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                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
1380                         struct vxlan_rdst *rd;
1381
1382                         if (rcu_access_pointer(f->nh)) {
1383                                 if (*idx < cb->args[2])
1384                                         goto skip_nh;
1385                                 err = vxlan_fdb_info(skb, vxlan, f,
1386                                                      NETLINK_CB(cb->skb).portid,
1387                                                      cb->nlh->nlmsg_seq,
1388                                                      RTM_NEWNEIGH,
1389                                                      NLM_F_MULTI, NULL);
1390                                 if (err < 0)
1391                                         goto out;
1392 skip_nh:
1393                                 *idx += 1;
1394                                 continue;
1395                         }
1396
1397                         list_for_each_entry_rcu(rd, &f->remotes, list) {
1398                                 if (*idx < cb->args[2])
1399                                         goto skip;
1400
1401                                 err = vxlan_fdb_info(skb, vxlan, f,
1402                                                      NETLINK_CB(cb->skb).portid,
1403                                                      cb->nlh->nlmsg_seq,
1404                                                      RTM_NEWNEIGH,
1405                                                      NLM_F_MULTI, rd);
1406                                 if (err < 0)
1407                                         goto out;
1408 skip:
1409                                 *idx += 1;
1410                         }
1411                 }
1412         }
1413 out:
1414         return err;
1415 }
1416
1417 static int vxlan_fdb_get(struct sk_buff *skb,
1418                          struct nlattr *tb[],
1419                          struct net_device *dev,
1420                          const unsigned char *addr,
1421                          u16 vid, u32 portid, u32 seq,
1422                          struct netlink_ext_ack *extack)
1423 {
1424         struct vxlan_dev *vxlan = netdev_priv(dev);
1425         struct vxlan_fdb *f;
1426         __be32 vni;
1427         int err;
1428
1429         if (tb[NDA_VNI])
1430                 vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
1431         else
1432                 vni = vxlan->default_dst.remote_vni;
1433
1434         rcu_read_lock();
1435
1436         f = __vxlan_find_mac(vxlan, addr, vni);
1437         if (!f) {
1438                 NL_SET_ERR_MSG(extack, "Fdb entry not found");
1439                 err = -ENOENT;
1440                 goto errout;
1441         }
1442
1443         err = vxlan_fdb_info(skb, vxlan, f, portid, seq,
1444                              RTM_NEWNEIGH, 0, first_remote_rcu(f));
1445 errout:
1446         rcu_read_unlock();
1447         return err;
1448 }
1449
1450 /* Watch incoming packets to learn mapping between Ethernet address
1451  * and Tunnel endpoint.
1452  * Return true if packet is bogus and should be dropped.
1453  */
1454 static bool vxlan_snoop(struct net_device *dev,
1455                         union vxlan_addr *src_ip, const u8 *src_mac,
1456                         u32 src_ifindex, __be32 vni)
1457 {
1458         struct vxlan_dev *vxlan = netdev_priv(dev);
1459         struct vxlan_fdb *f;
1460         u32 ifindex = 0;
1461
1462 #if IS_ENABLED(CONFIG_IPV6)
1463         if (src_ip->sa.sa_family == AF_INET6 &&
1464             (ipv6_addr_type(&src_ip->sin6.sin6_addr) & IPV6_ADDR_LINKLOCAL))
1465                 ifindex = src_ifindex;
1466 #endif
1467
1468         f = vxlan_find_mac(vxlan, src_mac, vni);
1469         if (likely(f)) {
1470                 struct vxlan_rdst *rdst = first_remote_rcu(f);
1471
1472                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip) &&
1473                            rdst->remote_ifindex == ifindex))
1474                         return false;
1475
1476                 /* Don't migrate static entries, drop packets */
1477                 if (f->state & (NUD_PERMANENT | NUD_NOARP))
1478                         return true;
1479
1480                 /* Don't override an fdb with nexthop with a learnt entry */
1481                 if (rcu_access_pointer(f->nh))
1482                         return true;
1483
1484                 if (net_ratelimit())
1485                         netdev_info(dev,
1486                                     "%pM migrated from %pIS to %pIS\n",
1487                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
1488
1489                 rdst->remote_ip = *src_ip;
1490                 f->updated = jiffies;
1491                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH, true, NULL);
1492         } else {
1493                 u32 hash_index = fdb_head_index(vxlan, src_mac, vni);
1494
1495                 /* learned new entry */
1496                 spin_lock(&vxlan->hash_lock[hash_index]);
1497
1498                 /* close off race between vxlan_flush and incoming packets */
1499                 if (netif_running(dev))
1500                         vxlan_fdb_update(vxlan, src_mac, src_ip,
1501                                          NUD_REACHABLE,
1502                                          NLM_F_EXCL|NLM_F_CREATE,
1503                                          vxlan->cfg.dst_port,
1504                                          vni,
1505                                          vxlan->default_dst.remote_vni,
1506                                          ifindex, NTF_SELF, 0, true, NULL);
1507                 spin_unlock(&vxlan->hash_lock[hash_index]);
1508         }
1509
1510         return false;
1511 }
1512
1513 /* See if multicast group is already in use by other ID */
1514 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
1515 {
1516         struct vxlan_dev *vxlan;
1517         struct vxlan_sock *sock4;
1518 #if IS_ENABLED(CONFIG_IPV6)
1519         struct vxlan_sock *sock6;
1520 #endif
1521         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
1522
1523         sock4 = rtnl_dereference(dev->vn4_sock);
1524
1525         /* The vxlan_sock is only used by dev, leaving group has
1526          * no effect on other vxlan devices.
1527          */
1528         if (family == AF_INET && sock4 && refcount_read(&sock4->refcnt) == 1)
1529                 return false;
1530 #if IS_ENABLED(CONFIG_IPV6)
1531         sock6 = rtnl_dereference(dev->vn6_sock);
1532         if (family == AF_INET6 && sock6 && refcount_read(&sock6->refcnt) == 1)
1533                 return false;
1534 #endif
1535
1536         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
1537                 if (!netif_running(vxlan->dev) || vxlan == dev)
1538                         continue;
1539
1540                 if (family == AF_INET &&
1541                     rtnl_dereference(vxlan->vn4_sock) != sock4)
1542                         continue;
1543 #if IS_ENABLED(CONFIG_IPV6)
1544                 if (family == AF_INET6 &&
1545                     rtnl_dereference(vxlan->vn6_sock) != sock6)
1546                         continue;
1547 #endif
1548
1549                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
1550                                       &dev->default_dst.remote_ip))
1551                         continue;
1552
1553                 if (vxlan->default_dst.remote_ifindex !=
1554                     dev->default_dst.remote_ifindex)
1555                         continue;
1556
1557                 return true;
1558         }
1559
1560         return false;
1561 }
1562
1563 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
1564 {
1565         struct vxlan_net *vn;
1566
1567         if (!vs)
1568                 return false;
1569         if (!refcount_dec_and_test(&vs->refcnt))
1570                 return false;
1571
1572         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1573         spin_lock(&vn->sock_lock);
1574         hlist_del_rcu(&vs->hlist);
1575         udp_tunnel_notify_del_rx_port(vs->sock,
1576                                       (vs->flags & VXLAN_F_GPE) ?
1577                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1578                                       UDP_TUNNEL_TYPE_VXLAN);
1579         spin_unlock(&vn->sock_lock);
1580
1581         return true;
1582 }
1583
1584 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1585 {
1586         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1587 #if IS_ENABLED(CONFIG_IPV6)
1588         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1589
1590         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
1591 #endif
1592
1593         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
1594         synchronize_net();
1595
1596         vxlan_vs_del_dev(vxlan);
1597
1598         if (__vxlan_sock_release_prep(sock4)) {
1599                 udp_tunnel_sock_release(sock4->sock);
1600                 kfree(sock4);
1601         }
1602
1603 #if IS_ENABLED(CONFIG_IPV6)
1604         if (__vxlan_sock_release_prep(sock6)) {
1605                 udp_tunnel_sock_release(sock6->sock);
1606                 kfree(sock6);
1607         }
1608 #endif
1609 }
1610
1611 /* Update multicast group membership when first VNI on
1612  * multicast address is brought up
1613  */
1614 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1615 {
1616         struct sock *sk;
1617         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1618         int ifindex = vxlan->default_dst.remote_ifindex;
1619         int ret = -EINVAL;
1620
1621         if (ip->sa.sa_family == AF_INET) {
1622                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1623                 struct ip_mreqn mreq = {
1624                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1625                         .imr_ifindex            = ifindex,
1626                 };
1627
1628                 sk = sock4->sock->sk;
1629                 lock_sock(sk);
1630                 ret = ip_mc_join_group(sk, &mreq);
1631                 release_sock(sk);
1632 #if IS_ENABLED(CONFIG_IPV6)
1633         } else {
1634                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1635
1636                 sk = sock6->sock->sk;
1637                 lock_sock(sk);
1638                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1639                                                    &ip->sin6.sin6_addr);
1640                 release_sock(sk);
1641 #endif
1642         }
1643
1644         return ret;
1645 }
1646
1647 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1648 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1649 {
1650         struct sock *sk;
1651         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1652         int ifindex = vxlan->default_dst.remote_ifindex;
1653         int ret = -EINVAL;
1654
1655         if (ip->sa.sa_family == AF_INET) {
1656                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1657                 struct ip_mreqn mreq = {
1658                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1659                         .imr_ifindex            = ifindex,
1660                 };
1661
1662                 sk = sock4->sock->sk;
1663                 lock_sock(sk);
1664                 ret = ip_mc_leave_group(sk, &mreq);
1665                 release_sock(sk);
1666 #if IS_ENABLED(CONFIG_IPV6)
1667         } else {
1668                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1669
1670                 sk = sock6->sock->sk;
1671                 lock_sock(sk);
1672                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1673                                                    &ip->sin6.sin6_addr);
1674                 release_sock(sk);
1675 #endif
1676         }
1677
1678         return ret;
1679 }
1680
1681 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1682                           struct sk_buff *skb, u32 vxflags)
1683 {
1684         size_t start, offset;
1685
1686         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1687                 goto out;
1688
1689         start = vxlan_rco_start(unparsed->vx_vni);
1690         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1691
1692         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1693                 return false;
1694
1695         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1696                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1697 out:
1698         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1699         unparsed->vx_vni &= VXLAN_VNI_MASK;
1700         return true;
1701 }
1702
1703 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1704                                 struct sk_buff *skb, u32 vxflags,
1705                                 struct vxlan_metadata *md)
1706 {
1707         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1708         struct metadata_dst *tun_dst;
1709
1710         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1711                 goto out;
1712
1713         md->gbp = ntohs(gbp->policy_id);
1714
1715         tun_dst = (struct metadata_dst *)skb_dst(skb);
1716         if (tun_dst) {
1717                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1718                 tun_dst->u.tun_info.options_len = sizeof(*md);
1719         }
1720         if (gbp->dont_learn)
1721                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1722
1723         if (gbp->policy_applied)
1724                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1725
1726         /* In flow-based mode, GBP is carried in dst_metadata */
1727         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1728                 skb->mark = md->gbp;
1729 out:
1730         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1731 }
1732
1733 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1734                                 __be16 *protocol,
1735                                 struct sk_buff *skb, u32 vxflags)
1736 {
1737         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1738
1739         /* Need to have Next Protocol set for interfaces in GPE mode. */
1740         if (!gpe->np_applied)
1741                 return false;
1742         /* "The initial version is 0. If a receiver does not support the
1743          * version indicated it MUST drop the packet.
1744          */
1745         if (gpe->version != 0)
1746                 return false;
1747         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1748          * processing MUST occur." However, we don't implement OAM
1749          * processing, thus drop the packet.
1750          */
1751         if (gpe->oam_flag)
1752                 return false;
1753
1754         *protocol = tun_p_to_eth_p(gpe->next_protocol);
1755         if (!*protocol)
1756                 return false;
1757
1758         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1759         return true;
1760 }
1761
1762 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1763                           struct vxlan_sock *vs,
1764                           struct sk_buff *skb, __be32 vni)
1765 {
1766         union vxlan_addr saddr;
1767         u32 ifindex = skb->dev->ifindex;
1768
1769         skb_reset_mac_header(skb);
1770         skb->protocol = eth_type_trans(skb, vxlan->dev);
1771         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1772
1773         /* Ignore packet loops (and multicast echo) */
1774         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1775                 return false;
1776
1777         /* Get address from the outer IP header */
1778         if (vxlan_get_sk_family(vs) == AF_INET) {
1779                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1780                 saddr.sa.sa_family = AF_INET;
1781 #if IS_ENABLED(CONFIG_IPV6)
1782         } else {
1783                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1784                 saddr.sa.sa_family = AF_INET6;
1785 #endif
1786         }
1787
1788         if ((vxlan->cfg.flags & VXLAN_F_LEARN) &&
1789             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source, ifindex, vni))
1790                 return false;
1791
1792         return true;
1793 }
1794
1795 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1796                                   struct sk_buff *skb)
1797 {
1798         int err = 0;
1799
1800         if (vxlan_get_sk_family(vs) == AF_INET)
1801                 err = IP_ECN_decapsulate(oiph, skb);
1802 #if IS_ENABLED(CONFIG_IPV6)
1803         else
1804                 err = IP6_ECN_decapsulate(oiph, skb);
1805 #endif
1806
1807         if (unlikely(err) && log_ecn_error) {
1808                 if (vxlan_get_sk_family(vs) == AF_INET)
1809                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1810                                              &((struct iphdr *)oiph)->saddr,
1811                                              ((struct iphdr *)oiph)->tos);
1812                 else
1813                         net_info_ratelimited("non-ECT from %pI6\n",
1814                                              &((struct ipv6hdr *)oiph)->saddr);
1815         }
1816         return err <= 1;
1817 }
1818
1819 /* Callback from net/ipv4/udp.c to receive packets */
1820 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1821 {
1822         struct pcpu_sw_netstats *stats;
1823         struct vxlan_dev *vxlan;
1824         struct vxlan_sock *vs;
1825         struct vxlanhdr unparsed;
1826         struct vxlan_metadata _md;
1827         struct vxlan_metadata *md = &_md;
1828         __be16 protocol = htons(ETH_P_TEB);
1829         bool raw_proto = false;
1830         void *oiph;
1831         __be32 vni = 0;
1832
1833         /* Need UDP and VXLAN header to be present */
1834         if (!pskb_may_pull(skb, VXLAN_HLEN))
1835                 goto drop;
1836
1837         unparsed = *vxlan_hdr(skb);
1838         /* VNI flag always required to be set */
1839         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1840                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1841                            ntohl(vxlan_hdr(skb)->vx_flags),
1842                            ntohl(vxlan_hdr(skb)->vx_vni));
1843                 /* Return non vxlan pkt */
1844                 goto drop;
1845         }
1846         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1847         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1848
1849         vs = rcu_dereference_sk_user_data(sk);
1850         if (!vs)
1851                 goto drop;
1852
1853         vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1854
1855         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1856         if (!vxlan)
1857                 goto drop;
1858
1859         /* For backwards compatibility, only allow reserved fields to be
1860          * used by VXLAN extensions if explicitly requested.
1861          */
1862         if (vs->flags & VXLAN_F_GPE) {
1863                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1864                         goto drop;
1865                 raw_proto = true;
1866         }
1867
1868         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1869                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1870                 goto drop;
1871
1872         if (vxlan_collect_metadata(vs)) {
1873                 struct metadata_dst *tun_dst;
1874
1875                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1876                                          key32_to_tunnel_id(vni), sizeof(*md));
1877
1878                 if (!tun_dst)
1879                         goto drop;
1880
1881                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1882
1883                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1884         } else {
1885                 memset(md, 0, sizeof(*md));
1886         }
1887
1888         if (vs->flags & VXLAN_F_REMCSUM_RX)
1889                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1890                         goto drop;
1891         if (vs->flags & VXLAN_F_GBP)
1892                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1893         /* Note that GBP and GPE can never be active together. This is
1894          * ensured in vxlan_dev_configure.
1895          */
1896
1897         if (unparsed.vx_flags || unparsed.vx_vni) {
1898                 /* If there are any unprocessed flags remaining treat
1899                  * this as a malformed packet. This behavior diverges from
1900                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1901                  * in reserved fields are to be ignored. The approach here
1902                  * maintains compatibility with previous stack code, and also
1903                  * is more robust and provides a little more security in
1904                  * adding extensions to VXLAN.
1905                  */
1906                 goto drop;
1907         }
1908
1909         if (!raw_proto) {
1910                 if (!vxlan_set_mac(vxlan, vs, skb, vni))
1911                         goto drop;
1912         } else {
1913                 skb_reset_mac_header(skb);
1914                 skb->dev = vxlan->dev;
1915                 skb->pkt_type = PACKET_HOST;
1916         }
1917
1918         oiph = skb_network_header(skb);
1919         skb_reset_network_header(skb);
1920
1921         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1922                 ++vxlan->dev->stats.rx_frame_errors;
1923                 ++vxlan->dev->stats.rx_errors;
1924                 goto drop;
1925         }
1926
1927         rcu_read_lock();
1928
1929         if (unlikely(!(vxlan->dev->flags & IFF_UP))) {
1930                 rcu_read_unlock();
1931                 atomic_long_inc(&vxlan->dev->rx_dropped);
1932                 goto drop;
1933         }
1934
1935         stats = this_cpu_ptr(vxlan->dev->tstats);
1936         u64_stats_update_begin(&stats->syncp);
1937         stats->rx_packets++;
1938         stats->rx_bytes += skb->len;
1939         u64_stats_update_end(&stats->syncp);
1940
1941         gro_cells_receive(&vxlan->gro_cells, skb);
1942
1943         rcu_read_unlock();
1944
1945         return 0;
1946
1947 drop:
1948         /* Consume bad packet */
1949         kfree_skb(skb);
1950         return 0;
1951 }
1952
1953 /* Callback from net/ipv{4,6}/udp.c to check that we have a VNI for errors */
1954 static int vxlan_err_lookup(struct sock *sk, struct sk_buff *skb)
1955 {
1956         struct vxlan_dev *vxlan;
1957         struct vxlan_sock *vs;
1958         struct vxlanhdr *hdr;
1959         __be32 vni;
1960
1961         if (!pskb_may_pull(skb, skb_transport_offset(skb) + VXLAN_HLEN))
1962                 return -EINVAL;
1963
1964         hdr = vxlan_hdr(skb);
1965
1966         if (!(hdr->vx_flags & VXLAN_HF_VNI))
1967                 return -EINVAL;
1968
1969         vs = rcu_dereference_sk_user_data(sk);
1970         if (!vs)
1971                 return -ENOENT;
1972
1973         vni = vxlan_vni(hdr->vx_vni);
1974         vxlan = vxlan_vs_find_vni(vs, skb->dev->ifindex, vni);
1975         if (!vxlan)
1976                 return -ENOENT;
1977
1978         return 0;
1979 }
1980
1981 static int arp_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
1982 {
1983         struct vxlan_dev *vxlan = netdev_priv(dev);
1984         struct arphdr *parp;
1985         u8 *arpptr, *sha;
1986         __be32 sip, tip;
1987         struct neighbour *n;
1988
1989         if (dev->flags & IFF_NOARP)
1990                 goto out;
1991
1992         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1993                 dev->stats.tx_dropped++;
1994                 goto out;
1995         }
1996         parp = arp_hdr(skb);
1997
1998         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1999              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
2000             parp->ar_pro != htons(ETH_P_IP) ||
2001             parp->ar_op != htons(ARPOP_REQUEST) ||
2002             parp->ar_hln != dev->addr_len ||
2003             parp->ar_pln != 4)
2004                 goto out;
2005         arpptr = (u8 *)parp + sizeof(struct arphdr);
2006         sha = arpptr;
2007         arpptr += dev->addr_len;        /* sha */
2008         memcpy(&sip, arpptr, sizeof(sip));
2009         arpptr += sizeof(sip);
2010         arpptr += dev->addr_len;        /* tha */
2011         memcpy(&tip, arpptr, sizeof(tip));
2012
2013         if (ipv4_is_loopback(tip) ||
2014             ipv4_is_multicast(tip))
2015                 goto out;
2016
2017         n = neigh_lookup(&arp_tbl, &tip, dev);
2018
2019         if (n) {
2020                 struct vxlan_fdb *f;
2021                 struct sk_buff  *reply;
2022
2023                 if (!(n->nud_state & NUD_CONNECTED)) {
2024                         neigh_release(n);
2025                         goto out;
2026                 }
2027
2028                 f = vxlan_find_mac(vxlan, n->ha, vni);
2029                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2030                         /* bridge-local neighbor */
2031                         neigh_release(n);
2032                         goto out;
2033                 }
2034
2035                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
2036                                 n->ha, sha);
2037
2038                 neigh_release(n);
2039
2040                 if (reply == NULL)
2041                         goto out;
2042
2043                 skb_reset_mac_header(reply);
2044                 __skb_pull(reply, skb_network_offset(reply));
2045                 reply->ip_summed = CHECKSUM_UNNECESSARY;
2046                 reply->pkt_type = PACKET_HOST;
2047
2048                 if (netif_rx_ni(reply) == NET_RX_DROP)
2049                         dev->stats.rx_dropped++;
2050         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2051                 union vxlan_addr ipa = {
2052                         .sin.sin_addr.s_addr = tip,
2053                         .sin.sin_family = AF_INET,
2054                 };
2055
2056                 vxlan_ip_miss(dev, &ipa);
2057         }
2058 out:
2059         consume_skb(skb);
2060         return NETDEV_TX_OK;
2061 }
2062
2063 #if IS_ENABLED(CONFIG_IPV6)
2064 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
2065         struct neighbour *n, bool isrouter)
2066 {
2067         struct net_device *dev = request->dev;
2068         struct sk_buff *reply;
2069         struct nd_msg *ns, *na;
2070         struct ipv6hdr *pip6;
2071         u8 *daddr;
2072         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
2073         int ns_olen;
2074         int i, len;
2075
2076         if (dev == NULL || !pskb_may_pull(request, request->len))
2077                 return NULL;
2078
2079         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
2080                 sizeof(*na) + na_olen + dev->needed_tailroom;
2081         reply = alloc_skb(len, GFP_ATOMIC);
2082         if (reply == NULL)
2083                 return NULL;
2084
2085         reply->protocol = htons(ETH_P_IPV6);
2086         reply->dev = dev;
2087         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
2088         skb_push(reply, sizeof(struct ethhdr));
2089         skb_reset_mac_header(reply);
2090
2091         ns = (struct nd_msg *)(ipv6_hdr(request) + 1);
2092
2093         daddr = eth_hdr(request)->h_source;
2094         ns_olen = request->len - skb_network_offset(request) -
2095                 sizeof(struct ipv6hdr) - sizeof(*ns);
2096         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
2097                 if (!ns->opt[i + 1]) {
2098                         kfree_skb(reply);
2099                         return NULL;
2100                 }
2101                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
2102                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
2103                         break;
2104                 }
2105         }
2106
2107         /* Ethernet header */
2108         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
2109         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
2110         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
2111         reply->protocol = htons(ETH_P_IPV6);
2112
2113         skb_pull(reply, sizeof(struct ethhdr));
2114         skb_reset_network_header(reply);
2115         skb_put(reply, sizeof(struct ipv6hdr));
2116
2117         /* IPv6 header */
2118
2119         pip6 = ipv6_hdr(reply);
2120         memset(pip6, 0, sizeof(struct ipv6hdr));
2121         pip6->version = 6;
2122         pip6->priority = ipv6_hdr(request)->priority;
2123         pip6->nexthdr = IPPROTO_ICMPV6;
2124         pip6->hop_limit = 255;
2125         pip6->daddr = ipv6_hdr(request)->saddr;
2126         pip6->saddr = *(struct in6_addr *)n->primary_key;
2127
2128         skb_pull(reply, sizeof(struct ipv6hdr));
2129         skb_reset_transport_header(reply);
2130
2131         /* Neighbor Advertisement */
2132         na = skb_put_zero(reply, sizeof(*na) + na_olen);
2133         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
2134         na->icmph.icmp6_router = isrouter;
2135         na->icmph.icmp6_override = 1;
2136         na->icmph.icmp6_solicited = 1;
2137         na->target = ns->target;
2138         ether_addr_copy(&na->opt[2], n->ha);
2139         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
2140         na->opt[1] = na_olen >> 3;
2141
2142         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
2143                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
2144                 csum_partial(na, sizeof(*na)+na_olen, 0));
2145
2146         pip6->payload_len = htons(sizeof(*na)+na_olen);
2147
2148         skb_push(reply, sizeof(struct ipv6hdr));
2149
2150         reply->ip_summed = CHECKSUM_UNNECESSARY;
2151
2152         return reply;
2153 }
2154
2155 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb, __be32 vni)
2156 {
2157         struct vxlan_dev *vxlan = netdev_priv(dev);
2158         const struct in6_addr *daddr;
2159         const struct ipv6hdr *iphdr;
2160         struct inet6_dev *in6_dev;
2161         struct neighbour *n;
2162         struct nd_msg *msg;
2163
2164         in6_dev = __in6_dev_get(dev);
2165         if (!in6_dev)
2166                 goto out;
2167
2168         iphdr = ipv6_hdr(skb);
2169         daddr = &iphdr->daddr;
2170         msg = (struct nd_msg *)(iphdr + 1);
2171
2172         if (ipv6_addr_loopback(daddr) ||
2173             ipv6_addr_is_multicast(&msg->target))
2174                 goto out;
2175
2176         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
2177
2178         if (n) {
2179                 struct vxlan_fdb *f;
2180                 struct sk_buff *reply;
2181
2182                 if (!(n->nud_state & NUD_CONNECTED)) {
2183                         neigh_release(n);
2184                         goto out;
2185                 }
2186
2187                 f = vxlan_find_mac(vxlan, n->ha, vni);
2188                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
2189                         /* bridge-local neighbor */
2190                         neigh_release(n);
2191                         goto out;
2192                 }
2193
2194                 reply = vxlan_na_create(skb, n,
2195                                         !!(f ? f->flags & NTF_ROUTER : 0));
2196
2197                 neigh_release(n);
2198
2199                 if (reply == NULL)
2200                         goto out;
2201
2202                 if (netif_rx_ni(reply) == NET_RX_DROP)
2203                         dev->stats.rx_dropped++;
2204
2205         } else if (vxlan->cfg.flags & VXLAN_F_L3MISS) {
2206                 union vxlan_addr ipa = {
2207                         .sin6.sin6_addr = msg->target,
2208                         .sin6.sin6_family = AF_INET6,
2209                 };
2210
2211                 vxlan_ip_miss(dev, &ipa);
2212         }
2213
2214 out:
2215         consume_skb(skb);
2216         return NETDEV_TX_OK;
2217 }
2218 #endif
2219
2220 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
2221 {
2222         struct vxlan_dev *vxlan = netdev_priv(dev);
2223         struct neighbour *n;
2224
2225         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
2226                 return false;
2227
2228         n = NULL;
2229         switch (ntohs(eth_hdr(skb)->h_proto)) {
2230         case ETH_P_IP:
2231         {
2232                 struct iphdr *pip;
2233
2234                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
2235                         return false;
2236                 pip = ip_hdr(skb);
2237                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
2238                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2239                         union vxlan_addr ipa = {
2240                                 .sin.sin_addr.s_addr = pip->daddr,
2241                                 .sin.sin_family = AF_INET,
2242                         };
2243
2244                         vxlan_ip_miss(dev, &ipa);
2245                         return false;
2246                 }
2247
2248                 break;
2249         }
2250 #if IS_ENABLED(CONFIG_IPV6)
2251         case ETH_P_IPV6:
2252         {
2253                 struct ipv6hdr *pip6;
2254
2255                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
2256                         return false;
2257                 pip6 = ipv6_hdr(skb);
2258                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
2259                 if (!n && (vxlan->cfg.flags & VXLAN_F_L3MISS)) {
2260                         union vxlan_addr ipa = {
2261                                 .sin6.sin6_addr = pip6->daddr,
2262                                 .sin6.sin6_family = AF_INET6,
2263                         };
2264
2265                         vxlan_ip_miss(dev, &ipa);
2266                         return false;
2267                 }
2268
2269                 break;
2270         }
2271 #endif
2272         default:
2273                 return false;
2274         }
2275
2276         if (n) {
2277                 bool diff;
2278
2279                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
2280                 if (diff) {
2281                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
2282                                 dev->addr_len);
2283                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
2284                 }
2285                 neigh_release(n);
2286                 return diff;
2287         }
2288
2289         return false;
2290 }
2291
2292 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
2293                                 struct vxlan_metadata *md)
2294 {
2295         struct vxlanhdr_gbp *gbp;
2296
2297         if (!md->gbp)
2298                 return;
2299
2300         gbp = (struct vxlanhdr_gbp *)vxh;
2301         vxh->vx_flags |= VXLAN_HF_GBP;
2302
2303         if (md->gbp & VXLAN_GBP_DONT_LEARN)
2304                 gbp->dont_learn = 1;
2305
2306         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
2307                 gbp->policy_applied = 1;
2308
2309         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
2310 }
2311
2312 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
2313                                __be16 protocol)
2314 {
2315         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
2316
2317         gpe->np_applied = 1;
2318         gpe->next_protocol = tun_p_from_eth_p(protocol);
2319         if (!gpe->next_protocol)
2320                 return -EPFNOSUPPORT;
2321         return 0;
2322 }
2323
2324 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
2325                            int iphdr_len, __be32 vni,
2326                            struct vxlan_metadata *md, u32 vxflags,
2327                            bool udp_sum)
2328 {
2329         struct vxlanhdr *vxh;
2330         int min_headroom;
2331         int err;
2332         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
2333         __be16 inner_protocol = htons(ETH_P_TEB);
2334
2335         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
2336             skb->ip_summed == CHECKSUM_PARTIAL) {
2337                 int csum_start = skb_checksum_start_offset(skb);
2338
2339                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
2340                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
2341                     (skb->csum_offset == offsetof(struct udphdr, check) ||
2342                      skb->csum_offset == offsetof(struct tcphdr, check)))
2343                         type |= SKB_GSO_TUNNEL_REMCSUM;
2344         }
2345
2346         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
2347                         + VXLAN_HLEN + iphdr_len;
2348
2349         /* Need space for new headers (invalidates iph ptr) */
2350         err = skb_cow_head(skb, min_headroom);
2351         if (unlikely(err))
2352                 return err;
2353
2354         err = iptunnel_handle_offloads(skb, type);
2355         if (err)
2356                 return err;
2357
2358         vxh = __skb_push(skb, sizeof(*vxh));
2359         vxh->vx_flags = VXLAN_HF_VNI;
2360         vxh->vx_vni = vxlan_vni_field(vni);
2361
2362         if (type & SKB_GSO_TUNNEL_REMCSUM) {
2363                 unsigned int start;
2364
2365                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
2366                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
2367                 vxh->vx_flags |= VXLAN_HF_RCO;
2368
2369                 if (!skb_is_gso(skb)) {
2370                         skb->ip_summed = CHECKSUM_NONE;
2371                         skb->encapsulation = 0;
2372                 }
2373         }
2374
2375         if (vxflags & VXLAN_F_GBP)
2376                 vxlan_build_gbp_hdr(vxh, vxflags, md);
2377         if (vxflags & VXLAN_F_GPE) {
2378                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
2379                 if (err < 0)
2380                         return err;
2381                 inner_protocol = skb->protocol;
2382         }
2383
2384         skb_set_inner_protocol(skb, inner_protocol);
2385         return 0;
2386 }
2387
2388 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
2389                                       struct vxlan_sock *sock4,
2390                                       struct sk_buff *skb, int oif, u8 tos,
2391                                       __be32 daddr, __be32 *saddr, __be16 dport, __be16 sport,
2392                                       struct dst_cache *dst_cache,
2393                                       const struct ip_tunnel_info *info)
2394 {
2395         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2396         struct rtable *rt = NULL;
2397         struct flowi4 fl4;
2398
2399         if (!sock4)
2400                 return ERR_PTR(-EIO);
2401
2402         if (tos && !info)
2403                 use_cache = false;
2404         if (use_cache) {
2405                 rt = dst_cache_get_ip4(dst_cache, saddr);
2406                 if (rt)
2407                         return rt;
2408         }
2409
2410         memset(&fl4, 0, sizeof(fl4));
2411         fl4.flowi4_oif = oif;
2412         fl4.flowi4_tos = RT_TOS(tos);
2413         fl4.flowi4_mark = skb->mark;
2414         fl4.flowi4_proto = IPPROTO_UDP;
2415         fl4.daddr = daddr;
2416         fl4.saddr = *saddr;
2417         fl4.fl4_dport = dport;
2418         fl4.fl4_sport = sport;
2419
2420         rt = ip_route_output_key(vxlan->net, &fl4);
2421         if (!IS_ERR(rt)) {
2422                 if (rt->dst.dev == dev) {
2423                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
2424                         ip_rt_put(rt);
2425                         return ERR_PTR(-ELOOP);
2426                 }
2427
2428                 *saddr = fl4.saddr;
2429                 if (use_cache)
2430                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
2431         } else {
2432                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
2433                 return ERR_PTR(-ENETUNREACH);
2434         }
2435         return rt;
2436 }
2437
2438 #if IS_ENABLED(CONFIG_IPV6)
2439 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
2440                                           struct net_device *dev,
2441                                           struct vxlan_sock *sock6,
2442                                           struct sk_buff *skb, int oif, u8 tos,
2443                                           __be32 label,
2444                                           const struct in6_addr *daddr,
2445                                           struct in6_addr *saddr,
2446                                           __be16 dport, __be16 sport,
2447                                           struct dst_cache *dst_cache,
2448                                           const struct ip_tunnel_info *info)
2449 {
2450         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
2451         struct dst_entry *ndst;
2452         struct flowi6 fl6;
2453
2454         if (!sock6)
2455                 return ERR_PTR(-EIO);
2456
2457         if (tos && !info)
2458                 use_cache = false;
2459         if (use_cache) {
2460                 ndst = dst_cache_get_ip6(dst_cache, saddr);
2461                 if (ndst)
2462                         return ndst;
2463         }
2464
2465         memset(&fl6, 0, sizeof(fl6));
2466         fl6.flowi6_oif = oif;
2467         fl6.daddr = *daddr;
2468         fl6.saddr = *saddr;
2469         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
2470         fl6.flowi6_mark = skb->mark;
2471         fl6.flowi6_proto = IPPROTO_UDP;
2472         fl6.fl6_dport = dport;
2473         fl6.fl6_sport = sport;
2474
2475         ndst = ipv6_stub->ipv6_dst_lookup_flow(vxlan->net, sock6->sock->sk,
2476                                                &fl6, NULL);
2477         if (unlikely(IS_ERR(ndst))) {
2478                 netdev_dbg(dev, "no route to %pI6\n", daddr);
2479                 return ERR_PTR(-ENETUNREACH);
2480         }
2481
2482         if (unlikely(ndst->dev == dev)) {
2483                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
2484                 dst_release(ndst);
2485                 return ERR_PTR(-ELOOP);
2486         }
2487
2488         *saddr = fl6.saddr;
2489         if (use_cache)
2490                 dst_cache_set_ip6(dst_cache, ndst, saddr);
2491         return ndst;
2492 }
2493 #endif
2494
2495 /* Bypass encapsulation if the destination is local */
2496 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
2497                                struct vxlan_dev *dst_vxlan, __be32 vni)
2498 {
2499         struct pcpu_sw_netstats *tx_stats, *rx_stats;
2500         union vxlan_addr loopback;
2501         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
2502         struct net_device *dev;
2503         int len = skb->len;
2504
2505         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
2506         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
2507         skb->pkt_type = PACKET_HOST;
2508         skb->encapsulation = 0;
2509         skb->dev = dst_vxlan->dev;
2510         __skb_pull(skb, skb_network_offset(skb));
2511
2512         if (remote_ip->sa.sa_family == AF_INET) {
2513                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
2514                 loopback.sa.sa_family =  AF_INET;
2515 #if IS_ENABLED(CONFIG_IPV6)
2516         } else {
2517                 loopback.sin6.sin6_addr = in6addr_loopback;
2518                 loopback.sa.sa_family =  AF_INET6;
2519 #endif
2520         }
2521
2522         rcu_read_lock();
2523         dev = skb->dev;
2524         if (unlikely(!(dev->flags & IFF_UP))) {
2525                 kfree_skb(skb);
2526                 goto drop;
2527         }
2528
2529         if (dst_vxlan->cfg.flags & VXLAN_F_LEARN)
2530                 vxlan_snoop(dev, &loopback, eth_hdr(skb)->h_source, 0, vni);
2531
2532         u64_stats_update_begin(&tx_stats->syncp);
2533         tx_stats->tx_packets++;
2534         tx_stats->tx_bytes += len;
2535         u64_stats_update_end(&tx_stats->syncp);
2536
2537         if (netif_rx(skb) == NET_RX_SUCCESS) {
2538                 u64_stats_update_begin(&rx_stats->syncp);
2539                 rx_stats->rx_packets++;
2540                 rx_stats->rx_bytes += len;
2541                 u64_stats_update_end(&rx_stats->syncp);
2542         } else {
2543 drop:
2544                 dev->stats.rx_dropped++;
2545         }
2546         rcu_read_unlock();
2547 }
2548
2549 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
2550                                  struct vxlan_dev *vxlan,
2551                                  union vxlan_addr *daddr,
2552                                  __be16 dst_port, int dst_ifindex, __be32 vni,
2553                                  struct dst_entry *dst,
2554                                  u32 rt_flags)
2555 {
2556 #if IS_ENABLED(CONFIG_IPV6)
2557         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
2558          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
2559          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
2560          */
2561         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
2562 #endif
2563         /* Bypass encapsulation if the destination is local */
2564         if (rt_flags & RTCF_LOCAL &&
2565             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2566                 struct vxlan_dev *dst_vxlan;
2567
2568                 dst_release(dst);
2569                 dst_vxlan = vxlan_find_vni(vxlan->net, dst_ifindex, vni,
2570                                            daddr->sa.sa_family, dst_port,
2571                                            vxlan->cfg.flags);
2572                 if (!dst_vxlan) {
2573                         dev->stats.tx_errors++;
2574                         kfree_skb(skb);
2575
2576                         return -ENOENT;
2577                 }
2578                 vxlan_encap_bypass(skb, vxlan, dst_vxlan, vni);
2579                 return 1;
2580         }
2581
2582         return 0;
2583 }
2584
2585 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
2586                            __be32 default_vni, struct vxlan_rdst *rdst,
2587                            bool did_rsc)
2588 {
2589         struct dst_cache *dst_cache;
2590         struct ip_tunnel_info *info;
2591         struct vxlan_dev *vxlan = netdev_priv(dev);
2592         const struct iphdr *old_iph = ip_hdr(skb);
2593         union vxlan_addr *dst;
2594         union vxlan_addr remote_ip, local_ip;
2595         struct vxlan_metadata _md;
2596         struct vxlan_metadata *md = &_md;
2597         __be16 src_port = 0, dst_port;
2598         struct dst_entry *ndst = NULL;
2599         __be32 vni, label;
2600         __u8 tos, ttl;
2601         int ifindex;
2602         int err;
2603         u32 flags = vxlan->cfg.flags;
2604         bool udp_sum = false;
2605         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2606
2607         info = skb_tunnel_info(skb);
2608
2609         if (rdst) {
2610                 dst = &rdst->remote_ip;
2611                 if (vxlan_addr_any(dst)) {
2612                         if (did_rsc) {
2613                                 /* short-circuited back to local bridge */
2614                                 vxlan_encap_bypass(skb, vxlan, vxlan, default_vni);
2615                                 return;
2616                         }
2617                         goto drop;
2618                 }
2619
2620                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2621                 vni = (rdst->remote_vni) ? : default_vni;
2622                 ifindex = rdst->remote_ifindex;
2623                 local_ip = vxlan->cfg.saddr;
2624                 dst_cache = &rdst->dst_cache;
2625                 md->gbp = skb->mark;
2626                 if (flags & VXLAN_F_TTL_INHERIT) {
2627                         ttl = ip_tunnel_get_ttl(old_iph, skb);
2628                 } else {
2629                         ttl = vxlan->cfg.ttl;
2630                         if (!ttl && vxlan_addr_multicast(dst))
2631                                 ttl = 1;
2632                 }
2633
2634                 tos = vxlan->cfg.tos;
2635                 if (tos == 1)
2636                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2637
2638                 if (dst->sa.sa_family == AF_INET)
2639                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2640                 else
2641                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2642                 label = vxlan->cfg.label;
2643         } else {
2644                 if (!info) {
2645                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2646                                   dev->name);
2647                         goto drop;
2648                 }
2649                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2650                 if (remote_ip.sa.sa_family == AF_INET) {
2651                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2652                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2653                 } else {
2654                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2655                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2656                 }
2657                 dst = &remote_ip;
2658                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2659                 vni = tunnel_id_to_key32(info->key.tun_id);
2660                 ifindex = 0;
2661                 dst_cache = &info->dst_cache;
2662                 if (info->key.tun_flags & TUNNEL_VXLAN_OPT) {
2663                         if (info->options_len < sizeof(*md))
2664                                 goto drop;
2665                         md = ip_tunnel_info_opts(info);
2666                 }
2667                 ttl = info->key.ttl;
2668                 tos = info->key.tos;
2669                 label = info->key.label;
2670                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2671         }
2672         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2673                                      vxlan->cfg.port_max, true);
2674
2675         rcu_read_lock();
2676         if (dst->sa.sa_family == AF_INET) {
2677                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2678                 struct rtable *rt;
2679                 __be16 df = 0;
2680
2681                 if (!ifindex)
2682                         ifindex = sock4->sock->sk->sk_bound_dev_if;
2683
2684                 rt = vxlan_get_route(vxlan, dev, sock4, skb, ifindex, tos,
2685                                      dst->sin.sin_addr.s_addr,
2686                                      &local_ip.sin.sin_addr.s_addr,
2687                                      dst_port, src_port,
2688                                      dst_cache, info);
2689                 if (IS_ERR(rt)) {
2690                         err = PTR_ERR(rt);
2691                         goto tx_error;
2692                 }
2693
2694                 if (!info) {
2695                         /* Bypass encapsulation if the destination is local */
2696                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2697                                                     dst_port, ifindex, vni,
2698                                                     &rt->dst, rt->rt_flags);
2699                         if (err)
2700                                 goto out_unlock;
2701
2702                         if (vxlan->cfg.df == VXLAN_DF_SET) {
2703                                 df = htons(IP_DF);
2704                         } else if (vxlan->cfg.df == VXLAN_DF_INHERIT) {
2705                                 struct ethhdr *eth = eth_hdr(skb);
2706
2707                                 if (ntohs(eth->h_proto) == ETH_P_IPV6 ||
2708                                     (ntohs(eth->h_proto) == ETH_P_IP &&
2709                                      old_iph->frag_off & htons(IP_DF)))
2710                                         df = htons(IP_DF);
2711                         }
2712                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2713                         df = htons(IP_DF);
2714                 }
2715
2716                 ndst = &rt->dst;
2717                 skb_tunnel_check_pmtu(skb, ndst, VXLAN_HEADROOM);
2718
2719                 tos = ip_tunnel_ecn_encap(RT_TOS(tos), old_iph, skb);
2720                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2721                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2722                                       vni, md, flags, udp_sum);
2723                 if (err < 0)
2724                         goto tx_error;
2725
2726                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, local_ip.sin.sin_addr.s_addr,
2727                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2728                                     src_port, dst_port, xnet, !udp_sum);
2729 #if IS_ENABLED(CONFIG_IPV6)
2730         } else {
2731                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2732
2733                 if (!ifindex)
2734                         ifindex = sock6->sock->sk->sk_bound_dev_if;
2735
2736                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, ifindex, tos,
2737                                         label, &dst->sin6.sin6_addr,
2738                                         &local_ip.sin6.sin6_addr,
2739                                         dst_port, src_port,
2740                                         dst_cache, info);
2741                 if (IS_ERR(ndst)) {
2742                         err = PTR_ERR(ndst);
2743                         ndst = NULL;
2744                         goto tx_error;
2745                 }
2746
2747                 if (!info) {
2748                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2749
2750                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2751                                                     dst_port, ifindex, vni,
2752                                                     ndst, rt6i_flags);
2753                         if (err)
2754                                 goto out_unlock;
2755                 }
2756
2757                 skb_tunnel_check_pmtu(skb, ndst, VXLAN6_HEADROOM);
2758
2759                 tos = ip_tunnel_ecn_encap(RT_TOS(tos), old_iph, skb);
2760                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2761                 skb_scrub_packet(skb, xnet);
2762                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2763                                       vni, md, flags, udp_sum);
2764                 if (err < 0)
2765                         goto tx_error;
2766
2767                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2768                                      &local_ip.sin6.sin6_addr,
2769                                      &dst->sin6.sin6_addr, tos, ttl,
2770                                      label, src_port, dst_port, !udp_sum);
2771 #endif
2772         }
2773 out_unlock:
2774         rcu_read_unlock();
2775         return;
2776
2777 drop:
2778         dev->stats.tx_dropped++;
2779         dev_kfree_skb(skb);
2780         return;
2781
2782 tx_error:
2783         rcu_read_unlock();
2784         if (err == -ELOOP)
2785                 dev->stats.collisions++;
2786         else if (err == -ENETUNREACH)
2787                 dev->stats.tx_carrier_errors++;
2788         dst_release(ndst);
2789         dev->stats.tx_errors++;
2790         kfree_skb(skb);
2791 }
2792
2793 static void vxlan_xmit_nh(struct sk_buff *skb, struct net_device *dev,
2794                           struct vxlan_fdb *f, __be32 vni, bool did_rsc)
2795 {
2796         struct vxlan_rdst nh_rdst;
2797         struct nexthop *nh;
2798         bool do_xmit;
2799         u32 hash;
2800
2801         memset(&nh_rdst, 0, sizeof(struct vxlan_rdst));
2802         hash = skb_get_hash(skb);
2803
2804         rcu_read_lock();
2805         nh = rcu_dereference(f->nh);
2806         if (!nh) {
2807                 rcu_read_unlock();
2808                 goto drop;
2809         }
2810         do_xmit = vxlan_fdb_nh_path_select(nh, hash, &nh_rdst);
2811         rcu_read_unlock();
2812
2813         if (likely(do_xmit))
2814                 vxlan_xmit_one(skb, dev, vni, &nh_rdst, did_rsc);
2815         else
2816                 goto drop;
2817
2818         return;
2819
2820 drop:
2821         dev->stats.tx_dropped++;
2822         dev_kfree_skb(skb);
2823 }
2824
2825 /* Transmit local packets over Vxlan
2826  *
2827  * Outer IP header inherits ECN and DF from inner header.
2828  * Outer UDP destination is the VXLAN assigned port.
2829  *           source port is based on hash of flow
2830  */
2831 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2832 {
2833         struct vxlan_dev *vxlan = netdev_priv(dev);
2834         struct vxlan_rdst *rdst, *fdst = NULL;
2835         const struct ip_tunnel_info *info;
2836         bool did_rsc = false;
2837         struct vxlan_fdb *f;
2838         struct ethhdr *eth;
2839         __be32 vni = 0;
2840
2841         info = skb_tunnel_info(skb);
2842
2843         skb_reset_mac_header(skb);
2844
2845         if (vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA) {
2846                 if (info && info->mode & IP_TUNNEL_INFO_BRIDGE &&
2847                     info->mode & IP_TUNNEL_INFO_TX) {
2848                         vni = tunnel_id_to_key32(info->key.tun_id);
2849                 } else {
2850                         if (info && info->mode & IP_TUNNEL_INFO_TX)
2851                                 vxlan_xmit_one(skb, dev, vni, NULL, false);
2852                         else
2853                                 kfree_skb(skb);
2854                         return NETDEV_TX_OK;
2855                 }
2856         }
2857
2858         if (vxlan->cfg.flags & VXLAN_F_PROXY) {
2859                 eth = eth_hdr(skb);
2860                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2861                         return arp_reduce(dev, skb, vni);
2862 #if IS_ENABLED(CONFIG_IPV6)
2863                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2864                          pskb_may_pull(skb, sizeof(struct ipv6hdr) +
2865                                             sizeof(struct nd_msg)) &&
2866                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2867                         struct nd_msg *m = (struct nd_msg *)(ipv6_hdr(skb) + 1);
2868
2869                         if (m->icmph.icmp6_code == 0 &&
2870                             m->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2871                                 return neigh_reduce(dev, skb, vni);
2872                 }
2873 #endif
2874         }
2875
2876         eth = eth_hdr(skb);
2877         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2878         did_rsc = false;
2879
2880         if (f && (f->flags & NTF_ROUTER) && (vxlan->cfg.flags & VXLAN_F_RSC) &&
2881             (ntohs(eth->h_proto) == ETH_P_IP ||
2882              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2883                 did_rsc = route_shortcircuit(dev, skb);
2884                 if (did_rsc)
2885                         f = vxlan_find_mac(vxlan, eth->h_dest, vni);
2886         }
2887
2888         if (f == NULL) {
2889                 f = vxlan_find_mac(vxlan, all_zeros_mac, vni);
2890                 if (f == NULL) {
2891                         if ((vxlan->cfg.flags & VXLAN_F_L2MISS) &&
2892                             !is_multicast_ether_addr(eth->h_dest))
2893                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2894
2895                         dev->stats.tx_dropped++;
2896                         kfree_skb(skb);
2897                         return NETDEV_TX_OK;
2898                 }
2899         }
2900
2901         if (rcu_access_pointer(f->nh)) {
2902                 vxlan_xmit_nh(skb, dev, f,
2903                               (vni ? : vxlan->default_dst.remote_vni), did_rsc);
2904         } else {
2905                 list_for_each_entry_rcu(rdst, &f->remotes, list) {
2906                         struct sk_buff *skb1;
2907
2908                         if (!fdst) {
2909                                 fdst = rdst;
2910                                 continue;
2911                         }
2912                         skb1 = skb_clone(skb, GFP_ATOMIC);
2913                         if (skb1)
2914                                 vxlan_xmit_one(skb1, dev, vni, rdst, did_rsc);
2915                 }
2916                 if (fdst)
2917                         vxlan_xmit_one(skb, dev, vni, fdst, did_rsc);
2918                 else
2919                         kfree_skb(skb);
2920         }
2921
2922         return NETDEV_TX_OK;
2923 }
2924
2925 /* Walk the forwarding table and purge stale entries */
2926 static void vxlan_cleanup(struct timer_list *t)
2927 {
2928         struct vxlan_dev *vxlan = from_timer(vxlan, t, age_timer);
2929         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2930         unsigned int h;
2931
2932         if (!netif_running(vxlan->dev))
2933                 return;
2934
2935         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2936                 struct hlist_node *p, *n;
2937
2938                 spin_lock(&vxlan->hash_lock[h]);
2939                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2940                         struct vxlan_fdb *f
2941                                 = container_of(p, struct vxlan_fdb, hlist);
2942                         unsigned long timeout;
2943
2944                         if (f->state & (NUD_PERMANENT | NUD_NOARP))
2945                                 continue;
2946
2947                         if (f->flags & NTF_EXT_LEARNED)
2948                                 continue;
2949
2950                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2951                         if (time_before_eq(timeout, jiffies)) {
2952                                 netdev_dbg(vxlan->dev,
2953                                            "garbage collect %pM\n",
2954                                            f->eth_addr);
2955                                 f->state = NUD_STALE;
2956                                 vxlan_fdb_destroy(vxlan, f, true, true);
2957                         } else if (time_before(timeout, next_timer))
2958                                 next_timer = timeout;
2959                 }
2960                 spin_unlock(&vxlan->hash_lock[h]);
2961         }
2962
2963         mod_timer(&vxlan->age_timer, next_timer);
2964 }
2965
2966 static void vxlan_vs_del_dev(struct vxlan_dev *vxlan)
2967 {
2968         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2969
2970         spin_lock(&vn->sock_lock);
2971         hlist_del_init_rcu(&vxlan->hlist4.hlist);
2972 #if IS_ENABLED(CONFIG_IPV6)
2973         hlist_del_init_rcu(&vxlan->hlist6.hlist);
2974 #endif
2975         spin_unlock(&vn->sock_lock);
2976 }
2977
2978 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan,
2979                              struct vxlan_dev_node *node)
2980 {
2981         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2982         __be32 vni = vxlan->default_dst.remote_vni;
2983
2984         node->vxlan = vxlan;
2985         spin_lock(&vn->sock_lock);
2986         hlist_add_head_rcu(&node->hlist, vni_head(vs, vni));
2987         spin_unlock(&vn->sock_lock);
2988 }
2989
2990 /* Setup stats when device is created */
2991 static int vxlan_init(struct net_device *dev)
2992 {
2993         struct vxlan_dev *vxlan = netdev_priv(dev);
2994         int err;
2995
2996         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2997         if (!dev->tstats)
2998                 return -ENOMEM;
2999
3000         err = gro_cells_init(&vxlan->gro_cells, dev);
3001         if (err) {
3002                 free_percpu(dev->tstats);
3003                 return err;
3004         }
3005
3006         return 0;
3007 }
3008
3009 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan, __be32 vni)
3010 {
3011         struct vxlan_fdb *f;
3012         u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, vni);
3013
3014         spin_lock_bh(&vxlan->hash_lock[hash_index]);
3015         f = __vxlan_find_mac(vxlan, all_zeros_mac, vni);
3016         if (f)
3017                 vxlan_fdb_destroy(vxlan, f, true, true);
3018         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
3019 }
3020
3021 static void vxlan_uninit(struct net_device *dev)
3022 {
3023         struct vxlan_dev *vxlan = netdev_priv(dev);
3024
3025         gro_cells_destroy(&vxlan->gro_cells);
3026
3027         vxlan_fdb_delete_default(vxlan, vxlan->cfg.vni);
3028
3029         free_percpu(dev->tstats);
3030 }
3031
3032 /* Start ageing timer and join group when device is brought up */
3033 static int vxlan_open(struct net_device *dev)
3034 {
3035         struct vxlan_dev *vxlan = netdev_priv(dev);
3036         int ret;
3037
3038         ret = vxlan_sock_add(vxlan);
3039         if (ret < 0)
3040                 return ret;
3041
3042         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
3043                 ret = vxlan_igmp_join(vxlan);
3044                 if (ret == -EADDRINUSE)
3045                         ret = 0;
3046                 if (ret) {
3047                         vxlan_sock_release(vxlan);
3048                         return ret;
3049                 }
3050         }
3051
3052         if (vxlan->cfg.age_interval)
3053                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
3054
3055         return ret;
3056 }
3057
3058 /* Purge the forwarding table */
3059 static void vxlan_flush(struct vxlan_dev *vxlan, bool do_all)
3060 {
3061         unsigned int h;
3062
3063         for (h = 0; h < FDB_HASH_SIZE; ++h) {
3064                 struct hlist_node *p, *n;
3065
3066                 spin_lock_bh(&vxlan->hash_lock[h]);
3067                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
3068                         struct vxlan_fdb *f
3069                                 = container_of(p, struct vxlan_fdb, hlist);
3070                         if (!do_all && (f->state & (NUD_PERMANENT | NUD_NOARP)))
3071                                 continue;
3072                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
3073                         if (!is_zero_ether_addr(f->eth_addr))
3074                                 vxlan_fdb_destroy(vxlan, f, true, true);
3075                 }
3076                 spin_unlock_bh(&vxlan->hash_lock[h]);
3077         }
3078 }
3079
3080 /* Cleanup timer and forwarding table on shutdown */
3081 static int vxlan_stop(struct net_device *dev)
3082 {
3083         struct vxlan_dev *vxlan = netdev_priv(dev);
3084         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3085         int ret = 0;
3086
3087         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
3088             !vxlan_group_used(vn, vxlan))
3089                 ret = vxlan_igmp_leave(vxlan);
3090
3091         del_timer_sync(&vxlan->age_timer);
3092
3093         vxlan_flush(vxlan, false);
3094         vxlan_sock_release(vxlan);
3095
3096         return ret;
3097 }
3098
3099 /* Stub, nothing needs to be done. */
3100 static void vxlan_set_multicast_list(struct net_device *dev)
3101 {
3102 }
3103
3104 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
3105 {
3106         struct vxlan_dev *vxlan = netdev_priv(dev);
3107         struct vxlan_rdst *dst = &vxlan->default_dst;
3108         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
3109                                                          dst->remote_ifindex);
3110         bool use_ipv6 = !!(vxlan->cfg.flags & VXLAN_F_IPV6);
3111
3112         /* This check is different than dev->max_mtu, because it looks at
3113          * the lowerdev->mtu, rather than the static dev->max_mtu
3114          */
3115         if (lowerdev) {
3116                 int max_mtu = lowerdev->mtu -
3117                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
3118                 if (new_mtu > max_mtu)
3119                         return -EINVAL;
3120         }
3121
3122         dev->mtu = new_mtu;
3123         return 0;
3124 }
3125
3126 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
3127 {
3128         struct vxlan_dev *vxlan = netdev_priv(dev);
3129         struct ip_tunnel_info *info = skb_tunnel_info(skb);
3130         __be16 sport, dport;
3131
3132         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
3133                                   vxlan->cfg.port_max, true);
3134         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
3135
3136         if (ip_tunnel_info_af(info) == AF_INET) {
3137                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
3138                 struct rtable *rt;
3139
3140                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
3141                                      info->key.u.ipv4.dst,
3142                                      &info->key.u.ipv4.src, dport, sport,
3143                                      &info->dst_cache, info);
3144                 if (IS_ERR(rt))
3145                         return PTR_ERR(rt);
3146                 ip_rt_put(rt);
3147         } else {
3148 #if IS_ENABLED(CONFIG_IPV6)
3149                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
3150                 struct dst_entry *ndst;
3151
3152                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
3153                                         info->key.label, &info->key.u.ipv6.dst,
3154                                         &info->key.u.ipv6.src, dport, sport,
3155                                         &info->dst_cache, info);
3156                 if (IS_ERR(ndst))
3157                         return PTR_ERR(ndst);
3158                 dst_release(ndst);
3159 #else /* !CONFIG_IPV6 */
3160                 return -EPFNOSUPPORT;
3161 #endif
3162         }
3163         info->key.tp_src = sport;
3164         info->key.tp_dst = dport;
3165         return 0;
3166 }
3167
3168 static const struct net_device_ops vxlan_netdev_ether_ops = {
3169         .ndo_init               = vxlan_init,
3170         .ndo_uninit             = vxlan_uninit,
3171         .ndo_open               = vxlan_open,
3172         .ndo_stop               = vxlan_stop,
3173         .ndo_start_xmit         = vxlan_xmit,
3174         .ndo_get_stats64        = ip_tunnel_get_stats64,
3175         .ndo_set_rx_mode        = vxlan_set_multicast_list,
3176         .ndo_change_mtu         = vxlan_change_mtu,
3177         .ndo_validate_addr      = eth_validate_addr,
3178         .ndo_set_mac_address    = eth_mac_addr,
3179         .ndo_fdb_add            = vxlan_fdb_add,
3180         .ndo_fdb_del            = vxlan_fdb_delete,
3181         .ndo_fdb_dump           = vxlan_fdb_dump,
3182         .ndo_fdb_get            = vxlan_fdb_get,
3183         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
3184         .ndo_change_proto_down  = dev_change_proto_down_generic,
3185 };
3186
3187 static const struct net_device_ops vxlan_netdev_raw_ops = {
3188         .ndo_init               = vxlan_init,
3189         .ndo_uninit             = vxlan_uninit,
3190         .ndo_open               = vxlan_open,
3191         .ndo_stop               = vxlan_stop,
3192         .ndo_start_xmit         = vxlan_xmit,
3193         .ndo_get_stats64        = ip_tunnel_get_stats64,
3194         .ndo_change_mtu         = vxlan_change_mtu,
3195         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
3196 };
3197
3198 /* Info for udev, that this is a virtual tunnel endpoint */
3199 static struct device_type vxlan_type = {
3200         .name = "vxlan",
3201 };
3202
3203 /* Calls the ndo_udp_tunnel_add of the caller in order to
3204  * supply the listening VXLAN udp ports. Callers are expected
3205  * to implement the ndo_udp_tunnel_add.
3206  */
3207 static void vxlan_offload_rx_ports(struct net_device *dev, bool push)
3208 {
3209         struct vxlan_sock *vs;
3210         struct net *net = dev_net(dev);
3211         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3212         unsigned int i;
3213
3214         spin_lock(&vn->sock_lock);
3215         for (i = 0; i < PORT_HASH_SIZE; ++i) {
3216                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist) {
3217                         unsigned short type;
3218
3219                         if (vs->flags & VXLAN_F_GPE)
3220                                 type = UDP_TUNNEL_TYPE_VXLAN_GPE;
3221                         else
3222                                 type = UDP_TUNNEL_TYPE_VXLAN;
3223
3224                         if (push)
3225                                 udp_tunnel_push_rx_port(dev, vs->sock, type);
3226                         else
3227                                 udp_tunnel_drop_rx_port(dev, vs->sock, type);
3228                 }
3229         }
3230         spin_unlock(&vn->sock_lock);
3231 }
3232
3233 /* Initialize the device structure. */
3234 static void vxlan_setup(struct net_device *dev)
3235 {
3236         struct vxlan_dev *vxlan = netdev_priv(dev);
3237         unsigned int h;
3238
3239         eth_hw_addr_random(dev);
3240         ether_setup(dev);
3241
3242         dev->needs_free_netdev = true;
3243         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
3244
3245         dev->features   |= NETIF_F_LLTX;
3246         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
3247         dev->features   |= NETIF_F_RXCSUM;
3248         dev->features   |= NETIF_F_GSO_SOFTWARE;
3249
3250         dev->vlan_features = dev->features;
3251         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
3252         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
3253         netif_keep_dst(dev);
3254         dev->priv_flags |= IFF_NO_QUEUE;
3255
3256         /* MTU range: 68 - 65535 */
3257         dev->min_mtu = ETH_MIN_MTU;
3258         dev->max_mtu = ETH_MAX_MTU;
3259
3260         INIT_LIST_HEAD(&vxlan->next);
3261
3262         timer_setup(&vxlan->age_timer, vxlan_cleanup, TIMER_DEFERRABLE);
3263
3264         vxlan->dev = dev;
3265
3266         for (h = 0; h < FDB_HASH_SIZE; ++h) {
3267                 spin_lock_init(&vxlan->hash_lock[h]);
3268                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
3269         }
3270 }
3271
3272 static void vxlan_ether_setup(struct net_device *dev)
3273 {
3274         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
3275         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3276         dev->netdev_ops = &vxlan_netdev_ether_ops;
3277 }
3278
3279 static void vxlan_raw_setup(struct net_device *dev)
3280 {
3281         dev->header_ops = NULL;
3282         dev->type = ARPHRD_NONE;
3283         dev->hard_header_len = 0;
3284         dev->addr_len = 0;
3285         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
3286         dev->netdev_ops = &vxlan_netdev_raw_ops;
3287 }
3288
3289 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
3290         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
3291         [IFLA_VXLAN_GROUP]      = { .len = sizeof_field(struct iphdr, daddr) },
3292         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
3293         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
3294         [IFLA_VXLAN_LOCAL]      = { .len = sizeof_field(struct iphdr, saddr) },
3295         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
3296         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
3297         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
3298         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
3299         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
3300         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
3301         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
3302         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
3303         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
3304         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
3305         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
3306         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
3307         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
3308         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
3309         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
3310         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
3311         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
3312         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
3313         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
3314         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
3315         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
3316         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
3317         [IFLA_VXLAN_TTL_INHERIT]        = { .type = NLA_FLAG },
3318         [IFLA_VXLAN_DF]         = { .type = NLA_U8 },
3319 };
3320
3321 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[],
3322                           struct netlink_ext_ack *extack)
3323 {
3324         if (tb[IFLA_ADDRESS]) {
3325                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
3326                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3327                                             "Provided link layer address is not Ethernet");
3328                         return -EINVAL;
3329                 }
3330
3331                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
3332                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_ADDRESS],
3333                                             "Provided Ethernet address is not unicast");
3334                         return -EADDRNOTAVAIL;
3335                 }
3336         }
3337
3338         if (tb[IFLA_MTU]) {
3339                 u32 mtu = nla_get_u32(tb[IFLA_MTU]);
3340
3341                 if (mtu < ETH_MIN_MTU || mtu > ETH_MAX_MTU) {
3342                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
3343                                             "MTU must be between 68 and 65535");
3344                         return -EINVAL;
3345                 }
3346         }
3347
3348         if (!data) {
3349                 NL_SET_ERR_MSG(extack,
3350                                "Required attributes not provided to perform the operation");
3351                 return -EINVAL;
3352         }
3353
3354         if (data[IFLA_VXLAN_ID]) {
3355                 u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
3356
3357                 if (id >= VXLAN_N_VID) {
3358                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_ID],
3359                                             "VXLAN ID must be lower than 16777216");
3360                         return -ERANGE;
3361                 }
3362         }
3363
3364         if (data[IFLA_VXLAN_PORT_RANGE]) {
3365                 const struct ifla_vxlan_port_range *p
3366                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3367
3368                 if (ntohs(p->high) < ntohs(p->low)) {
3369                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_PORT_RANGE],
3370                                             "Invalid source port range");
3371                         return -EINVAL;
3372                 }
3373         }
3374
3375         if (data[IFLA_VXLAN_DF]) {
3376                 enum ifla_vxlan_df df = nla_get_u8(data[IFLA_VXLAN_DF]);
3377
3378                 if (df < 0 || df > VXLAN_DF_MAX) {
3379                         NL_SET_ERR_MSG_ATTR(extack, data[IFLA_VXLAN_DF],
3380                                             "Invalid DF attribute");
3381                         return -EINVAL;
3382                 }
3383         }
3384
3385         return 0;
3386 }
3387
3388 static void vxlan_get_drvinfo(struct net_device *netdev,
3389                               struct ethtool_drvinfo *drvinfo)
3390 {
3391         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
3392         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
3393 }
3394
3395 static int vxlan_get_link_ksettings(struct net_device *dev,
3396                                     struct ethtool_link_ksettings *cmd)
3397 {
3398         struct vxlan_dev *vxlan = netdev_priv(dev);
3399         struct vxlan_rdst *dst = &vxlan->default_dst;
3400         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
3401                                                          dst->remote_ifindex);
3402
3403         if (!lowerdev) {
3404                 cmd->base.duplex = DUPLEX_UNKNOWN;
3405                 cmd->base.port = PORT_OTHER;
3406                 cmd->base.speed = SPEED_UNKNOWN;
3407
3408                 return 0;
3409         }
3410
3411         return __ethtool_get_link_ksettings(lowerdev, cmd);
3412 }
3413
3414 static const struct ethtool_ops vxlan_ethtool_ops = {
3415         .get_drvinfo            = vxlan_get_drvinfo,
3416         .get_link               = ethtool_op_get_link,
3417         .get_link_ksettings     = vxlan_get_link_ksettings,
3418 };
3419
3420 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
3421                                         __be16 port, u32 flags, int ifindex)
3422 {
3423         struct socket *sock;
3424         struct udp_port_cfg udp_conf;
3425         int err;
3426
3427         memset(&udp_conf, 0, sizeof(udp_conf));
3428
3429         if (ipv6) {
3430                 udp_conf.family = AF_INET6;
3431                 udp_conf.use_udp6_rx_checksums =
3432                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
3433                 udp_conf.ipv6_v6only = 1;
3434         } else {
3435                 udp_conf.family = AF_INET;
3436         }
3437
3438         udp_conf.local_udp_port = port;
3439         udp_conf.bind_ifindex = ifindex;
3440
3441         /* Open UDP socket */
3442         err = udp_sock_create(net, &udp_conf, &sock);
3443         if (err < 0)
3444                 return ERR_PTR(err);
3445
3446         return sock;
3447 }
3448
3449 /* Create new listen socket if needed */
3450 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
3451                                               __be16 port, u32 flags,
3452                                               int ifindex)
3453 {
3454         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3455         struct vxlan_sock *vs;
3456         struct socket *sock;
3457         unsigned int h;
3458         struct udp_tunnel_sock_cfg tunnel_cfg;
3459
3460         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
3461         if (!vs)
3462                 return ERR_PTR(-ENOMEM);
3463
3464         for (h = 0; h < VNI_HASH_SIZE; ++h)
3465                 INIT_HLIST_HEAD(&vs->vni_list[h]);
3466
3467         sock = vxlan_create_sock(net, ipv6, port, flags, ifindex);
3468         if (IS_ERR(sock)) {
3469                 kfree(vs);
3470                 return ERR_CAST(sock);
3471         }
3472
3473         vs->sock = sock;
3474         refcount_set(&vs->refcnt, 1);
3475         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
3476
3477         spin_lock(&vn->sock_lock);
3478         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
3479         udp_tunnel_notify_add_rx_port(sock,
3480                                       (vs->flags & VXLAN_F_GPE) ?
3481                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
3482                                       UDP_TUNNEL_TYPE_VXLAN);
3483         spin_unlock(&vn->sock_lock);
3484
3485         /* Mark socket as an encapsulation socket. */
3486         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
3487         tunnel_cfg.sk_user_data = vs;
3488         tunnel_cfg.encap_type = 1;
3489         tunnel_cfg.encap_rcv = vxlan_rcv;
3490         tunnel_cfg.encap_err_lookup = vxlan_err_lookup;
3491         tunnel_cfg.encap_destroy = NULL;
3492         tunnel_cfg.gro_receive = vxlan_gro_receive;
3493         tunnel_cfg.gro_complete = vxlan_gro_complete;
3494
3495         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
3496
3497         return vs;
3498 }
3499
3500 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
3501 {
3502         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3503         struct vxlan_sock *vs = NULL;
3504         struct vxlan_dev_node *node;
3505         int l3mdev_index = 0;
3506
3507         if (vxlan->cfg.remote_ifindex)
3508                 l3mdev_index = l3mdev_master_upper_ifindex_by_index(
3509                         vxlan->net, vxlan->cfg.remote_ifindex);
3510
3511         if (!vxlan->cfg.no_share) {
3512                 spin_lock(&vn->sock_lock);
3513                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
3514                                      vxlan->cfg.dst_port, vxlan->cfg.flags,
3515                                      l3mdev_index);
3516                 if (vs && !refcount_inc_not_zero(&vs->refcnt)) {
3517                         spin_unlock(&vn->sock_lock);
3518                         return -EBUSY;
3519                 }
3520                 spin_unlock(&vn->sock_lock);
3521         }
3522         if (!vs)
3523                 vs = vxlan_socket_create(vxlan->net, ipv6,
3524                                          vxlan->cfg.dst_port, vxlan->cfg.flags,
3525                                          l3mdev_index);
3526         if (IS_ERR(vs))
3527                 return PTR_ERR(vs);
3528 #if IS_ENABLED(CONFIG_IPV6)
3529         if (ipv6) {
3530                 rcu_assign_pointer(vxlan->vn6_sock, vs);
3531                 node = &vxlan->hlist6;
3532         } else
3533 #endif
3534         {
3535                 rcu_assign_pointer(vxlan->vn4_sock, vs);
3536                 node = &vxlan->hlist4;
3537         }
3538         vxlan_vs_add_dev(vs, vxlan, node);
3539         return 0;
3540 }
3541
3542 static int vxlan_sock_add(struct vxlan_dev *vxlan)
3543 {
3544         bool metadata = vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA;
3545         bool ipv6 = vxlan->cfg.flags & VXLAN_F_IPV6 || metadata;
3546         bool ipv4 = !ipv6 || metadata;
3547         int ret = 0;
3548
3549         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
3550 #if IS_ENABLED(CONFIG_IPV6)
3551         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
3552         if (ipv6) {
3553                 ret = __vxlan_sock_add(vxlan, true);
3554                 if (ret < 0 && ret != -EAFNOSUPPORT)
3555                         ipv4 = false;
3556         }
3557 #endif
3558         if (ipv4)
3559                 ret = __vxlan_sock_add(vxlan, false);
3560         if (ret < 0)
3561                 vxlan_sock_release(vxlan);
3562         return ret;
3563 }
3564
3565 static int vxlan_config_validate(struct net *src_net, struct vxlan_config *conf,
3566                                  struct net_device **lower,
3567                                  struct vxlan_dev *old,
3568                                  struct netlink_ext_ack *extack)
3569 {
3570         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
3571         struct vxlan_dev *tmp;
3572         bool use_ipv6 = false;
3573
3574         if (conf->flags & VXLAN_F_GPE) {
3575                 /* For now, allow GPE only together with
3576                  * COLLECT_METADATA. This can be relaxed later; in such
3577                  * case, the other side of the PtP link will have to be
3578                  * provided.
3579                  */
3580                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
3581                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3582                         NL_SET_ERR_MSG(extack,
3583                                        "VXLAN GPE does not support this combination of attributes");
3584                         return -EINVAL;
3585                 }
3586         }
3587
3588         if (!conf->remote_ip.sa.sa_family && !conf->saddr.sa.sa_family) {
3589                 /* Unless IPv6 is explicitly requested, assume IPv4 */
3590                 conf->remote_ip.sa.sa_family = AF_INET;
3591                 conf->saddr.sa.sa_family = AF_INET;
3592         } else if (!conf->remote_ip.sa.sa_family) {
3593                 conf->remote_ip.sa.sa_family = conf->saddr.sa.sa_family;
3594         } else if (!conf->saddr.sa.sa_family) {
3595                 conf->saddr.sa.sa_family = conf->remote_ip.sa.sa_family;
3596         }
3597
3598         if (conf->saddr.sa.sa_family != conf->remote_ip.sa.sa_family) {
3599                 NL_SET_ERR_MSG(extack,
3600                                "Local and remote address must be from the same family");
3601                 return -EINVAL;
3602         }
3603
3604         if (vxlan_addr_multicast(&conf->saddr)) {
3605                 NL_SET_ERR_MSG(extack, "Local address cannot be multicast");
3606                 return -EINVAL;
3607         }
3608
3609         if (conf->saddr.sa.sa_family == AF_INET6) {
3610                 if (!IS_ENABLED(CONFIG_IPV6)) {
3611                         NL_SET_ERR_MSG(extack,
3612                                        "IPv6 support not enabled in the kernel");
3613                         return -EPFNOSUPPORT;
3614                 }
3615                 use_ipv6 = true;
3616                 conf->flags |= VXLAN_F_IPV6;
3617
3618                 if (!(conf->flags & VXLAN_F_COLLECT_METADATA)) {
3619                         int local_type =
3620                                 ipv6_addr_type(&conf->saddr.sin6.sin6_addr);
3621                         int remote_type =
3622                                 ipv6_addr_type(&conf->remote_ip.sin6.sin6_addr);
3623
3624                         if (local_type & IPV6_ADDR_LINKLOCAL) {
3625                                 if (!(remote_type & IPV6_ADDR_LINKLOCAL) &&
3626                                     (remote_type != IPV6_ADDR_ANY)) {
3627                                         NL_SET_ERR_MSG(extack,
3628                                                        "Invalid combination of local and remote address scopes");
3629                                         return -EINVAL;
3630                                 }
3631
3632                                 conf->flags |= VXLAN_F_IPV6_LINKLOCAL;
3633                         } else {
3634                                 if (remote_type ==
3635                                     (IPV6_ADDR_UNICAST | IPV6_ADDR_LINKLOCAL)) {
3636                                         NL_SET_ERR_MSG(extack,
3637                                                        "Invalid combination of local and remote address scopes");
3638                                         return -EINVAL;
3639                                 }
3640
3641                                 conf->flags &= ~VXLAN_F_IPV6_LINKLOCAL;
3642                         }
3643                 }
3644         }
3645
3646         if (conf->label && !use_ipv6) {
3647                 NL_SET_ERR_MSG(extack,
3648                                "Label attribute only applies to IPv6 VXLAN devices");
3649                 return -EINVAL;
3650         }
3651
3652         if (conf->remote_ifindex) {
3653                 struct net_device *lowerdev;
3654
3655                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
3656                 if (!lowerdev) {
3657                         NL_SET_ERR_MSG(extack,
3658                                        "Invalid local interface, device not found");
3659                         return -ENODEV;
3660                 }
3661
3662 #if IS_ENABLED(CONFIG_IPV6)
3663                 if (use_ipv6) {
3664                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
3665                         if (idev && idev->cnf.disable_ipv6) {
3666                                 NL_SET_ERR_MSG(extack,
3667                                                "IPv6 support disabled by administrator");
3668                                 return -EPERM;
3669                         }
3670                 }
3671 #endif
3672
3673                 *lower = lowerdev;
3674         } else {
3675                 if (vxlan_addr_multicast(&conf->remote_ip)) {
3676                         NL_SET_ERR_MSG(extack,
3677                                        "Local interface required for multicast remote destination");
3678
3679                         return -EINVAL;
3680                 }
3681
3682 #if IS_ENABLED(CONFIG_IPV6)
3683                 if (conf->flags & VXLAN_F_IPV6_LINKLOCAL) {
3684                         NL_SET_ERR_MSG(extack,
3685                                        "Local interface required for link-local local/remote addresses");
3686                         return -EINVAL;
3687                 }
3688 #endif
3689
3690                 *lower = NULL;
3691         }
3692
3693         if (!conf->dst_port) {
3694                 if (conf->flags & VXLAN_F_GPE)
3695                         conf->dst_port = htons(4790); /* IANA VXLAN-GPE port */
3696                 else
3697                         conf->dst_port = htons(vxlan_port);
3698         }
3699
3700         if (!conf->age_interval)
3701                 conf->age_interval = FDB_AGE_DEFAULT;
3702
3703         list_for_each_entry(tmp, &vn->vxlan_list, next) {
3704                 if (tmp == old)
3705                         continue;
3706
3707                 if (tmp->cfg.vni != conf->vni)
3708                         continue;
3709                 if (tmp->cfg.dst_port != conf->dst_port)
3710                         continue;
3711                 if ((tmp->cfg.flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)) !=
3712                     (conf->flags & (VXLAN_F_RCV_FLAGS | VXLAN_F_IPV6)))
3713                         continue;
3714
3715                 if ((conf->flags & VXLAN_F_IPV6_LINKLOCAL) &&
3716                     tmp->cfg.remote_ifindex != conf->remote_ifindex)
3717                         continue;
3718
3719                 NL_SET_ERR_MSG(extack,
3720                                "A VXLAN device with the specified VNI already exists");
3721                 return -EEXIST;
3722         }
3723
3724         return 0;
3725 }
3726
3727 static void vxlan_config_apply(struct net_device *dev,
3728                                struct vxlan_config *conf,
3729                                struct net_device *lowerdev,
3730                                struct net *src_net,
3731                                bool changelink)
3732 {
3733         struct vxlan_dev *vxlan = netdev_priv(dev);
3734         struct vxlan_rdst *dst = &vxlan->default_dst;
3735         unsigned short needed_headroom = ETH_HLEN;
3736         bool use_ipv6 = !!(conf->flags & VXLAN_F_IPV6);
3737         int max_mtu = ETH_MAX_MTU;
3738
3739         if (!changelink) {
3740                 if (conf->flags & VXLAN_F_GPE)
3741                         vxlan_raw_setup(dev);
3742                 else
3743                         vxlan_ether_setup(dev);
3744
3745                 if (conf->mtu)
3746                         dev->mtu = conf->mtu;
3747
3748                 vxlan->net = src_net;
3749         }
3750
3751         dst->remote_vni = conf->vni;
3752
3753         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
3754
3755         if (lowerdev) {
3756                 dst->remote_ifindex = conf->remote_ifindex;
3757
3758                 dev->gso_max_size = lowerdev->gso_max_size;
3759                 dev->gso_max_segs = lowerdev->gso_max_segs;
3760
3761                 needed_headroom = lowerdev->hard_header_len;
3762
3763                 max_mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM :
3764                                            VXLAN_HEADROOM);
3765                 if (max_mtu < ETH_MIN_MTU)
3766                         max_mtu = ETH_MIN_MTU;
3767
3768                 if (!changelink && !conf->mtu)
3769                         dev->mtu = max_mtu;
3770         }
3771
3772         if (dev->mtu > max_mtu)
3773                 dev->mtu = max_mtu;
3774
3775         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
3776                 needed_headroom += VXLAN6_HEADROOM;
3777         else
3778                 needed_headroom += VXLAN_HEADROOM;
3779         dev->needed_headroom = needed_headroom;
3780
3781         memcpy(&vxlan->cfg, conf, sizeof(*conf));
3782 }
3783
3784 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
3785                                struct vxlan_config *conf, bool changelink,
3786                                struct netlink_ext_ack *extack)
3787 {
3788         struct vxlan_dev *vxlan = netdev_priv(dev);
3789         struct net_device *lowerdev;
3790         int ret;
3791
3792         ret = vxlan_config_validate(src_net, conf, &lowerdev, vxlan, extack);
3793         if (ret)
3794                 return ret;
3795
3796         vxlan_config_apply(dev, conf, lowerdev, src_net, changelink);
3797
3798         return 0;
3799 }
3800
3801 static int __vxlan_dev_create(struct net *net, struct net_device *dev,
3802                               struct vxlan_config *conf,
3803                               struct netlink_ext_ack *extack)
3804 {
3805         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3806         struct vxlan_dev *vxlan = netdev_priv(dev);
3807         struct net_device *remote_dev = NULL;
3808         struct vxlan_fdb *f = NULL;
3809         bool unregister = false;
3810         struct vxlan_rdst *dst;
3811         int err;
3812
3813         dst = &vxlan->default_dst;
3814         err = vxlan_dev_configure(net, dev, conf, false, extack);
3815         if (err)
3816                 return err;
3817
3818         dev->ethtool_ops = &vxlan_ethtool_ops;
3819
3820         /* create an fdb entry for a valid default destination */
3821         if (!vxlan_addr_any(&dst->remote_ip)) {
3822                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
3823                                        &dst->remote_ip,
3824                                        NUD_REACHABLE | NUD_PERMANENT,
3825                                        vxlan->cfg.dst_port,
3826                                        dst->remote_vni,
3827                                        dst->remote_vni,
3828                                        dst->remote_ifindex,
3829                                        NTF_SELF, 0, &f, extack);
3830                 if (err)
3831                         return err;
3832         }
3833
3834         err = register_netdevice(dev);
3835         if (err)
3836                 goto errout;
3837         unregister = true;
3838
3839         if (dst->remote_ifindex) {
3840                 remote_dev = __dev_get_by_index(net, dst->remote_ifindex);
3841                 if (!remote_dev)
3842                         goto errout;
3843
3844                 err = netdev_upper_dev_link(remote_dev, dev, extack);
3845                 if (err)
3846                         goto errout;
3847         }
3848
3849         err = rtnl_configure_link(dev, NULL);
3850         if (err)
3851                 goto unlink;
3852
3853         if (f) {
3854                 vxlan_fdb_insert(vxlan, all_zeros_mac, dst->remote_vni, f);
3855
3856                 /* notify default fdb entry */
3857                 err = vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f),
3858                                        RTM_NEWNEIGH, true, extack);
3859                 if (err) {
3860                         vxlan_fdb_destroy(vxlan, f, false, false);
3861                         if (remote_dev)
3862                                 netdev_upper_dev_unlink(remote_dev, dev);
3863                         goto unregister;
3864                 }
3865         }
3866
3867         list_add(&vxlan->next, &vn->vxlan_list);
3868         if (remote_dev)
3869                 dst->remote_dev = remote_dev;
3870         return 0;
3871 unlink:
3872         if (remote_dev)
3873                 netdev_upper_dev_unlink(remote_dev, dev);
3874 errout:
3875         /* unregister_netdevice() destroys the default FDB entry with deletion
3876          * notification. But the addition notification was not sent yet, so
3877          * destroy the entry by hand here.
3878          */
3879         if (f)
3880                 __vxlan_fdb_free(f);
3881 unregister:
3882         if (unregister)
3883                 unregister_netdevice(dev);
3884         return err;
3885 }
3886
3887 /* Set/clear flags based on attribute */
3888 static int vxlan_nl2flag(struct vxlan_config *conf, struct nlattr *tb[],
3889                           int attrtype, unsigned long mask, bool changelink,
3890                           bool changelink_supported,
3891                           struct netlink_ext_ack *extack)
3892 {
3893         unsigned long flags;
3894
3895         if (!tb[attrtype])
3896                 return 0;
3897
3898         if (changelink && !changelink_supported) {
3899                 vxlan_flag_attr_error(attrtype, extack);
3900                 return -EOPNOTSUPP;
3901         }
3902
3903         if (vxlan_policy[attrtype].type == NLA_FLAG)
3904                 flags = conf->flags | mask;
3905         else if (nla_get_u8(tb[attrtype]))
3906                 flags = conf->flags | mask;
3907         else
3908                 flags = conf->flags & ~mask;
3909
3910         conf->flags = flags;
3911
3912         return 0;
3913 }
3914
3915 static int vxlan_nl2conf(struct nlattr *tb[], struct nlattr *data[],
3916                          struct net_device *dev, struct vxlan_config *conf,
3917                          bool changelink, struct netlink_ext_ack *extack)
3918 {
3919         struct vxlan_dev *vxlan = netdev_priv(dev);
3920         int err = 0;
3921
3922         memset(conf, 0, sizeof(*conf));
3923
3924         /* if changelink operation, start with old existing cfg */
3925         if (changelink)
3926                 memcpy(conf, &vxlan->cfg, sizeof(*conf));
3927
3928         if (data[IFLA_VXLAN_ID]) {
3929                 __be32 vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3930
3931                 if (changelink && (vni != conf->vni)) {
3932                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_ID], "Cannot change VNI");
3933                         return -EOPNOTSUPP;
3934                 }
3935                 conf->vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
3936         }
3937
3938         if (data[IFLA_VXLAN_GROUP]) {
3939                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET)) {
3940                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP], "New group address family does not match old group");
3941                         return -EOPNOTSUPP;
3942                 }
3943
3944                 conf->remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
3945                 conf->remote_ip.sa.sa_family = AF_INET;
3946         } else if (data[IFLA_VXLAN_GROUP6]) {
3947                 if (!IS_ENABLED(CONFIG_IPV6)) {
3948                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "IPv6 support not enabled in the kernel");
3949                         return -EPFNOSUPPORT;
3950                 }
3951
3952                 if (changelink && (conf->remote_ip.sa.sa_family != AF_INET6)) {
3953                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_GROUP6], "New group address family does not match old group");
3954                         return -EOPNOTSUPP;
3955                 }
3956
3957                 conf->remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
3958                 conf->remote_ip.sa.sa_family = AF_INET6;
3959         }
3960
3961         if (data[IFLA_VXLAN_LOCAL]) {
3962                 if (changelink && (conf->saddr.sa.sa_family != AF_INET)) {
3963                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL], "New local address family does not match old");
3964                         return -EOPNOTSUPP;
3965                 }
3966
3967                 conf->saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
3968                 conf->saddr.sa.sa_family = AF_INET;
3969         } else if (data[IFLA_VXLAN_LOCAL6]) {
3970                 if (!IS_ENABLED(CONFIG_IPV6)) {
3971                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "IPv6 support not enabled in the kernel");
3972                         return -EPFNOSUPPORT;
3973                 }
3974
3975                 if (changelink && (conf->saddr.sa.sa_family != AF_INET6)) {
3976                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LOCAL6], "New local address family does not match old");
3977                         return -EOPNOTSUPP;
3978                 }
3979
3980                 /* TODO: respect scope id */
3981                 conf->saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
3982                 conf->saddr.sa.sa_family = AF_INET6;
3983         }
3984
3985         if (data[IFLA_VXLAN_LINK])
3986                 conf->remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
3987
3988         if (data[IFLA_VXLAN_TOS])
3989                 conf->tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
3990
3991         if (data[IFLA_VXLAN_TTL])
3992                 conf->ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
3993
3994         if (data[IFLA_VXLAN_TTL_INHERIT]) {
3995                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_TTL_INHERIT,
3996                                     VXLAN_F_TTL_INHERIT, changelink, false,
3997                                     extack);
3998                 if (err)
3999                         return err;
4000
4001         }
4002
4003         if (data[IFLA_VXLAN_LABEL])
4004                 conf->label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
4005                              IPV6_FLOWLABEL_MASK;
4006
4007         if (data[IFLA_VXLAN_LEARNING]) {
4008                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_LEARNING,
4009                                     VXLAN_F_LEARN, changelink, true,
4010                                     extack);
4011                 if (err)
4012                         return err;
4013         } else if (!changelink) {
4014                 /* default to learn on a new device */
4015                 conf->flags |= VXLAN_F_LEARN;
4016         }
4017
4018         if (data[IFLA_VXLAN_AGEING])
4019                 conf->age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
4020
4021         if (data[IFLA_VXLAN_PROXY]) {
4022                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_PROXY,
4023                                     VXLAN_F_PROXY, changelink, false,
4024                                     extack);
4025                 if (err)
4026                         return err;
4027         }
4028
4029         if (data[IFLA_VXLAN_RSC]) {
4030                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_RSC,
4031                                     VXLAN_F_RSC, changelink, false,
4032                                     extack);
4033                 if (err)
4034                         return err;
4035         }
4036
4037         if (data[IFLA_VXLAN_L2MISS]) {
4038                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L2MISS,
4039                                     VXLAN_F_L2MISS, changelink, false,
4040                                     extack);
4041                 if (err)
4042                         return err;
4043         }
4044
4045         if (data[IFLA_VXLAN_L3MISS]) {
4046                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_L3MISS,
4047                                     VXLAN_F_L3MISS, changelink, false,
4048                                     extack);
4049                 if (err)
4050                         return err;
4051         }
4052
4053         if (data[IFLA_VXLAN_LIMIT]) {
4054                 if (changelink) {
4055                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_LIMIT],
4056                                             "Cannot change limit");
4057                         return -EOPNOTSUPP;
4058                 }
4059                 conf->addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
4060         }
4061
4062         if (data[IFLA_VXLAN_COLLECT_METADATA]) {
4063                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_COLLECT_METADATA,
4064                                     VXLAN_F_COLLECT_METADATA, changelink, false,
4065                                     extack);
4066                 if (err)
4067                         return err;
4068         }
4069
4070         if (data[IFLA_VXLAN_PORT_RANGE]) {
4071                 if (!changelink) {
4072                         const struct ifla_vxlan_port_range *p
4073                                 = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
4074                         conf->port_min = ntohs(p->low);
4075                         conf->port_max = ntohs(p->high);
4076                 } else {
4077                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT_RANGE],
4078                                             "Cannot change port range");
4079                         return -EOPNOTSUPP;
4080                 }
4081         }
4082
4083         if (data[IFLA_VXLAN_PORT]) {
4084                 if (changelink) {
4085                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_PORT],
4086                                             "Cannot change port");
4087                         return -EOPNOTSUPP;
4088                 }
4089                 conf->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
4090         }
4091
4092         if (data[IFLA_VXLAN_UDP_CSUM]) {
4093                 if (changelink) {
4094                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_VXLAN_UDP_CSUM],
4095                                             "Cannot change UDP_CSUM flag");
4096                         return -EOPNOTSUPP;
4097                 }
4098                 if (!nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
4099                         conf->flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
4100         }
4101
4102         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]) {
4103                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4104                                     VXLAN_F_UDP_ZERO_CSUM6_TX, changelink,
4105                                     false, extack);
4106                 if (err)
4107                         return err;
4108         }
4109
4110         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]) {
4111                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4112                                     VXLAN_F_UDP_ZERO_CSUM6_RX, changelink,
4113                                     false, extack);
4114                 if (err)
4115                         return err;
4116         }
4117
4118         if (data[IFLA_VXLAN_REMCSUM_TX]) {
4119                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_TX,
4120                                     VXLAN_F_REMCSUM_TX, changelink, false,
4121                                     extack);
4122                 if (err)
4123                         return err;
4124         }
4125
4126         if (data[IFLA_VXLAN_REMCSUM_RX]) {
4127                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_RX,
4128                                     VXLAN_F_REMCSUM_RX, changelink, false,
4129                                     extack);
4130                 if (err)
4131                         return err;
4132         }
4133
4134         if (data[IFLA_VXLAN_GBP]) {
4135                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GBP,
4136                                     VXLAN_F_GBP, changelink, false, extack);
4137                 if (err)
4138                         return err;
4139         }
4140
4141         if (data[IFLA_VXLAN_GPE]) {
4142                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_GPE,
4143                                     VXLAN_F_GPE, changelink, false,
4144                                     extack);
4145                 if (err)
4146                         return err;
4147         }
4148
4149         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL]) {
4150                 err = vxlan_nl2flag(conf, data, IFLA_VXLAN_REMCSUM_NOPARTIAL,
4151                                     VXLAN_F_REMCSUM_NOPARTIAL, changelink,
4152                                     false, extack);
4153                 if (err)
4154                         return err;
4155         }
4156
4157         if (tb[IFLA_MTU]) {
4158                 if (changelink) {
4159                         NL_SET_ERR_MSG_ATTR(extack, tb[IFLA_MTU],
4160                                             "Cannot change mtu");
4161                         return -EOPNOTSUPP;
4162                 }
4163                 conf->mtu = nla_get_u32(tb[IFLA_MTU]);
4164         }
4165
4166         if (data[IFLA_VXLAN_DF])
4167                 conf->df = nla_get_u8(data[IFLA_VXLAN_DF]);
4168
4169         return 0;
4170 }
4171
4172 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
4173                          struct nlattr *tb[], struct nlattr *data[],
4174                          struct netlink_ext_ack *extack)
4175 {
4176         struct vxlan_config conf;
4177         int err;
4178
4179         err = vxlan_nl2conf(tb, data, dev, &conf, false, extack);
4180         if (err)
4181                 return err;
4182
4183         return __vxlan_dev_create(src_net, dev, &conf, extack);
4184 }
4185
4186 static int vxlan_changelink(struct net_device *dev, struct nlattr *tb[],
4187                             struct nlattr *data[],
4188                             struct netlink_ext_ack *extack)
4189 {
4190         struct vxlan_dev *vxlan = netdev_priv(dev);
4191         struct net_device *lowerdev;
4192         struct vxlan_config conf;
4193         struct vxlan_rdst *dst;
4194         int err;
4195
4196         dst = &vxlan->default_dst;
4197         err = vxlan_nl2conf(tb, data, dev, &conf, true, extack);
4198         if (err)
4199                 return err;
4200
4201         err = vxlan_config_validate(vxlan->net, &conf, &lowerdev,
4202                                     vxlan, extack);
4203         if (err)
4204                 return err;
4205
4206         if (dst->remote_dev == lowerdev)
4207                 lowerdev = NULL;
4208
4209         err = netdev_adjacent_change_prepare(dst->remote_dev, lowerdev, dev,
4210                                              extack);
4211         if (err)
4212                 return err;
4213
4214         /* handle default dst entry */
4215         if (!vxlan_addr_equal(&conf.remote_ip, &dst->remote_ip)) {
4216                 u32 hash_index = fdb_head_index(vxlan, all_zeros_mac, conf.vni);
4217
4218                 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4219                 if (!vxlan_addr_any(&conf.remote_ip)) {
4220                         err = vxlan_fdb_update(vxlan, all_zeros_mac,
4221                                                &conf.remote_ip,
4222                                                NUD_REACHABLE | NUD_PERMANENT,
4223                                                NLM_F_APPEND | NLM_F_CREATE,
4224                                                vxlan->cfg.dst_port,
4225                                                conf.vni, conf.vni,
4226                                                conf.remote_ifindex,
4227                                                NTF_SELF, 0, true, extack);
4228                         if (err) {
4229                                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4230                                 netdev_adjacent_change_abort(dst->remote_dev,
4231                                                              lowerdev, dev);
4232                                 return err;
4233                         }
4234                 }
4235                 if (!vxlan_addr_any(&dst->remote_ip))
4236                         __vxlan_fdb_delete(vxlan, all_zeros_mac,
4237                                            dst->remote_ip,
4238                                            vxlan->cfg.dst_port,
4239                                            dst->remote_vni,
4240                                            dst->remote_vni,
4241                                            dst->remote_ifindex,
4242                                            true);
4243                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4244         }
4245
4246         if (conf.age_interval != vxlan->cfg.age_interval)
4247                 mod_timer(&vxlan->age_timer, jiffies);
4248
4249         netdev_adjacent_change_commit(dst->remote_dev, lowerdev, dev);
4250         if (lowerdev && lowerdev != dst->remote_dev)
4251                 dst->remote_dev = lowerdev;
4252         vxlan_config_apply(dev, &conf, lowerdev, vxlan->net, true);
4253         return 0;
4254 }
4255
4256 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
4257 {
4258         struct vxlan_dev *vxlan = netdev_priv(dev);
4259
4260         vxlan_flush(vxlan, true);
4261
4262         list_del(&vxlan->next);
4263         unregister_netdevice_queue(dev, head);
4264         if (vxlan->default_dst.remote_dev)
4265                 netdev_upper_dev_unlink(vxlan->default_dst.remote_dev, dev);
4266 }
4267
4268 static size_t vxlan_get_size(const struct net_device *dev)
4269 {
4270
4271         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
4272                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
4273                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
4274                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
4275                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
4276                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL_INHERIT */
4277                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
4278                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_DF */
4279                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
4280                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
4281                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
4282                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
4283                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
4284                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
4285                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
4286                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
4287                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
4288                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
4289                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
4290                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
4291                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
4292                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
4293                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
4294                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
4295                 0;
4296 }
4297
4298 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
4299 {
4300         const struct vxlan_dev *vxlan = netdev_priv(dev);
4301         const struct vxlan_rdst *dst = &vxlan->default_dst;
4302         struct ifla_vxlan_port_range ports = {
4303                 .low =  htons(vxlan->cfg.port_min),
4304                 .high = htons(vxlan->cfg.port_max),
4305         };
4306
4307         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
4308                 goto nla_put_failure;
4309
4310         if (!vxlan_addr_any(&dst->remote_ip)) {
4311                 if (dst->remote_ip.sa.sa_family == AF_INET) {
4312                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
4313                                             dst->remote_ip.sin.sin_addr.s_addr))
4314                                 goto nla_put_failure;
4315 #if IS_ENABLED(CONFIG_IPV6)
4316                 } else {
4317                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
4318                                              &dst->remote_ip.sin6.sin6_addr))
4319                                 goto nla_put_failure;
4320 #endif
4321                 }
4322         }
4323
4324         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
4325                 goto nla_put_failure;
4326
4327         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
4328                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
4329                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
4330                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
4331                                 goto nla_put_failure;
4332 #if IS_ENABLED(CONFIG_IPV6)
4333                 } else {
4334                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
4335                                              &vxlan->cfg.saddr.sin6.sin6_addr))
4336                                 goto nla_put_failure;
4337 #endif
4338                 }
4339         }
4340
4341         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
4342             nla_put_u8(skb, IFLA_VXLAN_TTL_INHERIT,
4343                        !!(vxlan->cfg.flags & VXLAN_F_TTL_INHERIT)) ||
4344             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
4345             nla_put_u8(skb, IFLA_VXLAN_DF, vxlan->cfg.df) ||
4346             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
4347             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
4348                        !!(vxlan->cfg.flags & VXLAN_F_LEARN)) ||
4349             nla_put_u8(skb, IFLA_VXLAN_PROXY,
4350                        !!(vxlan->cfg.flags & VXLAN_F_PROXY)) ||
4351             nla_put_u8(skb, IFLA_VXLAN_RSC,
4352                        !!(vxlan->cfg.flags & VXLAN_F_RSC)) ||
4353             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
4354                        !!(vxlan->cfg.flags & VXLAN_F_L2MISS)) ||
4355             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
4356                        !!(vxlan->cfg.flags & VXLAN_F_L3MISS)) ||
4357             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
4358                        !!(vxlan->cfg.flags & VXLAN_F_COLLECT_METADATA)) ||
4359             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
4360             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
4361             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
4362             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
4363                        !(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
4364             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
4365                        !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
4366             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
4367                        !!(vxlan->cfg.flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
4368             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
4369                        !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_TX)) ||
4370             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
4371                        !!(vxlan->cfg.flags & VXLAN_F_REMCSUM_RX)))
4372                 goto nla_put_failure;
4373
4374         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
4375                 goto nla_put_failure;
4376
4377         if (vxlan->cfg.flags & VXLAN_F_GBP &&
4378             nla_put_flag(skb, IFLA_VXLAN_GBP))
4379                 goto nla_put_failure;
4380
4381         if (vxlan->cfg.flags & VXLAN_F_GPE &&
4382             nla_put_flag(skb, IFLA_VXLAN_GPE))
4383                 goto nla_put_failure;
4384
4385         if (vxlan->cfg.flags & VXLAN_F_REMCSUM_NOPARTIAL &&
4386             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
4387                 goto nla_put_failure;
4388
4389         return 0;
4390
4391 nla_put_failure:
4392         return -EMSGSIZE;
4393 }
4394
4395 static struct net *vxlan_get_link_net(const struct net_device *dev)
4396 {
4397         struct vxlan_dev *vxlan = netdev_priv(dev);
4398
4399         return vxlan->net;
4400 }
4401
4402 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
4403         .kind           = "vxlan",
4404         .maxtype        = IFLA_VXLAN_MAX,
4405         .policy         = vxlan_policy,
4406         .priv_size      = sizeof(struct vxlan_dev),
4407         .setup          = vxlan_setup,
4408         .validate       = vxlan_validate,
4409         .newlink        = vxlan_newlink,
4410         .changelink     = vxlan_changelink,
4411         .dellink        = vxlan_dellink,
4412         .get_size       = vxlan_get_size,
4413         .fill_info      = vxlan_fill_info,
4414         .get_link_net   = vxlan_get_link_net,
4415 };
4416
4417 struct net_device *vxlan_dev_create(struct net *net, const char *name,
4418                                     u8 name_assign_type,
4419                                     struct vxlan_config *conf)
4420 {
4421         struct nlattr *tb[IFLA_MAX + 1];
4422         struct net_device *dev;
4423         int err;
4424
4425         memset(&tb, 0, sizeof(tb));
4426
4427         dev = rtnl_create_link(net, name, name_assign_type,
4428                                &vxlan_link_ops, tb, NULL);
4429         if (IS_ERR(dev))
4430                 return dev;
4431
4432         err = __vxlan_dev_create(net, dev, conf, NULL);
4433         if (err < 0) {
4434                 free_netdev(dev);
4435                 return ERR_PTR(err);
4436         }
4437
4438         err = rtnl_configure_link(dev, NULL);
4439         if (err < 0) {
4440                 LIST_HEAD(list_kill);
4441
4442                 vxlan_dellink(dev, &list_kill);
4443                 unregister_netdevice_many(&list_kill);
4444                 return ERR_PTR(err);
4445         }
4446
4447         return dev;
4448 }
4449 EXPORT_SYMBOL_GPL(vxlan_dev_create);
4450
4451 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
4452                                              struct net_device *dev)
4453 {
4454         struct vxlan_dev *vxlan, *next;
4455         LIST_HEAD(list_kill);
4456
4457         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4458                 struct vxlan_rdst *dst = &vxlan->default_dst;
4459
4460                 /* In case we created vxlan device with carrier
4461                  * and we loose the carrier due to module unload
4462                  * we also need to remove vxlan device. In other
4463                  * cases, it's not necessary and remote_ifindex
4464                  * is 0 here, so no matches.
4465                  */
4466                 if (dst->remote_ifindex == dev->ifindex)
4467                         vxlan_dellink(vxlan->dev, &list_kill);
4468         }
4469
4470         unregister_netdevice_many(&list_kill);
4471 }
4472
4473 static int vxlan_netdevice_event(struct notifier_block *unused,
4474                                  unsigned long event, void *ptr)
4475 {
4476         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4477         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
4478
4479         if (event == NETDEV_UNREGISTER) {
4480                 vxlan_offload_rx_ports(dev, false);
4481                 vxlan_handle_lowerdev_unregister(vn, dev);
4482         } else if (event == NETDEV_REGISTER) {
4483                 vxlan_offload_rx_ports(dev, true);
4484         } else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO ||
4485                    event == NETDEV_UDP_TUNNEL_DROP_INFO) {
4486                 vxlan_offload_rx_ports(dev, event == NETDEV_UDP_TUNNEL_PUSH_INFO);
4487         }
4488
4489         return NOTIFY_DONE;
4490 }
4491
4492 static struct notifier_block vxlan_notifier_block __read_mostly = {
4493         .notifier_call = vxlan_netdevice_event,
4494 };
4495
4496 static void
4497 vxlan_fdb_offloaded_set(struct net_device *dev,
4498                         struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4499 {
4500         struct vxlan_dev *vxlan = netdev_priv(dev);
4501         struct vxlan_rdst *rdst;
4502         struct vxlan_fdb *f;
4503         u32 hash_index;
4504
4505         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4506
4507         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4508
4509         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4510         if (!f)
4511                 goto out;
4512
4513         rdst = vxlan_fdb_find_rdst(f, &fdb_info->remote_ip,
4514                                    fdb_info->remote_port,
4515                                    fdb_info->remote_vni,
4516                                    fdb_info->remote_ifindex);
4517         if (!rdst)
4518                 goto out;
4519
4520         rdst->offloaded = fdb_info->offloaded;
4521
4522 out:
4523         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4524 }
4525
4526 static int
4527 vxlan_fdb_external_learn_add(struct net_device *dev,
4528                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4529 {
4530         struct vxlan_dev *vxlan = netdev_priv(dev);
4531         struct netlink_ext_ack *extack;
4532         u32 hash_index;
4533         int err;
4534
4535         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4536         extack = switchdev_notifier_info_to_extack(&fdb_info->info);
4537
4538         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4539         err = vxlan_fdb_update(vxlan, fdb_info->eth_addr, &fdb_info->remote_ip,
4540                                NUD_REACHABLE,
4541                                NLM_F_CREATE | NLM_F_REPLACE,
4542                                fdb_info->remote_port,
4543                                fdb_info->vni,
4544                                fdb_info->remote_vni,
4545                                fdb_info->remote_ifindex,
4546                                NTF_USE | NTF_SELF | NTF_EXT_LEARNED,
4547                                0, false, extack);
4548         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4549
4550         return err;
4551 }
4552
4553 static int
4554 vxlan_fdb_external_learn_del(struct net_device *dev,
4555                              struct switchdev_notifier_vxlan_fdb_info *fdb_info)
4556 {
4557         struct vxlan_dev *vxlan = netdev_priv(dev);
4558         struct vxlan_fdb *f;
4559         u32 hash_index;
4560         int err = 0;
4561
4562         hash_index = fdb_head_index(vxlan, fdb_info->eth_addr, fdb_info->vni);
4563         spin_lock_bh(&vxlan->hash_lock[hash_index]);
4564
4565         f = vxlan_find_mac(vxlan, fdb_info->eth_addr, fdb_info->vni);
4566         if (!f)
4567                 err = -ENOENT;
4568         else if (f->flags & NTF_EXT_LEARNED)
4569                 err = __vxlan_fdb_delete(vxlan, fdb_info->eth_addr,
4570                                          fdb_info->remote_ip,
4571                                          fdb_info->remote_port,
4572                                          fdb_info->vni,
4573                                          fdb_info->remote_vni,
4574                                          fdb_info->remote_ifindex,
4575                                          false);
4576
4577         spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4578
4579         return err;
4580 }
4581
4582 static int vxlan_switchdev_event(struct notifier_block *unused,
4583                                  unsigned long event, void *ptr)
4584 {
4585         struct net_device *dev = switchdev_notifier_info_to_dev(ptr);
4586         struct switchdev_notifier_vxlan_fdb_info *fdb_info;
4587         int err = 0;
4588
4589         switch (event) {
4590         case SWITCHDEV_VXLAN_FDB_OFFLOADED:
4591                 vxlan_fdb_offloaded_set(dev, ptr);
4592                 break;
4593         case SWITCHDEV_VXLAN_FDB_ADD_TO_BRIDGE:
4594                 fdb_info = ptr;
4595                 err = vxlan_fdb_external_learn_add(dev, fdb_info);
4596                 if (err) {
4597                         err = notifier_from_errno(err);
4598                         break;
4599                 }
4600                 fdb_info->offloaded = true;
4601                 vxlan_fdb_offloaded_set(dev, fdb_info);
4602                 break;
4603         case SWITCHDEV_VXLAN_FDB_DEL_TO_BRIDGE:
4604                 fdb_info = ptr;
4605                 err = vxlan_fdb_external_learn_del(dev, fdb_info);
4606                 if (err) {
4607                         err = notifier_from_errno(err);
4608                         break;
4609                 }
4610                 fdb_info->offloaded = false;
4611                 vxlan_fdb_offloaded_set(dev, fdb_info);
4612                 break;
4613         }
4614
4615         return err;
4616 }
4617
4618 static struct notifier_block vxlan_switchdev_notifier_block __read_mostly = {
4619         .notifier_call = vxlan_switchdev_event,
4620 };
4621
4622 static void vxlan_fdb_nh_flush(struct nexthop *nh)
4623 {
4624         struct vxlan_fdb *fdb;
4625         struct vxlan_dev *vxlan;
4626         u32 hash_index;
4627
4628         rcu_read_lock();
4629         list_for_each_entry_rcu(fdb, &nh->fdb_list, nh_list) {
4630                 vxlan = rcu_dereference(fdb->vdev);
4631                 WARN_ON(!vxlan);
4632                 hash_index = fdb_head_index(vxlan, fdb->eth_addr,
4633                                             vxlan->default_dst.remote_vni);
4634                 spin_lock_bh(&vxlan->hash_lock[hash_index]);
4635                 if (!hlist_unhashed(&fdb->hlist))
4636                         vxlan_fdb_destroy(vxlan, fdb, false, false);
4637                 spin_unlock_bh(&vxlan->hash_lock[hash_index]);
4638         }
4639         rcu_read_unlock();
4640 }
4641
4642 static int vxlan_nexthop_event(struct notifier_block *nb,
4643                                unsigned long event, void *ptr)
4644 {
4645         struct nexthop *nh = ptr;
4646
4647         if (!nh || event != NEXTHOP_EVENT_DEL)
4648                 return NOTIFY_DONE;
4649
4650         vxlan_fdb_nh_flush(nh);
4651
4652         return NOTIFY_DONE;
4653 }
4654
4655 static struct notifier_block vxlan_nexthop_notifier_block __read_mostly = {
4656         .notifier_call = vxlan_nexthop_event,
4657 };
4658
4659 static __net_init int vxlan_init_net(struct net *net)
4660 {
4661         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4662         unsigned int h;
4663
4664         INIT_LIST_HEAD(&vn->vxlan_list);
4665         spin_lock_init(&vn->sock_lock);
4666
4667         for (h = 0; h < PORT_HASH_SIZE; ++h)
4668                 INIT_HLIST_HEAD(&vn->sock_list[h]);
4669
4670         return register_nexthop_notifier(net, &vxlan_nexthop_notifier_block);
4671 }
4672
4673 static void vxlan_destroy_tunnels(struct net *net, struct list_head *head)
4674 {
4675         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
4676         struct vxlan_dev *vxlan, *next;
4677         struct net_device *dev, *aux;
4678         unsigned int h;
4679
4680         for_each_netdev_safe(net, dev, aux)
4681                 if (dev->rtnl_link_ops == &vxlan_link_ops)
4682                         unregister_netdevice_queue(dev, head);
4683
4684         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
4685                 /* If vxlan->dev is in the same netns, it has already been added
4686                  * to the list by the previous loop.
4687                  */
4688                 if (!net_eq(dev_net(vxlan->dev), net))
4689                         unregister_netdevice_queue(vxlan->dev, head);
4690         }
4691
4692         for (h = 0; h < PORT_HASH_SIZE; ++h)
4693                 WARN_ON_ONCE(!hlist_empty(&vn->sock_list[h]));
4694 }
4695
4696 static void __net_exit vxlan_exit_batch_net(struct list_head *net_list)
4697 {
4698         struct net *net;
4699         LIST_HEAD(list);
4700
4701         rtnl_lock();
4702         list_for_each_entry(net, net_list, exit_list)
4703                 unregister_nexthop_notifier(net, &vxlan_nexthop_notifier_block);
4704         list_for_each_entry(net, net_list, exit_list)
4705                 vxlan_destroy_tunnels(net, &list);
4706
4707         unregister_netdevice_many(&list);
4708         rtnl_unlock();
4709 }
4710
4711 static struct pernet_operations vxlan_net_ops = {
4712         .init = vxlan_init_net,
4713         .exit_batch = vxlan_exit_batch_net,
4714         .id   = &vxlan_net_id,
4715         .size = sizeof(struct vxlan_net),
4716 };
4717
4718 static int __init vxlan_init_module(void)
4719 {
4720         int rc;
4721
4722         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
4723
4724         rc = register_pernet_subsys(&vxlan_net_ops);
4725         if (rc)
4726                 goto out1;
4727
4728         rc = register_netdevice_notifier(&vxlan_notifier_block);
4729         if (rc)
4730                 goto out2;
4731
4732         rc = register_switchdev_notifier(&vxlan_switchdev_notifier_block);
4733         if (rc)
4734                 goto out3;
4735
4736         rc = rtnl_link_register(&vxlan_link_ops);
4737         if (rc)
4738                 goto out4;
4739
4740         return 0;
4741 out4:
4742         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4743 out3:
4744         unregister_netdevice_notifier(&vxlan_notifier_block);
4745 out2:
4746         unregister_pernet_subsys(&vxlan_net_ops);
4747 out1:
4748         return rc;
4749 }
4750 late_initcall(vxlan_init_module);
4751
4752 static void __exit vxlan_cleanup_module(void)
4753 {
4754         rtnl_link_unregister(&vxlan_link_ops);
4755         unregister_switchdev_notifier(&vxlan_switchdev_notifier_block);
4756         unregister_netdevice_notifier(&vxlan_notifier_block);
4757         unregister_pernet_subsys(&vxlan_net_ops);
4758         /* rcu_barrier() is called by netns */
4759 }
4760 module_exit(vxlan_cleanup_module);
4761
4762 MODULE_LICENSE("GPL");
4763 MODULE_VERSION(VXLAN_VERSION);
4764 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
4765 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
4766 MODULE_ALIAS_RTNL_LINK("vxlan");