drivers/net/vxlan.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * VXLAN: Virtual eXtensible Local Area Network
   4  *
   5  * Copyright (c) 2012-2013 Vyatta Inc.
   6  */
   7
   8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   9
  10 #include <linux/kernel.h>
  11 #include <linux/module.h>
  12 #include <linux/errno.h>
  13 #include <linux/slab.h>
  14 #include <linux/udp.h>
  15 #include <linux/igmp.h>
  16 #include <linux/if_ether.h>
  17 #include <linux/ethtool.h>
  18 #include <net/arp.h>
  19 #include <net/ndisc.h>
  20 #include <net/ipv6_stubs.h>
  21 #include <net/ip.h>
  22 #include <net/icmp.h>
  23 #include <net/rtnetlink.h>
  24 #include <net/inet_ecn.h>
  25 #include <net/net_namespace.h>
  26 #include <net/netns/generic.h>
  27 #include <net/tun_proto.h>
  28 #include <net/vxlan.h>
  29 #include <net/nexthop.h>
  30
  31 #if IS_ENABLED(CONFIG_IPV6)
  32 #include <net/ip6_tunnel.h>
  33 #include <net/ip6_checksum.h>
  34 #endif
  35
  36 #define VXLAN_VERSION   "0.1"
  37
  38 #define PORT_HASH_BITS  8
  39 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  40 #define FDB_AGE_DEFAULT 300 /* 5 min */
  41 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  42
  43 /* UDP port for VXLAN traffic.
  44  * The IANA assigned port is 4789, but the Linux default is 8472
  45  * for compatibility with early adopters.
  46  */
  47 static unsigned short vxlan_port __read_mostly = 8472;
  48 module_param_named(udp_port, vxlan_port, ushort, 0444);
  49 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  50
  51 static bool log_ecn_error = true;
  52 module_param(log_ecn_error, bool, 0644);
  53 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  54
  55 static unsigned int vxlan_net_id;
  56 static struct rtnl_link_ops vxlan_link_ops;
  57
  58 static const u8 all_zeros_mac[ETH_ALEN + 2];
  59
  60 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  61
  62 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan);
  63
  64 /* per-network namespace private data for this module */
  65 struct vxlan_net {
  66         struct list_head  vxlan_list;
  67         struct hlist_head sock_list[PORT_HASH_SIZE];
  68         spinlock_t        sock_lock;
  69 };
  70
  71 /* Forwarding table entry */
  72 struct vxlan_fdb {
  73         struct hlist_node hlist;        /* linked list of entries */
  74         struct rcu_head   rcu;
  75         unsigned long     updated;      /* jiffies */
  76         unsigned long     used;
  77         struct list_head  remotes;
  78         u8                eth_addr[ETH_ALEN];
  79         u16               state;        /* see ndm_state */
  80         __be32            vni;
  81         u16               flags;        /* see ndm_flags and below */
  82         struct list_head  nh_list;
  83         struct nexthop __rcu *nh;
  84         struct vxlan_dev  __rcu *vdev;
  85 };
  86
  87 #define NTF_VXLAN_ADDED_BY_USER 0x100
  88
  89 /* salt for hash table */
  90 static u32 vxlan_salt __read_mostly;
  91
  92 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  93 {
  94         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  95                ip_tunnel_collect_metadata();
  96 }
  97
  98 #if IS_ENABLED(CONFIG_IPV6)
  99 static inline
 100 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 101 {
 102         if (a->sa.sa_family != b->sa.sa_family)
 103                 return false;
 104         if (a->sa.sa_family == AF_INET6)
 105                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
 106         else
 107                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 108 }
 109
 110 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 111 {
 112         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 113                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 114                 ip->sa.sa_family = AF_INET6;
 115                 return 0;
 116         } else if (nla_len(nla) >= sizeof(__be32)) {
 117                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 118                 ip->sa.sa_family = AF_INET;
 119                 return 0;
 120         } else {
 121                 return -EAFNOSUPPORT;
 122         }
 123 }
 124
 125 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 126                               const union vxlan_addr *ip)
 127 {
 128         if (ip->sa.sa_family == AF_INET6)
 129                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 130         else
 131                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 132 }
 133
 134 #else /* !CONFIG_IPV6 */
 135
 136 static inline
 137 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 138 {
 139         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 140 }
 141
 142 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 143 {
 144         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 145                 return -EAFNOSUPPORT;
 146         } else if (nla_len(nla) >= sizeof(__be32)) {
 147                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 148                 ip->sa.sa_family = AF_INET;
 149                 return 0;
 150         } else {
 151                 return -EAFNOSUPPORT;
 152         }
 153 }
 154
 155 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 156                               const union vxlan_addr *ip)
 157 {
 158         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 159 }
 160 #endif
 161
 162 /* Virtual Network hash table head */
 163 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 164 {
 165         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 166 }
 167
 168 /* Socket hash table head */
 169 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 170 {
 171         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 172
 173         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 174 }
 175
 176 /* First remote destination for a forwarding entry.
 177  * Guaranteed to be non-NULL because remotes are never deleted.
 178  */
 179 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 180 {
 181         if (rcu_access_pointer(fdb->nh))
 182                 return NULL;
 183         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 184 }
 185
 186 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 187 {
 188         if (rcu_access_pointer(fdb->nh))
 189                 return NULL;
 190         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 191 }
 192
 193 /* Find VXLAN socket based on network namespace, address family and UDP port
 194  * and enabled unshareable flags.
 195  */
 196 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 197                                           __be16 port, u32 flags, int ifindex)
 198 {
 199         struct vxlan_sock *vs;
 200
 201         flags &= VXLAN_F_RCV_FLAGS;
 202
 203         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 204                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 205                     vxlan_get_sk_family(vs) == family &&
 206                     vs->flags == flags &&
 207                     vs->sock->sk->sk_bound_dev_if == ifindex)
 208                         return vs;
 209         }
 210         return NULL;
 211 }
 212
 213 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, int ifindex,
 214                                            __be32 vni)
 215 {
 216         struct vxlan_dev_node *node;
 217
 218         /* For flow based devices, map all packets to VNI 0 */
 219         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 220                 vni = 0;
 221
 222         hlist_for_each_entry_rcu(node, vni_head(vs, vni), hlist) {
 223                 if (node->vxlan->default_dst.remote_vni != vni)
 224                         continue;
 225
 226                 if (IS_ENABLED(CONFIG_IPV6)) {
 227                         const struct vxlan_config *cfg = &node->vxlan->cfg;
 228
 229                         if ((cfg->flags & VXLAN_F_IPV6_LINKLOCAL) &&
 230                             cfg->remote_ifindex != ifindex)
 231                                 continue;
 232                 }
 233
 234                 return node->vxlan;
 235         }
 236
 237         return NULL;
 238 }
 239
 240 /* Look up VNI in a per net namespace table */
 241 static struct vxlan_dev *vxlan_find_vni(struct net *net, int ifindex,
 242                                         __be32 vni, sa_family_t family,
 243                                         __be16 port, u32 flags)
 244 {
 245         struct vxlan_sock *vs;
 246
 247         vs = vxlan_find_sock(net, family, port, flags, ifindex);
 248         if (!vs)
 249                 return NULL;
 250
 251         return vxlan_vs_find_vni(vs, ifindex, vni);
 252 }
 253
 254 /* Fill in neighbour message in skbuff. */
 255 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 256                           const struct vxlan_fdb *fdb,
 257                           u32 portid, u32 seq, int type, unsigned int flags,
 258                           const struct vxlan_rdst *rdst)
 259 {
 260         unsigned long now = jiffies;
 261         struct nda_cacheinfo ci;
 262         bool send_ip, send_eth;
 263         struct nlmsghdr *nlh;
 264         struct nexthop *nh;
 265         struct ndmsg *ndm;
 266         int nh_family;
 267         u32 nh_id;
 268
 269         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 270         if (nlh == NULL)
 271                 return -EMSGSIZE;
 272
 273         ndm = nlmsg_data(nlh);
 274         memset(ndm, 0, sizeof(*ndm));
 275
 276         send_eth = send_ip = true;
 277
 278         rcu_read_lock();
 279         nh = rcu_dereference(fdb->nh);
 280         if (nh) {
 281                 nh_family = nexthop_get_family(nh);
 282                 nh_id = nh->id;
 283         }
 284         rcu_read_unlock();
 285
 286         if (type == RTM_GETNEIGH) {
 287                 if (rdst) {
 288                         send_ip = !vxlan_addr_any(&rdst->remote_ip);
 289                         ndm->ndm_family = send_ip ? rdst->remote_ip.sa.sa_family : AF_INET;
 290                 } else if (nh) {
 291                         ndm->ndm_family = nh_family;
 292                 }
 293                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 294         } else
 295                 ndm->ndm_family = AF_BRIDGE;
 296         ndm->ndm_state = fdb->state;
 297         ndm->ndm_ifindex = vxlan->dev->ifindex;
 298         ndm->ndm_flags = fdb->flags;
 299         if (rdst && rdst->offloaded)
 300                 ndm->ndm_flags |= NTF_OFFLOADED;
 301         ndm->ndm_type = RTN_UNICAST;
 302
 303         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 304             nla_put_s32(skb, NDA_LINK_NETNSID,
 305                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 306                 goto nla_put_failure;
 307
 308         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 309                 goto nla_put_failure;
 310         if (nh) {
 311                 if (nla_put_u32(skb, NDA_NH_ID, nh_id))
 312                         goto nla_put_failure;
 313         } else if (rdst) {
 314                 if (send_ip && vxlan_nla_put_addr(skb, NDA_DST,
 315                                                   &rdst->remote_ip))
 316                         goto nla_put_failure;
 317
 318                 if (rdst->remote_port &&
 319                     rdst->remote_port != vxlan->cfg.dst_port &&
 320                     nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 321                         goto nla_put_failure;
 322                 if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 323                     nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 324                         goto nla_put_failure;
 325                 if (rdst->remote_ifindex &&
 326                     nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 327                         goto nla_put_failure;
 328         }
 329
 330         if ((vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) && fdb->vni &&
 331             nla_put_u32(skb, NDA_SRC_VNI,
 332                         be32_to_cpu(fdb->vni)))
 333                 goto nla_put_failure;
 334
 335         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 336         ci.ndm_confirmed = 0;
 337         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 338         ci.ndm_refcnt    = 0;
 339
 340         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 341                 goto nla_put_failure;
 342
 343         nlmsg_end(skb, nlh);
 344         return 0;
 345
 346 nla_put_failure:
 347         nlmsg_cancel(skb, nlh);
 348         return -EMSGSIZE;
 349 }
 350
 351 static inline size_t vxlan_nlmsg_size(void)
 352 {
 353         return NLMSG_ALIGN(sizeof(struct ndmsg))
 354                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 355                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 356                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 357                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 358                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 359                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 360                 + nla_total_size(sizeof(struct nda_cacheinfo));
 361 }
 362
 363 static void __vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 364                                struct vxlan_rdst *rd, int type)
 365 {
 366         struct net *net = dev_net(vxlan->dev);
 367         struct sk_buff *skb;
 368         int err = -ENOBUFS;
 369
 370         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 371         if (skb == NULL)
 372                 goto errout;
 373
 374         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 375         if (err < 0) {
 376                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 377                 WARN_ON(err == -EMSGSIZE);
 378                 kfree_skb(skb);
 379                 goto errout;
 380         }
 381
 382         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 383         return;
 384 errout:
 385         if (err < 0)
 386                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 387 }
 388
 389 static void vxlan_fdb_switchdev_notifier_info(const struct vxlan_dev *vxlan,
 390                             const struct vxlan_fdb *fdb,
 391                             const struct vxlan_rdst *rd,
 392                             struct netlink_ext_ack *extack,
 393                             struct switchdev_notifier_vxlan_fdb_info *fdb_info)
 394 {
 395         fdb_info->info.dev = vxlan->dev;
 396         fdb_info->info.extack = extack;
 397         fdb_info->remote_ip = rd->remote_ip;
 398         fdb_info->remote_port = rd->remote_port;
 399         fdb_info->remote_vni = rd->remote_vni;
 400         fdb_info->remote_ifindex = rd->remote_ifindex;
 401         memcpy(fdb_info->eth_addr, fdb->eth_addr, ETH_ALEN);
 402         fdb_info->vni = fdb->vni;
 403         fdb_info->offloaded = rd->offloaded;
 404         fdb_info->added_by_user = fdb->flags & NTF_VXLAN_ADDED_BY_USER;
 405 }
 406
 407 static int vxlan_fdb_switchdev_call_notifiers(struct vxlan_dev *vxlan,
 408                                               struct vxlan_fdb *fdb,
 409                                               struct vxlan_rdst *rd,
 410                                               bool adding,
 411                                               struct netlink_ext_ack *extack)
 412 {
 413         struct switchdev_notifier_vxlan_fdb_info info;
 414         enum switchdev_notifier_type notifier_type;
 415         int ret;
 416
 417         if (WARN_ON(!rd))
 418                 return 0;
 419
 420         notifier_type = adding ? SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE
 421                                : SWITCHDEV_VXLAN_FDB_DEL_TO_DEVICE;
 422         vxlan_fdb_switchdev_notifier_info(vxlan, fdb, rd, NULL, &info);
 423         ret = call_switchdev_notifiers(notifier_type, vxlan->dev,
 424                                        &info.info, extack);
 425         return notifier_to_errno(ret);
 426 }
 427
 428 static int vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 429                             struct vxlan_rdst *rd, int type, bool swdev_notify,
 430                             struct netlink_ext_ack *extack)
 431 {
 432         int err;
 433
 434         if (swdev_notify && rd) {
 435                 switch (type) {
 436                 case RTM_NEWNEIGH:
 437                         err = vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
 438                                                                  true, extack);
 439                         if (err)
 440                                 return err;
 441                         break;
 442                 case RTM_DELNEIGH:
 443                         vxlan_fdb_switchdev_call_notifiers(vxlan, fdb, rd,
 444                                                            false, extack);
 445                         break;
 446                 }
 447         }
 448
 449         __vxlan_fdb_notify(vxlan, fdb, rd, type);
 450         return 0;
 451 }
 452
 453 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 454 {
 455         struct vxlan_dev *vxlan = netdev_priv(dev);
 456         struct vxlan_fdb f = {
 457                 .state = NUD_STALE,
 458         };
 459         struct vxlan_rdst remote = {
 460                 .remote_ip = *ipa, /* goes to NDA_DST */
 461                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 462         };
 463
 464         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
 465 }
 466
 467 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 468 {
 469         struct vxlan_fdb f = {
 470                 .state = NUD_STALE,
 471         };
 472         struct vxlan_rdst remote = { };
 473
 474         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 475
 476         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH, true, NULL);
 477 }
 478
 479 /* Hash Ethernet address */
 480 static u32 eth_hash(const unsigned char *addr)
 481 {
 482         u64 value = get_unaligned((u64 *)addr);
 483
 484         /* only want 6 bytes */
 485 #ifdef __BIG_ENDIAN
 486         value >>= 16;
 487 #else
 488         value <<= 16;
 489 #endif
 490         return hash_64(value, FDB_HASH_BITS);
 491 }
 492
 493 static u32 eth_vni_hash(const unsigned char *addr, __be32 vni)
 494 {
 495         /* use 1 byte of OUI and 3 bytes of NIC */
 496         u32 key = get_unaligned((u32 *)(addr + 2));
 497
 498         return jhash_2words(key, vni, vxlan_salt) & (FDB_HASH_SIZE - 1);
 499 }
 500
 501 static u32 fdb_head_index(struct vxlan_dev *vxlan, const u8 *mac, __be32 vni)
 502 {
 503         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)
 504                 return eth_vni_hash(mac, vni);
 505         else
 506                 return eth_hash(mac);
 507 }
 508
 509 /* Hash chain to use given mac address */
 510 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 511                                                 const u8 *mac, __be32 vni)
 512 {
 513         return &vxlan->fdb_head[fdb_head_index(vxlan, mac, vni)];
 514 }
 515
 516 /* Look up Ethernet address in forwarding table */
 517 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 518                                           const u8 *mac, __be32 vni)
 519 {
 520         struct hlist_head *head = vxlan_fdb_head(vxlan, mac, vni);
 521         struct vxlan_fdb *f;
 522
 523         hlist_for_each_entry_rcu(f, head, hlist) {
 524                 if (ether_addr_equal(mac, f->eth_addr)) {
 525                         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
 526                                 if (vni == f->vni)
 527                                         return f;
 528                         } else {
 529                                 return f;
 530                         }
 531                 }
 532         }
 533
 534         return NULL;
 535 }
 536
 537 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 538                                         const u8 *mac, __be32 vni)
 539 {
 540         struct vxlan_fdb *f;
 541
 542         f = __vxlan_find_mac(vxlan, mac, vni);
 543         if (f && f->used != jiffies)
 544                 f->used = jiffies;
 545
 546         return f;
 547 }
 548
 549 /* caller should hold vxlan->hash_lock */
 550 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 551                                               union vxlan_addr *ip, __be16 port,
 552                                               __be32 vni, __u32 ifindex)
 553 {
 554         struct vxlan_rdst *rd;
 555
 556         list_for_each_entry(rd, &f->remotes, list) {
 557                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 558                     rd->remote_port == port &&
 559                     rd->remote_vni == vni &&
 560                     rd->remote_ifindex == ifindex)
 561                         return rd;
 562         }
 563
 564         return NULL;
 565 }
 566
 567 int vxlan_fdb_find_uc(struct net_device *dev, const u8 *mac, __be32 vni,
 568                       struct switchdev_notifier_vxlan_fdb_info *fdb_info)
 569 {
 570         struct vxlan_dev *vxlan = netdev_priv(dev);
 571         u8 eth_addr[ETH_ALEN + 2] = { 0 };
 572         struct vxlan_rdst *rdst;
 573         struct vxlan_fdb *f;
 574         int rc = 0;
 575
 576         if (is_multicast_ether_addr(mac) ||
 577             is_zero_ether_addr(mac))
 578                 return -EINVAL;
 579
 580         ether_addr_copy(eth_addr, mac);
 581
 582         rcu_read_lock();
 583
 584         f = __vxlan_find_mac(vxlan, eth_addr, vni);
 585         if (!f) {
 586                 rc = -ENOENT;
 587                 goto out;
 588         }
 589
 590         rdst = first_remote_rcu(f);
 591         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, NULL, fdb_info);
 592
 593 out:
 594         rcu_read_unlock();
 595         return rc;
 596 }
 597 EXPORT_SYMBOL_GPL(vxlan_fdb_find_uc);
 598
 599 static int vxlan_fdb_notify_one(struct notifier_block *nb,
 600                                 const struct vxlan_dev *vxlan,
 601                                 const struct vxlan_fdb *f,
 602                                 const struct vxlan_rdst *rdst,
 603                                 struct netlink_ext_ack *extack)
 604 {
 605         struct switchdev_notifier_vxlan_fdb_info fdb_info;
 606         int rc;
 607
 608         vxlan_fdb_switchdev_notifier_info(vxlan, f, rdst, extack, &fdb_info);
 609         rc = nb->notifier_call(nb, SWITCHDEV_VXLAN_FDB_ADD_TO_DEVICE,
 610                                &fdb_info);
 611         return notifier_to_errno(rc);
 612 }
 613
 614 int vxlan_fdb_replay(const struct net_device *dev, __be32 vni,
 615                      struct notifier_block *nb,
 616                      struct netlink_ext_ack *extack)
 617 {
 618         struct vxlan_dev *vxlan;
 619         struct vxlan_rdst *rdst;
 620         struct vxlan_fdb *f;
 621         unsigned int h;
 622         int rc = 0;
 623
 624         if (!netif_is_vxlan(dev))
 625                 return -EINVAL;
 626         vxlan = netdev_priv(dev);
 627
 628         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 629                 spin_lock_bh(&vxlan->hash_lock[h]);
 630                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist) {
 631                         if (f->vni == vni) {
 632                                 list_for_each_entry(rdst, &f->remotes, list) {
 633                                         rc = vxlan_fdb_notify_one(nb, vxlan,
 634                                                                   f, rdst,
 635                                                                   extack);
 636                                         if (rc)
 637                                                 goto unlock;
 638                                 }
 639                         }
 640                 }
 641                 spin_unlock_bh(&vxlan->hash_lock[h]);
 642         }
 643         return 0;
 644
 645 unlock:
 646         spin_unlock_bh(&vxlan->hash_lock[h]);
 647         return rc;
 648 }
 649 EXPORT_SYMBOL_GPL(vxlan_fdb_replay);
 650
 651 void vxlan_fdb_clear_offload(const struct net_device *dev, __be32 vni)
 652 {
 653         struct vxlan_dev *vxlan;
 654         struct vxlan_rdst *rdst;
 655         struct vxlan_fdb *f;
 656         unsigned int h;
 657
 658         if (!netif_is_vxlan(dev))
 659                 return;
 660         vxlan = netdev_priv(dev);
 661
 662         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 663                 spin_lock_bh(&vxlan->hash_lock[h]);
 664                 hlist_for_each_entry(f, &vxlan->fdb_head[h], hlist)
 665                         if (f->vni == vni)
 666                                 list_for_each_entry(rdst, &f->remotes, list)
 667                                         rdst->offloaded = false;
 668                 spin_unlock_bh(&vxlan->hash_lock[h]);
 669         }
 670
 671 }
 672 EXPORT_SYMBOL_GPL(vxlan_fdb_clear_offload);
 673
 674 /* Replace destination of unicast mac */
 675 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 676                              union vxlan_addr *ip, __be16 port, __be32 vni,
 677                              __u32 ifindex, struct vxlan_rdst *oldrd)
 678 {
 679         struct vxlan_rdst *rd;
 680
 681         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 682         if (rd)
 683                 return 0;
 684
 685         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 686         if (!rd)
 687                 return 0;
 688
 689         *oldrd = *rd;
 690         dst_cache_reset(&rd->dst_cache);
 691         rd->remote_ip = *ip;
 692         rd->remote_port = port;
 693         rd->remote_vni = vni;
 694         rd->remote_ifindex = ifindex;
 695         rd->offloaded = false;
 696         return 1;
 697 }
 698
 699 /* Add/update destinations for multicast */
 700 static int vxlan_fdb_append(struct vxlan_fdb *f,
 701                             union vxlan_addr *ip, __be16 port, __be32 vni,
 702                             __u32 ifindex, struct vxlan_rdst **rdp)
 703 {
 704         struct vxlan_rdst *rd;
 705
 706         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 707         if (rd)
 708                 return 0;
 709
 710         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 711         if (rd == NULL)
 712                 return -ENOBUFS;
 713
 714         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 715                 kfree(rd);
 716                 return -ENOBUFS;
 717         }
 718
 719         rd->remote_ip = *ip;
 720         rd->remote_port = port;
 721         rd->offloaded = false;
 722         rd->remote_vni = vni;
 723         rd->remote_ifindex = ifindex;
 724
 725         list_add_tail_rcu(&rd->list, &f->remotes);
 726
 727         *rdp = rd;
 728         return 1;
 729 }
 730
 731 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 732                                           unsigned int off,
 733                                           struct vxlanhdr *vh, size_t hdrlen,
 734                                           __be32 vni_field,
 735                                           struct gro_remcsum *grc,
 736                                           bool nopartial)
 737 {
 738         size_t start, offset;
 739
 740         if (skb->remcsum_offload)
 741                 return vh;
 742
 743         if (!NAPI_GRO_CB(skb)->csum_valid)
 744                 return NULL;
 745
 746         start = vxlan_rco_start(vni_field);
 747         offset = start + vxlan_rco_offset(vni_field);
 748
 749         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 750                                      start, offset, grc, nopartial);
 751
 752         skb->remcsum_offload = 1;
 753
 754         return vh;
 755 }
 756
 757 static struct sk_buff *vxlan_gro_receive(struct sock *sk,
 758                                          struct list_head *head,
 759                                          struct sk_buff *skb)
 760 {
 761         struct sk_buff *pp = NULL;
 762         struct sk_buff *p;
 763         struct vxlanhdr *vh, *vh2;
 764         unsigned int hlen, off_vx;
 765         int flush = 1;
 766         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 767         __be32 flags;
 768         struct gro_remcsum grc;
 769
 770         skb_gro_remcsum_init(&grc);
 771
 772         off_vx = skb_gro_offset(skb);
 773         hlen = off_vx + sizeof(*vh);
 774         vh   = skb_gro_header_fast(skb, off_vx);
 775         if (skb_gro_header_hard(skb, hlen)) {
 776                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 777                 if (unlikely(!vh))
 778                         goto out;
 779         }
 780
 781         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 782
 783         flags = vh->vx_flags;
 784
 785         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 786                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 787                                        vh->vx_vni, &grc,
 788                                        !!(vs->flags &
 789                                           VXLAN_F_REMCSUM_NOPARTIAL));
 790
 791                 if (!vh)
 792                         goto out;
 793         }
 794
 795         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 796
 797         list_for_each_entry(p, head, list) {
 798                 if (!NAPI_GRO_CB(p)->same_flow)
 799                         continue;
 800
 801                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 802                 if (vh->vx_flags != vh2->vx_flags ||
 803                     vh->vx_vni != vh2->vx_vni) {
 804                         NAPI_GRO_CB(p)->same_flow = 0;
 805                         continue;
 806                 }
 807         }
 808
 809         pp = call_gro_receive(eth_gro_receive, head, skb);
 810         flush = 0;
 811
 812 out:
 813         skb_gro_flush_final_remcsum(skb, pp, flush, &grc);
 814
 815         return pp;
 816 }
 817
 818 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 819 {
 820         /* Sets 'skb->inner_mac_header' since we are always called with
 821          * 'skb->encapsulation' set.
 822          */
 823         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 824 }
 825
 826 static struct vxlan_fdb *vxlan_fdb_alloc(struct vxlan_dev *vxlan, const u8 *mac,
 827                                          __u16 state, __be32 src_vni,
 828                                          __u16 ndm_flags)
 829 {
 830         struct vxlan_fdb *f;
 831
 832         f = kmalloc(sizeof(*f), GFP_ATOMIC);
 833         if (!f)
 834                 return NULL;
 835         f->state = state;
 836         f->flags = ndm_flags;
 837         f->updated = f->used = jiffies;
 838         f->vni = src_vni;
 839         f->nh = NULL;
 840         RCU_INIT_POINTER(f->vdev, vxlan);
 841         INIT_LIST_HEAD(&f->nh_list);
 842         INIT_LIST_HEAD(&f->remotes);
 843         memcpy(f->eth_addr, mac, ETH_ALEN);
 844
 845         return f;
 846 }
 847
 848 static void vxlan_fdb_insert(struct vxlan_dev *vxlan, const u8 *mac,
 849                              __be32 src_vni, struct vxlan_fdb *f)
 850 {
 851         ++vxlan->addrcnt;
 852         hlist_add_head_rcu(&f->hlist,
 853                            vxlan_fdb_head(vxlan, mac, src_vni));
 854 }
 855
 856 static int vxlan_fdb_nh_update(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 857                                u32 nhid, struct netlink_ext_ack *extack)
 858 {
 859         struct nexthop *old_nh = rtnl_dereference(fdb->nh);
 860         struct nexthop *nh;
 861         int err = -EINVAL;
 862
 863         if (old_nh && old_nh->id == nhid)
 864                 return 0;
 865
 866         nh = nexthop_find_by_id(vxlan->net, nhid);
 867         if (!nh) {
 868                 NL_SET_ERR_MSG(extack, "Nexthop id does not exist");
 869                 goto err_inval;
 870         }
 871
 872         if (nh) {
 873                 if (!nexthop_get(nh)) {
 874                         NL_SET_ERR_MSG(extack, "Nexthop has been deleted");
 875                         nh = NULL;
 876                         goto err_inval;
 877                 }
 878                 if (!nexthop_is_fdb(nh)) {
 879                         NL_SET_ERR_MSG(extack, "Nexthop is not a fdb nexthop");
 880                         goto err_inval;
 881                 }
 882
 883                 if (!nexthop_is_multipath(nh)) {
 884                         NL_SET_ERR_MSG(extack, "Nexthop is not a multipath group");
 885                         goto err_inval;
 886                 }
 887
 888                 /* check nexthop group family */
 889                 switch (vxlan->default_dst.remote_ip.sa.sa_family) {
 890                 case AF_INET:
 891                         if (!nexthop_has_v4(nh)) {
 892                                 err = -EAFNOSUPPORT;
 893                                 NL_SET_ERR_MSG(extack, "Nexthop group family not supported");
 894                                 goto err_inval;
 895                         }
 896                         break;
 897                 case AF_INET6:
 898                         if (nexthop_has_v4(nh)) {
 899                                 err = -EAFNOSUPPORT;
 900                                 NL_SET_ERR_MSG(extack, "Nexthop group family not supported");
 901                                 goto err_inval;
 902                         }
 903                 }
 904         }
 905
 906         if (old_nh) {
 907                 list_del_rcu(&fdb->nh_list);
 908                 nexthop_put(old_nh);
 909         }
 910         rcu_assign_pointer(fdb->nh, nh);
 911         list_add_tail_rcu(&fdb->nh_list, &nh->fdb_list);
 912         return 1;
 913
 914 err_inval:
 915         if (nh)
 916                 nexthop_put(nh);
 917         return err;
 918 }
 919
 920 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 921                             const u8 *mac, union vxlan_addr *ip,
 922                             __u16 state, __be16 port, __be32 src_vni,
 923                             __be32 vni, __u32 ifindex, __u16 ndm_flags,
 924                             u32 nhid, struct vxlan_fdb **fdb,
 925                             struct netlink_ext_ack *extack)
 926 {
 927         struct vxlan_rdst *rd = NULL;
 928         struct vxlan_fdb *f;
 929         int rc;
 930
 931         if (vxlan->cfg.addrmax &&
 932             vxlan->addrcnt >= vxlan->cfg.addrmax)
 933                 return -ENOSPC;
 934
 935         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 936         f = vxlan_fdb_alloc(vxlan, mac, state, src_vni, ndm_flags);
 937         if (!f)
 938                 return -ENOMEM;
 939
 940         if (nhid)
 941                 rc = vxlan_fdb_nh_update(vxlan, f, nhid, extack);
 942         else
 943                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 944         if (rc < 0)
 945                 goto errout;
 946
 947         *fdb = f;
 948
 949         return 0;
 950
 951 errout:
 952         kfree(f);
 953         return rc;
 954 }
 955
 956 static void __vxlan_fdb_free(struct vxlan_fdb *f)
 957 {
 958         struct vxlan_rdst *rd, *nd;
 959         struct nexthop *nh;
 960
 961         nh = rcu_dereference_raw(f->nh);
 962         if (nh) {
 963                 rcu_assign_pointer(f->nh, NULL);
 964                 rcu_assign_pointer(f->vdev, NULL);
 965                 nexthop_put(nh);
 966         }
 967
 968         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 969                 dst_cache_destroy(&rd->dst_cache);
 970                 kfree(rd);
 971         }
 972         kfree(f);
 973 }
 974
 975 static void vxlan_fdb_free(struct rcu_head *head)
 976 {
 977         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 978
 979         __vxlan_fdb_free(f);
 980 }
 981
 982 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
 983                               bool do_notify, bool swdev_notify)
 984 {
 985         struct vxlan_rdst *rd;
 986
 987         netdev_dbg(vxlan->dev, "delete %pM\n", f->eth_addr);
 988
 989         --vxlan->addrcnt;
 990         if (do_notify) {
 991                 if (rcu_access_pointer(f->nh))
 992                         vxlan_fdb_notify(vxlan, f, NULL, RTM_DELNEIGH,
 993                                          swdev_notify, NULL);
 994                 else
 995                         list_for_each_entry(rd, &f->remotes, list)
 996                                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH,
 997                                                  swdev_notify, NULL);
 998         }
 999
1000         hlist_del_rcu(&f->hlist);
1001         list_del_rcu(&f->nh_list);
1002         call_rcu(&f->rcu, vxlan_fdb_free);
1003 }
1004
1005 static void vxlan_dst_free(struct rcu_head *head)
1006 {
1007         struct vxlan_rdst *rd = container_of(head, struct vxlan_rdst, rcu);
1008
1009         dst_cache_destroy(&rd->dst_cache);
1010         kfree(rd);
1011 }
1012
1013 static int vxlan_fdb_update_existing(struct vxlan_dev *vxlan,
1014                                      union vxlan_addr *ip,
1015                                      __u16 state, __u16 flags,
1016                                      __be16 port, __be32 vni,
1017                                      __u32 ifindex, __u16 ndm_flags,
1018                                      struct vxlan_fdb *f, u32 nhid,
1019                                      bool swdev_notify,
1020                                      struct netlink_ext_ack *extack)
1021 {
1022         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
1023         struct vxlan_rdst *rd = NULL;
1024         struct vxlan_rdst oldrd;
1025         int notify = 0;
1026         int rc = 0;
1027         int err;
1028
1029         if (nhid && !rcu_access_pointer(f->nh)) {
1030                 NL_SET_ERR_MSG(extack,
1031                                "Cannot replace an existing non nexthop fdb with a nexthop");
1032                 return -EOPNOTSUPP;
1033         }
1034
1035         if (nhid && (flags & NLM_F_APPEND)) {
1036                 NL_SET_ERR_MSG(extack,
1037                                "Cannot append to a nexthop fdb");
1038                 return -EOPNOTSUPP;
1039         }
1040
1041         /* Do not allow an externally learned entry to take over an entry added
1042          * by the user.
1043          */
1044         if (!(fdb_flags & NTF_EXT_LEARNED) ||
1045             !(f->flags & NTF_VXLAN_ADDED_BY_USER)) {
1046                 if (f->state != state) {
1047                         f->state = state;
1048                         f->updated = jiffies;
1049                         notify = 1;
1050                 }
1051                 if (f->flags != fdb_flags) {
1052                         f->flags = fdb_flags;
1053                         f->updated = jiffies;
1054                         notify = 1;
1055                 }
1056         }
1057
1058         if ((flags & NLM_F_REPLACE)) {
1059                 /* Only change unicasts */
1060                 if (!(is_multicast_ether_addr(f->eth_addr) ||
1061                       is_zero_ether_addr(f->eth_addr))) {
1062                         if (nhid) {
1063                                 rc = vxlan_fdb_nh_update(vxlan, f, nhid, extack);
1064                                 if (rc < 0)
1065                                         return rc;
1066                         } else {
1067                                 rc = vxlan_fdb_replace(f, ip, port, vni,
1068                                                        ifindex, &oldrd);
1069                         }
1070                         notify |= rc;
1071                 } else {
1072                         NL_SET_ERR_MSG(extack, "Cannot replace non-unicast fdb entries");
1073                         return -EOPNOTSUPP;
1074                 }
1075         }
1076         if ((flags & NLM_F_APPEND) &&
1077             (is_multicast_ether_addr(f->eth_addr) ||
1078              is_zero_ether_addr(f->eth_addr))) {
1079                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
1080
1081                 if (rc < 0)
1082                         return rc;
1083                 notify |= rc;
1084         }
1085
1086         if (ndm_flags & NTF_USE)
1087                 f->used = jiffies;
1088
1089         if (notify) {
1090                 if (rd == NULL)
1091                         rd = first_remote_rtnl(f);
1092
1093                 err = vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH,
1094                                        swdev_notify, extack);
1095                 if (err)
1096                         goto err_notify;
1097         }
1098
1099         return 0;
1100
1101 err_notify:
1102         if (nhid)
1103                 return err;
1104         if ((flags & NLM_F_REPLACE) && rc)
1105                 *rd = oldrd;
1106         else if ((flags & NLM_F_APPEND) && rc) {
1107                 list_del_rcu(&rd->list);
1108                 call_rcu(&rd->rcu, vxlan_dst_free);
1109         }
1110         return err;
1111 }
1112
1113 static int vxlan_fdb_update_create(struct vxlan_dev *vxlan,
1114                                    const u8 *mac, union vxlan_addr *ip,
1115                                    __u16 state, __u16 flags,
1116                                    __be16 port, __be32 src_vni, __be32 vni,
1117                                    __u32 ifindex, __u16 ndm_flags, u32 nhid,
1118                                    bool swdev_notify,
1119                                    struct netlink_ext_ack *extack)
1120 {
1121         __u16 fdb_flags = (ndm_flags & ~NTF_USE);
1122         struct vxlan_fdb *f;
1123         int rc;
1124
1125         /* Disallow replace to add a multicast entry */
1126         if ((flags & NLM_F_REPLACE) &&
1127             (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
1128                 return -EOPNOTSUPP;
1129
1130         netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
1131         rc = vxlan_fdb_create(vxlan, mac, ip, state, port, src_vni,
1132                               vni, ifindex, fdb_flags, nhid, &f, extack);
1133         if (rc < 0)
1134                 return rc;
1135
1136         vxlan_fdb_insert(vxlan, mac, src_vni, f);
1137         rc = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_NEWNEIGH,
1138                               swdev_notify, extack);
1139         if (rc)
1140                 goto err_notify;
1141
1142         return 0;
1143
1144 err_notify:
1145         vxlan_fdb_destroy(vxlan, f, false, false);
1146         return rc;
1147 }
1148
1149 /* Add new entry to forwarding table -- assumes lock held */
1150 static int vxlan_fdb_update(struct vxlan_dev *vxlan,
1151                             const u8 *mac, union vxlan_addr *ip,
1152                             __u16 state, __u16 flags,
1153                             __be16 port, __be32 src_vni, __be32 vni,
1154                             __u32 ifindex, __u16 ndm_flags, u32 nhid,
1155                             bool swdev_notify,
1156                             struct netlink_ext_ack *extack)
1157 {
1158         struct vxlan_fdb *f;
1159
1160         f = __vxlan_find_mac(vxlan, mac, src_vni);
1161         if (f) {
1162                 if (flags & NLM_F_EXCL) {
1163                         netdev_dbg(vxlan->dev,
1164                                    "lost race to create %pM\n", mac);
1165                         return -EEXIST;
1166                 }
1167
1168                 return vxlan_fdb_update_existing(vxlan, ip, state, flags, port,
1169                                                  vni, ifindex, ndm_flags, f,
1170                                                  nhid, swdev_notify, extack);
1171         } else {
1172                 if (!(flags & NLM_F_CREATE))
1173                         return -ENOENT;
1174
1175                 return vxlan_fdb_update_create(vxlan, mac, ip, state, flags,
1176                                                port, src_vni, vni, ifindex,
1177                                                ndm_flags, nhid, swdev_notify,
1178                                                extack);
1179         }
1180 }
1181
1182 static void vxlan_fdb_dst_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f,
1183                                   struct vxlan_rdst *rd, bool swdev_notify)
1184 {
1185         list_del_rcu(&rd->list);
1186         vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH, swdev_notify, NULL);
1187         call_rcu(&rd->rcu, vxlan_dst_free);
1188 }
1189
1190 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
1191                            union vxlan_addr *ip, __be16 *port, __be32 *src_vni,
1192                            __be32 *vni, u32 *ifindex, u32 *nhid)
1193 {
1194         struct net *net = dev_net(vxlan->dev);
1195         int err;
1196
1197         if (tb[NDA_NH_ID] && (tb[NDA_DST] || tb[NDA_VNI] || tb[NDA_IFINDEX] ||
1198             tb[NDA_PORT]))
1199                 return -EINVAL;
1200
1201         if (tb[NDA_DST]) {
1202                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
1203                 if (err)
1204                         return err;
1205         } else {
1206                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
1207
1208                 if (remote->sa.sa_family == AF_INET) {
1209                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
1210                         ip->sa.sa_family = AF_INET;
1211 #if IS_ENABLED(CONFIG_IPV6)
1212                 } else {
1213                         ip->sin6.sin6_addr = in6addr_any;
1214                         ip->sa.sa_family = AF_INET6;
1215 #endif
1216                 }
1217         }
1218
1219         if (tb[NDA_PORT]) {
1220                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
1221                         return -EINVAL;
1222                 *port = nla_get_be16(tb[NDA_PORT]);
1223         } else {
1224                 *port = vxlan->cfg.dst_port;
1225         }
1226
1227         if (tb[NDA_VNI]) {
1228                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
1229                         return -EINVAL;
1230                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1231         } else {
1232                 *vni = vxlan->default_dst.remote_vni;
1233         }
1234
1235         if (tb[NDA_SRC_VNI]) {
1236                 if (nla_len(tb[NDA_SRC_VNI]) != sizeof(u32))
1237                         return -EINVAL;
1238                 *src_vni = cpu_to_be32(nla_get_u32(tb[NDA_SRC_VNI]));
1239         } else {
1240                 *src_vni = vxlan->default_dst.remote_vni;
1241         }
1242
1243         if (tb[NDA_IFINDEX]) {
1244                 struct net_device *tdev;
1245
1246                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
1247                         return -EINVAL;
1248                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
1249                 tdev = __dev_get_by_index(net, *ifindex);
1250                 if (!tdev)
1251                         return -EADDRNOTAVAIL;
1252         } else {
1253                 *ifindex = 0;
1254         }
1255
1256         if (tb[NDA_NH_ID])
1257                 *nhid = nla_get_u32(tb[NDA_NH_ID]);
1258         else
1259                 *nhid = 0;
1260
1261         return 0;
1262 }
1263
1264 /* Add static entry (via netlink) */
1265 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
1266                          struct net_device *dev,
1267                          const unsigned char *addr, u16 vid, u16 flags,
1268                          struct netlink_ext_ack *extack)
1269 {
1270         struct vxlan_dev *vxlan = netdev_priv(dev);
1271         /* struct net *net = dev_net(vxlan->dev); */
1272         union vxlan_addr ip;
1273         __be16 port;
1274         __be32 src_vni, vni;
1275         u32 ifindex, nhid;
1276         u32 hash_index;
1277         int err;
1278
1279         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
1280                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
1281                         ndm->ndm_state);
1282                 return -EINVAL;
1283         }
1284
1285         if (!tb || (!tb[NDA_DST] && !tb[NDA_NH_ID]))
1286                 return -EINVAL;
1287
1288         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex,
1289                               &nhid);
1290         if (err)
1291                 return err;
1292
1293         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
1294                 return -EAFNOSUPPORT;
1295
1296         hash_index = fdb_head_index(vxlan, addr, src_vni);
1297         spin_lock_bh(&vxlan->hash_lock[hash_index]);
1298         err = vxlan_fdb_update(vxlan, addr, &ip, ndm->ndm_state, flags,
1299                                port, src_vni, vni, ifindex,
1300                                ndm->ndm_flags | NTF_VXLAN_ADDED_BY_USER,
1301                                nhid, true, extack);
1302         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1303
1304         return err;
1305 }
1306
1307 static int __vxlan_fdb_delete(struct vxlan_dev *vxlan,
1308                               const unsigned char *addr, union vxlan_addr ip,
1309                               __be16 port, __be32 src_vni, __be32 vni,
1310                               u32 ifindex, bool swdev_notify)
1311 {
1312         struct vxlan_rdst *rd = NULL;
1313         struct vxlan_fdb *f;
1314         int err = -ENOENT;
1315
1316         f = vxlan_find_mac(vxlan, addr, src_vni);
1317         if (!f)
1318                 return err;
1319
1320         if (!vxlan_addr_any(&ip)) {
1321                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
1322                 if (!rd)
1323                         goto out;
1324         }
1325
1326         /* remove a destination if it's not the only one on the list,
1327          * otherwise destroy the fdb entry
1328          */
1329         if (rd && !list_is_singular(&f->remotes)) {
1330                 vxlan_fdb_dst_destroy(vxlan, f, rd, swdev_notify);
1331                 goto out;
1332         }
1333
1334         vxlan_fdb_destroy(vxlan, f, true, swdev_notify);
1335
1336 out:
1337         return 0;
1338 }
1339
1340 /* Delete entry (via netlink) */
1341 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
1342                             struct net_device *dev,
1343                             const unsigned char *addr, u16 vid)
1344 {
1345         struct vxlan_dev *vxlan = netdev_priv(dev);
1346         union vxlan_addr ip;
1347         __be32 src_vni, vni;
1348         u32 ifindex, nhid;
1349         u32 hash_index;
1350         __be16 port;
1351         int err;
1352
1353         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &src_vni, &vni, &ifindex,
1354                               &nhid);
1355         if (err)
1356                 return err;
1357
1358         hash_index = fdb_head_index(vxlan, addr, src_vni);
1359         spin_lock_bh(&vxlan->hash_lock[hash_index]);
1360         err = __vxlan_fdb_delete(vxlan, addr, ip, port, src_vni, vni, ifindex,
1361                                  true);
1362         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
1363
1364         return err;
1365 }
1366
1367 /* Dump forwarding table */
1368 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
1369                           struct net_device *dev,
1370                           struct net_device *filter_dev, int *idx)
1371 {
1372         struct vxlan_dev *vxlan = netdev_priv(dev);
1373         unsigned int h;
1374         int err = 0;
1375
1376         for (h = 0; h < FDB_HASH_SIZE; ++h) {
1377                 struct vxlan_fdb *f;
1378
1379                 rcu_read_lock();
1380                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
1381                         struct vxlan_rdst *rd;
1382
1383                         if (rcu_access_pointer(f->nh)) {
1384                                 if (*idx < cb->args[2])
1385                                         goto skip_nh;
1386                                 err = vxlan_fdb_info(skb, vxlan, f,
1387                                                      NETLINK_CB(cb->skb).portid,
1388                                                      cb->nlh->nlmsg_seq,
1389                                                      RTM_NEWNEIGH,
1390                                                      NLM_F_MULTI, NULL);
1391                                 if (err < 0) {
1392                                         rcu_read_unlock();
1393                                         goto out;
1394                                 }
1395 skip_nh:
1396                                 *idx += 1;
1397                                 continue;
1398                         }
1399
1400                         list_for_each_entry_rcu(rd, &f->remotes, list) {
1401                                 if (*idx < cb->args[2])
1402                                         goto skip;
1403
1404                                 err = vxlan_fdb_info(skb, vxlan, f,
1405                                                      NETLINK_CB(cb->skb).portid,
1406                                                      cb->nlh->nlmsg_seq,
1407                                                      RTM_NEWNEIGH,
1408                                                      NLM_F_MULTI, rd);
1409                                 if (err < 0) {
1410                                         rcu_read_unlock();
1411                                         goto out;
1412                                 }
1413 skip:
1414                                 *idx += 1;
1415                         }
1416                 }
1417                 rcu_read_unlock();
1418         }
1419 out:
1420         return err;
1421 }
1422
1423 static int vxlan_fdb_get(struct sk_buff *skb,
1424                          struct nlattr *tb[],
1425                          struct net_device *dev,
1426                          const unsigned char *addr,
1427                          u16 vid, u32 portid, u32 seq,
1428                          struct netlink_ext_ack *extack)
1429 {
1430         struct vxlan_dev *vxlan = netdev_priv(dev);
1431         struct vxlan_fdb *f;
1432         __be32 vni;
1433         int err;
1434
1435         if (tb[NDA_VNI])
1436                 vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1437         else
1438                 vni = vxlan->default_dst.remote_vni;
1439
1440         rcu_read_lock();
1441
1442         f = __vxlan_find_mac(vxlan, addr, vni);
1443         if (!f) {
1444                 NL_SET_ERR_MSG(extack, "Fdb entry not found");
1445                 err = -ENOENT;
1446                 goto errout;
1447         }
1448
1449         err = vxlan_fdb_info(skb, vxlan, f, portid, seq,
1450                              RTM_NEWNEIGH, 0, first_remote_rcu(f));
1451 errout:
1452         rcu_read_unlock();
1453         return err;
1454 }
1455
1456 /* Watch incoming packets to learn mapping between Ethernet address
1457  * and Tunnel endpoint.
1458  * Return true if packet is bogus and should be dropped.
1459  */
1460 static bool vxlan_snoop(struct net_device *dev,
1461                         union vxlan_addr *src_ip, const u8 *src_mac,
1462                         u32 src_ifindex, __be32 vni)
1463 {
1464         struct vxlan_dev *vxlan = netdev_priv(dev);
1465         struct vxlan_fdb *f;
1466         u32 ifindex = 0;
1467
1468 #if IS_ENABLED(CONFIG_IPV6)
1469         if (src_ip->sa.sa_family == AF_INET6 &&
1470             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
1471                 ifindex = src_ifindex;
1472 #endif
1473
1474         f = vxlan_find_mac(vxlan, src_mac, vni);
1475         if (likely(f)) {
1476                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1477
1478                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
1479                            rdst->remote_ifindex == ifindex))
1480                         return false;
1481
1482                 /* Don't migrate static entries, drop packets */
1483                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
1484                         return true;
1485
1486                 /* Don't override an fdb with nexthop with a learnt entry */
1487                 if (rcu_access_pointer(f->nh))
1488                         return true;
1489
1490                 if (net_ratelimit())
1491                         netdev_info(dev,
1492                                     "%pM migrated from %pIS to %pIS\n",
1493                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1494
1495                 rdst->remote_ip = *src_ip;
1496                 f->updated = jiffies;
1497                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH, true, NULL);
1498         } else {
1499                 u32 hash_index = fdb_head_index(vxlan, src_mac, vni);
1500
1501                 /* learned new entry */
1502                 spin_lock(&vxlan->hash_lock[hash_index]);
1503
1504                 /* close off race between vxlan_flush and incoming packets */
1505                 if (netif_running(dev))
1506                         vxlan_fdb_update(vxlan, src_mac, src_ip,
1507                                          NUD_REACHABLE,
1508                                          NLM_F_EXCL|NLM_F_CREATE,
1509                                          vxlan->cfg.dst_port,
1510                                          vni,
1511                                          vxlan->default_dst.remote_vni,
1512                                          ifindex, NTF_SELF, 0, true, NULL);
1513                 spin_unlock(&vxlan->hash_lock[hash_index]);
1514         }
1515
1516         return false;
1517 }
1518
1519 /* See if multicast group is already in use by other ID */
1520 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1521 {
1522         struct vxlan_dev *vxlan;
1523         struct vxlan_sock *sock4;
1524 #if IS_ENABLED(CONFIG_IPV6)
1525         struct vxlan_sock *sock6;
1526 #endif
1527         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1528
1529         sock4 = rtnl_dereference(dev->vn4_sock);
1530
1531         /* The vxlan_sock is only used by dev, leaving group has
1532          * no effect on other vxlan devices.
1533          */
1534         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1535                 return false;
1536 #if IS_ENABLED(CONFIG_IPV6)
1537         sock6 = rtnl_dereference(dev->vn6_sock);
1538         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1539                 return false;
1540 #endif
1541
1542         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1543                 if (!netif_running(vxlan->dev) || vxlan == dev)
1544                         continue;
1545
1546                 if (family == AF_INET &&
1547                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1548                         continue;
1549 #if IS_ENABLED(CONFIG_IPV6)
1550                 if (family == AF_INET6 &&
1551                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1552                         continue;
1553 #endif
1554
1555                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1556                                       &dev->default_dst.remote_ip))
1557                         continue;
1558
1559                 if (vxlan->default_dst.remote_ifindex !=
1560                     dev->default_dst.remote_ifindex)
1561                         continue;
1562
1563                 return true;
1564         }
1565
1566         return false;
1567 }
1568
1569 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1570 {
1571         struct vxlan_net *vn;
1572
1573         if (!vs)
1574                 return false;
1575         if (!refcount_dec_and_test(&vs->refcnt))
1576                 return false;
1577
1578         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1579         spin_lock(&vn->sock_lock);
1580         hlist_del_rcu(&vs->hlist);
1581         udp_tunnel_notify_del_rx_port(vs->sock,
1582                                       (vs->flags & VXLAN_F_GPE) ?
1583                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1584                                       UDP_TUNNEL_TYPE_VXLAN);
1585         spin_unlock(&vn->sock_lock);
1586
1587         return true;
1588 }
1589
1590 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1591 {
1592         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1593 #if IS_ENABLED(CONFIG_IPV6)
1594         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1595
1596         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1597 #endif
1598
1599         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1600         synchronize_net();
1601
1602         vxlan_vs_del_dev(vxlan);
1603
1604         if (__vxlan_sock_release_prep(sock4)) {
1605                 udp_tunnel_sock_release(sock4->sock);
1606                 kfree(sock4);
1607         }
1608
1609 #if IS_ENABLED(CONFIG_IPV6)
1610         if (__vxlan_sock_release_prep(sock6)) {
1611                 udp_tunnel_sock_release(sock6->sock);
1612                 kfree(sock6);
1613         }
1614 #endif
1615 }
1616
1617 /* Update multicast group membership when first VNI on
1618  * multicast address is brought up
1619  */
1620 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1621 {
1622         struct sock *sk;
1623         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1624         int ifindex = vxlan->default_dst.remote_ifindex;
1625         int ret = -EINVAL;
1626
1627         if (ip->sa.sa_family == AF_INET) {
1628                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1629                 struct ip_mreqn mreq = {
1630                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1631                         .imr_ifindex            = ifindex,
1632                 };
1633
1634                 sk = sock4->sock->sk;
1635                 lock_sock(sk);
1636                 ret = ip_mc_join_group(sk, &mreq);
1637                 release_sock(sk);
1638 #if IS_ENABLED(CONFIG_IPV6)
1639         } else {
1640                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1641
1642                 sk = sock6->sock->sk;
1643                 lock_sock(sk);
1644                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1645                                                    &ip->sin6.sin6_addr);
1646                 release_sock(sk);
1647 #endif
1648         }
1649
1650         return ret;
1651 }
1652
1653 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1654 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1655 {
1656         struct sock *sk;
1657         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1658         int ifindex = vxlan->default_dst.remote_ifindex;
1659         int ret = -EINVAL;
1660
1661         if (ip->sa.sa_family == AF_INET) {
1662                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1663                 struct ip_mreqn mreq = {
1664                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1665                         .imr_ifindex            = ifindex,
1666                 };
1667
1668                 sk = sock4->sock->sk;
1669                 lock_sock(sk);
1670                 ret = ip_mc_leave_group(sk, &mreq);
1671                 release_sock(sk);
1672 #if IS_ENABLED(CONFIG_IPV6)
1673         } else {
1674                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1675
1676                 sk = sock6->sock->sk;
1677                 lock_sock(sk);
1678                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1679                                                    &ip->sin6.sin6_addr);
1680                 release_sock(sk);
1681 #endif
1682         }
1683
1684         return ret;
1685 }
1686
1687 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1688                           struct sk_buff *skb, u32 vxflags)
1689 {
1690         size_t start, offset;
1691
1692         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1693                 goto out;
1694
1695         start = vxlan_rco_start(unparsed->vx_vni);
1696         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1697
1698         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1699                 return false;
1700
1701         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1702                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1703 out:
1704         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1705         unparsed->vx_vni &= VXLAN_VNI_MASK;
1706         return true;
1707 }
1708
1709 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1710                                 struct sk_buff *skb, u32 vxflags,
1711                                 struct vxlan_metadata *md)
1712 {
1713         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1714         struct metadata_dst *tun_dst;
1715
1716         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1717                 goto out;
1718
1719         md->gbp = ntohs(gbp->policy_id);
1720
1721         tun_dst = (struct metadata_dst *)skb_dst(skb);
1722         if (tun_dst) {
1723                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1724                 tun_dst->u.tun_info.options_len = sizeof(*md);
1725         }
1726         if (gbp->dont_learn)
1727                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1728
1729         if (gbp->policy_applied)
1730                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1731
1732         /* In flow-based mode, GBP is carried in dst_metadata */
1733         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1734                 skb->mark = md->gbp;
1735 out:
1736         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1737 }
1738
1739 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1740                                 __be16 *protocol,
1741                                 struct sk_buff *skb, u32 vxflags)
1742 {
1743         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1744
1745         /* Need to have Next Protocol set for interfaces in GPE mode. */
1746         if (!gpe->np_applied)
1747                 return false;
1748         /* "The initial version is 0. If a receiver does not support the
1749          * version indicated it MUST drop the packet.
1750          */
1751         if (gpe->version != 0)
1752                 return false;
1753         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1754          * processing MUST occur." However, we don't implement OAM
1755          * processing, thus drop the packet.
1756          */
1757         if (gpe->oam_flag)
1758                 return false;
1759
1760         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1761         if (!*protocol)
1762                 return false;
1763
1764         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1765         return true;
1766 }
1767
1768 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1769                           struct vxlan_sock *vs,
1770                           struct sk_buff *skb, __be32 vni)
1771 {
1772         union vxlan_addr saddr;
1773         u32 ifindex = skb->dev->ifindex;
1774
1775         skb_reset_mac_header(skb);
1776         skb->protocol = eth_type_trans(skb, vxlan->dev);
1777         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1778
1779         /* Ignore packet loops (and multicast echo) */
1780         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1781                 return false;
1782
1783         /* Get address from the outer IP header */
1784         if (vxlan_get_sk_family(vs) == AF_INET) {
1785                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1786                 saddr.sa.sa_family = AF_INET;
1787 #if IS_ENABLED(CONFIG_IPV6)
1788         } else {
1789                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1790                 saddr.sa.sa_family = AF_INET6;
1791 #endif
1792         }
1793
1794         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1795             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1796                 return false;
1797
1798         return true;
1799 }
1800
1801 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1802                                   struct sk_buff *skb)
1803 {
1804         int err = 0;
1805
1806         if (vxlan_get_sk_family(vs) == AF_INET)
1807                 err = IP_ECN_decapsulate(oiph, skb);
1808 #if IS_ENABLED(CONFIG_IPV6)
1809         else
1810                 err = IP6_ECN_decapsulate(oiph, skb);
1811 #endif
1812
1813         if (unlikely(err) && log_ecn_error) {
1814                 if (vxlan_get_sk_family(vs) == AF_INET)
1815                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1816                                              &((struct iphdr *)oiph)->saddr,
1817                                              ((struct iphdr *)oiph)->tos);
1818                 else
1819                         net_info_ratelimited("non-ECT from %pI6\n",
1820                                              &((struct ipv6hdr *)oiph)->saddr);
1821         }
1822         return err <= 1;
1823 }
1824
1825 /* Callback from net/ipv4/udp.c to receive packets */
1826 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1827 {
1828         struct pcpu_sw_netstats *stats;
1829         struct vxlan_dev *vxlan;
1830         struct vxlan_sock *vs;
1831         struct vxlanhdr unparsed;
1832         struct vxlan_metadata _md;
1833         struct vxlan_metadata *md = &_md;
1834         __be16 protocol = htons(ETH_P_TEB);
1835         bool raw_proto = false;
1836         void *oiph;
1837         __be32 vni = 0;
1838
1839         /* Need UDP and VXLAN header to be present */
1840         if (!pskb_may_pull(skb, VXLAN_HLEN))
1841                 goto drop;
1842
1843         unparsed = *vxlan_hdr(skb);
1844         /* VNI flag always required to be set */
1845         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1846                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1847                            ntohl(vxlan_hdr(skb)->vx_flags),
1848                            ntohl(vxlan_hdr(skb)->vx_vni));
1849                 /* Return non vxlan pkt */
1850                 goto drop;
1851         }
1852         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1853         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1854
1855         vs = rcu_dereference_sk_user_data(sk);
1856         if (!vs)
1857                 goto drop;
1858
1859         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1860
1861         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1862         if (!vxlan)
1863                 goto drop;
1864
1865         /* For backwards compatibility, only allow reserved fields to be
1866          * used by VXLAN extensions if explicitly requested.
1867          */
1868         if (vs->flags & VXLAN_F_GPE) {
1869                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1870                         goto drop;
1871                 raw_proto = true;
1872         }
1873
1874         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1875                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1876                 goto drop;
1877
1878         if (vxlan_collect_metadata(vs)) {
1879                 struct metadata_dst *tun_dst;
1880
1881                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1882                                          key32_to_tunnel_id(vni), sizeof(*md));
1883
1884                 if (!tun_dst)
1885                         goto drop;
1886
1887                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1888
1889                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1890         } else {
1891                 memset(md, 0, sizeof(*md));
1892         }
1893
1894         if (vs->flags & VXLAN_F_REMCSUM_RX)
1895                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1896                         goto drop;
1897         if (vs->flags & VXLAN_F_GBP)
1898                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1899         /* Note that GBP and GPE can never be active together. This is
1900          * ensured in vxlan_dev_configure.
1901          */
1902
1903         if (unparsed.vx_flags || unparsed.vx_vni) {
1904                 /* If there are any unprocessed flags remaining treat
1905                  * this as a malformed packet. This behavior diverges from
1906                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1907                  * in reserved fields are to be ignored. The approach here
1908                  * maintains compatibility with previous stack code, and also
1909                  * is more robust and provides a little more security in
1910                  * adding extensions to VXLAN.
1911                  */
1912                 goto drop;
1913         }
1914
1915         if (!raw_proto) {
1916                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1917                         goto drop;
1918         } else {
1919                 skb_reset_mac_header(skb);
1920                 skb->dev = vxlan->dev;
1921                 skb->pkt_type = PACKET_HOST;
1922         }
1923
1924         oiph = skb_network_header(skb);
1925         skb_reset_network_header(skb);
1926
1927         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1928                 ++vxlan->dev->stats.rx_frame_errors;
1929                 ++vxlan->dev->stats.rx_errors;
1930                 goto drop;
1931         }
1932
1933         rcu_read_lock();
1934
1935         if (unlikely(!(vxlan->dev->flags & IFF_UP))) {
1936                 rcu_read_unlock();
1937                 atomic_long_inc(&vxlan->dev->rx_dropped);
1938                 goto drop;
1939         }
1940
1941         stats = this_cpu_ptr(vxlan->dev->tstats);
1942         u64_stats_update_begin(&stats->syncp);
1943         stats->rx_packets++;
1944         stats->rx_bytes += skb->len;
1945         u64_stats_update_end(&stats->syncp);
1946
1947         gro_cells_receive(&vxlan->gro_cells, skb);
1948
1949         rcu_read_unlock();
1950
1951         return 0;
1952
1953 drop:
1954         /* Consume bad packet */
1955         kfree_skb(skb);
1956         return 0;
1957 }
1958
1959 /* Callback from net/ipv{4,6}/udp.c to check that we have a VNI for errors */
1960 static int vxlan_err_lookup(struct sock *sk, struct sk_buff *skb)
1961 {
1962         struct vxlan_dev *vxlan;
1963         struct vxlan_sock *vs;
1964         struct vxlanhdr *hdr;
1965         __be32 vni;
1966
1967         if (!pskb_may_pull(skb, skb_transport_offset(skb) + VXLAN_HLEN))
1968                 return -EINVAL;
1969
1970         hdr = vxlan_hdr(skb);
1971
1972         if (!(hdr->vx_flags & VXLAN_HF_VNI))
1973                 return -EINVAL;
1974
1975         vs = rcu_dereference_sk_user_data(sk);
1976         if (!vs)
1977                 return -ENOENT;
1978
1979         vni = vxlan_vni(hdr->vx_vni);
1980         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1981         if (!vxlan)
1982                 return -ENOENT;
1983
1984         return 0;
1985 }
1986
1987 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1988 {
1989         struct vxlan_dev *vxlan = netdev_priv(dev);
1990         struct arphdr *parp;
1991         u8 *arpptr, *sha;
1992         __be32 sip, tip;
1993         struct neighbour *n;
1994
1995         if (dev->flags & IFF_NOARP)
1996                 goto out;
1997
1998         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1999                 dev->stats.tx_dropped++;
2000                 goto out;
2001         }
2002         parp = arp_hdr(skb);
2003
2004         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
2005              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
2006             parp->ar_pro != htons(ETH_P_IP) ||
2007             parp->ar_op != htons(ARPOP_REQUEST) ||
2008             parp->ar_hln != dev->addr_len ||
2009             parp->ar_pln != 4)
2010                 goto out;
2011         arpptr = (u8 *)parp + sizeof(struct arphdr);
2012         sha = arpptr;
2013         arpptr += dev->addr_len;        /* sha */
2014         memcpy(&sip, arpptr, sizeof(sip));
2015         arpptr += sizeof(sip);
2016         arpptr += dev->addr_len;        /* tha */
2017         memcpy(&tip, arpptr, sizeof(tip));
2018
2019         if (ipv4_is_loopback(tip) ||
2020             ipv4_is_multicast(tip))
2021                 goto out;
2022
2023         n = neigh_lookup(&arp_tbl, &tip, dev);
2024
2025         if (n) {
2026                 struct vxlan_fdb *f;
2027                 struct sk_buff  *reply;
2028
2029                 if (!(n->nud_state & NUD_CONNECTED)) {
2030                         neigh_release(n);
2031                         goto out;
2032                 }
2033
2034                 f = vxlan_find_mac(vxlan, n->ha, vni);
2035                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2036                         /* bridge-local neighbor */
2037                         neigh_release(n);
2038                         goto out;
2039                 }
2040
2041                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
2042                                 n->ha, sha);
2043
2044                 neigh_release(n);
2045
2046                 if (reply == NULL)
2047                         goto out;
2048
2049                 skb_reset_mac_header(reply);
2050                 __skb_pull(reply, skb_network_offset(reply));
2051                 reply->ip_summed = CHECKSUM_UNNECESSARY;
2052                 reply->pkt_type = PACKET_HOST;
2053
2054                 if (netif_rx_ni(reply) == NET_RX_DROP)
2055                         dev->stats.rx_dropped++;
2056         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2057                 union vxlan_addr ipa = {
2058                         .sin.sin_addr.s_addr = tip,
2059                         .sin.sin_family = AF_INET,
2060                 };
2061
2062                 vxlan_ip_miss(dev, &ipa);
2063         }
2064 out:
2065         consume_skb(skb);
2066         return NETDEV_TX_OK;
2067 }
2068
2069 #if IS_ENABLED(CONFIG_IPV6)
2070 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
2071         struct neighbour *n, bool isrouter)
2072 {
2073         struct net_device *dev = request->dev;
2074         struct sk_buff *reply;
2075         struct nd_msg *ns, *na;
2076         struct ipv6hdr *pip6;
2077         u8 *daddr;
2078         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
2079         int ns_olen;
2080         int i, len;
2081
2082         if (dev == NULL || !pskb_may_pull(request, request->len))
2083                 return NULL;
2084
2085         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
2086                 sizeof(*na) + na_olen + dev->needed_tailroom;
2087         reply = alloc_skb(len, GFP_ATOMIC);
2088         if (reply == NULL)
2089                 return NULL;
2090
2091         reply->protocol = htons(ETH_P_IPV6);
2092         reply->dev = dev;
2093         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
2094         skb_push(reply, sizeof(struct ethhdr));
2095         skb_reset_mac_header(reply);
2096
2097         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
2098
2099         daddr = eth_hdr(request)->h_source;
2100         ns_olen = request->len - skb_network_offset(request) -
2101                 sizeof(struct ipv6hdr) - sizeof(*ns);
2102         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
2103                 if (!ns->opt[i + 1]) {
2104                         kfree_skb(reply);
2105                         return NULL;
2106                 }
2107                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
2108                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
2109                         break;
2110                 }
2111         }
2112
2113         /* Ethernet header */
2114         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
2115         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
2116         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
2117         reply->protocol = htons(ETH_P_IPV6);
2118
2119         skb_pull(reply, sizeof(struct ethhdr));
2120         skb_reset_network_header(reply);
2121         skb_put(reply, sizeof(struct ipv6hdr));
2122
2123         /* IPv6 header */
2124
2125         pip6 = ipv6_hdr(reply);
2126         memset(pip6, 0, sizeof(struct ipv6hdr));
2127         pip6->version = 6;
2128         pip6->priority = ipv6_hdr(request)->priority;
2129         pip6->nexthdr = IPPROTO_ICMPV6;
2130         pip6->hop_limit = 255;
2131         pip6->daddr = ipv6_hdr(request)->saddr;
2132         pip6->saddr = *(struct in6_addr *)n->primary_key;
2133
2134         skb_pull(reply, sizeof(struct ipv6hdr));
2135         skb_reset_transport_header(reply);
2136
2137         /* Neighbor Advertisement */
2138         na = skb_put_zero(reply, sizeof(*na) + na_olen);
2139         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
2140         na->icmph.icmp6_router = isrouter;
2141         na->icmph.icmp6_override = 1;
2142         na->icmph.icmp6_solicited = 1;
2143         na->target = ns->target;
2144         ether_addr_copy(&na->opt[2], n->ha);
2145         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
2146         na->opt[1] = na_olen >> 3;
2147
2148         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
2149                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
2150                 csum_partial(na, sizeof(*na)+na_olen, 0));
2151
2152         pip6->payload_len = htons(sizeof(*na)+na_olen);
2153
2154         skb_push(reply, sizeof(struct ipv6hdr));
2155
2156         reply->ip_summed = CHECKSUM_UNNECESSARY;
2157
2158         return reply;
2159 }
2160
2161 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
2162 {
2163         struct vxlan_dev *vxlan = netdev_priv(dev);
2164         const struct in6_addr *daddr;
2165         const struct ipv6hdr *iphdr;
2166         struct inet6_dev *in6_dev;
2167         struct neighbour *n;
2168         struct nd_msg *msg;
2169
2170         in6_dev = __in6_dev_get(dev);
2171         if (!in6_dev)
2172                 goto out;
2173
2174         iphdr = ipv6_hdr(skb);
2175         daddr = &iphdr->daddr;
2176         msg = (struct nd_msg *)(iphdr + 1);
2177
2178         if (ipv6_addr_loopback(daddr) ||
2179             ipv6_addr_is_multicast(&msg->target))
2180                 goto out;
2181
2182         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
2183
2184         if (n) {
2185                 struct vxlan_fdb *f;
2186                 struct sk_buff *reply;
2187
2188                 if (!(n->nud_state & NUD_CONNECTED)) {
2189                         neigh_release(n);
2190                         goto out;
2191                 }
2192
2193                 f = vxlan_find_mac(vxlan, n->ha, vni);
2194                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2195                         /* bridge-local neighbor */
2196                         neigh_release(n);
2197                         goto out;
2198                 }
2199
2200                 reply = vxlan_na_create(skb, n,
2201                                         !!(f ? f->flags & NTF_ROUTER : 0));
2202
2203                 neigh_release(n);
2204
2205                 if (reply == NULL)
2206                         goto out;
2207
2208                 if (netif_rx_ni(reply) == NET_RX_DROP)
2209                         dev->stats.rx_dropped++;
2210
2211         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2212                 union vxlan_addr ipa = {
2213                         .sin6.sin6_addr = msg->target,
2214                         .sin6.sin6_family = AF_INET6,
2215                 };
2216
2217                 vxlan_ip_miss(dev, &ipa);
2218         }
2219
2220 out:
2221         consume_skb(skb);
2222         return NETDEV_TX_OK;
2223 }
2224 #endif
2225
2226 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
2227 {
2228         struct vxlan_dev *vxlan = netdev_priv(dev);
2229         struct neighbour *n;
2230
2231         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
2232                 return false;
2233
2234         n = NULL;
2235         switch (ntohs(eth_hdr(skb)->h_proto)) {
2236         case ETH_P_IP:
2237         {
2238                 struct iphdr *pip;
2239
2240                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
2241                         return false;
2242                 pip = ip_hdr(skb);
2243                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
2244                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2245                         union vxlan_addr ipa = {
2246                                 .sin.sin_addr.s_addr = pip->daddr,
2247                                 .sin.sin_family = AF_INET,
2248                         };
2249
2250                         vxlan_ip_miss(dev, &ipa);
2251                         return false;
2252                 }
2253
2254                 break;
2255         }
2256 #if IS_ENABLED(CONFIG_IPV6)
2257         case ETH_P_IPV6:
2258         {
2259                 struct ipv6hdr *pip6;
2260
2261                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
2262                         return false;
2263                 pip6 = ipv6_hdr(skb);
2264                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
2265                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2266                         union vxlan_addr ipa = {
2267                                 .sin6.sin6_addr = pip6->daddr,
2268                                 .sin6.sin6_family = AF_INET6,
2269                         };
2270
2271                         vxlan_ip_miss(dev, &ipa);
2272                         return false;
2273                 }
2274
2275                 break;
2276         }
2277 #endif
2278         default:
2279                 return false;
2280         }
2281
2282         if (n) {
2283                 bool diff;
2284
2285                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
2286                 if (diff) {
2287                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
2288                                 dev->addr_len);
2289                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
2290                 }
2291                 neigh_release(n);
2292                 return diff;
2293         }
2294
2295         return false;
2296 }
2297
2298 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
2299                                 struct vxlan_metadata *md)
2300 {
2301         struct vxlanhdr_gbp *gbp;
2302
2303         if (!md->gbp)
2304                 return;
2305
2306         gbp = (struct vxlanhdr_gbp *)vxh;
2307         vxh->vx_flags |= VXLAN_HF_GBP;
2308
2309         if (md->gbp & VXLAN_GBP_DONT_LEARN)
2310                 gbp->dont_learn = 1;
2311
2312         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
2313                 gbp->policy_applied = 1;
2314
2315         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
2316 }
2317
2318 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
2319                                __be16 protocol)
2320 {
2321         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
2322
2323         gpe->np_applied = 1;
2324         gpe->next_protocol = tun_p_from_eth_p(protocol);
2325         if (!gpe->next_protocol)
2326                 return -EPFNOSUPPORT;
2327         return 0;
2328 }
2329
2330 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
2331                            int iphdr_len, __be32 vni,
2332                            struct vxlan_metadata *md, u32 vxflags,
2333                            bool udp_sum)
2334 {
2335         struct vxlanhdr *vxh;
2336         int min_headroom;
2337         int err;
2338         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
2339         __be16 inner_protocol = htons(ETH_P_TEB);
2340
2341         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
2342             skb->ip_summed == CHECKSUM_PARTIAL) {
2343                 int csum_start = skb_checksum_start_offset(skb);
2344
2345                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
2346                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
2347                     (skb->csum_offset == offsetof(struct udphdr, check) ||
2348                      skb->csum_offset == offsetof(struct tcphdr, check)))
2349                         type |= SKB_GSO_TUNNEL_REMCSUM;
2350         }
2351
2352         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
2353                         + VXLAN_HLEN + iphdr_len;
2354
2355         /* Need space for new headers (invalidates iph ptr) */
2356         err = skb_cow_head(skb, min_headroom);
2357         if (unlikely(err))
2358                 return err;
2359
2360         err = iptunnel_handle_offloads(skb, type);
2361         if (err)
2362                 return err;
2363
2364         vxh = __skb_push(skb, sizeof(*vxh));
2365         vxh->vx_flags = VXLAN_HF_VNI;
2366         vxh->vx_vni = vxlan_vni_field(vni);
2367
2368         if (type & SKB_GSO_TUNNEL_REMCSUM) {
2369                 unsigned int start;
2370
2371                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
2372                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
2373                 vxh->vx_flags |= VXLAN_HF_RCO;
2374
2375                 if (!skb_is_gso(skb)) {
2376                         skb->ip_summed = CHECKSUM_NONE;
2377                         skb->encapsulation = 0;
2378                 }
2379         }
2380
2381         if (vxflags & VXLAN_F_GBP)
2382                 vxlan_build_gbp_hdr(vxh, vxflags, md);
2383         if (vxflags & VXLAN_F_GPE) {
2384                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
2385                 if (err < 0)
2386                         return err;
2387                 inner_protocol = skb->protocol;
2388         }
2389
2390         skb_set_inner_protocol(skb, inner_protocol);
2391         return 0;
2392 }
2393
2394 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
2395                                       struct vxlan_sock *sock4,
2396                                       struct sk_buff *skb, int oif, u8 tos,
2397                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
2398                                       struct dst_cache *dst_cache,
2399                                       const struct ip_tunnel_info *info)
2400 {
2401         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2402         struct rtable *rt = NULL;
2403         struct flowi4 fl4;
2404
2405         if (!sock4)
2406                 return ERR_PTR(-EIO);
2407
2408         if (tos && !info)
2409                 use_cache = false;
2410         if (use_cache) {
2411                 rt = dst_cache_get_ip4(dst_cache, saddr);
2412                 if (rt)
2413                         return rt;
2414         }
2415
2416         memset(&fl4, 0, sizeof(fl4));
2417         fl4.flowi4_oif = oif;
2418         fl4.flowi4_tos = RT_TOS(tos);
2419         fl4.flowi4_mark = skb->mark;
2420         fl4.flowi4_proto = IPPROTO_UDP;
2421         fl4.daddr = daddr;
2422         fl4.saddr = *saddr;
2423         fl4.fl4_dport = dport;
2424         fl4.fl4_sport = sport;
2425
2426         rt = ip_route_output_key(vxlan->net, &fl4);
2427         if (!IS_ERR(rt)) {
2428                 if (rt->dst.dev == dev) {
2429                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
2430                         ip_rt_put(rt);
2431                         return ERR_PTR(-ELOOP);
2432                 }
2433
2434                 *saddr = fl4.saddr;
2435                 if (use_cache)
2436                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2437         } else {
2438                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
2439                 return ERR_PTR(-ENETUNREACH);
2440         }
2441         return rt;
2442 }
2443
2444 #if IS_ENABLED(CONFIG_IPV6)
2445 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
2446                                           struct net_device *dev,
2447                                           struct vxlan_sock *sock6,
2448                                           struct sk_buff *skb, int oif, u8 tos,
2449                                           __be32 label,
2450                                           const struct in6_addr *daddr,
2451                                           struct in6_addr *saddr,
2452                                           __be16 dport, __be16 sport,
2453                                           struct dst_cache *dst_cache,
2454                                           const struct ip_tunnel_info *info)
2455 {
2456         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2457         struct dst_entry *ndst;
2458         struct flowi6 fl6;
2459
2460         if (!sock6)
2461                 return ERR_PTR(-EIO);
2462
2463         if (tos && !info)
2464                 use_cache = false;
2465         if (use_cache) {
2466                 ndst = dst_cache_get_ip6(dst_cache, saddr);
2467                 if (ndst)
2468                         return ndst;
2469         }
2470
2471         memset(&fl6, 0, sizeof(fl6));
2472         fl6.flowi6_oif = oif;
2473         fl6.daddr = *daddr;
2474         fl6.saddr = *saddr;
2475         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
2476         fl6.flowi6_mark = skb->mark;
2477         fl6.flowi6_proto = IPPROTO_UDP;
2478         fl6.fl6_dport = dport;
2479         fl6.fl6_sport = sport;
2480
2481         ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk,
2482                                                &fl6, NULL);
2483         if (unlikely(IS_ERR(ndst))) {
2484                 netdev_dbg(dev, "no route to %pI6\n", daddr);
2485                 return ERR_PTR(-ENETUNREACH);
2486         }
2487
2488         if (unlikely(ndst->dev == dev)) {
2489                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
2490                 dst_release(ndst);
2491                 return ERR_PTR(-ELOOP);
2492         }
2493
2494         *saddr = fl6.saddr;
2495         if (use_cache)
2496                 dst_cache_set_ip6(dst_cache, ndst, saddr);
2497         return ndst;
2498 }
2499 #endif
2500
2501 /* Bypass encapsulation if the destination is local */
2502 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2503                                struct vxlan_dev *dst_vxlan, __be32 vni)
2504 {
2505         struct pcpu_sw_netstats *tx_stats, *rx_stats;
2506         union vxlan_addr loopback;
2507         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2508         struct net_device *dev;
2509         int len = skb->len;
2510
2511         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2512         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2513         skb->pkt_type = PACKET_HOST;
2514         skb->encapsulation = 0;
2515         skb->dev = dst_vxlan->dev;
2516         __skb_pull(skb, skb_network_offset(skb));
2517
2518         if (remote_ip->sa.sa_family == AF_INET) {
2519                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2520                 loopback.sa.sa_family =  AF_INET;
2521 #if IS_ENABLED(CONFIG_IPV6)
2522         } else {
2523                 loopback.sin6.sin6_addr = in6addr_loopback;
2524                 loopback.sa.sa_family =  AF_INET6;
2525 #endif
2526         }
2527
2528         rcu_read_lock();
2529         dev = skb->dev;
2530         if (unlikely(!(dev->flags & IFF_UP))) {
2531                 kfree_skb(skb);
2532                 goto drop;
2533         }
2534
2535         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
2536                 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni);
2537
2538         u64_stats_update_begin(&tx_stats->syncp);
2539         tx_stats->tx_packets++;
2540         tx_stats->tx_bytes += len;
2541         u64_stats_update_end(&tx_stats->syncp);
2542
2543         if (netif_rx(skb) == NET_RX_SUCCESS) {
2544                 u64_stats_update_begin(&rx_stats->syncp);
2545                 rx_stats->rx_packets++;
2546                 rx_stats->rx_bytes += len;
2547                 u64_stats_update_end(&rx_stats->syncp);
2548         } else {
2549 drop:
2550                 dev->stats.rx_dropped++;
2551         }
2552         rcu_read_unlock();
2553 }
2554
2555 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2556                                  struct vxlan_dev *vxlan,
2557                                  union vxlan_addr *daddr,
2558                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2559                                  struct dst_entry *dst,
2560                                  u32 rt_flags)
2561 {
2562 #if IS_ENABLED(CONFIG_IPV6)
2563         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2564          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2565          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2566          */
2567         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2568 #endif
2569         /* Bypass encapsulation if the destination is local */
2570         if (rt_flags & RTCF_LOCAL &&
2571             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2572                 struct vxlan_dev *dst_vxlan;
2573
2574                 dst_release(dst);
2575                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2576                                            daddr->sa.sa_family, dst_port,
2577                                            vxlan->cfg.flags);
2578                 if (!dst_vxlan) {
2579                         dev->stats.tx_errors++;
2580                         kfree_skb(skb);
2581
2582                         return -ENOENT;
2583                 }
2584                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2585                 return 1;
2586         }
2587
2588         return 0;
2589 }
2590
2591 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2592                            __be32 default_vni, struct vxlan_rdst *rdst,
2593                            bool did_rsc)
2594 {
2595         struct dst_cache *dst_cache;
2596         struct ip_tunnel_info *info;
2597         struct vxlan_dev *vxlan = netdev_priv(dev);
2598         const struct iphdr *old_iph = ip_hdr(skb);
2599         union vxlan_addr *dst;
2600         union vxlan_addr remote_ip, local_ip;
2601         struct vxlan_metadata _md;
2602         struct vxlan_metadata *md = &_md;
2603         __be16 src_port = 0, dst_port;
2604         struct dst_entry *ndst = NULL;
2605         __be32 vni, label;
2606         __u8 tos, ttl;
2607         int ifindex;
2608         int err;
2609         u32 flags = vxlan->cfg.flags;
2610         bool udp_sum = false;
2611         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2612
2613         info = skb_tunnel_info(skb);
2614
2615         if (rdst) {
2616                 dst = &rdst->remote_ip;
2617                 if (vxlan_addr_any(dst)) {
2618                         if (did_rsc) {
2619                                 /* short-circuited back to local bridge */
2620                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2621                                 return;
2622                         }
2623                         goto drop;
2624                 }
2625
2626                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2627                 vni = (rdst->remote_vni) ? : default_vni;
2628                 ifindex = rdst->remote_ifindex;
2629                 local_ip = vxlan->cfg.saddr;
2630                 dst_cache = &rdst->dst_cache;
2631                 md->gbp = skb->mark;
2632                 if (flags & VXLAN_F_TTL_INHERIT) {
2633                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2634                 } else {
2635                         ttl = vxlan->cfg.ttl;
2636                         if (!ttl && vxlan_addr_multicast(dst))
2637                                 ttl = 1;
2638                 }
2639
2640                 tos = vxlan->cfg.tos;
2641                 if (tos == 1)
2642                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2643
2644                 if (dst->sa.sa_family == AF_INET)
2645                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2646                 else
2647                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2648                 label = vxlan->cfg.label;
2649         } else {
2650                 if (!info) {
2651                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2652                                   dev->name);
2653                         goto drop;
2654                 }
2655                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2656                 if (remote_ip.sa.sa_family == AF_INET) {
2657                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2658                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2659                 } else {
2660                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2661                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2662                 }
2663                 dst = &remote_ip;
2664                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2665                 vni = tunnel_id_to_key32(info->key.tun_id);
2666                 ifindex = 0;
2667                 dst_cache = &info->dst_cache;
2668                 if (info->key.tun_flags & TUNNEL_VXLAN_OPT) {
2669                         if (info->options_len < sizeof(*md))
2670                                 goto drop;
2671                         md = ip_tunnel_info_opts(info);
2672                 }
2673                 ttl = info->key.ttl;
2674                 tos = info->key.tos;
2675                 label = info->key.label;
2676                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2677         }
2678         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2679                                      vxlan->cfg.port_max, true);
2680
2681         rcu_read_lock();
2682         if (dst->sa.sa_family == AF_INET) {
2683                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2684                 struct rtable *rt;
2685                 __be16 df = 0;
2686
2687                 if (!ifindex)
2688                         ifindex = sock4->sock->sk->sk_bound_dev_if;
2689
2690                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2691                                      dst->sin.sin_addr.s_addr,
2692                                      &local_ip.sin.sin_addr.s_addr,
2693                                      dst_port, src_port,
2694                                      dst_cache, info);
2695                 if (IS_ERR(rt)) {
2696                         err = PTR_ERR(rt);
2697                         goto tx_error;
2698                 }
2699
2700                 if (!info) {
2701                         /* Bypass encapsulation if the destination is local */
2702                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2703                                                     dst_port, ifindex, vni,
2704                                                     &rt->dst, rt->rt_flags);
2705                         if (err)
2706                                 goto out_unlock;
2707
2708                         if (vxlan->cfg.df == VXLAN_DF_SET) {
2709                                 df = htons(IP_DF);
2710                         } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) {
2711                                 struct ethhdr *eth = eth_hdr(skb);
2712
2713                                 if (ntohs(eth->h_proto) == ETH_P_IPV6 ||
2714                                     (ntohs(eth->h_proto) == ETH_P_IP &&
2715                                      old_iph->frag_off & htons(IP_DF)))
2716                                         df = htons(IP_DF);
2717                         }
2718                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2719                         df = htons(IP_DF);
2720                 }
2721
2722                 ndst = &rt->dst;
2723                 skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM);
2724
2725                 tos = ip_tunnel_ecn_encap(RT_TOS(tos), old_iph, skb);
2726                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2727                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2728                                       vni, md, flags, udp_sum);
2729                 if (err < 0)
2730                         goto tx_error;
2731
2732                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2733                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2734                                     src_port, dst_port, xnet, !udp_sum);
2735 #if IS_ENABLED(CONFIG_IPV6)
2736         } else {
2737                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2738
2739                 if (!ifindex)
2740                         ifindex = sock6->sock->sk->sk_bound_dev_if;
2741
2742                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2743                                         label, &dst->sin6.sin6_addr,
2744                                         &local_ip.sin6.sin6_addr,
2745                                         dst_port, src_port,
2746                                         dst_cache, info);
2747                 if (IS_ERR(ndst)) {
2748                         err = PTR_ERR(ndst);
2749                         ndst = NULL;
2750                         goto tx_error;
2751                 }
2752
2753                 if (!info) {
2754                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2755
2756                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2757                                                     dst_port, ifindex, vni,
2758                                                     ndst, rt6i_flags);
2759                         if (err)
2760                                 goto out_unlock;
2761                 }
2762
2763                 skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM);
2764
2765                 tos = ip_tunnel_ecn_encap(RT_TOS(tos), old_iph, skb);
2766                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2767                 skb_scrub_packet(skb, xnet);
2768                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2769                                       vni, md, flags, udp_sum);
2770                 if (err < 0)
2771                         goto tx_error;
2772
2773                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2774                                      &local_ip.sin6.sin6_addr,
2775                                      &dst->sin6.sin6_addr, tos, ttl,
2776                                      label, src_port, dst_port, !udp_sum);
2777 #endif
2778         }
2779 out_unlock:
2780         rcu_read_unlock();
2781         return;
2782
2783 drop:
2784         dev->stats.tx_dropped++;
2785         dev_kfree_skb(skb);
2786         return;
2787
2788 tx_error:
2789         rcu_read_unlock();
2790         if (err == -ELOOP)
2791                 dev->stats.collisions++;
2792         else if (err == -ENETUNREACH)
2793                 dev->stats.tx_carrier_errors++;
2794         dst_release(ndst);
2795         dev->stats.tx_errors++;
2796         kfree_skb(skb);
2797 }
2798
2799 static void vxlan_xmit_nh(struct sk_buff *skb, struct net_device *dev,
2800                           struct vxlan_fdb *f, __be32 vni, bool did_rsc)
2801 {
2802         struct vxlan_rdst nh_rdst;
2803         struct nexthop *nh;
2804         bool do_xmit;
2805         u32 hash;
2806
2807         memset(&nh_rdst, 0, sizeof(struct vxlan_rdst));
2808         hash = skb_get_hash(skb);
2809
2810         rcu_read_lock();
2811         nh = rcu_dereference(f->nh);
2812         if (!nh) {
2813                 rcu_read_unlock();
2814                 goto drop;
2815         }
2816         do_xmit = vxlan_fdb_nh_path_select(nh, hash, &nh_rdst);
2817         rcu_read_unlock();
2818
2819         if (likely(do_xmit))
2820                 vxlan_xmit_one(skb, dev, vni, &nh_rdst, did_rsc);
2821         else
2822                 goto drop;
2823
2824         return;
2825
2826 drop:
2827         dev->stats.tx_dropped++;
2828         dev_kfree_skb(skb);
2829 }
2830
2831 /* Transmit local packets over Vxlan
2832  *
2833  * Outer IP header inherits ECN and DF from inner header.
2834  * Outer UDP destination is the VXLAN assigned port.
2835  *           source port is based on hash of flow
2836  */
2837 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2838 {
2839         struct vxlan_dev *vxlan = netdev_priv(dev);
2840         struct vxlan_rdst *rdst, *fdst = NULL;
2841         const struct ip_tunnel_info *info;
2842         bool did_rsc = false;
2843         struct vxlan_fdb *f;
2844         struct ethhdr *eth;
2845         __be32 vni = 0;
2846
2847         info = skb_tunnel_info(skb);
2848
2849         skb_reset_mac_header(skb);
2850
2851         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2852                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2853                     info->mode & IP_TUNNEL_INFO_TX) {
2854                         vni = tunnel_id_to_key32(info->key.tun_id);
2855                 } else {
2856                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2857                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2858                         else
2859                                 kfree_skb(skb);
2860                         return NETDEV_TX_OK;
2861                 }
2862         }
2863
2864         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2865                 eth = eth_hdr(skb);
2866                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2867                         return arp_reduce(dev, skb, vni);
2868 #if IS_ENABLED(CONFIG_IPV6)
2869                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2870                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2871                                             sizeof(struct nd_msg)) &&
2872                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2873                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2874
2875                         if (m->icmph.icmp6_code == 0 &&
2876                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2877                                 return neigh_reduce(dev, skb, vni);
2878                 }
2879 #endif
2880         }
2881
2882         eth = eth_hdr(skb);
2883         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2884         did_rsc = false;
2885
2886         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2887             (ntohs(eth->h_proto) == ETH_P_IP ||
2888              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2889                 did_rsc = route_shortcircuit(dev, skb);
2890                 if (did_rsc)
2891                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2892         }
2893
2894         if (f == NULL) {
2895                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2896                 if (f == NULL) {
2897                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2898                             !is_multicast_ether_addr(eth->h_dest))
2899                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2900
2901                         dev->stats.tx_dropped++;
2902                         kfree_skb(skb);
2903                         return NETDEV_TX_OK;
2904                 }
2905         }
2906
2907         if (rcu_access_pointer(f->nh)) {
2908                 vxlan_xmit_nh(skb, dev, f,
2909                               (vni ? : vxlan->default_dst.remote_vni), did_rsc);
2910         } else {
2911                 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2912                         struct sk_buff *skb1;
2913
2914                         if (!fdst) {
2915                                 fdst = rdst;
2916                                 continue;
2917                         }
2918                         skb1 = skb_clone(skb, GFP_ATOMIC);
2919                         if (skb1)
2920                                 vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2921                 }
2922                 if (fdst)
2923                         vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2924                 else
2925                         kfree_skb(skb);
2926         }
2927
2928         return NETDEV_TX_OK;
2929 }
2930
2931 /* Walk the forwarding table and purge stale entries */
2932 static void vxlan_cleanup(struct timer_list *t)
2933 {
2934         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2935         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2936         unsigned int h;
2937
2938         if (!netif_running(vxlan->dev))
2939                 return;
2940
2941         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2942                 struct hlist_node *p, *n;
2943
2944                 spin_lock(&vxlan->hash_lock[h]);
2945                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2946                         struct vxlan_fdb *f
2947                                 = container_of(p, struct vxlan_fdb, hlist);
2948                         unsigned long timeout;
2949
2950                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2951                                 continue;
2952
2953                         if (f->flags & NTF_EXT_LEARNED)
2954                                 continue;
2955
2956                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2957                         if (time_before_eq(timeout, jiffies)) {
2958                                 netdev_dbg(vxlan->dev,
2959                                            "garbage collect %pM\n",
2960                                            f->eth_addr);
2961                                 f->state = NUD_STALE;
2962                                 vxlan_fdb_destroy(vxlan, f, true, true);
2963                         } else if (time_before(timeout, next_timer))
2964                                 next_timer = timeout;
2965                 }
2966                 spin_unlock(&vxlan->hash_lock[h]);
2967         }
2968
2969         mod_timer(&vxlan->age_timer, next_timer);
2970 }
2971
2972 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2973 {
2974         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2975
2976         spin_lock(&vn->sock_lock);
2977         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2978 #if IS_ENABLED(CONFIG_IPV6)
2979         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2980 #endif
2981         spin_unlock(&vn->sock_lock);
2982 }
2983
2984 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2985                              struct vxlan_dev_node *node)
2986 {
2987         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2988         __be32 vni = vxlan->default_dst.remote_vni;
2989
2990         node->vxlan = vxlan;
2991         spin_lock(&vn->sock_lock);
2992         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2993         spin_unlock(&vn->sock_lock);
2994 }
2995
2996 /* Setup stats when device is created */
2997 static int vxlan_init(struct net_device *dev)
2998 {
2999         struct vxlan_dev *vxlan = netdev_priv(dev);
3000         int err;
3001
3002         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
3003         if (!dev->tstats)
3004                 return -ENOMEM;
3005
3006         err = gro_cells_init(&vxlan->gro_cells, dev);
3007         if (err) {
3008                 free_percpu(dev->tstats);
3009                 return err;
3010         }
3011
3012         return 0;
3013 }
3014
3015 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
3016 {
3017         struct vxlan_fdb *f;
3018         u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, vni);
3019
3020         spin_lock_bh(&vxlan->hash_lock[hash_index]);
3021         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
3022         if (f)
3023                 vxlan_fdb_destroy(vxlan, f, true, true);
3024         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
3025 }
3026
3027 static void vxlan_uninit(struct net_device *dev)
3028 {
3029         struct vxlan_dev *vxlan = netdev_priv(dev);
3030
3031         gro_cells_destroy(&vxlan->gro_cells);
3032
3033         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
3034
3035         free_percpu(dev->tstats);
3036 }
3037
3038 /* Start ageing timer and join group when device is brought up */
3039 static int vxlan_open(struct net_device *dev)
3040 {
3041         struct vxlan_dev *vxlan = netdev_priv(dev);
3042         int ret;
3043
3044         ret = vxlan_sock_add(vxlan);
3045         if (ret < 0)
3046                 return ret;
3047
3048         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
3049                 ret = vxlan_igmp_join(vxlan);
3050                 if (ret == -EADDRINUSE)
3051                         ret = 0;
3052                 if (ret) {
3053                         vxlan_sock_release(vxlan);
3054                         return ret;
3055                 }
3056         }
3057
3058         if (vxlan->cfg.age_interval)
3059                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
3060
3061         return ret;
3062 }
3063
3064 /* Purge the forwarding table */
3065 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
3066 {
3067         unsigned int h;
3068
3069         for (h = 0; h < FDB_HASH_SIZE; ++h) {
3070                 struct hlist_node *p, *n;
3071
3072                 spin_lock_bh(&vxlan->hash_lock[h]);
3073                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
3074                         struct vxlan_fdb *f
3075                                 = container_of(p, struct vxlan_fdb, hlist);
3076                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
3077                                 continue;
3078                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
3079                         if (!is_zero_ether_addr(f->eth_addr))
3080                                 vxlan_fdb_destroy(vxlan, f, true, true);
3081                 }
3082                 spin_unlock_bh(&vxlan->hash_lock[h]);
3083         }
3084 }
3085
3086 /* Cleanup timer and forwarding table on shutdown */
3087 static int vxlan_stop(struct net_device *dev)
3088 {
3089         struct vxlan_dev *vxlan = netdev_priv(dev);
3090         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3091         int ret = 0;
3092
3093         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
3094             !vxlan_group_used(vn, vxlan))
3095                 ret = vxlan_igmp_leave(vxlan);
3096
3097         del_timer_sync(&vxlan->age_timer);
3098
3099         vxlan_flush(vxlan, false);
3100         vxlan_sock_release(vxlan);
3101
3102         return ret;
3103 }
3104
3105 /* Stub, nothing needs to be done. */
3106 static void vxlan_set_multicast_list(struct net_device *dev)
3107 {
3108 }
3109
3110 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
3111 {
3112         struct vxlan_dev *vxlan = netdev_priv(dev);
3113         struct vxlan_rdst *dst = &vxlan->default_dst;
3114         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
3115                                                          dst->remote_ifindex);
3116         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
3117
3118         /* This check is different than dev->max_mtu, because it looks at
3119          * the lowerdev->mtu, rather than the static dev->max_mtu
3120          */
3121         if (lowerdev) {
3122                 int max_mtu = lowerdev->mtu -
3123                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
3124                 if (new_mtu > max_mtu)
3125                         return -EINVAL;
3126         }
3127
3128         dev->mtu = new_mtu;
3129         return 0;
3130 }
3131
3132 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
3133 {
3134         struct vxlan_dev *vxlan = netdev_priv(dev);
3135         struct ip_tunnel_info *info = skb_tunnel_info(skb);
3136         __be16 sport, dport;
3137
3138         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
3139                                   vxlan->cfg.port_max, true);
3140         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
3141
3142         if (ip_tunnel_info_af(info) == AF_INET) {
3143                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
3144                 struct rtable *rt;
3145
3146                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
3147                                      info->key.u.ipv4.dst,
3148                                      &info->key.u.ipv4.src, dport, sport,
3149                                      &info->dst_cache, info);
3150                 if (IS_ERR(rt))
3151                         return PTR_ERR(rt);
3152                 ip_rt_put(rt);
3153         } else {
3154 #if IS_ENABLED(CONFIG_IPV6)
3155                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
3156                 struct dst_entry *ndst;
3157
3158                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
3159                                         info->key.label, &info->key.u.ipv6.dst,
3160                                         &info->key.u.ipv6.src, dport, sport,
3161                                         &info->dst_cache, info);
3162                 if (IS_ERR(ndst))
3163                         return PTR_ERR(ndst);
3164                 dst_release(ndst);
3165 #else /* !CONFIG_IPV6 */
3166                 return -EPFNOSUPPORT;
3167 #endif
3168         }
3169         info->key.tp_src = sport;
3170         info->key.tp_dst = dport;
3171         return 0;
3172 }
3173
3174 static const struct net_device_ops vxlan_netdev_ether_ops = {
3175         .ndo_init               = vxlan_init,
3176         .ndo_uninit             = vxlan_uninit,
3177         .ndo_open               = vxlan_open,
3178         .ndo_stop               = vxlan_stop,
3179         .ndo_start_xmit         = vxlan_xmit,
3180         .ndo_get_stats64        = ip_tunnel_get_stats64,
3181         .ndo_set_rx_mode        = vxlan_set_multicast_list,
3182         .ndo_change_mtu         = vxlan_change_mtu,
3183         .ndo_validate_addr      = eth_validate_addr,
3184         .ndo_set_mac_address    = eth_mac_addr,
3185         .ndo_fdb_add            = vxlan_fdb_add,
3186         .ndo_fdb_del            = vxlan_fdb_delete,
3187         .ndo_fdb_dump           = vxlan_fdb_dump,
3188         .ndo_fdb_get            = vxlan_fdb_get,
3189         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
3190         .ndo_change_proto_down  = dev_change_proto_down_generic,
3191 };
3192
3193 static const struct net_device_ops vxlan_netdev_raw_ops = {
3194         .ndo_init               = vxlan_init,
3195         .ndo_uninit             = vxlan_uninit,
3196         .ndo_open               = vxlan_open,
3197         .ndo_stop               = vxlan_stop,
3198         .ndo_start_xmit         = vxlan_xmit,
3199         .ndo_get_stats64        = ip_tunnel_get_stats64,
3200         .ndo_change_mtu         = vxlan_change_mtu,
3201         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
3202 };
3203
3204 /* Info for udev, that this is a virtual tunnel endpoint */
3205 static struct device_type vxlan_type = {
3206         .name = "vxlan",
3207 };
3208
3209 /* Calls the ndo_udp_tunnel_add of the caller in order to
3210  * supply the listening VXLAN udp ports. Callers are expected
3211  * to implement the ndo_udp_tunnel_add.
3212  */
3213 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
3214 {
3215         struct vxlan_sock *vs;
3216         struct net *net = dev_net(dev);
3217         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3218         unsigned int i;
3219
3220         spin_lock(&vn->sock_lock);
3221         for (i = 0; i < PORT_HASH_SIZE; ++i) {
3222                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
3223                         unsigned short type;
3224
3225                         if (vs->flags & VXLAN_F_GPE)
3226                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
3227                         else
3228                                 type = UDP_TUNNEL_TYPE_VXLAN;
3229
3230                         if (push)
3231                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
3232                         else
3233                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
3234                 }
3235         }
3236         spin_unlock(&vn->sock_lock);
3237 }
3238
3239 /* Initialize the device structure. */
3240 static void vxlan_setup(struct net_device *dev)
3241 {
3242         struct vxlan_dev *vxlan = netdev_priv(dev);
3243         unsigned int h;
3244
3245         eth_hw_addr_random(dev);
3246         ether_setup(dev);
3247
3248         dev->needs_free_netdev = true;
3249         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
3250
3251         dev->features   |= NETIF_F_LLTX;
3252         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
3253         dev->features   |= NETIF_F_RXCSUM;
3254         dev->features   |= NETIF_F_GSO_SOFTWARE;
3255
3256         dev->vlan_features = dev->features;
3257         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
3258         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
3259         netif_keep_dst(dev);
3260         dev->priv_flags |= IFF_NO_QUEUE;
3261
3262         /* MTU range: 68 - 65535 */
3263         dev->min_mtu = ETH_MIN_MTU;
3264         dev->max_mtu = ETH_MAX_MTU;
3265
3266         INIT_LIST_HEAD(&vxlan->next);
3267
3268         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
3269
3270         vxlan->dev = dev;
3271
3272         for (h = 0; h < FDB_HASH_SIZE; ++h) {
3273                 spin_lock_init(&vxlan->hash_lock[h]);
3274                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
3275         }
3276 }
3277
3278 static void vxlan_ether_setup(struct net_device *dev)
3279 {
3280         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
3281         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3282         dev->netdev_ops = &vxlan_netdev_ether_ops;
3283 }
3284
3285 static void vxlan_raw_setup(struct net_device *dev)
3286 {
3287         dev->header_ops = NULL;
3288         dev->type = ARPHRD_NONE;
3289         dev->hard_header_len = 0;
3290         dev->addr_len = 0;
3291         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
3292         dev->netdev_ops = &vxlan_netdev_raw_ops;
3293 }
3294
3295 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
3296         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
3297         [IFLA_VXLAN_GROUP]      = { .len = sizeof_field(struct iphdr, daddr) },
3298         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
3299         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
3300         [IFLA_VXLAN_LOCAL]      = { .len = sizeof_field(struct iphdr, saddr) },
3301         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
3302         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
3303         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
3304         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
3305         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
3306         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
3307         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
3308         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
3309         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
3310         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
3311         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
3312         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
3313         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
3314         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
3315         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
3316         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
3317         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
3318         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
3319         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
3320         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
3321         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
3322         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
3323         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
3324         [IFLA_VXLAN_DF]         = { .type = NLA_U8 },
3325 };
3326
3327 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
3328                           struct netlink_ext_ack *extack)
3329 {
3330         if (tb[IFLA_ADDRESS]) {
3331                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
3332                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3333                                             "Provided link layer address is not Ethernet");
3334                         return -EINVAL;
3335                 }
3336
3337                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
3338                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3339                                             "Provided Ethernet address is not unicast");
3340                         return -EADDRNOTAVAIL;
3341                 }
3342         }
3343
3344         if (tb[IFLA_MTU]) {
3345                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
3346
3347                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
3348                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3349                                             "MTU must be between 68 and 65535");
3350                         return -EINVAL;
3351                 }
3352         }
3353
3354         if (!data) {
3355                 NL_SET_ERR_MSG(extack,
3356                                "Required attributes not provided to perform the operation");
3357                 return -EINVAL;
3358         }
3359
3360         if (data[IFLA_VXLAN_ID]) {
3361                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
3362
3363                 if (id >= VXLAN_N_VID) {
3364                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_ID],
3365                                             "VXLAN ID must be lower than 16777216");
3366                         return -ERANGE;
3367                 }
3368         }
3369
3370         if (data[IFLA_VXLAN_PORT_RANGE]) {
3371                 const struct ifla_vxlan_port_range *p
3372                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3373
3374                 if (ntohs(p->high) < ntohs(p->low)) {
3375                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_PORT_RANGE],
3376                                             "Invalid source port range");
3377                         return -EINVAL;
3378                 }
3379         }
3380
3381         if (data[IFLA_VXLAN_DF]) {
3382                 enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]);
3383
3384                 if (df < 0 || df > VXLAN_DF_MAX) {
3385                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_DF],
3386                                             "Invalid DF attribute");
3387                         return -EINVAL;
3388                 }
3389         }
3390
3391         return 0;
3392 }
3393
3394 static void vxlan_get_drvinfo(struct net_device *netdev,
3395                               struct ethtool_drvinfo *drvinfo)
3396 {
3397         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
3398         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
3399 }
3400
3401 static int vxlan_get_link_ksettings(struct net_device *dev,
3402                                     struct ethtool_link_ksettings *cmd)
3403 {
3404         struct vxlan_dev *vxlan = netdev_priv(dev);
3405         struct vxlan_rdst *dst = &vxlan->default_dst;
3406         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
3407                                                          dst->remote_ifindex);
3408
3409         if (!lowerdev) {
3410                 cmd->base.duplex = DUPLEX_UNKNOWN;
3411                 cmd->base.port = PORT_OTHER;
3412                 cmd->base.speed = SPEED_UNKNOWN;
3413
3414                 return 0;
3415         }
3416
3417         return __ethtool_get_link_ksettings(lowerdev, cmd);
3418 }
3419
3420 static const struct ethtool_ops vxlan_ethtool_ops = {
3421         .get_drvinfo            = vxlan_get_drvinfo,
3422         .get_link               = ethtool_op_get_link,
3423         .get_link_ksettings     = vxlan_get_link_ksettings,
3424 };
3425
3426 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
3427                                         __be16 port, u32 flags, int ifindex)
3428 {
3429         struct socket *sock;
3430         struct udp_port_cfg udp_conf;
3431         int err;
3432
3433         memset(&udp_conf, 0, sizeof(udp_conf));
3434
3435         if (ipv6) {
3436                 udp_conf.family = AF_INET6;
3437                 udp_conf.use_udp6_rx_checksums =
3438                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
3439                 udp_conf.ipv6_v6only = 1;
3440         } else {
3441                 udp_conf.family = AF_INET;
3442         }
3443
3444         udp_conf.local_udp_port = port;
3445         udp_conf.bind_ifindex = ifindex;
3446
3447         /* Open UDP socket */
3448         err = udp_sock_create(net, &udp_conf, &sock);
3449         if (err < 0)
3450                 return ERR_PTR(err);
3451
3452         return sock;
3453 }
3454
3455 /* Create new listen socket if needed */
3456 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
3457                                               __be16 port, u32 flags,
3458                                               int ifindex)
3459 {
3460         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3461         struct vxlan_sock *vs;
3462         struct socket *sock;
3463         unsigned int h;
3464         struct udp_tunnel_sock_cfg tunnel_cfg;
3465
3466         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
3467         if (!vs)
3468                 return ERR_PTR(-ENOMEM);
3469
3470         for (h = 0; h < VNI_HASH_SIZE; ++h)
3471                 INIT_HLIST_HEAD(&vs->vni_list[h]);
3472
3473         sock = vxlan_create_sock(net, ipv6, port, flags, ifindex);
3474         if (IS_ERR(sock)) {
3475                 kfree(vs);
3476                 return ERR_CAST(sock);
3477         }
3478
3479         vs->sock = sock;
3480         refcount_set(&vs->refcnt, 1);
3481         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
3482
3483         spin_lock(&vn->sock_lock);
3484         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
3485         udp_tunnel_notify_add_rx_port(sock,
3486                                       (vs->flags & VXLAN_F_GPE) ?
3487                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
3488                                       UDP_TUNNEL_TYPE_VXLAN);
3489         spin_unlock(&vn->sock_lock);
3490
3491         /* Mark socket as an encapsulation socket. */
3492         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
3493         tunnel_cfg.sk_user_data = vs;
3494         tunnel_cfg.encap_type = 1;
3495         tunnel_cfg.encap_rcv = vxlan_rcv;
3496         tunnel_cfg.encap_err_lookup = vxlan_err_lookup;
3497         tunnel_cfg.encap_destroy = NULL;
3498         tunnel_cfg.gro_receive = vxlan_gro_receive;
3499         tunnel_cfg.gro_complete = vxlan_gro_complete;
3500
3501         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
3502
3503         return vs;
3504 }
3505
3506 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
3507 {
3508         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3509         struct vxlan_sock *vs = NULL;
3510         struct vxlan_dev_node *node;
3511         int l3mdev_index = 0;
3512
3513         if (vxlan->cfg.remote_ifindex)
3514                 l3mdev_index = l3mdev_master_upper_ifindex_by_index(
3515                         vxlan->net, vxlan->cfg.remote_ifindex);
3516
3517         if (!vxlan->cfg.no_share) {
3518                 spin_lock(&vn->sock_lock);
3519                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
3520                                      vxlan->cfg.dst_port, vxlan->cfg.flags,
3521                                      l3mdev_index);
3522                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
3523                         spin_unlock(&vn->sock_lock);
3524                         return -EBUSY;
3525                 }
3526                 spin_unlock(&vn->sock_lock);
3527         }
3528         if (!vs)
3529                 vs = vxlan_socket_create(vxlan->net, ipv6,
3530                                          vxlan->cfg.dst_port, vxlan->cfg.flags,
3531                                          l3mdev_index);
3532         if (IS_ERR(vs))
3533                 return PTR_ERR(vs);
3534 #if IS_ENABLED(CONFIG_IPV6)
3535         if (ipv6) {
3536                 rcu_assign_pointer(vxlan->vn6_sock, vs);
3537                 node = &vxlan->hlist6;
3538         } else
3539 #endif
3540         {
3541                 rcu_assign_pointer(vxlan->vn4_sock, vs);
3542                 node = &vxlan->hlist4;
3543         }
3544         vxlan_vs_add_dev(vs, vxlan, node);
3545         return 0;
3546 }
3547
3548 static int vxlan_sock_add(struct vxlan_dev *vxlan)
3549 {
3550         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
3551         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
3552         bool ipv4 = !ipv6 || metadata;
3553         int ret = 0;
3554
3555         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
3556 #if IS_ENABLED(CONFIG_IPV6)
3557         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
3558         if (ipv6) {
3559                 ret = __vxlan_sock_add(vxlan, true);
3560                 if (ret < 0 && ret != -EAFNOSUPPORT)
3561                         ipv4 = false;
3562         }
3563 #endif
3564         if (ipv4)
3565                 ret = __vxlan_sock_add(vxlan, false);
3566         if (ret < 0)
3567                 vxlan_sock_release(vxlan);
3568         return ret;
3569 }
3570
3571 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
3572                                  struct net_device **lower,
3573                                  struct vxlan_dev *old,
3574                                  struct netlink_ext_ack *extack)
3575 {
3576         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
3577         struct vxlan_dev *tmp;
3578         bool use_ipv6 = false;
3579
3580         if (conf->flags & VXLAN_F_GPE) {
3581                 /* For now, allow GPE only together with
3582                  * COLLECT_METADATA. This can be relaxed later; in such
3583                  * case, the other side of the PtP link will have to be
3584                  * provided.
3585                  */
3586                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
3587                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3588                         NL_SET_ERR_MSG(extack,
3589                                        "VXLAN GPE does not support this combination of attributes");
3590                         return -EINVAL;
3591                 }
3592         }
3593
3594         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
3595                 /* Unless IPv6 is explicitly requested, assume IPv4 */
3596                 conf->remote_ip.sa.sa_family = AF_INET;
3597                 conf->saddr.sa.sa_family = AF_INET;
3598         } else if (!conf->remote_ip.sa.sa_family) {
3599                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
3600         } else if (!conf->saddr.sa.sa_family) {
3601                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
3602         }
3603
3604         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
3605                 NL_SET_ERR_MSG(extack,
3606                                "Local and remote address must be from the same family");
3607                 return -EINVAL;
3608         }
3609
3610         if (vxlan_addr_multicast(&conf->saddr)) {
3611                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
3612                 return -EINVAL;
3613         }
3614
3615         if (conf->saddr.sa.sa_family == AF_INET6) {
3616                 if (!IS_ENABLED(CONFIG_IPV6)) {
3617                         NL_SET_ERR_MSG(extack,
3618                                        "IPv6 support not enabled in the kernel");
3619                         return -EPFNOSUPPORT;
3620                 }
3621                 use_ipv6 = true;
3622                 conf->flags |= VXLAN_F_IPV6;
3623
3624                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3625                         int local_type =
3626                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3627                         int remote_type =
3628                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3629
3630                         if (local_type & IPV6_ADDR_LINKLOCAL) {
3631                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3632                                     (remote_type != IPV6_ADDR_ANY)) {
3633                                         NL_SET_ERR_MSG(extack,
3634                                                        "Invalid combination of local and remote address scopes");
3635                                         return -EINVAL;
3636                                 }
3637
3638                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3639                         } else {
3640                                 if (remote_type ==
3641                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3642                                         NL_SET_ERR_MSG(extack,
3643                                                        "Invalid combination of local and remote address scopes");
3644                                         return -EINVAL;
3645                                 }
3646
3647                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3648                         }
3649                 }
3650         }
3651
3652         if (conf->label && !use_ipv6) {
3653                 NL_SET_ERR_MSG(extack,
3654                                "Label attribute only applies to IPv6 VXLAN devices");
3655                 return -EINVAL;
3656         }
3657
3658         if (conf->remote_ifindex) {
3659                 struct net_device *lowerdev;
3660
3661                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3662                 if (!lowerdev) {
3663                         NL_SET_ERR_MSG(extack,
3664                                        "Invalid local interface, device not found");
3665                         return -ENODEV;
3666                 }
3667
3668 #if IS_ENABLED(CONFIG_IPV6)
3669                 if (use_ipv6) {
3670                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3671                         if (idev && idev->cnf.disable_ipv6) {
3672                                 NL_SET_ERR_MSG(extack,
3673                                                "IPv6 support disabled by administrator");
3674                                 return -EPERM;
3675                         }
3676                 }
3677 #endif
3678
3679                 *lower = lowerdev;
3680         } else {
3681                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3682                         NL_SET_ERR_MSG(extack,
3683                                        "Local interface required for multicast remote destination");
3684
3685                         return -EINVAL;
3686                 }
3687
3688 #if IS_ENABLED(CONFIG_IPV6)
3689                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3690                         NL_SET_ERR_MSG(extack,
3691                                        "Local interface required for link-local local/remote addresses");
3692                         return -EINVAL;
3693                 }
3694 #endif
3695
3696                 *lower = NULL;
3697         }
3698
3699         if (!conf->dst_port) {
3700                 if (conf->flags & VXLAN_F_GPE)
3701                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3702                 else
3703                         conf->dst_port = htons(vxlan_port);
3704         }
3705
3706         if (!conf->age_interval)
3707                 conf->age_interval = FDB_AGE_DEFAULT;
3708
3709         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3710                 if (tmp == old)
3711                         continue;
3712
3713                 if (tmp->cfg.vni != conf->vni)
3714                         continue;
3715                 if (tmp->cfg.dst_port != conf->dst_port)
3716                         continue;
3717                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3718                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3719                         continue;
3720
3721                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3722                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3723                         continue;
3724
3725                 NL_SET_ERR_MSG(extack,
3726                                "A VXLAN device with the specified VNI already exists");
3727                 return -EEXIST;
3728         }
3729
3730         return 0;
3731 }
3732
3733 static void vxlan_config_apply(struct net_device *dev,
3734                                struct vxlan_config *conf,
3735                                struct net_device *lowerdev,
3736                                struct net *src_net,
3737                                bool changelink)
3738 {
3739         struct vxlan_dev *vxlan = netdev_priv(dev);
3740         struct vxlan_rdst *dst = &vxlan->default_dst;
3741         unsigned short needed_headroom = ETH_HLEN;
3742         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3743         int max_mtu = ETH_MAX_MTU;
3744
3745         if (!changelink) {
3746                 if (conf->flags & VXLAN_F_GPE)
3747                         vxlan_raw_setup(dev);
3748                 else
3749                         vxlan_ether_setup(dev);
3750
3751                 if (conf->mtu)
3752                         dev->mtu = conf->mtu;
3753
3754                 vxlan->net = src_net;
3755         }
3756
3757         dst->remote_vni = conf->vni;
3758
3759         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3760
3761         if (lowerdev) {
3762                 dst->remote_ifindex = conf->remote_ifindex;
3763
3764                 dev->gso_max_size = lowerdev->gso_max_size;
3765                 dev->gso_max_segs = lowerdev->gso_max_segs;
3766
3767                 needed_headroom = lowerdev->hard_header_len;
3768
3769                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3770                                            VXLAN_HEADROOM);
3771                 if (max_mtu < ETH_MIN_MTU)
3772                         max_mtu = ETH_MIN_MTU;
3773
3774                 if (!changelink && !conf->mtu)
3775                         dev->mtu = max_mtu;
3776         }
3777
3778         if (dev->mtu > max_mtu)
3779                 dev->mtu = max_mtu;
3780
3781         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3782                 needed_headroom += VXLAN6_HEADROOM;
3783         else
3784                 needed_headroom += VXLAN_HEADROOM;
3785         dev->needed_headroom = needed_headroom;
3786
3787         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3788 }
3789
3790 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3791                                struct vxlan_config *conf, bool changelink,
3792                                struct netlink_ext_ack *extack)
3793 {
3794         struct vxlan_dev *vxlan = netdev_priv(dev);
3795         struct net_device *lowerdev;
3796         int ret;
3797
3798         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3799         if (ret)
3800                 return ret;
3801
3802         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3803
3804         return 0;
3805 }
3806
3807 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3808                               struct vxlan_config *conf,
3809                               struct netlink_ext_ack *extack)
3810 {
3811         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3812         struct vxlan_dev *vxlan = netdev_priv(dev);
3813         struct net_device *remote_dev = NULL;
3814         struct vxlan_fdb *f = NULL;
3815         bool unregister = false;
3816         struct vxlan_rdst *dst;
3817         int err;
3818
3819         dst = &vxlan->default_dst;
3820         err = vxlan_dev_configure(net, dev, conf, false, extack);
3821         if (err)
3822                 return err;
3823
3824         dev->ethtool_ops = &vxlan_ethtool_ops;
3825
3826         /* create an fdb entry for a valid default destination */
3827         if (!vxlan_addr_any(&dst->remote_ip)) {
3828                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3829                                        &dst->remote_ip,
3830                                        NUD_REACHABLE | NUD_PERMANENT,
3831                                        vxlan->cfg.dst_port,
3832                                        dst->remote_vni,
3833                                        dst->remote_vni,
3834                                        dst->remote_ifindex,
3835                                        NTF_SELF, 0, &f, extack);
3836                 if (err)
3837                         return err;
3838         }
3839
3840         err = register_netdevice(dev);
3841         if (err)
3842                 goto errout;
3843         unregister = true;
3844
3845         if (dst->remote_ifindex) {
3846                 remote_dev = __dev_get_by_index(net, dst->remote_ifindex);
3847                 if (!remote_dev)
3848                         goto errout;
3849
3850                 err = netdev_upper_dev_link(remote_dev, dev, extack);
3851                 if (err)
3852                         goto errout;
3853         }
3854
3855         err = rtnl_configure_link(dev, NULL);
3856         if (err)
3857                 goto unlink;
3858
3859         if (f) {
3860                 vxlan_fdb_insert(vxlan, all_zeros_mac, dst->remote_vni, f);
3861
3862                 /* notify default fdb entry */
3863                 err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f),
3864                                        RTM_NEWNEIGH, true, extack);
3865                 if (err) {
3866                         vxlan_fdb_destroy(vxlan, f, false, false);
3867                         if (remote_dev)
3868                                 netdev_upper_dev_unlink(remote_dev, dev);
3869                         goto unregister;
3870                 }
3871         }
3872
3873         list_add(&vxlan->next, &vn->vxlan_list);
3874         if (remote_dev)
3875                 dst->remote_dev = remote_dev;
3876         return 0;
3877 unlink:
3878         if (remote_dev)
3879                 netdev_upper_dev_unlink(remote_dev, dev);
3880 errout:
3881         /* unregister_netdevice() destroys the default FDB entry with deletion
3882          * notification. But the addition notification was not sent yet, so
3883          * destroy the entry by hand here.
3884          */
3885         if (f)
3886                 __vxlan_fdb_free(f);
3887 unregister:
3888         if (unregister)
3889                 unregister_netdevice(dev);
3890         return err;
3891 }
3892
3893 /* Set/clear flags based on attribute */
3894 static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[],
3895                           int attrtype, unsigned long mask, bool changelink,
3896                           bool changelink_supported,
3897                           struct netlink_ext_ack *extack)
3898 {
3899         unsigned long flags;
3900
3901         if (!tb[attrtype])
3902                 return 0;
3903
3904         if (changelink && !changelink_supported) {
3905                 vxlan_flag_attr_error(attrtype, extack);
3906                 return -EOPNOTSUPP;
3907         }
3908
3909         if (vxlan_policy[attrtype].type == NLA_FLAG)
3910                 flags = conf->flags | mask;
3911         else if (nla_get_u8(tb[attrtype]))
3912                 flags = conf->flags | mask;
3913         else
3914                 flags = conf->flags & ~mask;
3915
3916         conf->flags = flags;
3917
3918         return 0;
3919 }
3920
3921 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3922                          struct net_device *dev, struct vxlan_config *conf,
3923                          bool changelink, struct netlink_ext_ack *extack)
3924 {
3925         struct vxlan_dev *vxlan = netdev_priv(dev);
3926         int err = 0;
3927
3928         memset(conf, 0, sizeof(*conf));
3929
3930         /* if changelink operation, start with old existing cfg */
3931         if (changelink)
3932                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3933
3934         if (data[IFLA_VXLAN_ID]) {
3935                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3936
3937                 if (changelink && (vni != conf->vni)) {
3938                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI");
3939                         return -EOPNOTSUPP;
3940                 }
3941                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3942         }
3943
3944         if (data[IFLA_VXLAN_GROUP]) {
3945                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) {
3946                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group");
3947                         return -EOPNOTSUPP;
3948                 }
3949
3950                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3951                 conf->remote_ip.sa.sa_family = AF_INET;
3952         } else if (data[IFLA_VXLAN_GROUP6]) {
3953                 if (!IS_ENABLED(CONFIG_IPV6)) {
3954                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel");
3955                         return -EPFNOSUPPORT;
3956                 }
3957
3958                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) {
3959                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group");
3960                         return -EOPNOTSUPP;
3961                 }
3962
3963                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3964                 conf->remote_ip.sa.sa_family = AF_INET6;
3965         }
3966
3967         if (data[IFLA_VXLAN_LOCAL]) {
3968                 if (changelink && (conf->saddr.sa.sa_family != AF_INET)) {
3969                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old");
3970                         return -EOPNOTSUPP;
3971                 }
3972
3973                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3974                 conf->saddr.sa.sa_family = AF_INET;
3975         } else if (data[IFLA_VXLAN_LOCAL6]) {
3976                 if (!IS_ENABLED(CONFIG_IPV6)) {
3977                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel");
3978                         return -EPFNOSUPPORT;
3979                 }
3980
3981                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) {
3982                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old");
3983                         return -EOPNOTSUPP;
3984                 }
3985
3986                 /* TODO: respect scope id */
3987                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3988                 conf->saddr.sa.sa_family = AF_INET6;
3989         }
3990
3991         if (data[IFLA_VXLAN_LINK])
3992                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3993
3994         if (data[IFLA_VXLAN_TOS])
3995                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3996
3997         if (data[IFLA_VXLAN_TTL])
3998                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3999
4000         if (data[IFLA_VXLAN_TTL_INHERIT]) {
4001                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT,
4002                                     VXLAN_F_TTL_INHERIT, changelink, false,
4003                                     extack);
4004                 if (err)
4005                         return err;
4006
4007         }
4008
4009         if (data[IFLA_VXLAN_LABEL])
4010                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
4011                              IPV6_FLOWLABEL_MASK;
4012
4013         if (data[IFLA_VXLAN_LEARNING]) {
4014                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING,
4015                                     VXLAN_F_LEARN, changelink, true,
4016                                     extack);
4017                 if (err)
4018                         return err;
4019         } else if (!changelink) {
4020                 /* default to learn on a new device */
4021                 conf->flags |= VXLAN_F_LEARN;
4022         }
4023
4024         if (data[IFLA_VXLAN_AGEING])
4025                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
4026
4027         if (data[IFLA_VXLAN_PROXY]) {
4028                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY,
4029                                     VXLAN_F_PROXY, changelink, false,
4030                                     extack);
4031                 if (err)
4032                         return err;
4033         }
4034
4035         if (data[IFLA_VXLAN_RSC]) {
4036                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC,
4037                                     VXLAN_F_RSC, changelink, false,
4038                                     extack);
4039                 if (err)
4040                         return err;
4041         }
4042
4043         if (data[IFLA_VXLAN_L2MISS]) {
4044                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS,
4045                                     VXLAN_F_L2MISS, changelink, false,
4046                                     extack);
4047                 if (err)
4048                         return err;
4049         }
4050
4051         if (data[IFLA_VXLAN_L3MISS]) {
4052                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS,
4053                                     VXLAN_F_L3MISS, changelink, false,
4054                                     extack);
4055                 if (err)
4056                         return err;
4057         }
4058
4059         if (data[IFLA_VXLAN_LIMIT]) {
4060                 if (changelink) {
4061                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT],
4062                                             "Cannot change limit");
4063                         return -EOPNOTSUPP;
4064                 }
4065                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
4066         }
4067
4068         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
4069                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA,
4070                                     VXLAN_F_COLLECT_METADATA, changelink, false,
4071                                     extack);
4072                 if (err)
4073                         return err;
4074         }
4075
4076         if (data[IFLA_VXLAN_PORT_RANGE]) {
4077                 if (!changelink) {
4078                         const struct ifla_vxlan_port_range *p
4079                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
4080                         conf->port_min = ntohs(p->low);
4081                         conf->port_max = ntohs(p->high);
4082                 } else {
4083                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
4084                                             "Cannot change port range");
4085                         return -EOPNOTSUPP;
4086                 }
4087         }
4088
4089         if (data[IFLA_VXLAN_PORT]) {
4090                 if (changelink) {
4091                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT],
4092                                             "Cannot change port");
4093                         return -EOPNOTSUPP;
4094                 }
4095                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
4096         }
4097
4098         if (data[IFLA_VXLAN_UDP_CSUM]) {
4099                 if (changelink) {
4100                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM],
4101                                             "Cannot change UDP_CSUM flag");
4102                         return -EOPNOTSUPP;
4103                 }
4104                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
4105                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
4106         }
4107
4108         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
4109                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4110                                     VXLAN_F_UDP_ZERO_CSUM6_TX, changelink,
4111                                     false, extack);
4112                 if (err)
4113                         return err;
4114         }
4115
4116         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
4117                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4118                                     VXLAN_F_UDP_ZERO_CSUM6_RX, changelink,
4119                                     false, extack);
4120                 if (err)
4121                         return err;
4122         }
4123
4124         if (data[IFLA_VXLAN_REMCSUM_TX]) {
4125                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX,
4126                                     VXLAN_F_REMCSUM_TX, changelink, false,
4127                                     extack);
4128                 if (err)
4129                         return err;
4130         }
4131
4132         if (data[IFLA_VXLAN_REMCSUM_RX]) {
4133                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX,
4134                                     VXLAN_F_REMCSUM_RX, changelink, false,
4135                                     extack);
4136                 if (err)
4137                         return err;
4138         }
4139
4140         if (data[IFLA_VXLAN_GBP]) {
4141                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP,
4142                                     VXLAN_F_GBP, changelink, false, extack);
4143                 if (err)
4144                         return err;
4145         }
4146
4147         if (data[IFLA_VXLAN_GPE]) {
4148                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE,
4149                                     VXLAN_F_GPE, changelink, false,
4150                                     extack);
4151                 if (err)
4152                         return err;
4153         }
4154
4155         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
4156                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL,
4157                                     VXLAN_F_REMCSUM_NOPARTIAL, changelink,
4158                                     false, extack);
4159                 if (err)
4160                         return err;
4161         }
4162
4163         if (tb[IFLA_MTU]) {
4164                 if (changelink) {
4165                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
4166                                             "Cannot change mtu");
4167                         return -EOPNOTSUPP;
4168                 }
4169                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
4170         }
4171
4172         if (data[IFLA_VXLAN_DF])
4173                 conf->df = nla_get_u8(data[IFLA_VXLAN_DF]);
4174
4175         return 0;
4176 }
4177
4178 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
4179                          struct nlattr *tb[], struct nlattr *data[],
4180                          struct netlink_ext_ack *extack)
4181 {
4182         struct vxlan_config conf;
4183         int err;
4184
4185         err = vxlan_nl2conf(tb, data, dev, &conf, false, extack);
4186         if (err)
4187                 return err;
4188
4189         return __vxlan_dev_create(src_net, dev, &conf, extack);
4190 }
4191
4192 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
4193                             struct nlattr *data[],
4194                             struct netlink_ext_ack *extack)
4195 {
4196         struct vxlan_dev *vxlan = netdev_priv(dev);
4197         struct net_device *lowerdev;
4198         struct vxlan_config conf;
4199         struct vxlan_rdst *dst;
4200         int err;
4201
4202         dst = &vxlan->default_dst;
4203         err = vxlan_nl2conf(tb, data, dev, &conf, true, extack);
4204         if (err)
4205                 return err;
4206
4207         err = vxlan_config_validate(vxlan->net, &conf, &lowerdev,
4208                                     vxlan, extack);
4209         if (err)
4210                 return err;
4211
4212         if (dst->remote_dev == lowerdev)
4213                 lowerdev = NULL;
4214
4215         err = netdev_adjacent_change_prepare(dst->remote_dev, lowerdev, dev,
4216                                              extack);
4217         if (err)
4218                 return err;
4219
4220         /* handle default dst entry */
4221         if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) {
4222                 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, conf.vni);
4223
4224                 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4225                 if (!vxlan_addr_any(&conf.remote_ip)) {
4226                         err = vxlan_fdb_update(vxlan, all_zeros_mac,
4227                                                &conf.remote_ip,
4228                                                NUD_REACHABLE | NUD_PERMANENT,
4229                                                NLM_F_APPEND | NLM_F_CREATE,
4230                                                vxlan->cfg.dst_port,
4231                                                conf.vni, conf.vni,
4232                                                conf.remote_ifindex,
4233                                                NTF_SELF, 0, true, extack);
4234                         if (err) {
4235                                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4236                                 netdev_adjacent_change_abort(dst->remote_dev,
4237                                                              lowerdev, dev);
4238                                 return err;
4239                         }
4240                 }
4241                 if (!vxlan_addr_any(&dst->remote_ip))
4242                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
4243                                            dst->remote_ip,
4244                                            vxlan->cfg.dst_port,
4245                                            dst->remote_vni,
4246                                            dst->remote_vni,
4247                                            dst->remote_ifindex,
4248                                            true);
4249                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4250         }
4251
4252         if (conf.age_interval != vxlan->cfg.age_interval)
4253                 mod_timer(&vxlan->age_timer, jiffies);
4254
4255         netdev_adjacent_change_commit(dst->remote_dev, lowerdev, dev);
4256         if (lowerdev && lowerdev != dst->remote_dev)
4257                 dst->remote_dev = lowerdev;
4258         vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true);
4259         return 0;
4260 }
4261
4262 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
4263 {
4264         struct vxlan_dev *vxlan = netdev_priv(dev);
4265
4266         vxlan_flush(vxlan, true);
4267
4268         list_del(&vxlan->next);
4269         unregister_netdevice_queue(dev, head);
4270         if (vxlan->default_dst.remote_dev)
4271                 netdev_upper_dev_unlink(vxlan->default_dst.remote_dev, dev);
4272 }
4273
4274 static size_t vxlan_get_size(const struct net_device *dev)
4275 {
4276
4277         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
4278                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
4279                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
4280                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
4281                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
4282                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL_INHERIT */
4283                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
4284                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_DF */
4285                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
4286                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
4287                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
4288                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
4289                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
4290                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
4291                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
4292                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
4293                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
4294                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
4295                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
4296                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
4297                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
4298                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
4299                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
4300                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
4301                 0;
4302 }
4303
4304 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
4305 {
4306         const struct vxlan_dev *vxlan = netdev_priv(dev);
4307         const struct vxlan_rdst *dst = &vxlan->default_dst;
4308         struct ifla_vxlan_port_range ports = {
4309                 .low =  htons(vxlan->cfg.port_min),
4310                 .high = htons(vxlan->cfg.port_max),
4311         };
4312
4313         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
4314                 goto nla_put_failure;
4315
4316         if (!vxlan_addr_any(&dst->remote_ip)) {
4317                 if (dst->remote_ip.sa.sa_family == AF_INET) {
4318                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
4319                                             dst->remote_ip.sin.sin_addr.s_addr))
4320                                 goto nla_put_failure;
4321 #if IS_ENABLED(CONFIG_IPV6)
4322                 } else {
4323                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
4324                                              &dst->remote_ip.sin6.sin6_addr))
4325                                 goto nla_put_failure;
4326 #endif
4327                 }
4328         }
4329
4330         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
4331                 goto nla_put_failure;
4332
4333         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
4334                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
4335                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
4336                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
4337                                 goto nla_put_failure;
4338 #if IS_ENABLED(CONFIG_IPV6)
4339                 } else {
4340                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
4341                                              &vxlan->cfg.saddr.sin6.sin6_addr))
4342                                 goto nla_put_failure;
4343 #endif
4344                 }
4345         }
4346
4347         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
4348             nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT,
4349                        !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) ||
4350             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
4351             nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) ||
4352             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
4353             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
4354                        !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
4355             nla_put_u8(skb, IFLA_VXLAN_PROXY,
4356                        !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
4357             nla_put_u8(skb, IFLA_VXLAN_RSC,
4358                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
4359             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
4360                        !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
4361             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
4362                        !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
4363             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
4364                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
4365             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
4366             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
4367             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
4368             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
4369                        !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
4370             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4371                        !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
4372             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4373                        !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
4374             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
4375                        !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
4376             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
4377                        !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
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         return 0;
4396
4397 nla_put_failure:
4398         return -EMSGSIZE;
4399 }
4400
4401 static struct net *vxlan_get_link_net(const struct net_device *dev)
4402 {
4403         struct vxlan_dev *vxlan = netdev_priv(dev);
4404
4405         return vxlan->net;
4406 }
4407
4408 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
4409         .kind           = "vxlan",
4410         .maxtype        = IFLA_VXLAN_MAX,
4411         .policy         = vxlan_policy,
4412         .priv_size      = sizeof(struct vxlan_dev),
4413         .setup          = vxlan_setup,
4414         .validate       = vxlan_validate,
4415         .newlink        = vxlan_newlink,
4416         .changelink     = vxlan_changelink,
4417         .dellink        = vxlan_dellink,
4418         .get_size       = vxlan_get_size,
4419         .fill_info      = vxlan_fill_info,
4420         .get_link_net   = vxlan_get_link_net,
4421 };
4422
4423 struct net_device *vxlan_dev_create(struct net *net, const char *name,
4424                                     u8 name_assign_type,
4425                                     struct vxlan_config *conf)
4426 {
4427         struct nlattr *tb[IFLA_MAX + 1];
4428         struct net_device *dev;
4429         int err;
4430
4431         memset(&tb, 0, sizeof(tb));
4432
4433         dev = rtnl_create_link(net, name, name_assign_type,
4434                                &vxlan_link_ops, tb, NULL);
4435         if (IS_ERR(dev))
4436                 return dev;
4437
4438         err = __vxlan_dev_create(net, dev, conf, NULL);
4439         if (err < 0) {
4440                 free_netdev(dev);
4441                 return ERR_PTR(err);
4442         }
4443
4444         err = rtnl_configure_link(dev, NULL);
4445         if (err < 0) {
4446                 LIST_HEAD(list_kill);
4447
4448                 vxlan_dellink(dev, &list_kill);
4449                 unregister_netdevice_many(&list_kill);
4450                 return ERR_PTR(err);
4451         }
4452
4453         return dev;
4454 }
4455 EXPORT_SYMBOL_GPL(vxlan_dev_create);
4456
4457 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
4458                                              struct net_device *dev)
4459 {
4460         struct vxlan_dev *vxlan, *next;
4461         LIST_HEAD(list_kill);
4462
4463         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4464                 struct vxlan_rdst *dst = &vxlan->default_dst;
4465
4466                 /* In case we created vxlan device with carrier
4467                  * and we loose the carrier due to module unload
4468                  * we also need to remove vxlan device. In other
4469                  * cases, it's not necessary and remote_ifindex
4470                  * is 0 here, so no matches.
4471                  */
4472                 if (dst->remote_ifindex == dev->ifindex)
4473                         vxlan_dellink(vxlan->dev, &list_kill);
4474         }
4475
4476         unregister_netdevice_many(&list_kill);
4477 }
4478
4479 static int vxlan_netdevice_event(struct notifier_block *unused,
4480                                  unsigned long event, void *ptr)
4481 {
4482         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4483         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
4484
4485         if (event == NETDEV_UNREGISTER) {
4486                 vxlan_offload_rx_ports(dev, false);
4487                 vxlan_handle_lowerdev_unregister(vn, dev);
4488         } else if (event == NETDEV_REGISTER) {
4489                 vxlan_offload_rx_ports(dev, true);
4490         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
4491                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
4492                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
4493         }
4494
4495         return NOTIFY_DONE;
4496 }
4497
4498 static struct notifier_block vxlan_notifier_block __read_mostly = {
4499         .notifier_call = vxlan_netdevice_event,
4500 };
4501
4502 static void
4503 vxlan_fdb_offloaded_set(struct net_device *dev,
4504                         struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4505 {
4506         struct vxlan_dev *vxlan = netdev_priv(dev);
4507         struct vxlan_rdst *rdst;
4508         struct vxlan_fdb *f;
4509         u32 hash_index;
4510
4511         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4512
4513         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4514
4515         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4516         if (!f)
4517                 goto out;
4518
4519         rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip,
4520                                    fdb_info->remote_port,
4521                                    fdb_info->remote_vni,
4522                                    fdb_info->remote_ifindex);
4523         if (!rdst)
4524                 goto out;
4525
4526         rdst->offloaded = fdb_info->offloaded;
4527
4528 out:
4529         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4530 }
4531
4532 static int
4533 vxlan_fdb_external_learn_add(struct net_device *dev,
4534                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4535 {
4536         struct vxlan_dev *vxlan = netdev_priv(dev);
4537         struct netlink_ext_ack *extack;
4538         u32 hash_index;
4539         int err;
4540
4541         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4542         extack = switchdev_notifier_info_to_extack(&fdb_info->info);
4543
4544         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4545         err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip,
4546                                NUD_REACHABLE,
4547                                NLM_F_CREATE | NLM_F_REPLACE,
4548                                fdb_info->remote_port,
4549                                fdb_info->vni,
4550                                fdb_info->remote_vni,
4551                                fdb_info->remote_ifindex,
4552                                NTF_USE | NTF_SELF | NTF_EXT_LEARNED,
4553                                0, false, extack);
4554         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4555
4556         return err;
4557 }
4558
4559 static int
4560 vxlan_fdb_external_learn_del(struct net_device *dev,
4561                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4562 {
4563         struct vxlan_dev *vxlan = netdev_priv(dev);
4564         struct vxlan_fdb *f;
4565         u32 hash_index;
4566         int err = 0;
4567
4568         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4569         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4570
4571         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4572         if (!f)
4573                 err = -ENOENT;
4574         else if (f->flags & NTF_EXT_LEARNED)
4575                 err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr,
4576                                          fdb_info->remote_ip,
4577                                          fdb_info->remote_port,
4578                                          fdb_info->vni,
4579                                          fdb_info->remote_vni,
4580                                          fdb_info->remote_ifindex,
4581                                          false);
4582
4583         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4584
4585         return err;
4586 }
4587
4588 static int vxlan_switchdev_event(struct notifier_block *unused,
4589                                  unsigned long event, void *ptr)
4590 {
4591         struct net_device *dev = switchdev_notifier_info_to_dev(ptr);
4592         struct switchdev_notifier_vxlan_fdb_info *fdb_info;
4593         int err = 0;
4594
4595         switch (event) {
4596         case SWITCHDEV_VXLAN_FDB_OFFLOADED:
4597                 vxlan_fdb_offloaded_set(dev, ptr);
4598                 break;
4599         case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE:
4600                 fdb_info = ptr;
4601                 err = vxlan_fdb_external_learn_add(dev, fdb_info);
4602                 if (err) {
4603                         err = notifier_from_errno(err);
4604                         break;
4605                 }
4606                 fdb_info->offloaded = true;
4607                 vxlan_fdb_offloaded_set(dev, fdb_info);
4608                 break;
4609         case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE:
4610                 fdb_info = ptr;
4611                 err = vxlan_fdb_external_learn_del(dev, fdb_info);
4612                 if (err) {
4613                         err = notifier_from_errno(err);
4614                         break;
4615                 }
4616                 fdb_info->offloaded = false;
4617                 vxlan_fdb_offloaded_set(dev, fdb_info);
4618                 break;
4619         }
4620
4621         return err;
4622 }
4623
4624 static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = {
4625         .notifier_call = vxlan_switchdev_event,
4626 };
4627
4628 static void vxlan_fdb_nh_flush(struct nexthop *nh)
4629 {
4630         struct vxlan_fdb *fdb;
4631         struct vxlan_dev *vxlan;
4632         u32 hash_index;
4633
4634         rcu_read_lock();
4635         list_for_each_entry_rcu(fdb, &nh->fdb_list, nh_list) {
4636                 vxlan = rcu_dereference(fdb->vdev);
4637                 WARN_ON(!vxlan);
4638                 hash_index = fdb_head_index(vxlan, fdb->eth_addr,
4639                                             vxlan->default_dst.remote_vni);
4640                 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4641                 if (!hlist_unhashed(&fdb->hlist))
4642                         vxlan_fdb_destroy(vxlan, fdb, false, false);
4643                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4644         }
4645         rcu_read_unlock();
4646 }
4647
4648 static int vxlan_nexthop_event(struct notifier_block *nb,
4649                                unsigned long event, void *ptr)
4650 {
4651         struct nexthop *nh = ptr;
4652
4653         if (!nh || event != NEXTHOP_EVENT_DEL)
4654                 return NOTIFY_DONE;
4655
4656         vxlan_fdb_nh_flush(nh);
4657
4658         return NOTIFY_DONE;
4659 }
4660
4661 static struct notifier_block vxlan_nexthop_notifier_block __read_mostly = {
4662         .notifier_call = vxlan_nexthop_event,
4663 };
4664
4665 static __net_init int vxlan_init_net(struct net *net)
4666 {
4667         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4668         unsigned int h;
4669
4670         INIT_LIST_HEAD(&vn->vxlan_list);
4671         spin_lock_init(&vn->sock_lock);
4672
4673         for (h = 0; h < PORT_HASH_SIZE; ++h)
4674                 INIT_HLIST_HEAD(&vn->sock_list[h]);
4675
4676         return register_nexthop_notifier(net, &vxlan_nexthop_notifier_block);
4677 }
4678
4679 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
4680 {
4681         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4682         struct vxlan_dev *vxlan, *next;
4683         struct net_device *dev, *aux;
4684         unsigned int h;
4685
4686         for_each_netdev_safe(net, dev, aux)
4687                 if (dev->rtnl_link_ops == &vxlan_link_ops)
4688                         unregister_netdevice_queue(dev, head);
4689
4690         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4691                 /* If vxlan->dev is in the same netns, it has already been added
4692                  * to the list by the previous loop.
4693                  */
4694                 if (!net_eq(dev_net(vxlan->dev), net))
4695                         unregister_netdevice_queue(vxlan->dev, head);
4696         }
4697
4698         for (h = 0; h < PORT_HASH_SIZE; ++h)
4699                 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
4700 }
4701
4702 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
4703 {
4704         struct net *net;
4705         LIST_HEAD(list);
4706
4707         rtnl_lock();
4708         list_for_each_entry(net, net_list, exit_list)
4709                 unregister_nexthop_notifier(net, &vxlan_nexthop_notifier_block);
4710         list_for_each_entry(net, net_list, exit_list)
4711                 vxlan_destroy_tunnels(net, &list);
4712
4713         unregister_netdevice_many(&list);
4714         rtnl_unlock();
4715 }
4716
4717 static struct pernet_operations vxlan_net_ops = {
4718         .init = vxlan_init_net,
4719         .exit_batch = vxlan_exit_batch_net,
4720         .id   = &vxlan_net_id,
4721         .size = sizeof(struct vxlan_net),
4722 };
4723
4724 static int __init vxlan_init_module(void)
4725 {
4726         int rc;
4727
4728         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
4729
4730         rc = register_pernet_subsys(&vxlan_net_ops);
4731         if (rc)
4732                 goto out1;
4733
4734         rc = register_netdevice_notifier(&vxlan_notifier_block);
4735         if (rc)
4736                 goto out2;
4737
4738         rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block);
4739         if (rc)
4740                 goto out3;
4741
4742         rc = rtnl_link_register(&vxlan_link_ops);
4743         if (rc)
4744                 goto out4;
4745
4746         return 0;
4747 out4:
4748         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4749 out3:
4750         unregister_netdevice_notifier(&vxlan_notifier_block);
4751 out2:
4752         unregister_pernet_subsys(&vxlan_net_ops);
4753 out1:
4754         return rc;
4755 }
4756 late_initcall(vxlan_init_module);
4757
4758 static void __exit vxlan_cleanup_module(void)
4759 {
4760         rtnl_link_unregister(&vxlan_link_ops);
4761         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4762         unregister_netdevice_notifier(&vxlan_notifier_block);
4763         unregister_pernet_subsys(&vxlan_net_ops);
4764         /* rcu_barrier() is called by netns */
4765 }
4766 module_exit(vxlan_cleanup_module);
4767
4768 MODULE_LICENSE("GPL");
4769 MODULE_VERSION(VXLAN_VERSION);
4770 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
4771 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
4772 MODULE_ALIAS_RTNL_LINK("vxlan");