {
2473                         ttl = vxlan->cfg.ttl;
2474                         if (!ttl && vxlan_addr_multicast(dst))
2475                                 ttl = 1;
2476                 }
2477
2478                 tos = vxlan->cfg.tos;
2479                 if (tos == 1)
2480                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2481
2482                 if (dst->sa.sa_family == AF_INET)
2483                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2484                 else
2485                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2486 #if IS_ENABLED(CONFIG_IPV6)
2487                 label = vxlan->cfg.label;
2488 #endif
2489         } else {
2490                 if (!info) {
2491                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2492                                   dev->name);
2493                         goto drop;
2494                 }
2495                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2496                 if (remote_ip.sa.sa_family == AF_INET) {
2497                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2498                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2499                 } else {
2500                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2501                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2502                 }
2503                 dst = &remote_ip;
2504                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2505                 flow_flags = info->key.flow_flags;
2506                 vni = tunnel_id_to_key32(info->key.tun_id);
2507                 ifindex = 0;
2508                 dst_cache = &info->dst_cache;
2509                 if (info->key.tun_flags & TUNNEL_VXLAN_OPT) {
2510                         if (info->options_len < sizeof(*md))
2511                                 goto drop;
2512                         md = ip_tunnel_info_opts(info);
2513                 }
2514                 ttl = info->key.ttl;
2515                 tos = info->key.tos;
2516 #if IS_ENABLED(CONFIG_IPV6)
2517                 label = info->key.label;
2518 #endif
2519                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2520         }
2521         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2522                                      vxlan->cfg.port_max, true);
2523
2524         rcu_read_lock();
2525         if (dst->sa.sa_family == AF_INET) {
2526                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2527                 struct rtable *rt;
2528                 __be16 df = 0;
2529
2530                 if (!ifindex)
2531                         ifindex = sock4->sock->sk->sk_bound_dev_if;
2532
2533                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2534                                      dst->sin.sin_addr.s_addr,
2535                                      &local_ip.sin.sin_addr.s_addr,
2536                                      dst_port, src_port, flow_flags,
2537                                      dst_cache, info);
2538                 if (IS_ERR(rt)) {
2539                         err = PTR_ERR(rt);
2540                         goto tx_error;
2541                 }
2542
2543                 if (!info) {
2544                         /* Bypass encapsulation if the destination is local */
2545                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2546                                                     dst_port, ifindex, vni,
2547                                                     &rt->dst, rt->rt_flags);
2548                         if (err)
2549                                 goto out_unlock;
2550
2551                         if (vxlan->cfg.df == VXLAN_DF_SET) {
2552                                 df = htons(IP_DF);
2553                         } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) {
2554                                 struct ethhdr *eth = eth_hdr(skb);
2555
2556                                 if (ntohs(eth->h_proto) == ETH_P_IPV6 ||
2557                                     (ntohs(eth->h_proto) == ETH_P_IP &&
2558                                      old_iph->frag_off & htons(IP_DF)))
2559                                         df = htons(IP_DF);
2560                         }
2561                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2562                         df = htons(IP_DF);
2563                 }
2564
2565                 ndst = &rt->dst;
2566                 err = skb_tunnel_check_pmtu(skb, ndst, vxlan_headroom(flags & VXLAN_F_GPE),
2567                                             netif_is_any_bridge_port(dev));
2568                 if (err < 0) {
2569                         goto tx_error;
2570                 } else if (err) {
2571                         if (info) {
2572                                 struct ip_tunnel_info *unclone;
2573                                 struct in_addr src, dst;
2574
2575                                 unclone = skb_tunnel_info_unclone(skb);
2576                                 if (unlikely(!unclone))
2577                                         goto tx_error;
2578
2579                                 src = remote_ip.sin.sin_addr;
2580                                 dst = local_ip.sin.sin_addr;
2581                                 unclone->key.u.ipv4.src = src.s_addr;
2582                                 unclone->key.u.ipv4.dst = dst.s_addr;
2583                         }
2584                         vxlan_encap_bypass(skb, vxlan, vxlan, vni, false);
2585                         dst_release(ndst);
2586                         goto out_unlock;
2587                 }
2588
2589                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2590                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2591                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2592                                       vni, md, flags, udp_sum);
2593                 if (err < 0)
2594                         goto tx_error;
2595
2596                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2597                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2598                                     src_port, dst_port, xnet, !udp_sum);
2599 #if IS_ENABLED(CONFIG_IPV6)
2600         } else {
2601                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2602
2603                 if (!ifindex)
2604                         ifindex = sock6->sock->sk->sk_bound_dev_if;
2605
2606                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2607                                         label, &dst->sin6.sin6_addr,
2608                                         &local_ip.sin6.sin6_addr,
2609                                         dst_port, src_port,
2610                                         dst_cache, info);
2611                 if (IS_ERR(ndst)) {
2612                         err = PTR_ERR(ndst);
2613                         ndst = NULL;
2614                         goto tx_error;
2615                 }
2616
2617                 if (!info) {
2618                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2619
2620                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2621                                                     dst_port, ifindex, vni,
2622                                                     ndst, rt6i_flags);
2623                         if (err)
2624                                 goto out_unlock;
2625                 }
2626
2627                 err = skb_tunnel_check_pmtu(skb, ndst,
2628                                             vxlan_headroom((flags & VXLAN_F_GPE) | VXLAN_F_IPV6),
2629                                             netif_is_any_bridge_port(dev));
2630                 if (err < 0) {
2631                         goto tx_error;
2632                 } else if (err) {
2633                         if (info) {
2634                                 struct ip_tunnel_info *unclone;
2635                                 struct in6_addr src, dst;
2636
2637                                 unclone = skb_tunnel_info_unclone(skb);
2638                                 if (unlikely(!unclone))
2639                                         goto tx_error;
2640
2641                                 src = remote_ip.sin6.sin6_addr;
2642                                 dst = local_ip.sin6.sin6_addr;
2643                                 unclone->key.u.ipv6.src = src;
2644                                 unclone->key.u.ipv6.dst = dst;
2645                         }
2646
2647                         vxlan_encap_bypass(skb, vxlan, vxlan, vni, false);
2648                         dst_release(ndst);
2649                         goto out_unlock;
2650                 }
2651
2652                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2653                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2654                 skb_scrub_packet(skb, xnet);
2655                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2656                                       vni, md, flags, udp_sum);
2657                 if (err < 0)
2658                         goto tx_error;
2659
2660                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2661                                      &local_ip.sin6.sin6_addr,
2662                                      &dst->sin6.sin6_addr, tos, ttl,
2663                                      label, src_port, dst_port, !udp_sum);
2664 #endif
2665         }
2666         vxlan_vnifilter_count(vxlan, vni, NULL, VXLAN_VNI_STATS_TX, pkt_len);
2667 out_unlock:
2668         rcu_read_unlock();
2669         return;
2670
2671 drop:
2672         dev->stats.tx_dropped++;
2673         vxlan_vnifilter_count(vxlan, vni, NULL, VXLAN_VNI_STATS_TX_DROPS, 0);
2674         dev_kfree_skb(skb);
2675         return;
2676
2677 tx_error:
2678         rcu_read_unlock();
2679         if (err == -ELOOP)
2680                 dev->stats.collisions++;
2681         else if (err == -ENETUNREACH)
2682                 dev->stats.tx_carrier_errors++;
2683         dst_release(ndst);
2684         dev->stats.tx_errors++;
2685         vxlan_vnifilter_count(vxlan, vni, NULL, VXLAN_VNI_STATS_TX_ERRORS, 0);
2686         kfree_skb(skb);
2687 }
2688
2689 static void vxlan_xmit_nh(struct sk_buff *skb, struct net_device *dev,
2690                           struct vxlan_fdb *f, __be32 vni, bool did_rsc)
2691 {
2692         struct vxlan_rdst nh_rdst;
2693         struct nexthop *nh;
2694         bool do_xmit;
2695         u32 hash;
2696
2697         memset(&nh_rdst, 0, sizeof(struct vxlan_rdst));
2698         hash = skb_get_hash(skb);
2699
2700         rcu_read_lock();
2701         nh = rcu_dereference(f->nh);
2702         if (!nh) {
2703                 rcu_read_unlock();
2704                 goto drop;
2705         }
2706         do_xmit = vxlan_fdb_nh_path_select(nh, hash, &nh_rdst);
2707         rcu_read_unlock();
2708
2709         if (likely(do_xmit))
2710                 vxlan_xmit_one(skb, dev, vni, &nh_rdst, did_rsc);
2711         else
2712                 goto drop;
2713
2714         return;
2715
2716 drop:
2717         dev->stats.tx_dropped++;
2718         vxlan_vnifilter_count(netdev_priv(dev), vni, NULL,
2719                               VXLAN_VNI_STATS_TX_DROPS, 0);
2720         dev_kfree_skb(skb);
2721 }
2722
2723 /* Transmit local packets over Vxlan
2724  *
2725  * Outer IP header inherits ECN and DF from inner header.
2726  * Outer UDP destination is the VXLAN assigned port.
2727  *           source port is based on hash of flow
2728  */
2729 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2730 {
2731         struct vxlan_dev *vxlan = netdev_priv(dev);
2732         struct vxlan_rdst *rdst, *fdst = NULL;
2733         const struct ip_tunnel_info *info;
2734         bool did_rsc = false;
2735         struct vxlan_fdb *f;
2736         struct ethhdr *eth;
2737         __be32 vni = 0;
2738
2739         info = skb_tunnel_info(skb);
2740
2741         skb_reset_mac_header(skb);
2742
2743         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2744                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2745                     info->mode & IP_TUNNEL_INFO_TX) {
2746                         vni = tunnel_id_to_key32(info->key.tun_id);
2747                 } else {
2748                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2749                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2750                         else
2751                                 kfree_skb(skb);
2752                         return NETDEV_TX_OK;
2753                 }
2754         }
2755
2756         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2757                 eth = eth_hdr(skb);
2758                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2759                         return arp_reduce(dev, skb, vni);
2760 #if IS_ENABLED(CONFIG_IPV6)
2761                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2762                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2763                                             sizeof(struct nd_msg)) &&
2764                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2765                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2766
2767                         if (m->icmph.icmp6_code == 0 &&
2768                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2769                                 return neigh_reduce(dev, skb, vni);
2770                 }
2771 #endif
2772         }
2773
2774         if (vxlan->cfg.flags & VXLAN_F_MDB) {
2775                 struct vxlan_mdb_entry *mdb_entry;
2776
2777                 rcu_read_lock();
2778                 mdb_entry = vxlan_mdb_entry_skb_get(vxlan, skb, vni);
2779                 if (mdb_entry) {
2780                         netdev_tx_t ret;
2781
2782                         ret = vxlan_mdb_xmit(vxlan, mdb_entry, skb);
2783                         rcu_read_unlock();
2784                         return ret;
2785                 }
2786                 rcu_read_unlock();
2787         }
2788
2789         eth = eth_hdr(skb);
2790         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2791         did_rsc = false;
2792
2793         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2794             (ntohs(eth->h_proto) == ETH_P_IP ||
2795              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2796                 did_rsc = route_shortcircuit(dev, skb);
2797                 if (did_rsc)
2798                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2799         }
2800
2801         if (f == NULL) {
2802                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2803                 if (f == NULL) {
2804                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2805                             !is_multicast_ether_addr(eth->h_dest))
2806                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2807
2808                         dev->stats.tx_dropped++;
2809                         vxlan_vnifilter_count(vxlan, vni, NULL,
2810                                               VXLAN_VNI_STATS_TX_DROPS, 0);
2811                         kfree_skb(skb);
2812                         return NETDEV_TX_OK;
2813                 }
2814         }
2815
2816         if (rcu_access_pointer(f->nh)) {
2817                 vxlan_xmit_nh(skb, dev, f,
2818                               (vni ? : vxlan->default_dst.remote_vni), did_rsc);
2819         } else {
2820                 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2821                         struct sk_buff *skb1;
2822
2823                         if (!fdst) {
2824                                 fdst = rdst;
2825                                 continue;
2826                         }
2827                         skb1 = skb_clone(skb, GFP_ATOMIC);
2828                         if (skb1)
2829                                 vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2830                 }
2831                 if (fdst)
2832                         vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2833                 else
2834                         kfree_skb(skb);
2835         }
2836
2837         return NETDEV_TX_OK;
2838 }
2839
2840 /* Walk the forwarding table and purge stale entries */
2841 static void vxlan_cleanup(struct timer_list *t)
2842 {
2843         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2844         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2845         unsigned int h;
2846
2847         if (!netif_running(vxlan->dev))
2848                 return;
2849
2850         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2851                 struct hlist_node *p, *n;
2852
2853                 spin_lock(&vxlan->hash_lock[h]);
2854                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2855                         struct vxlan_fdb *f
2856                                 = container_of(p, struct vxlan_fdb, hlist);
2857                         unsigned long timeout;
2858
2859                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2860                                 continue;
2861
2862                         if (f->flags & NTF_EXT_LEARNED)
2863                                 continue;
2864
2865                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2866                         if (time_before_eq(timeout, jiffies)) {
2867                                 netdev_dbg(vxlan->dev,
2868                                            "garbage collect %pM\n",
2869                                            f->eth_addr);
2870                                 f->state = NUD_STALE;
2871                                 vxlan_fdb_destroy(vxlan, f, true, true);
2872                         } else if (time_before(timeout, next_timer))
2873                                 next_timer = timeout;
2874                 }
2875                 spin_unlock(&vxlan->hash_lock[h]);
2876         }
2877
2878         mod_timer(&vxlan->age_timer, next_timer);
2879 }
2880
2881 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2882 {
2883         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2884
2885         spin_lock(&vn->sock_lock);
2886         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2887 #if IS_ENABLED(CONFIG_IPV6)
2888         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2889 #endif
2890         spin_unlock(&vn->sock_lock);
2891 }
2892
2893 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2894                              struct vxlan_dev_node *node)
2895 {
2896         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2897         __be32 vni = vxlan->default_dst.remote_vni;
2898
2899         node->vxlan = vxlan;
2900         spin_lock(&vn->sock_lock);
2901         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2902         spin_unlock(&vn->sock_lock);
2903 }
2904
2905 /* Setup stats when device is created */
2906 static int vxlan_init(struct net_device *dev)
2907 {
2908         struct vxlan_dev *vxlan = netdev_priv(dev);
2909         int err;
2910
2911         if (vxlan->cfg.flags & VXLAN_F_VNIFILTER)
2912                 vxlan_vnigroup_init(vxlan);
2913
2914         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2915         if (!dev->tstats) {
2916                 err = -ENOMEM;
2917                 goto err_vnigroup_uninit;
2918         }
2919
2920         err = gro_cells_init(&vxlan->gro_cells, dev);
2921         if (err)
2922                 goto err_free_percpu;
2923
2924         err = vxlan_mdb_init(vxlan);
2925         if (err)
2926                 goto err_gro_cells_destroy;
2927
2928         return 0;
2929
2930 err_gro_cells_destroy:
2931         gro_cells_destroy(&vxlan->gro_cells);
2932 err_free_percpu:
2933         free_percpu(dev->tstats);
2934 err_vnigroup_uninit:
2935         if (vxlan->cfg.flags & VXLAN_F_VNIFILTER)
2936                 vxlan_vnigroup_uninit(vxlan);
2937         return err;
2938 }
2939
2940 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
2941 {
2942         struct vxlan_fdb *f;
2943         u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, vni);
2944
2945         spin_lock_bh(&vxlan->hash_lock[hash_index]);
2946         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
2947         if (f)
2948                 vxlan_fdb_destroy(vxlan, f, true, true);
2949         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
2950 }
2951
2952 static void vxlan_uninit(struct net_device *dev)
2953 {
2954         struct vxlan_dev *vxlan = netdev_priv(dev);
2955
2956         vxlan_mdb_fini(vxlan);
2957
2958         if (vxlan->cfg.flags & VXLAN_F_VNIFILTER)
2959                 vxlan_vnigroup_uninit(vxlan);
2960
2961         gro_cells_destroy(&vxlan->gro_cells);
2962
2963         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
2964
2965         free_percpu(dev->tstats);
2966 }
2967
2968 /* Start ageing timer and join group when device is brought up */
2969 static int vxlan_open(struct net_device *dev)
2970 {
2971         struct vxlan_dev *vxlan = netdev_priv(dev);
2972         int ret;
2973
2974         ret = vxlan_sock_add(vxlan);
2975         if (ret < 0)
2976                 return ret;
2977
2978         ret = vxlan_multicast_join(vxlan);
2979         if (ret) {
2980                 vxlan_sock_release(vxlan);
2981                 return ret;
2982         }
2983
2984         if (vxlan->cfg.age_interval)
2985                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2986
2987         return ret;
2988 }
2989
2990 /* Purge the forwarding table */
2991 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
2992 {
2993         unsigned int h;
2994
2995         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2996                 struct hlist_node *p, *n;
2997
2998                 spin_lock_bh(&vxlan->hash_lock[h]);
2999                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
3000                         struct vxlan_fdb *f
3001                                 = container_of(p, struct vxlan_fdb, hlist);
3002                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
3003                                 continue;
3004                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
3005                         if (is_zero_ether_addr(f->eth_addr) &&
3006                             f->vni == vxlan->cfg.vni)
3007                                 continue;
3008                         vxlan_fdb_destroy(vxlan, f, true, true);
3009                 }
3010                 spin_unlock_bh(&vxlan->hash_lock[h]);
3011         }
3012 }
3013
3014 /* Cleanup timer and forwarding table on shutdown */
3015 static int vxlan_stop(struct net_device *dev)
3016 {
3017         struct vxlan_dev *vxlan = netdev_priv(dev);
3018
3019         vxlan_multicast_leave(vxlan);
3020
3021         del_timer_sync(&vxlan->age_timer);
3022
3023         vxlan_flush(vxlan, false);
3024         vxlan_sock_release(vxlan);
3025
3026         return 0;
3027 }
3028
3029 /* Stub, nothing needs to be done. */
3030 static void vxlan_set_multicast_list(struct net_device *dev)
3031 {
3032 }
3033
3034 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
3035 {
3036         struct vxlan_dev *vxlan = netdev_priv(dev);
3037         struct vxlan_rdst *dst = &vxlan->default_dst;
3038         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
3039                                                          dst->remote_ifindex);
3040
3041         /* This check is different than dev->max_mtu, because it looks at
3042          * the lowerdev->mtu, rather than the static dev->max_mtu
3043          */
3044         if (lowerdev) {
3045                 int max_mtu = lowerdev->mtu - vxlan_headroom(vxlan->cfg.flags);
3046                 if (new_mtu > max_mtu)
3047                         return -EINVAL;
3048         }
3049
3050         dev->mtu = new_mtu;
3051         return 0;
3052 }
3053
3054 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
3055 {
3056         struct vxlan_dev *vxlan = netdev_priv(dev);
3057         struct ip_tunnel_info *info = skb_tunnel_info(skb);
3058         __be16 sport, dport;
3059
3060         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
3061                                   vxlan->cfg.port_max, true);
3062         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
3063
3064         if (ip_tunnel_info_af(info) == AF_INET) {
3065                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
3066                 struct rtable *rt;
3067
3068                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
3069                                      info->key.u.ipv4.dst,
3070                                      &info->key.u.ipv4.src, dport, sport,
3071                                      info->key.flow_flags, &info->dst_cache,
3072                                      info);
3073                 if (IS_ERR(rt))
3074                         return PTR_ERR(rt);
3075                 ip_rt_put(rt);
3076         } else {
3077 #if IS_ENABLED(CONFIG_IPV6)
3078                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
3079                 struct dst_entry *ndst;
3080
3081                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
3082                                         info->key.label, &info->key.u.ipv6.dst,
3083                                         &info->key.u.ipv6.src, dport, sport,
3084                                         &info->dst_cache, info);
3085                 if (IS_ERR(ndst))
3086                         return PTR_ERR(ndst);
3087                 dst_release(ndst);
3088 #else /* !CONFIG_IPV6 */
3089                 return -EPFNOSUPPORT;
3090 #endif
3091         }
3092         info->key.tp_src = sport;
3093         info->key.tp_dst = dport;
3094         return 0;
3095 }
3096
3097 static const struct net_device_ops vxlan_netdev_ether_ops = {
3098         .ndo_init               = vxlan_init,
3099         .ndo_uninit             = vxlan_uninit,
3100         .ndo_open               = vxlan_open,
3101         .ndo_stop               = vxlan_stop,
3102         .ndo_start_xmit         = vxlan_xmit,
3103         .ndo_get_stats64        = dev_get_tstats64,
3104         .ndo_set_rx_mode        = vxlan_set_multicast_list,
3105         .ndo_change_mtu         = vxlan_change_mtu,
3106         .ndo_validate_addr      = eth_validate_addr,
3107         .ndo_set_mac_address    = eth_mac_addr,
3108         .ndo_fdb_add            = vxlan_fdb_add,
3109         .ndo_fdb_del            = vxlan_fdb_delete,
3110         .ndo_fdb_dump           = vxlan_fdb_dump,
3111         .ndo_fdb_get            = vxlan_fdb_get,
3112         .ndo_mdb_add            = vxlan_mdb_add,
3113         .ndo_mdb_del            = vxlan_mdb_del,
3114         .ndo_mdb_dump           = vxlan_mdb_dump,
3115         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
3116 };
3117
3118 static const struct net_device_ops vxlan_netdev_raw_ops = {
3119         .ndo_init               = vxlan_init,
3120         .ndo_uninit             = vxlan_uninit,
3121         .ndo_open               = vxlan_open,
3122         .ndo_stop               = vxlan_stop,
3123         .ndo_start_xmit         = vxlan_xmit,
3124         .ndo_get_stats64        = dev_get_tstats64,
3125         .ndo_change_mtu         = vxlan_change_mtu,
3126         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
3127 };
3128
3129 /* Info for udev, that this is a virtual tunnel endpoint */
3130 static struct device_type vxlan_type = {
3131         .name = "vxlan",
3132 };
3133
3134 /* Calls the ndo_udp_tunnel_add of the caller in order to
3135  * supply the listening VXLAN udp ports. Callers are expected
3136  * to implement the ndo_udp_tunnel_add.
3137  */
3138 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
3139 {
3140         struct vxlan_sock *vs;
3141         struct net *net = dev_net(dev);
3142         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3143         unsigned int i;
3144
3145         spin_lock(&vn->sock_lock);
3146         for (i = 0; i < PORT_HASH_SIZE; ++i) {
3147                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
3148                         unsigned short type;
3149
3150                         if (vs->flags & VXLAN_F_GPE)
3151                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
3152                         else
3153                                 type = UDP_TUNNEL_TYPE_VXLAN;
3154
3155                         if (push)
3156                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
3157                         else
3158                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
3159                 }
3160         }
3161         spin_unlock(&vn->sock_lock);
3162 }
3163
3164 /* Initialize the device structure. */
3165 static void vxlan_setup(struct net_device *dev)
3166 {
3167         struct vxlan_dev *vxlan = netdev_priv(dev);
3168         unsigned int h;
3169
3170         eth_hw_addr_random(dev);
3171         ether_setup(dev);
3172
3173         dev->needs_free_netdev = true;
3174         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
3175
3176         dev->features   |= NETIF_F_LLTX;
3177         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
3178         dev->features   |= NETIF_F_RXCSUM;
3179         dev->features   |= NETIF_F_GSO_SOFTWARE;
3180
3181         dev->vlan_features = dev->features;
3182         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_FRAGLIST;
3183         dev->hw_features |= NETIF_F_RXCSUM;
3184         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
3185         netif_keep_dst(dev);
3186         dev->priv_flags |= IFF_NO_QUEUE | IFF_CHANGE_PROTO_DOWN;
3187
3188         /* MTU range: 68 - 65535 */
3189         dev->min_mtu = ETH_MIN_MTU;
3190         dev->max_mtu = ETH_MAX_MTU;
3191
3192         INIT_LIST_HEAD(&vxlan->next);
3193
3194         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
3195
3196         vxlan->dev = dev;
3197
3198         for (h = 0; h < FDB_HASH_SIZE; ++h) {
3199                 spin_lock_init(&vxlan->hash_lock[h]);
3200                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
3201         }
3202 }
3203
3204 static void vxlan_ether_setup(struct net_device *dev)
3205 {
3206         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
3207         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3208         dev->netdev_ops = &vxlan_netdev_ether_ops;
3209 }
3210
3211 static void vxlan_raw_setup(struct net_device *dev)
3212 {
3213         dev->header_ops = NULL;
3214         dev->type = ARPHRD_NONE;
3215         dev->hard_header_len = 0;
3216         dev->addr_len = 0;
3217         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
3218         dev->netdev_ops = &vxlan_netdev_raw_ops;
3219 }
3220
3221 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
3222         [IFLA_VXLAN_UNSPEC]     = { .strict_start_type = IFLA_VXLAN_LOCALBYPASS },
3223         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
3224         [IFLA_VXLAN_GROUP]      = { .len = sizeof_field(struct iphdr, daddr) },
3225         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
3226         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
3227         [IFLA_VXLAN_LOCAL]      = { .len = sizeof_field(struct iphdr, saddr) },
3228         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
3229         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
3230         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
3231         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
3232         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
3233         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
3234         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
3235         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
3236         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
3237         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
3238         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
3239         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
3240         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
3241         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
3242         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
3243         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
3244         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
3245         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
3246         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
3247         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
3248         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
3249         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
3250         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
3251         [IFLA_VXLAN_DF]         = { .type = NLA_U8 },
3252         [IFLA_VXLAN_VNIFILTER]  = { .type = NLA_U8 },
3253         [IFLA_VXLAN_LOCALBYPASS]        = NLA_POLICY_MAX(NLA_U8, 1),
3254 };
3255
3256 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
3257                           struct netlink_ext_ack *extack)
3258 {
3259         if (tb[IFLA_ADDRESS]) {
3260                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
3261                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3262                                             "Provided link layer address is not Ethernet");
3263                         return -EINVAL;
3264                 }
3265
3266                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
3267                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3268                                             "Provided Ethernet address is not unicast");
3269                         return -EADDRNOTAVAIL;
3270                 }
3271         }
3272
3273         if (tb[IFLA_MTU]) {
3274                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
3275
3276                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
3277                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3278                                             "MTU must be between 68 and 65535");
3279                         return -EINVAL;
3280                 }
3281         }
3282
3283         if (!data) {
3284                 NL_SET_ERR_MSG(extack,
3285                                "Required attributes not provided to perform the operation");
3286                 return -EINVAL;
3287         }
3288
3289         if (data[IFLA_VXLAN_ID]) {
3290                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
3291
3292                 if (id >= VXLAN_N_VID) {
3293                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_ID],
3294                                             "VXLAN ID must be lower than 16777216");
3295                         return -ERANGE;
3296                 }
3297         }
3298
3299         if (data[IFLA_VXLAN_PORT_RANGE]) {
3300                 const struct ifla_vxlan_port_range *p
3301                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3302
3303                 if (ntohs(p->high) < ntohs(p->low)) {
3304                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_PORT_RANGE],
3305                                             "Invalid source port range");
3306                         return -EINVAL;
3307                 }
3308         }
3309
3310         if (data[IFLA_VXLAN_DF]) {
3311                 enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]);
3312
3313                 if (df < 0 || df > VXLAN_DF_MAX) {
3314                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_DF],
3315                                             "Invalid DF attribute");
3316                         return -EINVAL;
3317                 }
3318         }
3319
3320         return 0;
3321 }
3322
3323 static void vxlan_get_drvinfo(struct net_device *netdev,
3324                               struct ethtool_drvinfo *drvinfo)
3325 {
3326         strscpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
3327         strscpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
3328 }
3329
3330 static int vxlan_get_link_ksettings(struct net_device *dev,
3331                                     struct ethtool_link_ksettings *cmd)
3332 {
3333         struct vxlan_dev *vxlan = netdev_priv(dev);
3334         struct vxlan_rdst *dst = &vxlan->default_dst;
3335         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
3336                                                          dst->remote_ifindex);
3337
3338         if (!lowerdev) {
3339                 cmd->base.duplex = DUPLEX_UNKNOWN;
3340                 cmd->base.port = PORT_OTHER;
3341                 cmd->base.speed = SPEED_UNKNOWN;
3342
3343                 return 0;
3344         }
3345
3346         return __ethtool_get_link_ksettings(lowerdev, cmd);
3347 }
3348
3349 static const struct ethtool_ops vxlan_ethtool_ops = {
3350         .get_drvinfo            = vxlan_get_drvinfo,
3351         .get_link               = ethtool_op_get_link,
3352         .get_link_ksettings     = vxlan_get_link_ksettings,
3353 };
3354
3355 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
3356                                         __be16 port, u32 flags, int ifindex)
3357 {
3358         struct socket *sock;
3359         struct udp_port_cfg udp_conf;
3360         int err;
3361
3362         memset(&udp_conf, 0, sizeof(udp_conf));
3363
3364         if (ipv6) {
3365                 udp_conf.family = AF_INET6;
3366                 udp_conf.use_udp6_rx_checksums =
3367                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
3368                 udp_conf.ipv6_v6only = 1;
3369         } else {
3370                 udp_conf.family = AF_INET;
3371         }
3372
3373         udp_conf.local_udp_port = port;
3374         udp_conf.bind_ifindex = ifindex;
3375
3376         /* Open UDP socket */
3377         err = udp_sock_create(net, &udp_conf, &sock);
3378         if (err < 0)
3379                 return ERR_PTR(err);
3380
3381         udp_allow_gso(sock->sk);
3382         return sock;
3383 }
3384
3385 /* Create new listen socket if needed */
3386 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
3387                                               __be16 port, u32 flags,
3388                                               int ifindex)
3389 {
3390         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3391         struct vxlan_sock *vs;
3392         struct socket *sock;
3393         unsigned int h;
3394         struct udp_tunnel_sock_cfg tunnel_cfg;
3395
3396         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
3397         if (!vs)
3398                 return ERR_PTR(-ENOMEM);
3399
3400         for (h = 0; h < VNI_HASH_SIZE; ++h)
3401                 INIT_HLIST_HEAD(&vs->vni_list[h]);
3402
3403         sock = vxlan_create_sock(net, ipv6, port, flags, ifindex);
3404         if (IS_ERR(sock)) {
3405                 kfree(vs);
3406                 return ERR_CAST(sock);
3407         }
3408
3409         vs->sock = sock;
3410         refcount_set(&vs->refcnt, 1);
3411         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
3412
3413         spin_lock(&vn->sock_lock);
3414         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
3415         udp_tunnel_notify_add_rx_port(sock,
3416                                       (vs->flags & VXLAN_F_GPE) ?
3417                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
3418                                       UDP_TUNNEL_TYPE_VXLAN);
3419         spin_unlock(&vn->sock_lock);
3420
3421         /* Mark socket as an encapsulation socket. */
3422         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
3423         tunnel_cfg.sk_user_data = vs;
3424         tunnel_cfg.encap_type = 1;
3425         tunnel_cfg.encap_rcv = vxlan_rcv;
3426         tunnel_cfg.encap_err_lookup = vxlan_err_lookup;
3427         tunnel_cfg.encap_destroy = NULL;
3428         if (vs->flags & VXLAN_F_GPE) {
3429                 tunnel_cfg.gro_receive = vxlan_gpe_gro_receive;
3430                 tunnel_cfg.gro_complete = vxlan_gpe_gro_complete;
3431         } else {
3432                 tunnel_cfg.gro_receive = vxlan_gro_receive;
3433                 tunnel_cfg.gro_complete = vxlan_gro_complete;
3434         }
3435
3436         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
3437
3438         return vs;
3439 }
3440
3441 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
3442 {
3443         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3444         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
3445         struct vxlan_sock *vs = NULL;
3446         struct vxlan_dev_node *node;
3447         int l3mdev_index = 0;
3448
3449         if (vxlan->cfg.remote_ifindex)
3450                 l3mdev_index = l3mdev_master_upper_ifindex_by_index(
3451                         vxlan->net, vxlan->cfg.remote_ifindex);
3452
3453         if (!vxlan->cfg.no_share) {
3454                 spin_lock(&vn->sock_lock);
3455                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
3456                                      vxlan->cfg.dst_port, vxlan->cfg.flags,
3457                                      l3mdev_index);
3458                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
3459                         spin_unlock(&vn->sock_lock);
3460                         return -EBUSY;
3461                 }
3462                 spin_unlock(&vn->sock_lock);
3463         }
3464         if (!vs)
3465                 vs = vxlan_socket_create(vxlan->net, ipv6,
3466                                          vxlan->cfg.dst_port, vxlan->cfg.flags,
3467                                          l3mdev_index);
3468         if (IS_ERR(vs))
3469                 return PTR_ERR(vs);
3470 #if IS_ENABLED(CONFIG_IPV6)
3471         if (ipv6) {
3472                 rcu_assign_pointer(vxlan->vn6_sock, vs);
3473                 node = &vxlan->hlist6;
3474         } else
3475 #endif
3476         {
3477                 rcu_assign_pointer(vxlan->vn4_sock, vs);
3478                 node = &vxlan->hlist4;
3479         }
3480
3481         if (metadata && (vxlan->cfg.flags & VXLAN_F_VNIFILTER))
3482                 vxlan_vs_add_vnigrp(vxlan, vs, ipv6);
3483         else
3484                 vxlan_vs_add_dev(vs, vxlan, node);
3485
3486         return 0;
3487 }
3488
3489 static int vxlan_sock_add(struct vxlan_dev *vxlan)
3490 {
3491         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
3492         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
3493         bool ipv4 = !ipv6 || metadata;
3494         int ret = 0;
3495
3496         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
3497 #if IS_ENABLED(CONFIG_IPV6)
3498         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
3499         if (ipv6) {
3500                 ret = __vxlan_sock_add(vxlan, true);
3501                 if (ret < 0 && ret != -EAFNOSUPPORT)
3502                         ipv4 = false;
3503         }
3504 #endif
3505         if (ipv4)
3506                 ret = __vxlan_sock_add(vxlan, false);
3507         if (ret < 0)
3508                 vxlan_sock_release(vxlan);
3509         return ret;
3510 }
3511
3512 int vxlan_vni_in_use(struct net *src_net, struct vxlan_dev *vxlan,
3513                      struct vxlan_config *conf, __be32 vni)
3514 {
3515         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
3516         struct vxlan_dev *tmp;
3517
3518         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3519                 if (tmp == vxlan)
3520                         continue;
3521                 if (tmp->cfg.flags & VXLAN_F_VNIFILTER) {
3522                         if (!vxlan_vnifilter_lookup(tmp, vni))
3523                                 continue;
3524                 } else if (tmp->cfg.vni != vni) {
3525                         continue;
3526                 }
3527                 if (tmp->cfg.dst_port != conf->dst_port)
3528                         continue;
3529                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3530                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3531                         continue;
3532
3533                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3534                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3535                         continue;
3536
3537                 return -EEXIST;
3538         }
3539
3540         return 0;
3541 }
3542
3543 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
3544                                  struct net_device **lower,
3545                                  struct vxlan_dev *old,
3546                                  struct netlink_ext_ack *extack)
3547 {
3548         bool use_ipv6 = false;
3549
3550         if (conf->flags & VXLAN_F_GPE) {
3551                 /* For now, allow GPE only together with
3552                  * COLLECT_METADATA. This can be relaxed later; in such
3553                  * case, the other side of the PtP link will have to be
3554                  * provided.
3555                  */
3556                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
3557                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3558                         NL_SET_ERR_MSG(extack,
3559                                        "VXLAN GPE does not support this combination of attributes");
3560                         return -EINVAL;
3561                 }
3562         }
3563
3564         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
3565                 /* Unless IPv6 is explicitly requested, assume IPv4 */
3566                 conf->remote_ip.sa.sa_family = AF_INET;
3567                 conf->saddr.sa.sa_family = AF_INET;
3568         } else if (!conf->remote_ip.sa.sa_family) {
3569                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
3570         } else if (!conf->saddr.sa.sa_family) {
3571                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
3572         }
3573
3574         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
3575                 NL_SET_ERR_MSG(extack,
3576                                "Local and remote address must be from the same family");
3577                 return -EINVAL;
3578         }
3579
3580         if (vxlan_addr_multicast(&conf->saddr)) {
3581                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
3582                 return -EINVAL;
3583         }
3584
3585         if (conf->saddr.sa.sa_family == AF_INET6) {
3586                 if (!IS_ENABLED(CONFIG_IPV6)) {
3587                         NL_SET_ERR_MSG(extack,
3588                                        "IPv6 support not enabled in the kernel");
3589                         return -EPFNOSUPPORT;
3590                 }
3591                 use_ipv6 = true;
3592                 conf->flags |= VXLAN_F_IPV6;
3593
3594                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3595                         int local_type =
3596                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3597                         int remote_type =
3598                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3599
3600                         if (local_type & IPV6_ADDR_LINKLOCAL) {
3601                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3602                                     (remote_type != IPV6_ADDR_ANY)) {
3603                                         NL_SET_ERR_MSG(extack,
3604                                                        "Invalid combination of local and remote address scopes");
3605                                         return -EINVAL;
3606                                 }
3607
3608                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3609                         } else {
3610                                 if (remote_type ==
3611                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3612                                         NL_SET_ERR_MSG(extack,
3613                                                        "Invalid combination of local and remote address scopes");
3614                                         return -EINVAL;
3615                                 }
3616
3617                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3618                         }
3619                 }
3620         }
3621
3622         if (conf->label && !use_ipv6) {
3623                 NL_SET_ERR_MSG(extack,
3624                                "Label attribute only applies to IPv6 VXLAN devices");
3625                 return -EINVAL;
3626         }
3627
3628         if (conf->remote_ifindex) {
3629                 struct net_device *lowerdev;
3630
3631                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3632                 if (!lowerdev) {
3633                         NL_SET_ERR_MSG(extack,
3634                                        "Invalid local interface, device not found");
3635                         return -ENODEV;
3636                 }
3637
3638 #if IS_ENABLED(CONFIG_IPV6)
3639                 if (use_ipv6) {
3640                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3641
3642                         if (idev && idev->cnf.disable_ipv6) {
3643                                 NL_SET_ERR_MSG(extack,
3644                                                "IPv6 support disabled by administrator");
3645                                 return -EPERM;
3646                         }
3647                 }
3648 #endif
3649
3650                 *lower = lowerdev;
3651         } else {
3652                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3653                         NL_SET_ERR_MSG(extack,
3654                                        "Local interface required for multicast remote destination");
3655
3656                         return -EINVAL;
3657                 }
3658
3659 #if IS_ENABLED(CONFIG_IPV6)
3660                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3661                         NL_SET_ERR_MSG(extack,
3662                                        "Local interface required for link-local local/remote addresses");
3663                         return -EINVAL;
3664                 }
3665 #endif
3666
3667                 *lower = NULL;
3668         }
3669
3670         if (!conf->dst_port) {
3671                 if (conf->flags & VXLAN_F_GPE)
3672                         conf->dst_port = htons(IANA_VXLAN_GPE_UDP_PORT);
3673                 else
3674                         conf->dst_port = htons(vxlan_port);
3675         }
3676
3677         if (!conf->age_interval)
3678                 conf->age_interval = FDB_AGE_DEFAULT;
3679
3680         if (vxlan_vni_in_use(src_net, old, conf, conf->vni)) {
3681                 NL_SET_ERR_MSG(extack,
3682                                "A VXLAN device with the specified VNI already exists");
3683                 return -EEXIST;
3684         }
3685
3686         return 0;
3687 }
3688
3689 static void vxlan_config_apply(struct net_device *dev,
3690                                struct vxlan_config *conf,
3691                                struct net_device *lowerdev,
3692                                struct net *src_net,
3693                                bool changelink)
3694 {
3695         struct vxlan_dev *vxlan = netdev_priv(dev);
3696         struct vxlan_rdst *dst = &vxlan->default_dst;
3697         unsigned short needed_headroom = ETH_HLEN;
3698         int max_mtu = ETH_MAX_MTU;
3699         u32 flags = conf->flags;
3700
3701         if (!changelink) {
3702                 if (flags & VXLAN_F_GPE)
3703                         vxlan_raw_setup(dev);
3704                 else
3705                         vxlan_ether_setup(dev);
3706
3707                 if (conf->mtu)
3708                         dev->mtu = conf->mtu;
3709
3710                 vxlan->net = src_net;
3711         }
3712
3713         dst->remote_vni = conf->vni;
3714
3715         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3716
3717         if (lowerdev) {
3718                 dst->remote_ifindex = conf->remote_ifindex;
3719
3720                 netif_inherit_tso_max(dev, lowerdev);
3721
3722                 needed_headroom = lowerdev->hard_header_len;
3723                 needed_headroom += lowerdev->needed_headroom;
3724
3725                 dev->needed_tailroom = lowerdev->needed_tailroom;
3726
3727                 max_mtu = lowerdev->mtu - vxlan_headroom(flags);
3728                 if (max_mtu < ETH_MIN_MTU)
3729                         max_mtu = ETH_MIN_MTU;
3730
3731                 if (!changelink && !conf->mtu)
3732                         dev->mtu = max_mtu;
3733         }
3734
3735         if (dev->mtu > max_mtu)
3736                 dev->mtu = max_mtu;
3737
3738         if (flags & VXLAN_F_COLLECT_METADATA)
3739                 flags |= VXLAN_F_IPV6;
3740         needed_headroom += vxlan_headroom(flags);
3741         dev->needed_headroom = needed_headroom;
3742
3743         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3744 }
3745
3746 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3747                                struct vxlan_config *conf, bool changelink,
3748                                struct netlink_ext_ack *extack)
3749 {
3750         struct vxlan_dev *vxlan = netdev_priv(dev);
3751         struct net_device *lowerdev;
3752         int ret;
3753
3754         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3755         if (ret)
3756                 return ret;
3757
3758         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3759
3760         return 0;
3761 }
3762
3763 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3764                               struct vxlan_config *conf,
3765                               struct netlink_ext_ack *extack)
3766 {
3767         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3768         struct vxlan_dev *vxlan = netdev_priv(dev);
3769         struct net_device *remote_dev = NULL;
3770         struct vxlan_fdb *f = NULL;
3771         bool unregister = false;
3772         struct vxlan_rdst *dst;
3773         int err;
3774
3775         dst = &vxlan->default_dst;
3776         err = vxlan_dev_configure(net, dev, conf, false, extack);
3777         if (err)
3778                 return err;
3779
3780         dev->ethtool_ops = &vxlan_ethtool_ops;
3781
3782         /* create an fdb entry for a valid default destination */
3783         if (!vxlan_addr_any(&dst->remote_ip)) {
3784                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3785                                        &dst->remote_ip,
3786                                        NUD_REACHABLE | NUD_PERMANENT,
3787                                        vxlan->cfg.dst_port,
3788                                        dst->remote_vni,
3789                                        dst->remote_vni,
3790                                        dst->remote_ifindex,
3791                                        NTF_SELF, 0, &f, extack);
3792                 if (err)
3793                         return err;
3794         }
3795
3796         err = register_netdevice(dev);
3797         if (err)
3798                 goto errout;
3799         unregister = true;
3800
3801         if (dst->remote_ifindex) {
3802                 remote_dev = __dev_get_by_index(net, dst->remote_ifindex);
3803                 if (!remote_dev) {
3804                         err = -ENODEV;
3805                         goto errout;
3806                 }
3807
3808                 err = netdev_upper_dev_link(remote_dev, dev, extack);
3809                 if (err)
3810                         goto errout;
3811         }
3812
3813         err = rtnl_configure_link(dev, NULL, 0, NULL);
3814         if (err < 0)
3815                 goto unlink;
3816
3817         if (f) {
3818                 vxlan_fdb_insert(vxlan, all_zeros_mac, dst->remote_vni, f);
3819
3820                 /* notify default fdb entry */
3821                 err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f),
3822                                        RTM_NEWNEIGH, true, extack);
3823                 if (err) {
3824                         vxlan_fdb_destroy(vxlan, f, false, false);
3825                         if (remote_dev)
3826                                 netdev_upper_dev_unlink(remote_dev, dev);
3827                         goto unregister;
3828                 }
3829         }
3830
3831         list_add(&vxlan->next, &vn->vxlan_list);
3832         if (remote_dev)
3833                 dst->remote_dev = remote_dev;
3834         return 0;
3835 unlink:
3836         if (remote_dev)
3837                 netdev_upper_dev_unlink(remote_dev, dev);
3838 errout:
3839         /* unregister_netdevice() destroys the default FDB entry with deletion
3840          * notification. But the addition notification was not sent yet, so
3841          * destroy the entry by hand here.
3842          */
3843         if (f)
3844                 __vxlan_fdb_free(f);
3845 unregister:
3846         if (unregister)
3847                 unregister_netdevice(dev);
3848         return err;
3849 }
3850
3851 /* Set/clear flags based on attribute */
3852 static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[],
3853                           int attrtype, unsigned long mask, bool changelink,
3854                           bool changelink_supported,
3855                           struct netlink_ext_ack *extack)
3856 {
3857         unsigned long flags;
3858
3859         if (!tb[attrtype])
3860                 return 0;
3861
3862         if (changelink && !changelink_supported) {
3863                 vxlan_flag_attr_error(attrtype, extack);
3864                 return -EOPNOTSUPP;
3865         }
3866
3867         if (vxlan_policy[attrtype].type == NLA_FLAG)
3868                 flags = conf->flags | mask;
3869         else if (nla_get_u8(tb[attrtype]))
3870                 flags = conf->flags | mask;
3871         else
3872                 flags = conf->flags & ~mask;
3873
3874         conf->flags = flags;
3875
3876         return 0;
3877 }
3878
3879 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3880                          struct net_device *dev, struct vxlan_config *conf,
3881                          bool changelink, struct netlink_ext_ack *extack)
3882 {
3883         struct vxlan_dev *vxlan = netdev_priv(dev);
3884         int err = 0;
3885
3886         memset(conf, 0, sizeof(*conf));
3887
3888         /* if changelink operation, start with old existing cfg */
3889         if (changelink)
3890                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3891
3892         if (data[IFLA_VXLAN_ID]) {
3893                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3894
3895                 if (changelink && (vni != conf->vni)) {
3896                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI");
3897                         return -EOPNOTSUPP;
3898                 }
3899                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3900         }
3901
3902         if (data[IFLA_VXLAN_GROUP]) {
3903                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) {
3904                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group");
3905                         return -EOPNOTSUPP;
3906                 }
3907
3908                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3909                 conf->remote_ip.sa.sa_family = AF_INET;
3910         } else if (data[IFLA_VXLAN_GROUP6]) {
3911                 if (!IS_ENABLED(CONFIG_IPV6)) {
3912                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel");
3913                         return -EPFNOSUPPORT;
3914                 }
3915
3916                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) {
3917                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group");
3918                         return -EOPNOTSUPP;
3919                 }
3920
3921                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3922                 conf->remote_ip.sa.sa_family = AF_INET6;
3923         }
3924
3925         if (data[IFLA_VXLAN_LOCAL]) {
3926                 if (changelink && (conf->saddr.sa.sa_family != AF_INET)) {
3927                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old");
3928                         return -EOPNOTSUPP;
3929                 }
3930
3931                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3932                 conf->saddr.sa.sa_family = AF_INET;
3933         } else if (data[IFLA_VXLAN_LOCAL6]) {
3934                 if (!IS_ENABLED(CONFIG_IPV6)) {
3935                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel");
3936                         return -EPFNOSUPPORT;
3937                 }
3938
3939                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) {
3940                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old");
3941                         return -EOPNOTSUPP;
3942                 }
3943
3944                 /* TODO: respect scope id */
3945                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3946                 conf->saddr.sa.sa_family = AF_INET6;
3947         }
3948
3949         if (data[IFLA_VXLAN_LINK])
3950                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3951
3952         if (data[IFLA_VXLAN_TOS])
3953                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3954
3955         if (data[IFLA_VXLAN_TTL])
3956                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3957
3958         if (data[IFLA_VXLAN_TTL_INHERIT]) {
3959                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT,
3960                                     VXLAN_F_TTL_INHERIT, changelink, false,
3961                                     extack);
3962                 if (err)
3963                         return err;
3964
3965         }
3966
3967         if (data[IFLA_VXLAN_LABEL])
3968                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
3969                              IPV6_FLOWLABEL_MASK;
3970
3971         if (data[IFLA_VXLAN_LEARNING]) {
3972                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING,
3973                                     VXLAN_F_LEARN, changelink, true,
3974                                     extack);
3975                 if (err)
3976                         return err;
3977         } else if (!changelink) {
3978                 /* default to learn on a new device */
3979                 conf->flags |= VXLAN_F_LEARN;
3980         }
3981
3982         if (data[IFLA_VXLAN_AGEING])
3983                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
3984
3985         if (data[IFLA_VXLAN_PROXY]) {
3986                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY,
3987                                     VXLAN_F_PROXY, changelink, false,
3988                                     extack);
3989                 if (err)
3990                         return err;
3991         }
3992
3993         if (data[IFLA_VXLAN_RSC]) {
3994                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC,
3995                                     VXLAN_F_RSC, changelink, false,
3996                                     extack);
3997                 if (err)
3998                         return err;
3999         }
4000
4001         if (data[IFLA_VXLAN_L2MISS]) {
4002                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS,
4003                                     VXLAN_F_L2MISS, changelink, false,
4004                                     extack);
4005                 if (err)
4006                         return err;
4007         }
4008
4009         if (data[IFLA_VXLAN_L3MISS]) {
4010                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS,
4011                                     VXLAN_F_L3MISS, changelink, false,
4012                                     extack);
4013                 if (err)
4014                         return err;
4015         }
4016
4017         if (data[IFLA_VXLAN_LIMIT]) {
4018                 if (changelink) {
4019                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT],
4020                                             "Cannot change limit");
4021                         return -EOPNOTSUPP;
4022                 }
4023                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
4024         }
4025
4026         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
4027                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA,
4028                                     VXLAN_F_COLLECT_METADATA, changelink, false,
4029                                     extack);
4030                 if (err)
4031                         return err;
4032         }
4033
4034         if (data[IFLA_VXLAN_PORT_RANGE]) {
4035                 if (!changelink) {
4036                         const struct ifla_vxlan_port_range *p
4037                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
4038                         conf->port_min = ntohs(p->low);
4039                         conf->port_max = ntohs(p->high);
4040                 } else {
4041                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
4042                                             "Cannot change port range");
4043                         return -EOPNOTSUPP;
4044                 }
4045         }
4046
4047         if (data[IFLA_VXLAN_PORT]) {
4048                 if (changelink) {
4049                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT],
4050                                             "Cannot change port");
4051                         return -EOPNOTSUPP;
4052                 }
4053                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
4054         }
4055
4056         if (data[IFLA_VXLAN_UDP_CSUM]) {
4057                 if (changelink) {
4058                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM],
4059                                             "Cannot change UDP_CSUM flag");
4060                         return -EOPNOTSUPP;
4061                 }
4062                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
4063                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
4064         }
4065
4066         if (data[IFLA_VXLAN_LOCALBYPASS]) {
4067                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LOCALBYPASS,
4068                                     VXLAN_F_LOCALBYPASS, changelink,
4069                                     true, extack);
4070                 if (err)
4071                         return err;
4072         } else if (!changelink) {
4073                 /* default to local bypass on a new device */
4074                 conf->flags |= VXLAN_F_LOCALBYPASS;
4075         }
4076
4077         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
4078                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4079                                     VXLAN_F_UDP_ZERO_CSUM6_TX, changelink,
4080                                     false, extack);
4081                 if (err)
4082                         return err;
4083         }
4084
4085         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
4086                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4087                                     VXLAN_F_UDP_ZERO_CSUM6_RX, changelink,
4088                                     false, extack);
4089                 if (err)
4090                         return err;
4091         }
4092
4093         if (data[IFLA_VXLAN_REMCSUM_TX]) {
4094                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX,
4095                                     VXLAN_F_REMCSUM_TX, changelink, false,
4096                                     extack);
4097                 if (err)
4098                         return err;
4099         }
4100
4101         if (data[IFLA_VXLAN_REMCSUM_RX]) {
4102                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX,
4103                                     VXLAN_F_REMCSUM_RX, changelink, false,
4104                                     extack);
4105                 if (err)
4106                         return err;
4107         }
4108
4109         if (data[IFLA_VXLAN_GBP]) {
4110                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP,
4111                                     VXLAN_F_GBP, changelink, false, extack);
4112                 if (err)
4113                         return err;
4114         }
4115
4116         if (data[IFLA_VXLAN_GPE]) {
4117                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE,
4118                                     VXLAN_F_GPE, changelink, false,
4119                                     extack);
4120                 if (err)
4121                         return err;
4122         }
4123
4124         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
4125                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL,
4126                                     VXLAN_F_REMCSUM_NOPARTIAL, changelink,
4127                                     false, extack);
4128                 if (err)
4129                         return err;
4130         }
4131
4132         if (tb[IFLA_MTU]) {
4133                 if (changelink) {
4134                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
4135                                             "Cannot change mtu");
4136                         return -EOPNOTSUPP;
4137                 }
4138                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
4139         }
4140
4141         if (data[IFLA_VXLAN_DF])
4142                 conf->df = nla_get_u8(data[IFLA_VXLAN_DF]);
4143
4144         if (data[IFLA_VXLAN_VNIFILTER]) {
4145                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_VNIFILTER,
4146                                     VXLAN_F_VNIFILTER, changelink, false,
4147                                     extack);
4148                 if (err)
4149                         return err;
4150
4151                 if ((conf->flags & VXLAN_F_VNIFILTER) &&
4152                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
4153                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_VNIFILTER],
4154                                             "vxlan vnifilter only valid in collect metadata mode");
4155                         return -EINVAL;
4156                 }
4157         }
4158
4159         return 0;
4160 }
4161
4162 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
4163                          struct nlattr *tb[], struct nlattr *data[],
4164                          struct netlink_ext_ack *extack)
4165 {
4166         struct vxlan_config conf;
4167         int err;
4168
4169         err = vxlan_nl2conf(tb, data, dev, &conf, false, extack);
4170         if (err)
4171                 return err;
4172
4173         return __vxlan_dev_create(src_net, dev, &conf, extack);
4174 }
4175
4176 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
4177                             struct nlattr *data[],
4178                             struct netlink_ext_ack *extack)
4179 {
4180         struct vxlan_dev *vxlan = netdev_priv(dev);
4181         struct net_device *lowerdev;
4182         struct vxlan_config conf;
4183         struct vxlan_rdst *dst;
4184         int err;
4185
4186         dst = &vxlan->default_dst;
4187         err = vxlan_nl2conf(tb, data, dev, &conf, true, extack);
4188         if (err)
4189                 return err;
4190
4191         err = vxlan_config_validate(vxlan->net, &conf, &lowerdev,
4192                                     vxlan, extack);
4193         if (err)
4194                 return err;
4195
4196         if (dst->remote_dev == lowerdev)
4197                 lowerdev = NULL;
4198
4199         err = netdev_adjacent_change_prepare(dst->remote_dev, lowerdev, dev,
4200                                              extack);
4201         if (err)
4202                 return err;
4203
4204         /* handle default dst entry */
4205         if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) {
4206                 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, conf.vni);
4207
4208                 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4209                 if (!vxlan_addr_any(&conf.remote_ip)) {
4210                         err = vxlan_fdb_update(vxlan, all_zeros_mac,
4211                                                &conf.remote_ip,
4212                                                NUD_REACHABLE | NUD_PERMANENT,
4213                                                NLM_F_APPEND | NLM_F_CREATE,
4214                                                vxlan->cfg.dst_port,
4215                                                conf.vni, conf.vni,
4216                                                conf.remote_ifindex,
4217                                                NTF_SELF, 0, true, extack);
4218                         if (err) {
4219                                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4220                                 netdev_adjacent_change_abort(dst->remote_dev,
4221                                                              lowerdev, dev);
4222                                 return err;
4223                         }
4224                 }
4225                 if (!vxlan_addr_any(&dst->remote_ip))
4226                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
4227                                            dst->remote_ip,
4228                                            vxlan->cfg.dst_port,
4229                                            dst->remote_vni,
4230                                            dst->remote_vni,
4231                                            dst->remote_ifindex,
4232                                            true);
4233                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4234
4235                 /* If vni filtering device, also update fdb entries of
4236                  * all vnis that were using default remote ip
4237                  */
4238                 if (vxlan->cfg.flags & VXLAN_F_VNIFILTER) {
4239                         err = vxlan_vnilist_update_group(vxlan, &dst->remote_ip,
4240                                                          &conf.remote_ip, extack);
4241                         if (err) {
4242                                 netdev_adjacent_change_abort(dst->remote_dev,
4243                                                              lowerdev, dev);
4244                                 return err;
4245                         }
4246                 }
4247         }
4248
4249         if (conf.age_interval != vxlan->cfg.age_interval)
4250                 mod_timer(&vxlan->age_timer, jiffies);
4251
4252         netdev_adjacent_change_commit(dst->remote_dev, lowerdev, dev);
4253         if (lowerdev && lowerdev != dst->remote_dev)
4254                 dst->remote_dev = lowerdev;
4255         vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true);
4256         return 0;
4257 }
4258
4259 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
4260 {
4261         struct vxlan_dev *vxlan = netdev_priv(dev);
4262
4263         vxlan_flush(vxlan, true);
4264
4265         list_del(&vxlan->next);
4266         unregister_netdevice_queue(dev, head);
4267         if (vxlan->default_dst.remote_dev)
4268                 netdev_upper_dev_unlink(vxlan->default_dst.remote_dev, dev);
4269 }
4270
4271 static size_t vxlan_get_size(const struct net_device *dev)
4272 {
4273
4274         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
4275                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
4276                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
4277                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
4278                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
4279                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL_INHERIT */
4280                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
4281                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_DF */
4282                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
4283                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
4284                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
4285                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
4286                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
4287                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
4288                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
4289                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
4290                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
4291                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
4292                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
4293                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
4294                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
4295                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
4296                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
4297                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
4298                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_LOCALBYPASS */
4299                 0;
4300 }
4301
4302 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
4303 {
4304         const struct vxlan_dev *vxlan = netdev_priv(dev);
4305         const struct vxlan_rdst *dst = &vxlan->default_dst;
4306         struct ifla_vxlan_port_range ports = {
4307                 .low =  htons(vxlan->cfg.port_min),
4308                 .high = htons(vxlan->cfg.port_max),
4309         };
4310
4311         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
4312                 goto nla_put_failure;
4313
4314         if (!vxlan_addr_any(&dst->remote_ip)) {
4315                 if (dst->remote_ip.sa.sa_family == AF_INET) {
4316                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
4317                                             dst->remote_ip.sin.sin_addr.s_addr))
4318                                 goto nla_put_failure;
4319 #if IS_ENABLED(CONFIG_IPV6)
4320                 } else {
4321                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
4322                                              &dst->remote_ip.sin6.sin6_addr))
4323                                 goto nla_put_failure;
4324 #endif
4325                 }
4326         }
4327
4328         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
4329                 goto nla_put_failure;
4330
4331         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
4332                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
4333                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
4334                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
4335                                 goto nla_put_failure;
4336 #if IS_ENABLED(CONFIG_IPV6)
4337                 } else {
4338                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
4339                                              &vxlan->cfg.saddr.sin6.sin6_addr))
4340                                 goto nla_put_failure;
4341 #endif
4342                 }
4343         }
4344
4345         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
4346             nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT,
4347                        !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) ||
4348             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
4349             nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) ||
4350             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
4351             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
4352                        !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
4353             nla_put_u8(skb, IFLA_VXLAN_PROXY,
4354                        !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
4355             nla_put_u8(skb, IFLA_VXLAN_RSC,
4356                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
4357             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
4358                        !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
4359             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
4360                        !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
4361             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
4362                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
4363             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
4364             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
4365             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
4366             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
4367                        !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
4368             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4369                        !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
4370             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4371                        !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
4372             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
4373                        !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
4374             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
4375                        !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)) ||
4376             nla_put_u8(skb, IFLA_VXLAN_LOCALBYPASS,
4377                        !!(vxlan->cfg.flags & VXLAN_F_LOCALBYPASS)))
4378                 goto nla_put_failure;
4379
4380         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
4381                 goto nla_put_failure;
4382
4383         if (vxlan->cfg.flags & VXLAN_F_GBP &&
4384             nla_put_flag(skb, IFLA_VXLAN_GBP))
4385                 goto nla_put_failure;
4386
4387         if (vxlan->cfg.flags & VXLAN_F_GPE &&
4388             nla_put_flag(skb, IFLA_VXLAN_GPE))
4389                 goto nla_put_failure;
4390
4391         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
4392             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
4393                 goto nla_put_failure;
4394
4395         if (vxlan->cfg.flags & VXLAN_F_VNIFILTER &&
4396             nla_put_u8(skb, IFLA_VXLAN_VNIFILTER,
4397                        !!(vxlan->cfg.flags & VXLAN_F_VNIFILTER)))
4398                 goto nla_put_failure;
4399
4400         return 0;
4401
4402 nla_put_failure:
4403         return -EMSGSIZE;
4404 }
4405
4406 static struct net *vxlan_get_link_net(const struct net_device *dev)
4407 {
4408         struct vxlan_dev *vxlan = netdev_priv(dev);
4409
4410         return vxlan->net;
4411 }
4412
4413 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
4414         .kind           = "vxlan",
4415         .maxtype        = IFLA_VXLAN_MAX,
4416         .policy         = vxlan_policy,
4417         .priv_size      = sizeof(struct vxlan_dev),
4418         .setup          = vxlan_setup,
4419         .validate       = vxlan_validate,
4420         .newlink        = vxlan_newlink,
4421         .changelink     = vxlan_changelink,
4422         .dellink        = vxlan_dellink,
4423         .get_size       = vxlan_get_size,
4424         .fill_info      = vxlan_fill_info,
4425         .get_link_net   = vxlan_get_link_net,
4426 };
4427
4428 struct net_device *vxlan_dev_create(struct net *net, const char *name,
4429                                     u8 name_assign_type,
4430                                     struct vxlan_config *conf)
4431 {
4432         struct nlattr *tb[IFLA_MAX + 1];
4433         struct net_device *dev;
4434         int err;
4435
4436         memset(&tb, 0, sizeof(tb));
4437
4438         dev = rtnl_create_link(net, name, name_assign_type,
4439                                &vxlan_link_ops, tb, NULL);
4440         if (IS_ERR(dev))
4441                 return dev;
4442
4443         err = __vxlan_dev_create(net, dev, conf, NULL);
4444         if (err < 0) {
4445                 free_netdev(dev);
4446                 return ERR_PTR(err);
4447         }
4448
4449         err = rtnl_configure_link(dev, NULL, 0, NULL);
4450         if (err < 0) {
4451                 LIST_HEAD(list_kill);
4452
4453                 vxlan_dellink(dev, &list_kill);
4454                 unregister_netdevice_many(&list_kill);
4455                 return ERR_PTR(err);
4456         }
4457
4458         return dev;
4459 }
4460 EXPORT_SYMBOL_GPL(vxlan_dev_create);
4461
4462 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
4463                                              struct net_device *dev)
4464 {
4465         struct vxlan_dev *vxlan, *next;
4466         LIST_HEAD(list_kill);
4467
4468         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4469                 struct vxlan_rdst *dst = &vxlan->default_dst;
4470
4471                 /* In case we created vxlan device with carrier
4472                  * and we loose the carrier due to module unload
4473                  * we also need to remove vxlan device. In other
4474                  * cases, it's not necessary and remote_ifindex
4475                  * is 0 here, so no matches.
4476                  */
4477                 if (dst->remote_ifindex == dev->ifindex)
4478                         vxlan_dellink(vxlan->dev, &list_kill);
4479         }
4480
4481         unregister_netdevice_many(&list_kill);
4482 }
4483
4484 static int vxlan_netdevice_event(struct notifier_block *unused,
4485                                  unsigned long event, void *ptr)
4486 {
4487         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4488         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
4489
4490         if (event == NETDEV_UNREGISTER)
4491                 vxlan_handle_lowerdev_unregister(vn, dev);
4492         else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
4493                 vxlan_offload_rx_ports(dev, true);
4494         else if (event == NETDEV_UDP_TUNNEL_DROP_INFO)
4495                 vxlan_offload_rx_ports(dev, false);
4496
4497         return NOTIFY_DONE;
4498 }
4499
4500 static struct notifier_block vxlan_notifier_block __read_mostly = {
4501         .notifier_call = vxlan_netdevice_event,
4502 };
4503
4504 static void
4505 vxlan_fdb_offloaded_set(struct net_device *dev,
4506                         struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4507 {
4508         struct vxlan_dev *vxlan = netdev_priv(dev);
4509         struct vxlan_rdst *rdst;
4510         struct vxlan_fdb *f;
4511         u32 hash_index;
4512
4513         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4514
4515         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4516
4517         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4518         if (!f)
4519                 goto out;
4520
4521         rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip,
4522                                    fdb_info->remote_port,
4523                                    fdb_info->remote_vni,
4524                                    fdb_info->remote_ifindex);
4525         if (!rdst)
4526                 goto out;
4527
4528         rdst->offloaded = fdb_info->offloaded;
4529
4530 out:
4531         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4532 }
4533
4534 static int
4535 vxlan_fdb_external_learn_add(struct net_device *dev,
4536                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4537 {
4538         struct vxlan_dev *vxlan = netdev_priv(dev);
4539         struct netlink_ext_ack *extack;
4540         u32 hash_index;
4541         int err;
4542
4543         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4544         extack = switchdev_notifier_info_to_extack(&fdb_info->info);
4545
4546         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4547         err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip,
4548                                NUD_REACHABLE,
4549                                NLM_F_CREATE | NLM_F_REPLACE,
4550                                fdb_info->remote_port,
4551                                fdb_info->vni,
4552                                fdb_info->remote_vni,
4553                                fdb_info->remote_ifindex,
4554                                NTF_USE | NTF_SELF | NTF_EXT_LEARNED,
4555                                0, false, extack);
4556         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4557
4558         return err;
4559 }
4560
4561 static int
4562 vxlan_fdb_external_learn_del(struct net_device *dev,
4563                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4564 {
4565         struct vxlan_dev *vxlan = netdev_priv(dev);
4566         struct vxlan_fdb *f;
4567         u32 hash_index;
4568         int err = 0;
4569
4570         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4571         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4572
4573         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4574         if (!f)
4575                 err = -ENOENT;
4576         else if (f->flags & NTF_EXT_LEARNED)
4577                 err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr,
4578                                          fdb_info->remote_ip,
4579                                          fdb_info->remote_port,
4580                                          fdb_info->vni,
4581                                          fdb_info->remote_vni,
4582                                          fdb_info->remote_ifindex,
4583                                          false);
4584
4585         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4586
4587         return err;
4588 }
4589
4590 static int vxlan_switchdev_event(struct notifier_block *unused,
4591                                  unsigned long event, void *ptr)
4592 {
4593         struct net_device *dev = switchdev_notifier_info_to_dev(ptr);
4594         struct switchdev_notifier_vxlan_fdb_info *fdb_info;
4595         int err = 0;
4596
4597         switch (event) {
4598         case SWITCHDEV_VXLAN_FDB_OFFLOADED:
4599                 vxlan_fdb_offloaded_set(dev, ptr);
4600                 break;
4601         case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE:
4602                 fdb_info = ptr;
4603                 err = vxlan_fdb_external_learn_add(dev, fdb_info);
4604                 if (err) {
4605                         err = notifier_from_errno(err);
4606                         break;
4607                 }
4608                 fdb_info->offloaded = true;
4609                 vxlan_fdb_offloaded_set(dev, fdb_info);
4610                 break;
4611         case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE:
4612                 fdb_info = ptr;
4613                 err = vxlan_fdb_external_learn_del(dev, fdb_info);
4614                 if (err) {
4615                         err = notifier_from_errno(err);
4616                         break;
4617                 }
4618                 fdb_info->offloaded = false;
4619                 vxlan_fdb_offloaded_set(dev, fdb_info);
4620                 break;
4621         }
4622
4623         return err;
4624 }
4625
4626 static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = {
4627         .notifier_call = vxlan_switchdev_event,
4628 };
4629
4630 static void vxlan_fdb_nh_flush(struct nexthop *nh)
4631 {
4632         struct vxlan_fdb *fdb;
4633         struct vxlan_dev *vxlan;
4634         u32 hash_index;
4635
4636         rcu_read_lock();
4637         list_for_each_entry_rcu(fdb, &nh->fdb_list, nh_list) {
4638                 vxlan = rcu_dereference(fdb->vdev);
4639                 WARN_ON(!vxlan);
4640                 hash_index = fdb_head_index(vxlan, fdb->eth_addr,
4641                                             vxlan->default_dst.remote_vni);
4642                 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4643                 if (!hlist_unhashed(&fdb->hlist))
4644                         vxlan_fdb_destroy(vxlan, fdb, false, false);
4645                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4646         }
4647         rcu_read_unlock();
4648 }
4649
4650 static int vxlan_nexthop_event(struct notifier_block *nb,
4651                                unsigned long event, void *ptr)
4652 {
4653         struct nh_notifier_info *info = ptr;
4654         struct nexthop *nh;
4655
4656         if (event != NEXTHOP_EVENT_DEL)
4657                 return NOTIFY_DONE;
4658
4659         nh = nexthop_find_by_id(info->net, info->id);
4660         if (!nh)
4661                 return NOTIFY_DONE;
4662
4663         vxlan_fdb_nh_flush(nh);
4664
4665         return NOTIFY_DONE;
4666 }
4667
4668 static __net_init int vxlan_init_net(struct net *net)
4669 {
4670         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4671         unsigned int h;
4672
4673         INIT_LIST_HEAD(&vn->vxlan_list);
4674         spin_lock_init(&vn->sock_lock);
4675         vn->nexthop_notifier_block.notifier_call = vxlan_nexthop_event;
4676
4677         for (h = 0; h < PORT_HASH_SIZE; ++h)
4678                 INIT_HLIST_HEAD(&vn->sock_list[h]);
4679
4680         return register_nexthop_notifier(net, &vn->nexthop_notifier_block,
4681                                          NULL);
4682 }
4683
4684 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
4685 {
4686         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4687         struct vxlan_dev *vxlan, *next;
4688         struct net_device *dev, *aux;
4689
4690         for_each_netdev_safe(net, dev, aux)
4691                 if (dev->rtnl_link_ops == &vxlan_link_ops)
4692                         unregister_netdevice_queue(dev, head);
4693
4694         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4695                 /* If vxlan->dev is in the same netns, it has already been added
4696                  * to the list by the previous loop.
4697                  */
4698                 if (!net_eq(dev_net(vxlan->dev), net))
4699                         unregister_netdevice_queue(vxlan->dev, head);
4700         }
4701
4702 }
4703
4704 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
4705 {
4706         struct net *net;
4707         LIST_HEAD(list);
4708         unsigned int h;
4709
4710         list_for_each_entry(net, net_list, exit_list) {
4711                 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4712
4713                 unregister_nexthop_notifier(net, &vn->nexthop_notifier_block);
4714         }
4715         rtnl_lock();
4716         list_for_each_entry(net, net_list, exit_list)
4717                 vxlan_destroy_tunnels(net, &list);
4718
4719         unregister_netdevice_many(&list);
4720         rtnl_unlock();
4721
4722         list_for_each_entry(net, net_list, exit_list) {
4723                 struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4724
4725                 for (h = 0; h < PORT_HASH_SIZE; ++h)
4726                         WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
4727         }
4728 }
4729
4730 static struct pernet_operations vxlan_net_ops = {
4731         .init = vxlan_init_net,
4732         .exit_batch = vxlan_exit_batch_net,
4733         .id   = &vxlan_net_id,
4734         .size = sizeof(struct vxlan_net),
4735 };
4736
4737 static int __init vxlan_init_module(void)
4738 {
4739         int rc;
4740
4741         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
4742
4743         rc = register_pernet_subsys(&vxlan_net_ops);
4744         if (rc)
4745                 goto out1;
4746
4747         rc = register_netdevice_notifier(&vxlan_notifier_block);
4748         if (rc)
4749                 goto out2;
4750
4751         rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block);
4752         if (rc)
4753                 goto out3;
4754
4755         rc = rtnl_link_register(&vxlan_link_ops);
4756         if (rc)
4757                 goto out4;
4758
4759         vxlan_vnifilter_init();
4760
4761         return 0;
4762 out4:
4763         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4764 out3:
4765         unregister_netdevice_notifier(&vxlan_notifier_block);
4766 out2:
4767         unregister_pernet_subsys(&vxlan_net_ops);
4768 out1:
4769         return rc;
4770 }
4771 late_initcall(vxlan_init_module);
4772
4773 static void __exit vxlan_cleanup_module(void)
4774 {
4775         vxlan_vnifilter_uninit();
4776         rtnl_link_unregister(&vxlan_link_ops);
4777         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4778         unregister_netdevice_notifier(&vxlan_notifier_block);
4779         unregister_pernet_subsys(&vxlan_net_ops);
4780         /* rcu_barrier() is called by netns */
4781 }
4782 module_exit(vxlan_cleanup_module);
4783
4784 MODULE_LICENSE("GPL");
4785 MODULE_VERSION(VXLAN_VERSION);
4786 MODULE_AUTHOR("Stephen Hemminger <[email protected]>");
4787 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
4788 MODULE_ALIAS_RTNL_LINK("vxlan");