drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/errno.h>
  16 #include <linux/slab.h>
  17 #include <linux/udp.h>
  18 #include <linux/igmp.h>
  19 #include <linux/if_ether.h>
  20 #include <linux/ethtool.h>
  21 #include <net/arp.h>
  22 #include <net/ndisc.h>
  23 #include <net/ip.h>
  24 #include <net/icmp.h>
  25 #include <net/rtnetlink.h>
  26 #include <net/inet_ecn.h>
  27 #include <net/net_namespace.h>
  28 #include <net/netns/generic.h>
  29 #include <net/vxlan.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 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 /* per-network namespace private data for this module */
  63 struct vxlan_net {
  64         struct list_head  vxlan_list;
  65         struct hlist_head sock_list[PORT_HASH_SIZE];
  66         spinlock_t        sock_lock;
  67 };
  68
  69 /* Forwarding table entry */
  70 struct vxlan_fdb {
  71         struct hlist_node hlist;        /* linked list of entries */
  72         struct rcu_head   rcu;
  73         unsigned long     updated;      /* jiffies */
  74         unsigned long     used;
  75         struct list_head  remotes;
  76         u8                eth_addr[ETH_ALEN];
  77         u16               state;        /* see ndm_state */
  78         u8                flags;        /* see ndm_flags */
  79 };
  80
  81 /* salt for hash table */
  82 static u32 vxlan_salt __read_mostly;
  83
  84 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  85 {
  86         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  87                ip_tunnel_collect_metadata();
  88 }
  89
  90 #if IS_ENABLED(CONFIG_IPV6)
  91 static inline
  92 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
  93 {
  94         if (a->sa.sa_family != b->sa.sa_family)
  95                 return false;
  96         if (a->sa.sa_family == AF_INET6)
  97                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
  98         else
  99                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 100 }
 101
 102 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 103 {
 104         if (ipa->sa.sa_family == AF_INET6)
 105                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 106         else
 107                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 108 }
 109
 110 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 111 {
 112         if (ipa->sa.sa_family == AF_INET6)
 113                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 114         else
 115                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 116 }
 117
 118 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 119 {
 120         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 121                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 122                 ip->sa.sa_family = AF_INET6;
 123                 return 0;
 124         } else if (nla_len(nla) >= sizeof(__be32)) {
 125                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 126                 ip->sa.sa_family = AF_INET;
 127                 return 0;
 128         } else {
 129                 return -EAFNOSUPPORT;
 130         }
 131 }
 132
 133 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 134                               const union vxlan_addr *ip)
 135 {
 136         if (ip->sa.sa_family == AF_INET6)
 137                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 138         else
 139                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 140 }
 141
 142 #else /* !CONFIG_IPV6 */
 143
 144 static inline
 145 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 146 {
 147         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 148 }
 149
 150 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 151 {
 152         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 153 }
 154
 155 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 156 {
 157         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 158 }
 159
 160 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 161 {
 162         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 163                 return -EAFNOSUPPORT;
 164         } else if (nla_len(nla) >= sizeof(__be32)) {
 165                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 166                 ip->sa.sa_family = AF_INET;
 167                 return 0;
 168         } else {
 169                 return -EAFNOSUPPORT;
 170         }
 171 }
 172
 173 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 174                               const union vxlan_addr *ip)
 175 {
 176         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 177 }
 178 #endif
 179
 180 /* Virtual Network hash table head */
 181 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 182 {
 183         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 184 }
 185
 186 /* Socket hash table head */
 187 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 188 {
 189         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 190
 191         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 192 }
 193
 194 /* First remote destination for a forwarding entry.
 195  * Guaranteed to be non-NULL because remotes are never deleted.
 196  */
 197 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 198 {
 199         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 200 }
 201
 202 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 203 {
 204         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 205 }
 206
 207 /* Find VXLAN socket based on network namespace, address family and UDP port
 208  * and enabled unshareable flags.
 209  */
 210 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 211                                           __be16 port, u32 flags)
 212 {
 213         struct vxlan_sock *vs;
 214
 215         flags &= VXLAN_F_RCV_FLAGS;
 216
 217         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 218                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 219                     vxlan_get_sk_family(vs) == family &&
 220                     vs->flags == flags)
 221                         return vs;
 222         }
 223         return NULL;
 224 }
 225
 226 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
 227 {
 228         struct vxlan_dev *vxlan;
 229
 230         /* For flow based devices, map all packets to VNI 0 */
 231         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 232                 vni = 0;
 233
 234         hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
 235                 if (vxlan->default_dst.remote_vni == vni)
 236                         return vxlan;
 237         }
 238
 239         return NULL;
 240 }
 241
 242 /* Look up VNI in a per net namespace table */
 243 static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
 244                                         sa_family_t family, __be16 port,
 245                                         u32 flags)
 246 {
 247         struct vxlan_sock *vs;
 248
 249         vs = vxlan_find_sock(net, family, port, flags);
 250         if (!vs)
 251                 return NULL;
 252
 253         return vxlan_vs_find_vni(vs, vni);
 254 }
 255
 256 /* Fill in neighbour message in skbuff. */
 257 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 258                           const struct vxlan_fdb *fdb,
 259                           u32 portid, u32 seq, int type, unsigned int flags,
 260                           const struct vxlan_rdst *rdst)
 261 {
 262         unsigned long now = jiffies;
 263         struct nda_cacheinfo ci;
 264         struct nlmsghdr *nlh;
 265         struct ndmsg *ndm;
 266         bool send_ip, send_eth;
 267
 268         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 269         if (nlh == NULL)
 270                 return -EMSGSIZE;
 271
 272         ndm = nlmsg_data(nlh);
 273         memset(ndm, 0, sizeof(*ndm));
 274
 275         send_eth = send_ip = true;
 276
 277         if (type == RTM_GETNEIGH) {
 278                 ndm->ndm_family = AF_INET;
 279                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 280                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 281         } else
 282                 ndm->ndm_family = AF_BRIDGE;
 283         ndm->ndm_state = fdb->state;
 284         ndm->ndm_ifindex = vxlan->dev->ifindex;
 285         ndm->ndm_flags = fdb->flags;
 286         ndm->ndm_type = RTN_UNICAST;
 287
 288         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 289             nla_put_s32(skb, NDA_LINK_NETNSID,
 290                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 291                 goto nla_put_failure;
 292
 293         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 294                 goto nla_put_failure;
 295
 296         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 297                 goto nla_put_failure;
 298
 299         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 300             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 301                 goto nla_put_failure;
 302         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 303             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 304                 goto nla_put_failure;
 305         if (rdst->remote_ifindex &&
 306             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 307                 goto nla_put_failure;
 308
 309         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 310         ci.ndm_confirmed = 0;
 311         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 312         ci.ndm_refcnt    = 0;
 313
 314         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 315                 goto nla_put_failure;
 316
 317         nlmsg_end(skb, nlh);
 318         return 0;
 319
 320 nla_put_failure:
 321         nlmsg_cancel(skb, nlh);
 322         return -EMSGSIZE;
 323 }
 324
 325 static inline size_t vxlan_nlmsg_size(void)
 326 {
 327         return NLMSG_ALIGN(sizeof(struct ndmsg))
 328                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 329                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 330                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 331                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 332                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 333                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 334                 + nla_total_size(sizeof(struct nda_cacheinfo));
 335 }
 336
 337 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 338                              struct vxlan_rdst *rd, int type)
 339 {
 340         struct net *net = dev_net(vxlan->dev);
 341         struct sk_buff *skb;
 342         int err = -ENOBUFS;
 343
 344         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 345         if (skb == NULL)
 346                 goto errout;
 347
 348         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 349         if (err < 0) {
 350                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 351                 WARN_ON(err == -EMSGSIZE);
 352                 kfree_skb(skb);
 353                 goto errout;
 354         }
 355
 356         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 357         return;
 358 errout:
 359         if (err < 0)
 360                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 361 }
 362
 363 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 364 {
 365         struct vxlan_dev *vxlan = netdev_priv(dev);
 366         struct vxlan_fdb f = {
 367                 .state = NUD_STALE,
 368         };
 369         struct vxlan_rdst remote = {
 370                 .remote_ip = *ipa, /* goes to NDA_DST */
 371                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 372         };
 373
 374         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 375 }
 376
 377 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 378 {
 379         struct vxlan_fdb f = {
 380                 .state = NUD_STALE,
 381         };
 382         struct vxlan_rdst remote = { };
 383
 384         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 385
 386         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 387 }
 388
 389 /* Hash Ethernet address */
 390 static u32 eth_hash(const unsigned char *addr)
 391 {
 392         u64 value = get_unaligned((u64 *)addr);
 393
 394         /* only want 6 bytes */
 395 #ifdef __BIG_ENDIAN
 396         value >>= 16;
 397 #else
 398         value <<= 16;
 399 #endif
 400         return hash_64(value, FDB_HASH_BITS);
 401 }
 402
 403 /* Hash chain to use given mac address */
 404 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 405                                                 const u8 *mac)
 406 {
 407         return &vxlan->fdb_head[eth_hash(mac)];
 408 }
 409
 410 /* Look up Ethernet address in forwarding table */
 411 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 412                                         const u8 *mac)
 413 {
 414         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 415         struct vxlan_fdb *f;
 416
 417         hlist_for_each_entry_rcu(f, head, hlist) {
 418                 if (ether_addr_equal(mac, f->eth_addr))
 419                         return f;
 420         }
 421
 422         return NULL;
 423 }
 424
 425 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 426                                         const u8 *mac)
 427 {
 428         struct vxlan_fdb *f;
 429
 430         f = __vxlan_find_mac(vxlan, mac);
 431         if (f)
 432                 f->used = jiffies;
 433
 434         return f;
 435 }
 436
 437 /* caller should hold vxlan->hash_lock */
 438 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 439                                               union vxlan_addr *ip, __be16 port,
 440                                               __be32 vni, __u32 ifindex)
 441 {
 442         struct vxlan_rdst *rd;
 443
 444         list_for_each_entry(rd, &f->remotes, list) {
 445                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 446                     rd->remote_port == port &&
 447                     rd->remote_vni == vni &&
 448                     rd->remote_ifindex == ifindex)
 449                         return rd;
 450         }
 451
 452         return NULL;
 453 }
 454
 455 /* Replace destination of unicast mac */
 456 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 457                              union vxlan_addr *ip, __be16 port, __be32 vni,
 458                              __u32 ifindex)
 459 {
 460         struct vxlan_rdst *rd;
 461
 462         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 463         if (rd)
 464                 return 0;
 465
 466         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 467         if (!rd)
 468                 return 0;
 469
 470         dst_cache_reset(&rd->dst_cache);
 471         rd->remote_ip = *ip;
 472         rd->remote_port = port;
 473         rd->remote_vni = vni;
 474         rd->remote_ifindex = ifindex;
 475         return 1;
 476 }
 477
 478 /* Add/update destinations for multicast */
 479 static int vxlan_fdb_append(struct vxlan_fdb *f,
 480                             union vxlan_addr *ip, __be16 port, __be32 vni,
 481                             __u32 ifindex, struct vxlan_rdst **rdp)
 482 {
 483         struct vxlan_rdst *rd;
 484
 485         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 486         if (rd)
 487                 return 0;
 488
 489         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 490         if (rd == NULL)
 491                 return -ENOBUFS;
 492
 493         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 494                 kfree(rd);
 495                 return -ENOBUFS;
 496         }
 497
 498         rd->remote_ip = *ip;
 499         rd->remote_port = port;
 500         rd->remote_vni = vni;
 501         rd->remote_ifindex = ifindex;
 502
 503         list_add_tail_rcu(&rd->list, &f->remotes);
 504
 505         *rdp = rd;
 506         return 1;
 507 }
 508
 509 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 510                                           unsigned int off,
 511                                           struct vxlanhdr *vh, size_t hdrlen,
 512                                           __be32 vni_field,
 513                                           struct gro_remcsum *grc,
 514                                           bool nopartial)
 515 {
 516         size_t start, offset;
 517
 518         if (skb->remcsum_offload)
 519                 return vh;
 520
 521         if (!NAPI_GRO_CB(skb)->csum_valid)
 522                 return NULL;
 523
 524         start = vxlan_rco_start(vni_field);
 525         offset = start + vxlan_rco_offset(vni_field);
 526
 527         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 528                                      start, offset, grc, nopartial);
 529
 530         skb->remcsum_offload = 1;
 531
 532         return vh;
 533 }
 534
 535 static struct sk_buff **vxlan_gro_receive(struct sock *sk,
 536                                           struct sk_buff **head,
 537                                           struct sk_buff *skb)
 538 {
 539         struct sk_buff *p, **pp = NULL;
 540         struct vxlanhdr *vh, *vh2;
 541         unsigned int hlen, off_vx;
 542         int flush = 1;
 543         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 544         __be32 flags;
 545         struct gro_remcsum grc;
 546
 547         skb_gro_remcsum_init(&grc);
 548
 549         off_vx = skb_gro_offset(skb);
 550         hlen = off_vx + sizeof(*vh);
 551         vh   = skb_gro_header_fast(skb, off_vx);
 552         if (skb_gro_header_hard(skb, hlen)) {
 553                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 554                 if (unlikely(!vh))
 555                         goto out;
 556         }
 557
 558         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 559
 560         flags = vh->vx_flags;
 561
 562         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 563                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 564                                        vh->vx_vni, &grc,
 565                                        !!(vs->flags &
 566                                           VXLAN_F_REMCSUM_NOPARTIAL));
 567
 568                 if (!vh)
 569                         goto out;
 570         }
 571
 572         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 573
 574         for (p = *head; p; p = p->next) {
 575                 if (!NAPI_GRO_CB(p)->same_flow)
 576                         continue;
 577
 578                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 579                 if (vh->vx_flags != vh2->vx_flags ||
 580                     vh->vx_vni != vh2->vx_vni) {
 581                         NAPI_GRO_CB(p)->same_flow = 0;
 582                         continue;
 583                 }
 584         }
 585
 586         pp = call_gro_receive(eth_gro_receive, head, skb);
 587         flush = 0;
 588
 589 out:
 590         skb_gro_remcsum_cleanup(skb, &grc);
 591         NAPI_GRO_CB(skb)->flush |= flush;
 592
 593         return pp;
 594 }
 595
 596 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 597 {
 598         /* Sets 'skb->inner_mac_header' since we are always called with
 599          * 'skb->encapsulation' set.
 600          */
 601         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 602 }
 603
 604 /* Add new entry to forwarding table -- assumes lock held */
 605 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 606                             const u8 *mac, union vxlan_addr *ip,
 607                             __u16 state, __u16 flags,
 608                             __be16 port, __be32 vni, __u32 ifindex,
 609                             __u8 ndm_flags)
 610 {
 611         struct vxlan_rdst *rd = NULL;
 612         struct vxlan_fdb *f;
 613         int notify = 0;
 614
 615         f = __vxlan_find_mac(vxlan, mac);
 616         if (f) {
 617                 if (flags & NLM_F_EXCL) {
 618                         netdev_dbg(vxlan->dev,
 619                                    "lost race to create %pM\n", mac);
 620                         return -EEXIST;
 621                 }
 622                 if (f->state != state) {
 623                         f->state = state;
 624                         f->updated = jiffies;
 625                         notify = 1;
 626                 }
 627                 if (f->flags != ndm_flags) {
 628                         f->flags = ndm_flags;
 629                         f->updated = jiffies;
 630                         notify = 1;
 631                 }
 632                 if ((flags & NLM_F_REPLACE)) {
 633                         /* Only change unicasts */
 634                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 635                              is_zero_ether_addr(f->eth_addr))) {
 636                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 637                                                            ifindex);
 638                         } else
 639                                 return -EOPNOTSUPP;
 640                 }
 641                 if ((flags & NLM_F_APPEND) &&
 642                     (is_multicast_ether_addr(f->eth_addr) ||
 643                      is_zero_ether_addr(f->eth_addr))) {
 644                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 645                                                   &rd);
 646
 647                         if (rc < 0)
 648                                 return rc;
 649                         notify |= rc;
 650                 }
 651         } else {
 652                 if (!(flags & NLM_F_CREATE))
 653                         return -ENOENT;
 654
 655                 if (vxlan->cfg.addrmax &&
 656                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 657                         return -ENOSPC;
 658
 659                 /* Disallow replace to add a multicast entry */
 660                 if ((flags & NLM_F_REPLACE) &&
 661                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 662                         return -EOPNOTSUPP;
 663
 664                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 665                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 666                 if (!f)
 667                         return -ENOMEM;
 668
 669                 notify = 1;
 670                 f->state = state;
 671                 f->flags = ndm_flags;
 672                 f->updated = f->used = jiffies;
 673                 INIT_LIST_HEAD(&f->remotes);
 674                 memcpy(f->eth_addr, mac, ETH_ALEN);
 675
 676                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 677
 678                 ++vxlan->addrcnt;
 679                 hlist_add_head_rcu(&f->hlist,
 680                                    vxlan_fdb_head(vxlan, mac));
 681         }
 682
 683         if (notify) {
 684                 if (rd == NULL)
 685                         rd = first_remote_rtnl(f);
 686                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 687         }
 688
 689         return 0;
 690 }
 691
 692 static void vxlan_fdb_free(struct rcu_head *head)
 693 {
 694         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 695         struct vxlan_rdst *rd, *nd;
 696
 697         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 698                 dst_cache_destroy(&rd->dst_cache);
 699                 kfree(rd);
 700         }
 701         kfree(f);
 702 }
 703
 704 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 705 {
 706         netdev_dbg(vxlan->dev,
 707                     "delete %pM\n", f->eth_addr);
 708
 709         --vxlan->addrcnt;
 710         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 711
 712         hlist_del_rcu(&f->hlist);
 713         call_rcu(&f->rcu, vxlan_fdb_free);
 714 }
 715
 716 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 717                            union vxlan_addr *ip, __be16 *port, __be32 *vni,
 718                            u32 *ifindex)
 719 {
 720         struct net *net = dev_net(vxlan->dev);
 721         int err;
 722
 723         if (tb[NDA_DST]) {
 724                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 725                 if (err)
 726                         return err;
 727         } else {
 728                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 729                 if (remote->sa.sa_family == AF_INET) {
 730                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 731                         ip->sa.sa_family = AF_INET;
 732 #if IS_ENABLED(CONFIG_IPV6)
 733                 } else {
 734                         ip->sin6.sin6_addr = in6addr_any;
 735                         ip->sa.sa_family = AF_INET6;
 736 #endif
 737                 }
 738         }
 739
 740         if (tb[NDA_PORT]) {
 741                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 742                         return -EINVAL;
 743                 *port = nla_get_be16(tb[NDA_PORT]);
 744         } else {
 745                 *port = vxlan->cfg.dst_port;
 746         }
 747
 748         if (tb[NDA_VNI]) {
 749                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 750                         return -EINVAL;
 751                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 752         } else {
 753                 *vni = vxlan->default_dst.remote_vni;
 754         }
 755
 756         if (tb[NDA_IFINDEX]) {
 757                 struct net_device *tdev;
 758
 759                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 760                         return -EINVAL;
 761                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 762                 tdev = __dev_get_by_index(net, *ifindex);
 763                 if (!tdev)
 764                         return -EADDRNOTAVAIL;
 765         } else {
 766                 *ifindex = 0;
 767         }
 768
 769         return 0;
 770 }
 771
 772 /* Add static entry (via netlink) */
 773 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 774                          struct net_device *dev,
 775                          const unsigned char *addr, u16 vid, u16 flags)
 776 {
 777         struct vxlan_dev *vxlan = netdev_priv(dev);
 778         /* struct net *net = dev_net(vxlan->dev); */
 779         union vxlan_addr ip;
 780         __be16 port;
 781         __be32 vni;
 782         u32 ifindex;
 783         int err;
 784
 785         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 786                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 787                         ndm->ndm_state);
 788                 return -EINVAL;
 789         }
 790
 791         if (tb[NDA_DST] == NULL)
 792                 return -EINVAL;
 793
 794         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 795         if (err)
 796                 return err;
 797
 798         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 799                 return -EAFNOSUPPORT;
 800
 801         spin_lock_bh(&vxlan->hash_lock);
 802         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 803                                port, vni, ifindex, ndm->ndm_flags);
 804         spin_unlock_bh(&vxlan->hash_lock);
 805
 806         return err;
 807 }
 808
 809 /* Delete entry (via netlink) */
 810 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 811                             struct net_device *dev,
 812                             const unsigned char *addr, u16 vid)
 813 {
 814         struct vxlan_dev *vxlan = netdev_priv(dev);
 815         struct vxlan_fdb *f;
 816         struct vxlan_rdst *rd = NULL;
 817         union vxlan_addr ip;
 818         __be16 port;
 819         __be32 vni;
 820         u32 ifindex;
 821         int err;
 822
 823         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 824         if (err)
 825                 return err;
 826
 827         err = -ENOENT;
 828
 829         spin_lock_bh(&vxlan->hash_lock);
 830         f = vxlan_find_mac(vxlan, addr);
 831         if (!f)
 832                 goto out;
 833
 834         if (!vxlan_addr_any(&ip)) {
 835                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 836                 if (!rd)
 837                         goto out;
 838         }
 839
 840         err = 0;
 841
 842         /* remove a destination if it's not the only one on the list,
 843          * otherwise destroy the fdb entry
 844          */
 845         if (rd && !list_is_singular(&f->remotes)) {
 846                 list_del_rcu(&rd->list);
 847                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 848                 kfree_rcu(rd, rcu);
 849                 goto out;
 850         }
 851
 852         vxlan_fdb_destroy(vxlan, f);
 853
 854 out:
 855         spin_unlock_bh(&vxlan->hash_lock);
 856
 857         return err;
 858 }
 859
 860 /* Dump forwarding table */
 861 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 862                           struct net_device *dev,
 863                           struct net_device *filter_dev, int *idx)
 864 {
 865         struct vxlan_dev *vxlan = netdev_priv(dev);
 866         unsigned int h;
 867         int err = 0;
 868
 869         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 870                 struct vxlan_fdb *f;
 871
 872                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 873                         struct vxlan_rdst *rd;
 874
 875                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 876                                 if (*idx < cb->args[2])
 877                                         goto skip;
 878
 879                                 err = vxlan_fdb_info(skb, vxlan, f,
 880                                                      NETLINK_CB(cb->skb).portid,
 881                                                      cb->nlh->nlmsg_seq,
 882                                                      RTM_NEWNEIGH,
 883                                                      NLM_F_MULTI, rd);
 884                                 if (err < 0)
 885                                         goto out;
 886 skip:
 887                                 *idx += 1;
 888                         }
 889                 }
 890         }
 891 out:
 892         return err;
 893 }
 894
 895 /* Watch incoming packets to learn mapping between Ethernet address
 896  * and Tunnel endpoint.
 897  * Return true if packet is bogus and should be dropped.
 898  */
 899 static bool vxlan_snoop(struct net_device *dev,
 900                         union vxlan_addr *src_ip, const u8 *src_mac)
 901 {
 902         struct vxlan_dev *vxlan = netdev_priv(dev);
 903         struct vxlan_fdb *f;
 904
 905         f = vxlan_find_mac(vxlan, src_mac);
 906         if (likely(f)) {
 907                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 908
 909                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 910                         return false;
 911
 912                 /* Don't migrate static entries, drop packets */
 913                 if (f->state & NUD_NOARP)
 914                         return true;
 915
 916                 if (net_ratelimit())
 917                         netdev_info(dev,
 918                                     "%pM migrated from %pIS to %pIS\n",
 919                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 920
 921                 rdst->remote_ip = *src_ip;
 922                 f->updated = jiffies;
 923                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 924         } else {
 925                 /* learned new entry */
 926                 spin_lock(&vxlan->hash_lock);
 927
 928                 /* close off race between vxlan_flush and incoming packets */
 929                 if (netif_running(dev))
 930                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 931                                          NUD_REACHABLE,
 932                                          NLM_F_EXCL|NLM_F_CREATE,
 933                                          vxlan->cfg.dst_port,
 934                                          vxlan->default_dst.remote_vni,
 935                                          0, NTF_SELF);
 936                 spin_unlock(&vxlan->hash_lock);
 937         }
 938
 939         return false;
 940 }
 941
 942 /* See if multicast group is already in use by other ID */
 943 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
 944 {
 945         struct vxlan_dev *vxlan;
 946         struct vxlan_sock *sock4;
 947 #if IS_ENABLED(CONFIG_IPV6)
 948         struct vxlan_sock *sock6;
 949 #endif
 950         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
 951
 952         sock4 = rtnl_dereference(dev->vn4_sock);
 953
 954         /* The vxlan_sock is only used by dev, leaving group has
 955          * no effect on other vxlan devices.
 956          */
 957         if (family == AF_INET && sock4 && atomic_read(&sock4->refcnt) == 1)
 958                 return false;
 959 #if IS_ENABLED(CONFIG_IPV6)
 960         sock6 = rtnl_dereference(dev->vn6_sock);
 961         if (family == AF_INET6 && sock6 && atomic_read(&sock6->refcnt) == 1)
 962                 return false;
 963 #endif
 964
 965         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
 966                 if (!netif_running(vxlan->dev) || vxlan == dev)
 967                         continue;
 968
 969                 if (family == AF_INET &&
 970                     rtnl_dereference(vxlan->vn4_sock) != sock4)
 971                         continue;
 972 #if IS_ENABLED(CONFIG_IPV6)
 973                 if (family == AF_INET6 &&
 974                     rtnl_dereference(vxlan->vn6_sock) != sock6)
 975                         continue;
 976 #endif
 977
 978                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
 979                                       &dev->default_dst.remote_ip))
 980                         continue;
 981
 982                 if (vxlan->default_dst.remote_ifindex !=
 983                     dev->default_dst.remote_ifindex)
 984                         continue;
 985
 986                 return true;
 987         }
 988
 989         return false;
 990 }
 991
 992 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
 993 {
 994         struct vxlan_net *vn;
 995
 996         if (!vs)
 997                 return false;
 998         if (!atomic_dec_and_test(&vs->refcnt))
 999                 return false;
1000
1001         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1002         spin_lock(&vn->sock_lock);
1003         hlist_del_rcu(&vs->hlist);
1004         udp_tunnel_notify_del_rx_port(vs->sock,
1005                                       (vs->flags & VXLAN_F_GPE) ?
1006                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1007                                       UDP_TUNNEL_TYPE_VXLAN);
1008         spin_unlock(&vn->sock_lock);
1009
1010         return true;
1011 }
1012
1013 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1014 {
1015         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1016 #if IS_ENABLED(CONFIG_IPV6)
1017         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1018
1019         rcu_assign_pointer(vxlan->vn6_sock, NULL);
1020 #endif
1021
1022         rcu_assign_pointer(vxlan->vn4_sock, NULL);
1023         synchronize_net();
1024
1025         if (__vxlan_sock_release_prep(sock4)) {
1026                 udp_tunnel_sock_release(sock4->sock);
1027                 kfree(sock4);
1028         }
1029
1030 #if IS_ENABLED(CONFIG_IPV6)
1031         if (__vxlan_sock_release_prep(sock6)) {
1032                 udp_tunnel_sock_release(sock6->sock);
1033                 kfree(sock6);
1034         }
1035 #endif
1036 }
1037
1038 /* Update multicast group membership when first VNI on
1039  * multicast address is brought up
1040  */
1041 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1042 {
1043         struct sock *sk;
1044         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1045         int ifindex = vxlan->default_dst.remote_ifindex;
1046         int ret = -EINVAL;
1047
1048         if (ip->sa.sa_family == AF_INET) {
1049                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1050                 struct ip_mreqn mreq = {
1051                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1052                         .imr_ifindex            = ifindex,
1053                 };
1054
1055                 sk = sock4->sock->sk;
1056                 lock_sock(sk);
1057                 ret = ip_mc_join_group(sk, &mreq);
1058                 release_sock(sk);
1059 #if IS_ENABLED(CONFIG_IPV6)
1060         } else {
1061                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1062
1063                 sk = sock6->sock->sk;
1064                 lock_sock(sk);
1065                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1066                                                    &ip->sin6.sin6_addr);
1067                 release_sock(sk);
1068 #endif
1069         }
1070
1071         return ret;
1072 }
1073
1074 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1075 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1076 {
1077         struct sock *sk;
1078         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1079         int ifindex = vxlan->default_dst.remote_ifindex;
1080         int ret = -EINVAL;
1081
1082         if (ip->sa.sa_family == AF_INET) {
1083                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1084                 struct ip_mreqn mreq = {
1085                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1086                         .imr_ifindex            = ifindex,
1087                 };
1088
1089                 sk = sock4->sock->sk;
1090                 lock_sock(sk);
1091                 ret = ip_mc_leave_group(sk, &mreq);
1092                 release_sock(sk);
1093 #if IS_ENABLED(CONFIG_IPV6)
1094         } else {
1095                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1096
1097                 sk = sock6->sock->sk;
1098                 lock_sock(sk);
1099                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1100                                                    &ip->sin6.sin6_addr);
1101                 release_sock(sk);
1102 #endif
1103         }
1104
1105         return ret;
1106 }
1107
1108 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1109                           struct sk_buff *skb, u32 vxflags)
1110 {
1111         size_t start, offset;
1112
1113         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1114                 goto out;
1115
1116         start = vxlan_rco_start(unparsed->vx_vni);
1117         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1118
1119         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1120                 return false;
1121
1122         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1123                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1124 out:
1125         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1126         unparsed->vx_vni &= VXLAN_VNI_MASK;
1127         return true;
1128 }
1129
1130 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1131                                 struct sk_buff *skb, u32 vxflags,
1132                                 struct vxlan_metadata *md)
1133 {
1134         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1135         struct metadata_dst *tun_dst;
1136
1137         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1138                 goto out;
1139
1140         md->gbp = ntohs(gbp->policy_id);
1141
1142         tun_dst = (struct metadata_dst *)skb_dst(skb);
1143         if (tun_dst) {
1144                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1145                 tun_dst->u.tun_info.options_len = sizeof(*md);
1146         }
1147         if (gbp->dont_learn)
1148                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1149
1150         if (gbp->policy_applied)
1151                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1152
1153         /* In flow-based mode, GBP is carried in dst_metadata */
1154         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1155                 skb->mark = md->gbp;
1156 out:
1157         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1158 }
1159
1160 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1161                                 __be16 *protocol,
1162                                 struct sk_buff *skb, u32 vxflags)
1163 {
1164         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1165
1166         /* Need to have Next Protocol set for interfaces in GPE mode. */
1167         if (!gpe->np_applied)
1168                 return false;
1169         /* "The initial version is 0. If a receiver does not support the
1170          * version indicated it MUST drop the packet.
1171          */
1172         if (gpe->version != 0)
1173                 return false;
1174         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1175          * processing MUST occur." However, we don't implement OAM
1176          * processing, thus drop the packet.
1177          */
1178         if (gpe->oam_flag)
1179                 return false;
1180
1181         switch (gpe->next_protocol) {
1182         case VXLAN_GPE_NP_IPV4:
1183                 *protocol = htons(ETH_P_IP);
1184                 break;
1185         case VXLAN_GPE_NP_IPV6:
1186                 *protocol = htons(ETH_P_IPV6);
1187                 break;
1188         case VXLAN_GPE_NP_ETHERNET:
1189                 *protocol = htons(ETH_P_TEB);
1190                 break;
1191         default:
1192                 return false;
1193         }
1194
1195         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1196         return true;
1197 }
1198
1199 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1200                           struct vxlan_sock *vs,
1201                           struct sk_buff *skb)
1202 {
1203         union vxlan_addr saddr;
1204
1205         skb_reset_mac_header(skb);
1206         skb->protocol = eth_type_trans(skb, vxlan->dev);
1207         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1208
1209         /* Ignore packet loops (and multicast echo) */
1210         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1211                 return false;
1212
1213         /* Get address from the outer IP header */
1214         if (vxlan_get_sk_family(vs) == AF_INET) {
1215                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1216                 saddr.sa.sa_family = AF_INET;
1217 #if IS_ENABLED(CONFIG_IPV6)
1218         } else {
1219                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1220                 saddr.sa.sa_family = AF_INET6;
1221 #endif
1222         }
1223
1224         if ((vxlan->flags & VXLAN_F_LEARN) &&
1225             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1226                 return false;
1227
1228         return true;
1229 }
1230
1231 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1232                                   struct sk_buff *skb)
1233 {
1234         int err = 0;
1235
1236         if (vxlan_get_sk_family(vs) == AF_INET)
1237                 err = IP_ECN_decapsulate(oiph, skb);
1238 #if IS_ENABLED(CONFIG_IPV6)
1239         else
1240                 err = IP6_ECN_decapsulate(oiph, skb);
1241 #endif
1242
1243         if (unlikely(err) && log_ecn_error) {
1244                 if (vxlan_get_sk_family(vs) == AF_INET)
1245                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1246                                              &((struct iphdr *)oiph)->saddr,
1247                                              ((struct iphdr *)oiph)->tos);
1248                 else
1249                         net_info_ratelimited("non-ECT from %pI6\n",
1250                                              &((struct ipv6hdr *)oiph)->saddr);
1251         }
1252         return err <= 1;
1253 }
1254
1255 /* Callback from net/ipv4/udp.c to receive packets */
1256 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1257 {
1258         struct pcpu_sw_netstats *stats;
1259         struct vxlan_dev *vxlan;
1260         struct vxlan_sock *vs;
1261         struct vxlanhdr unparsed;
1262         struct vxlan_metadata _md;
1263         struct vxlan_metadata *md = &_md;
1264         __be16 protocol = htons(ETH_P_TEB);
1265         bool raw_proto = false;
1266         void *oiph;
1267
1268         /* Need UDP and VXLAN header to be present */
1269         if (!pskb_may_pull(skb, VXLAN_HLEN))
1270                 goto drop;
1271
1272         unparsed = *vxlan_hdr(skb);
1273         /* VNI flag always required to be set */
1274         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1275                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1276                            ntohl(vxlan_hdr(skb)->vx_flags),
1277                            ntohl(vxlan_hdr(skb)->vx_vni));
1278                 /* Return non vxlan pkt */
1279                 goto drop;
1280         }
1281         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1282         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1283
1284         vs = rcu_dereference_sk_user_data(sk);
1285         if (!vs)
1286                 goto drop;
1287
1288         vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1289         if (!vxlan)
1290                 goto drop;
1291
1292         /* For backwards compatibility, only allow reserved fields to be
1293          * used by VXLAN extensions if explicitly requested.
1294          */
1295         if (vs->flags & VXLAN_F_GPE) {
1296                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1297                         goto drop;
1298                 raw_proto = true;
1299         }
1300
1301         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1302                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1303                         goto drop;
1304
1305         if (vxlan_collect_metadata(vs)) {
1306                 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1307                 struct metadata_dst *tun_dst;
1308
1309                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1310                                          key32_to_tunnel_id(vni), sizeof(*md));
1311
1312                 if (!tun_dst)
1313                         goto drop;
1314
1315                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1316
1317                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1318         } else {
1319                 memset(md, 0, sizeof(*md));
1320         }
1321
1322         if (vs->flags & VXLAN_F_REMCSUM_RX)
1323                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1324                         goto drop;
1325         if (vs->flags & VXLAN_F_GBP)
1326                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1327         /* Note that GBP and GPE can never be active together. This is
1328          * ensured in vxlan_dev_configure.
1329          */
1330
1331         if (unparsed.vx_flags || unparsed.vx_vni) {
1332                 /* If there are any unprocessed flags remaining treat
1333                  * this as a malformed packet. This behavior diverges from
1334                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1335                  * in reserved fields are to be ignored. The approach here
1336                  * maintains compatibility with previous stack code, and also
1337                  * is more robust and provides a little more security in
1338                  * adding extensions to VXLAN.
1339                  */
1340                 goto drop;
1341         }
1342
1343         if (!raw_proto) {
1344                 if (!vxlan_set_mac(vxlan, vs, skb))
1345                         goto drop;
1346         } else {
1347                 skb_reset_mac_header(skb);
1348                 skb->dev = vxlan->dev;
1349                 skb->pkt_type = PACKET_HOST;
1350         }
1351
1352         oiph = skb_network_header(skb);
1353         skb_reset_network_header(skb);
1354
1355         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1356                 ++vxlan->dev->stats.rx_frame_errors;
1357                 ++vxlan->dev->stats.rx_errors;
1358                 goto drop;
1359         }
1360
1361         stats = this_cpu_ptr(vxlan->dev->tstats);
1362         u64_stats_update_begin(&stats->syncp);
1363         stats->rx_packets++;
1364         stats->rx_bytes += skb->len;
1365         u64_stats_update_end(&stats->syncp);
1366
1367         gro_cells_receive(&vxlan->gro_cells, skb);
1368         return 0;
1369
1370 drop:
1371         /* Consume bad packet */
1372         kfree_skb(skb);
1373         return 0;
1374 }
1375
1376 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1377 {
1378         struct vxlan_dev *vxlan = netdev_priv(dev);
1379         struct arphdr *parp;
1380         u8 *arpptr, *sha;
1381         __be32 sip, tip;
1382         struct neighbour *n;
1383
1384         if (dev->flags & IFF_NOARP)
1385                 goto out;
1386
1387         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1388                 dev->stats.tx_dropped++;
1389                 goto out;
1390         }
1391         parp = arp_hdr(skb);
1392
1393         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1394              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1395             parp->ar_pro != htons(ETH_P_IP) ||
1396             parp->ar_op != htons(ARPOP_REQUEST) ||
1397             parp->ar_hln != dev->addr_len ||
1398             parp->ar_pln != 4)
1399                 goto out;
1400         arpptr = (u8 *)parp + sizeof(struct arphdr);
1401         sha = arpptr;
1402         arpptr += dev->addr_len;        /* sha */
1403         memcpy(&sip, arpptr, sizeof(sip));
1404         arpptr += sizeof(sip);
1405         arpptr += dev->addr_len;        /* tha */
1406         memcpy(&tip, arpptr, sizeof(tip));
1407
1408         if (ipv4_is_loopback(tip) ||
1409             ipv4_is_multicast(tip))
1410                 goto out;
1411
1412         n = neigh_lookup(&arp_tbl, &tip, dev);
1413
1414         if (n) {
1415                 struct vxlan_fdb *f;
1416                 struct sk_buff  *reply;
1417
1418                 if (!(n->nud_state & NUD_CONNECTED)) {
1419                         neigh_release(n);
1420                         goto out;
1421                 }
1422
1423                 f = vxlan_find_mac(vxlan, n->ha);
1424                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1425                         /* bridge-local neighbor */
1426                         neigh_release(n);
1427                         goto out;
1428                 }
1429
1430                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1431                                 n->ha, sha);
1432
1433                 neigh_release(n);
1434
1435                 if (reply == NULL)
1436                         goto out;
1437
1438                 skb_reset_mac_header(reply);
1439                 __skb_pull(reply, skb_network_offset(reply));
1440                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1441                 reply->pkt_type = PACKET_HOST;
1442
1443                 if (netif_rx_ni(reply) == NET_RX_DROP)
1444                         dev->stats.rx_dropped++;
1445         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1446                 union vxlan_addr ipa = {
1447                         .sin.sin_addr.s_addr = tip,
1448                         .sin.sin_family = AF_INET,
1449                 };
1450
1451                 vxlan_ip_miss(dev, &ipa);
1452         }
1453 out:
1454         consume_skb(skb);
1455         return NETDEV_TX_OK;
1456 }
1457
1458 #if IS_ENABLED(CONFIG_IPV6)
1459 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1460         struct neighbour *n, bool isrouter)
1461 {
1462         struct net_device *dev = request->dev;
1463         struct sk_buff *reply;
1464         struct nd_msg *ns, *na;
1465         struct ipv6hdr *pip6;
1466         u8 *daddr;
1467         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1468         int ns_olen;
1469         int i, len;
1470
1471         if (dev == NULL)
1472                 return NULL;
1473
1474         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1475                 sizeof(*na) + na_olen + dev->needed_tailroom;
1476         reply = alloc_skb(len, GFP_ATOMIC);
1477         if (reply == NULL)
1478                 return NULL;
1479
1480         reply->protocol = htons(ETH_P_IPV6);
1481         reply->dev = dev;
1482         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1483         skb_push(reply, sizeof(struct ethhdr));
1484         skb_reset_mac_header(reply);
1485
1486         ns = (struct nd_msg *)skb_transport_header(request);
1487
1488         daddr = eth_hdr(request)->h_source;
1489         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1490         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1491                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1492                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1493                         break;
1494                 }
1495         }
1496
1497         /* Ethernet header */
1498         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1499         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1500         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1501         reply->protocol = htons(ETH_P_IPV6);
1502
1503         skb_pull(reply, sizeof(struct ethhdr));
1504         skb_reset_network_header(reply);
1505         skb_put(reply, sizeof(struct ipv6hdr));
1506
1507         /* IPv6 header */
1508
1509         pip6 = ipv6_hdr(reply);
1510         memset(pip6, 0, sizeof(struct ipv6hdr));
1511         pip6->version = 6;
1512         pip6->priority = ipv6_hdr(request)->priority;
1513         pip6->nexthdr = IPPROTO_ICMPV6;
1514         pip6->hop_limit = 255;
1515         pip6->daddr = ipv6_hdr(request)->saddr;
1516         pip6->saddr = *(struct in6_addr *)n->primary_key;
1517
1518         skb_pull(reply, sizeof(struct ipv6hdr));
1519         skb_reset_transport_header(reply);
1520
1521         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1522
1523         /* Neighbor Advertisement */
1524         memset(na, 0, sizeof(*na)+na_olen);
1525         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1526         na->icmph.icmp6_router = isrouter;
1527         na->icmph.icmp6_override = 1;
1528         na->icmph.icmp6_solicited = 1;
1529         na->target = ns->target;
1530         ether_addr_copy(&na->opt[2], n->ha);
1531         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1532         na->opt[1] = na_olen >> 3;
1533
1534         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1535                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1536                 csum_partial(na, sizeof(*na)+na_olen, 0));
1537
1538         pip6->payload_len = htons(sizeof(*na)+na_olen);
1539
1540         skb_push(reply, sizeof(struct ipv6hdr));
1541
1542         reply->ip_summed = CHECKSUM_UNNECESSARY;
1543
1544         return reply;
1545 }
1546
1547 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1548 {
1549         struct vxlan_dev *vxlan = netdev_priv(dev);
1550         struct nd_msg *msg;
1551         const struct ipv6hdr *iphdr;
1552         const struct in6_addr *saddr, *daddr;
1553         struct neighbour *n;
1554         struct inet6_dev *in6_dev;
1555
1556         in6_dev = __in6_dev_get(dev);
1557         if (!in6_dev)
1558                 goto out;
1559
1560         iphdr = ipv6_hdr(skb);
1561         saddr = &iphdr->saddr;
1562         daddr = &iphdr->daddr;
1563
1564         msg = (struct nd_msg *)skb_transport_header(skb);
1565         if (msg->icmph.icmp6_code != 0 ||
1566             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1567                 goto out;
1568
1569         if (ipv6_addr_loopback(daddr) ||
1570             ipv6_addr_is_multicast(&msg->target))
1571                 goto out;
1572
1573         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1574
1575         if (n) {
1576                 struct vxlan_fdb *f;
1577                 struct sk_buff *reply;
1578
1579                 if (!(n->nud_state & NUD_CONNECTED)) {
1580                         neigh_release(n);
1581                         goto out;
1582                 }
1583
1584                 f = vxlan_find_mac(vxlan, n->ha);
1585                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1586                         /* bridge-local neighbor */
1587                         neigh_release(n);
1588                         goto out;
1589                 }
1590
1591                 reply = vxlan_na_create(skb, n,
1592                                         !!(f ? f->flags & NTF_ROUTER : 0));
1593
1594                 neigh_release(n);
1595
1596                 if (reply == NULL)
1597                         goto out;
1598
1599                 if (netif_rx_ni(reply) == NET_RX_DROP)
1600                         dev->stats.rx_dropped++;
1601
1602         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1603                 union vxlan_addr ipa = {
1604                         .sin6.sin6_addr = msg->target,
1605                         .sin6.sin6_family = AF_INET6,
1606                 };
1607
1608                 vxlan_ip_miss(dev, &ipa);
1609         }
1610
1611 out:
1612         consume_skb(skb);
1613         return NETDEV_TX_OK;
1614 }
1615 #endif
1616
1617 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1618 {
1619         struct vxlan_dev *vxlan = netdev_priv(dev);
1620         struct neighbour *n;
1621
1622         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1623                 return false;
1624
1625         n = NULL;
1626         switch (ntohs(eth_hdr(skb)->h_proto)) {
1627         case ETH_P_IP:
1628         {
1629                 struct iphdr *pip;
1630
1631                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1632                         return false;
1633                 pip = ip_hdr(skb);
1634                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1635                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1636                         union vxlan_addr ipa = {
1637                                 .sin.sin_addr.s_addr = pip->daddr,
1638                                 .sin.sin_family = AF_INET,
1639                         };
1640
1641                         vxlan_ip_miss(dev, &ipa);
1642                         return false;
1643                 }
1644
1645                 break;
1646         }
1647 #if IS_ENABLED(CONFIG_IPV6)
1648         case ETH_P_IPV6:
1649         {
1650                 struct ipv6hdr *pip6;
1651
1652                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1653                         return false;
1654                 pip6 = ipv6_hdr(skb);
1655                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1656                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1657                         union vxlan_addr ipa = {
1658                                 .sin6.sin6_addr = pip6->daddr,
1659                                 .sin6.sin6_family = AF_INET6,
1660                         };
1661
1662                         vxlan_ip_miss(dev, &ipa);
1663                         return false;
1664                 }
1665
1666                 break;
1667         }
1668 #endif
1669         default:
1670                 return false;
1671         }
1672
1673         if (n) {
1674                 bool diff;
1675
1676                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1677                 if (diff) {
1678                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1679                                 dev->addr_len);
1680                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1681                 }
1682                 neigh_release(n);
1683                 return diff;
1684         }
1685
1686         return false;
1687 }
1688
1689 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1690                                 struct vxlan_metadata *md)
1691 {
1692         struct vxlanhdr_gbp *gbp;
1693
1694         if (!md->gbp)
1695                 return;
1696
1697         gbp = (struct vxlanhdr_gbp *)vxh;
1698         vxh->vx_flags |= VXLAN_HF_GBP;
1699
1700         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1701                 gbp->dont_learn = 1;
1702
1703         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1704                 gbp->policy_applied = 1;
1705
1706         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1707 }
1708
1709 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1710                                __be16 protocol)
1711 {
1712         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1713
1714         gpe->np_applied = 1;
1715
1716         switch (protocol) {
1717         case htons(ETH_P_IP):
1718                 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1719                 return 0;
1720         case htons(ETH_P_IPV6):
1721                 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1722                 return 0;
1723         case htons(ETH_P_TEB):
1724                 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1725                 return 0;
1726         }
1727         return -EPFNOSUPPORT;
1728 }
1729
1730 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1731                            int iphdr_len, __be32 vni,
1732                            struct vxlan_metadata *md, u32 vxflags,
1733                            bool udp_sum)
1734 {
1735         struct vxlanhdr *vxh;
1736         int min_headroom;
1737         int err;
1738         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1739         __be16 inner_protocol = htons(ETH_P_TEB);
1740
1741         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1742             skb->ip_summed == CHECKSUM_PARTIAL) {
1743                 int csum_start = skb_checksum_start_offset(skb);
1744
1745                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1746                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1747                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1748                      skb->csum_offset == offsetof(struct tcphdr, check)))
1749                         type |= SKB_GSO_TUNNEL_REMCSUM;
1750         }
1751
1752         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1753                         + VXLAN_HLEN + iphdr_len
1754                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1755
1756         /* Need space for new headers (invalidates iph ptr) */
1757         err = skb_cow_head(skb, min_headroom);
1758         if (unlikely(err))
1759                 goto out_free;
1760
1761         skb = vlan_hwaccel_push_inside(skb);
1762         if (WARN_ON(!skb))
1763                 return -ENOMEM;
1764
1765         err = iptunnel_handle_offloads(skb, type);
1766         if (err)
1767                 goto out_free;
1768
1769         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1770         vxh->vx_flags = VXLAN_HF_VNI;
1771         vxh->vx_vni = vxlan_vni_field(vni);
1772
1773         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1774                 unsigned int start;
1775
1776                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1777                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1778                 vxh->vx_flags |= VXLAN_HF_RCO;
1779
1780                 if (!skb_is_gso(skb)) {
1781                         skb->ip_summed = CHECKSUM_NONE;
1782                         skb->encapsulation = 0;
1783                 }
1784         }
1785
1786         if (vxflags & VXLAN_F_GBP)
1787                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1788         if (vxflags & VXLAN_F_GPE) {
1789                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1790                 if (err < 0)
1791                         goto out_free;
1792                 inner_protocol = skb->protocol;
1793         }
1794
1795         skb_set_inner_protocol(skb, inner_protocol);
1796         return 0;
1797
1798 out_free:
1799         kfree_skb(skb);
1800         return err;
1801 }
1802
1803 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1804                                       struct sk_buff *skb, int oif, u8 tos,
1805                                       __be32 daddr, __be32 *saddr,
1806                                       struct dst_cache *dst_cache,
1807                                       const struct ip_tunnel_info *info)
1808 {
1809         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1810         struct rtable *rt = NULL;
1811         struct flowi4 fl4;
1812
1813         if (tos && !info)
1814                 use_cache = false;
1815         if (use_cache) {
1816                 rt = dst_cache_get_ip4(dst_cache, saddr);
1817                 if (rt)
1818                         return rt;
1819         }
1820
1821         memset(&fl4, 0, sizeof(fl4));
1822         fl4.flowi4_oif = oif;
1823         fl4.flowi4_tos = RT_TOS(tos);
1824         fl4.flowi4_mark = skb->mark;
1825         fl4.flowi4_proto = IPPROTO_UDP;
1826         fl4.daddr = daddr;
1827         fl4.saddr = *saddr;
1828
1829         rt = ip_route_output_key(vxlan->net, &fl4);
1830         if (!IS_ERR(rt)) {
1831                 *saddr = fl4.saddr;
1832                 if (use_cache)
1833                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1834         }
1835         return rt;
1836 }
1837
1838 #if IS_ENABLED(CONFIG_IPV6)
1839 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1840                                           struct sk_buff *skb, int oif, u8 tos,
1841                                           __be32 label,
1842                                           const struct in6_addr *daddr,
1843                                           struct in6_addr *saddr,
1844                                           struct dst_cache *dst_cache,
1845                                           const struct ip_tunnel_info *info)
1846 {
1847         struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
1848         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1849         struct dst_entry *ndst;
1850         struct flowi6 fl6;
1851         int err;
1852
1853         if (!sock6)
1854                 return ERR_PTR(-EIO);
1855
1856         if (tos && !info)
1857                 use_cache = false;
1858         if (use_cache) {
1859                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1860                 if (ndst)
1861                         return ndst;
1862         }
1863
1864         memset(&fl6, 0, sizeof(fl6));
1865         fl6.flowi6_oif = oif;
1866         fl6.daddr = *daddr;
1867         fl6.saddr = *saddr;
1868         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1869         fl6.flowi6_mark = skb->mark;
1870         fl6.flowi6_proto = IPPROTO_UDP;
1871
1872         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1873                                          sock6->sock->sk,
1874                                          &ndst, &fl6);
1875         if (err < 0)
1876                 return ERR_PTR(err);
1877
1878         *saddr = fl6.saddr;
1879         if (use_cache)
1880                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1881         return ndst;
1882 }
1883 #endif
1884
1885 /* Bypass encapsulation if the destination is local */
1886 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1887                                struct vxlan_dev *dst_vxlan)
1888 {
1889         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1890         union vxlan_addr loopback;
1891         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1892         struct net_device *dev = skb->dev;
1893         int len = skb->len;
1894
1895         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1896         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1897         skb->pkt_type = PACKET_HOST;
1898         skb->encapsulation = 0;
1899         skb->dev = dst_vxlan->dev;
1900         __skb_pull(skb, skb_network_offset(skb));
1901
1902         if (remote_ip->sa.sa_family == AF_INET) {
1903                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1904                 loopback.sa.sa_family =  AF_INET;
1905 #if IS_ENABLED(CONFIG_IPV6)
1906         } else {
1907                 loopback.sin6.sin6_addr = in6addr_loopback;
1908                 loopback.sa.sa_family =  AF_INET6;
1909 #endif
1910         }
1911
1912         if (dst_vxlan->flags & VXLAN_F_LEARN)
1913                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1914
1915         u64_stats_update_begin(&tx_stats->syncp);
1916         tx_stats->tx_packets++;
1917         tx_stats->tx_bytes += len;
1918         u64_stats_update_end(&tx_stats->syncp);
1919
1920         if (netif_rx(skb) == NET_RX_SUCCESS) {
1921                 u64_stats_update_begin(&rx_stats->syncp);
1922                 rx_stats->rx_packets++;
1923                 rx_stats->rx_bytes += len;
1924                 u64_stats_update_end(&rx_stats->syncp);
1925         } else {
1926                 dev->stats.rx_dropped++;
1927         }
1928 }
1929
1930 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1931                            struct vxlan_rdst *rdst, bool did_rsc)
1932 {
1933         struct dst_cache *dst_cache;
1934         struct ip_tunnel_info *info;
1935         struct vxlan_dev *vxlan = netdev_priv(dev);
1936         struct sock *sk;
1937         struct rtable *rt = NULL;
1938         const struct iphdr *old_iph;
1939         union vxlan_addr *dst;
1940         union vxlan_addr remote_ip, local_ip;
1941         union vxlan_addr *src;
1942         struct vxlan_metadata _md;
1943         struct vxlan_metadata *md = &_md;
1944         __be16 src_port = 0, dst_port;
1945         __be32 vni, label;
1946         __be16 df = 0;
1947         __u8 tos, ttl;
1948         int err;
1949         u32 flags = vxlan->flags;
1950         bool udp_sum = false;
1951         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1952
1953         info = skb_tunnel_info(skb);
1954
1955         if (rdst) {
1956                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1957                 vni = rdst->remote_vni;
1958                 dst = &rdst->remote_ip;
1959                 src = &vxlan->cfg.saddr;
1960                 dst_cache = &rdst->dst_cache;
1961         } else {
1962                 if (!info) {
1963                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1964                                   dev->name);
1965                         goto drop;
1966                 }
1967                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1968                 vni = tunnel_id_to_key32(info->key.tun_id);
1969                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1970                 if (remote_ip.sa.sa_family == AF_INET) {
1971                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1972                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
1973                 } else {
1974                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1975                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
1976                 }
1977                 dst = &remote_ip;
1978                 src = &local_ip;
1979                 dst_cache = &info->dst_cache;
1980         }
1981
1982         if (vxlan_addr_any(dst)) {
1983                 if (did_rsc) {
1984                         /* short-circuited back to local bridge */
1985                         vxlan_encap_bypass(skb, vxlan, vxlan);
1986                         return;
1987                 }
1988                 goto drop;
1989         }
1990
1991         old_iph = ip_hdr(skb);
1992
1993         ttl = vxlan->cfg.ttl;
1994         if (!ttl && vxlan_addr_multicast(dst))
1995                 ttl = 1;
1996
1997         tos = vxlan->cfg.tos;
1998         if (tos == 1)
1999                 tos = ip_tunnel_get_dsfield(old_iph, skb);
2000
2001         label = vxlan->cfg.label;
2002         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2003                                      vxlan->cfg.port_max, true);
2004
2005         if (info) {
2006                 ttl = info->key.ttl;
2007                 tos = info->key.tos;
2008                 label = info->key.label;
2009                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2010
2011                 if (info->options_len)
2012                         md = ip_tunnel_info_opts(info);
2013         } else {
2014                 md->gbp = skb->mark;
2015         }
2016
2017         if (dst->sa.sa_family == AF_INET) {
2018                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2019
2020                 if (!sock4)
2021                         goto drop;
2022                 sk = sock4->sock->sk;
2023
2024                 rt = vxlan_get_route(vxlan, skb,
2025                                      rdst ? rdst->remote_ifindex : 0, tos,
2026                                      dst->sin.sin_addr.s_addr,
2027                                      &src->sin.sin_addr.s_addr,
2028                                      dst_cache, info);
2029                 if (IS_ERR(rt)) {
2030                         netdev_dbg(dev, "no route to %pI4\n",
2031                                    &dst->sin.sin_addr.s_addr);
2032                         dev->stats.tx_carrier_errors++;
2033                         goto tx_error;
2034                 }
2035
2036                 if (rt->dst.dev == dev) {
2037                         netdev_dbg(dev, "circular route to %pI4\n",
2038                                    &dst->sin.sin_addr.s_addr);
2039                         dev->stats.collisions++;
2040                         goto rt_tx_error;
2041                 }
2042
2043                 /* Bypass encapsulation if the destination is local */
2044                 if (!info && rt->rt_flags & RTCF_LOCAL &&
2045                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2046                         struct vxlan_dev *dst_vxlan;
2047
2048                         ip_rt_put(rt);
2049                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2050                                                    dst->sa.sa_family, dst_port,
2051                                                    vxlan->flags);
2052                         if (!dst_vxlan)
2053                                 goto tx_error;
2054                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2055                         return;
2056                 }
2057
2058                 if (!info)
2059                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2060                 else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2061                         df = htons(IP_DF);
2062
2063                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2064                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2065                 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2066                                       vni, md, flags, udp_sum);
2067                 if (err < 0)
2068                         goto xmit_tx_error;
2069
2070                 udp_tunnel_xmit_skb(rt, sk, skb, src->sin.sin_addr.s_addr,
2071                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2072                                     src_port, dst_port, xnet, !udp_sum);
2073 #if IS_ENABLED(CONFIG_IPV6)
2074         } else {
2075                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2076                 struct dst_entry *ndst;
2077                 u32 rt6i_flags;
2078
2079                 if (!sock6)
2080                         goto drop;
2081                 sk = sock6->sock->sk;
2082
2083                 ndst = vxlan6_get_route(vxlan, skb,
2084                                         rdst ? rdst->remote_ifindex : 0, tos,
2085                                         label, &dst->sin6.sin6_addr,
2086                                         &src->sin6.sin6_addr,
2087                                         dst_cache, info);
2088                 if (IS_ERR(ndst)) {
2089                         netdev_dbg(dev, "no route to %pI6\n",
2090                                    &dst->sin6.sin6_addr);
2091                         dev->stats.tx_carrier_errors++;
2092                         goto tx_error;
2093                 }
2094
2095                 if (ndst->dev == dev) {
2096                         netdev_dbg(dev, "circular route to %pI6\n",
2097                                    &dst->sin6.sin6_addr);
2098                         dst_release(ndst);
2099                         dev->stats.collisions++;
2100                         goto tx_error;
2101                 }
2102
2103                 /* Bypass encapsulation if the destination is local */
2104                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2105                 if (!info && rt6i_flags & RTF_LOCAL &&
2106                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2107                         struct vxlan_dev *dst_vxlan;
2108
2109                         dst_release(ndst);
2110                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2111                                                    dst->sa.sa_family, dst_port,
2112                                                    vxlan->flags);
2113                         if (!dst_vxlan)
2114                                 goto tx_error;
2115                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2116                         return;
2117                 }
2118
2119                 if (!info)
2120                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2121
2122                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2123                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2124                 skb_scrub_packet(skb, xnet);
2125                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2126                                       vni, md, flags, udp_sum);
2127                 if (err < 0) {
2128                         dst_release(ndst);
2129                         dev->stats.tx_errors++;
2130                         return;
2131                 }
2132                 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2133                                      &src->sin6.sin6_addr,
2134                                      &dst->sin6.sin6_addr, tos, ttl,
2135                                      label, src_port, dst_port, !udp_sum);
2136 #endif
2137         }
2138
2139         return;
2140
2141 drop:
2142         dev->stats.tx_dropped++;
2143         goto tx_free;
2144
2145 xmit_tx_error:
2146         /* skb is already freed. */
2147         skb = NULL;
2148 rt_tx_error:
2149         ip_rt_put(rt);
2150 tx_error:
2151         dev->stats.tx_errors++;
2152 tx_free:
2153         dev_kfree_skb(skb);
2154 }
2155
2156 /* Transmit local packets over Vxlan
2157  *
2158  * Outer IP header inherits ECN and DF from inner header.
2159  * Outer UDP destination is the VXLAN assigned port.
2160  *           source port is based on hash of flow
2161  */
2162 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2163 {
2164         struct vxlan_dev *vxlan = netdev_priv(dev);
2165         const struct ip_tunnel_info *info;
2166         struct ethhdr *eth;
2167         bool did_rsc = false;
2168         struct vxlan_rdst *rdst, *fdst = NULL;
2169         struct vxlan_fdb *f;
2170
2171         info = skb_tunnel_info(skb);
2172
2173         skb_reset_mac_header(skb);
2174
2175         if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2176                 if (info && info->mode & IP_TUNNEL_INFO_TX)
2177                         vxlan_xmit_one(skb, dev, NULL, false);
2178                 else
2179                         kfree_skb(skb);
2180                 return NETDEV_TX_OK;
2181         }
2182
2183         if (vxlan->flags & VXLAN_F_PROXY) {
2184                 eth = eth_hdr(skb);
2185                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2186                         return arp_reduce(dev, skb);
2187 #if IS_ENABLED(CONFIG_IPV6)
2188                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2189                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2190                                        + sizeof(struct nd_msg)) &&
2191                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2192                                 struct nd_msg *msg;
2193
2194                                 msg = (struct nd_msg *)skb_transport_header(skb);
2195                                 if (msg->icmph.icmp6_code == 0 &&
2196                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2197                                         return neigh_reduce(dev, skb);
2198                 }
2199 #endif
2200         }
2201
2202         eth = eth_hdr(skb);
2203         f = vxlan_find_mac(vxlan, eth->h_dest);
2204         did_rsc = false;
2205
2206         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2207             (ntohs(eth->h_proto) == ETH_P_IP ||
2208              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2209                 did_rsc = route_shortcircuit(dev, skb);
2210                 if (did_rsc)
2211                         f = vxlan_find_mac(vxlan, eth->h_dest);
2212         }
2213
2214         if (f == NULL) {
2215                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2216                 if (f == NULL) {
2217                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2218                             !is_multicast_ether_addr(eth->h_dest))
2219                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2220
2221                         dev->stats.tx_dropped++;
2222                         kfree_skb(skb);
2223                         return NETDEV_TX_OK;
2224                 }
2225         }
2226
2227         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2228                 struct sk_buff *skb1;
2229
2230                 if (!fdst) {
2231                         fdst = rdst;
2232                         continue;
2233                 }
2234                 skb1 = skb_clone(skb, GFP_ATOMIC);
2235                 if (skb1)
2236                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2237         }
2238
2239         if (fdst)
2240                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2241         else
2242                 kfree_skb(skb);
2243         return NETDEV_TX_OK;
2244 }
2245
2246 /* Walk the forwarding table and purge stale entries */
2247 static void vxlan_cleanup(unsigned long arg)
2248 {
2249         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2250         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2251         unsigned int h;
2252
2253         if (!netif_running(vxlan->dev))
2254                 return;
2255
2256         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2257                 struct hlist_node *p, *n;
2258
2259                 spin_lock_bh(&vxlan->hash_lock);
2260                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2261                         struct vxlan_fdb *f
2262                                 = container_of(p, struct vxlan_fdb, hlist);
2263                         unsigned long timeout;
2264
2265                         if (f->state & NUD_PERMANENT)
2266                                 continue;
2267
2268                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2269                         if (time_before_eq(timeout, jiffies)) {
2270                                 netdev_dbg(vxlan->dev,
2271                                            "garbage collect %pM\n",
2272                                            f->eth_addr);
2273                                 f->state = NUD_STALE;
2274                                 vxlan_fdb_destroy(vxlan, f);
2275                         } else if (time_before(timeout, next_timer))
2276                                 next_timer = timeout;
2277                 }
2278                 spin_unlock_bh(&vxlan->hash_lock);
2279         }
2280
2281         mod_timer(&vxlan->age_timer, next_timer);
2282 }
2283
2284 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2285 {
2286         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2287         __be32 vni = vxlan->default_dst.remote_vni;
2288
2289         spin_lock(&vn->sock_lock);
2290         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2291         spin_unlock(&vn->sock_lock);
2292 }
2293
2294 /* Setup stats when device is created */
2295 static int vxlan_init(struct net_device *dev)
2296 {
2297         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2298         if (!dev->tstats)
2299                 return -ENOMEM;
2300
2301         return 0;
2302 }
2303
2304 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2305 {
2306         struct vxlan_fdb *f;
2307
2308         spin_lock_bh(&vxlan->hash_lock);
2309         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2310         if (f)
2311                 vxlan_fdb_destroy(vxlan, f);
2312         spin_unlock_bh(&vxlan->hash_lock);
2313 }
2314
2315 static void vxlan_uninit(struct net_device *dev)
2316 {
2317         struct vxlan_dev *vxlan = netdev_priv(dev);
2318
2319         vxlan_fdb_delete_default(vxlan);
2320
2321         free_percpu(dev->tstats);
2322 }
2323
2324 /* Start ageing timer and join group when device is brought up */
2325 static int vxlan_open(struct net_device *dev)
2326 {
2327         struct vxlan_dev *vxlan = netdev_priv(dev);
2328         int ret;
2329
2330         ret = vxlan_sock_add(vxlan);
2331         if (ret < 0)
2332                 return ret;
2333
2334         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2335                 ret = vxlan_igmp_join(vxlan);
2336                 if (ret == -EADDRINUSE)
2337                         ret = 0;
2338                 if (ret) {
2339                         vxlan_sock_release(vxlan);
2340                         return ret;
2341                 }
2342         }
2343
2344         if (vxlan->cfg.age_interval)
2345                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2346
2347         return ret;
2348 }
2349
2350 /* Purge the forwarding table */
2351 static void vxlan_flush(struct vxlan_dev *vxlan)
2352 {
2353         unsigned int h;
2354
2355         spin_lock_bh(&vxlan->hash_lock);
2356         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2357                 struct hlist_node *p, *n;
2358                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2359                         struct vxlan_fdb *f
2360                                 = container_of(p, struct vxlan_fdb, hlist);
2361                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2362                         if (!is_zero_ether_addr(f->eth_addr))
2363                                 vxlan_fdb_destroy(vxlan, f);
2364                 }
2365         }
2366         spin_unlock_bh(&vxlan->hash_lock);
2367 }
2368
2369 /* Cleanup timer and forwarding table on shutdown */
2370 static int vxlan_stop(struct net_device *dev)
2371 {
2372         struct vxlan_dev *vxlan = netdev_priv(dev);
2373         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2374         int ret = 0;
2375
2376         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2377             !vxlan_group_used(vn, vxlan))
2378                 ret = vxlan_igmp_leave(vxlan);
2379
2380         del_timer_sync(&vxlan->age_timer);
2381
2382         vxlan_flush(vxlan);
2383         vxlan_sock_release(vxlan);
2384
2385         return ret;
2386 }
2387
2388 /* Stub, nothing needs to be done. */
2389 static void vxlan_set_multicast_list(struct net_device *dev)
2390 {
2391 }
2392
2393 static int __vxlan_change_mtu(struct net_device *dev,
2394                               struct net_device *lowerdev,
2395                               struct vxlan_rdst *dst, int new_mtu, bool strict)
2396 {
2397         int max_mtu = IP_MAX_MTU;
2398
2399         if (lowerdev)
2400                 max_mtu = lowerdev->mtu;
2401
2402         if (dst->remote_ip.sa.sa_family == AF_INET6)
2403                 max_mtu -= VXLAN6_HEADROOM;
2404         else
2405                 max_mtu -= VXLAN_HEADROOM;
2406
2407         if (new_mtu < 68)
2408                 return -EINVAL;
2409
2410         if (new_mtu > max_mtu) {
2411                 if (strict)
2412                         return -EINVAL;
2413
2414                 new_mtu = max_mtu;
2415         }
2416
2417         dev->mtu = new_mtu;
2418         return 0;
2419 }
2420
2421 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2422 {
2423         struct vxlan_dev *vxlan = netdev_priv(dev);
2424         struct vxlan_rdst *dst = &vxlan->default_dst;
2425         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2426                                                          dst->remote_ifindex);
2427         return __vxlan_change_mtu(dev, lowerdev, dst, new_mtu, true);
2428 }
2429
2430 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2431 {
2432         struct vxlan_dev *vxlan = netdev_priv(dev);
2433         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2434         __be16 sport, dport;
2435
2436         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2437                                   vxlan->cfg.port_max, true);
2438         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2439
2440         if (ip_tunnel_info_af(info) == AF_INET) {
2441                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2442                 struct rtable *rt;
2443
2444                 if (!sock4)
2445                         return -EINVAL;
2446                 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2447                                      info->key.u.ipv4.dst,
2448                                      &info->key.u.ipv4.src, NULL, info);
2449                 if (IS_ERR(rt))
2450                         return PTR_ERR(rt);
2451                 ip_rt_put(rt);
2452         } else {
2453 #if IS_ENABLED(CONFIG_IPV6)
2454                 struct dst_entry *ndst;
2455
2456                 ndst = vxlan6_get_route(vxlan, skb, 0, info->key.tos,
2457                                         info->key.label, &info->key.u.ipv6.dst,
2458                                         &info->key.u.ipv6.src, NULL, info);
2459                 if (IS_ERR(ndst))
2460                         return PTR_ERR(ndst);
2461                 dst_release(ndst);
2462 #else /* !CONFIG_IPV6 */
2463                 return -EPFNOSUPPORT;
2464 #endif
2465         }
2466         info->key.tp_src = sport;
2467         info->key.tp_dst = dport;
2468         return 0;
2469 }
2470
2471 static const struct net_device_ops vxlan_netdev_ether_ops = {
2472         .ndo_init               = vxlan_init,
2473         .ndo_uninit             = vxlan_uninit,
2474         .ndo_open               = vxlan_open,
2475         .ndo_stop               = vxlan_stop,
2476         .ndo_start_xmit         = vxlan_xmit,
2477         .ndo_get_stats64        = ip_tunnel_get_stats64,
2478         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2479         .ndo_change_mtu         = vxlan_change_mtu,
2480         .ndo_validate_addr      = eth_validate_addr,
2481         .ndo_set_mac_address    = eth_mac_addr,
2482         .ndo_fdb_add            = vxlan_fdb_add,
2483         .ndo_fdb_del            = vxlan_fdb_delete,
2484         .ndo_fdb_dump           = vxlan_fdb_dump,
2485         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2486 };
2487
2488 static const struct net_device_ops vxlan_netdev_raw_ops = {
2489         .ndo_init               = vxlan_init,
2490         .ndo_uninit             = vxlan_uninit,
2491         .ndo_open               = vxlan_open,
2492         .ndo_stop               = vxlan_stop,
2493         .ndo_start_xmit         = vxlan_xmit,
2494         .ndo_get_stats64        = ip_tunnel_get_stats64,
2495         .ndo_change_mtu         = vxlan_change_mtu,
2496         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2497 };
2498
2499 /* Info for udev, that this is a virtual tunnel endpoint */
2500 static struct device_type vxlan_type = {
2501         .name = "vxlan",
2502 };
2503
2504 /* Calls the ndo_udp_tunnel_add of the caller in order to
2505  * supply the listening VXLAN udp ports. Callers are expected
2506  * to implement the ndo_udp_tunnel_add.
2507  */
2508 static void vxlan_push_rx_ports(struct net_device *dev)
2509 {
2510         struct vxlan_sock *vs;
2511         struct net *net = dev_net(dev);
2512         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2513         unsigned int i;
2514
2515         spin_lock(&vn->sock_lock);
2516         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2517                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2518                         udp_tunnel_push_rx_port(dev, vs->sock,
2519                                                 (vs->flags & VXLAN_F_GPE) ?
2520                                                 UDP_TUNNEL_TYPE_VXLAN_GPE :
2521                                                 UDP_TUNNEL_TYPE_VXLAN);
2522         }
2523         spin_unlock(&vn->sock_lock);
2524 }
2525
2526 /* Initialize the device structure. */
2527 static void vxlan_setup(struct net_device *dev)
2528 {
2529         struct vxlan_dev *vxlan = netdev_priv(dev);
2530         unsigned int h;
2531
2532         eth_hw_addr_random(dev);
2533         ether_setup(dev);
2534
2535         dev->destructor = free_netdev;
2536         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2537
2538         dev->features   |= NETIF_F_LLTX;
2539         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2540         dev->features   |= NETIF_F_RXCSUM;
2541         dev->features   |= NETIF_F_GSO_SOFTWARE;
2542
2543         dev->vlan_features = dev->features;
2544         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2545         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2546         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2547         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2548         netif_keep_dst(dev);
2549         dev->priv_flags |= IFF_NO_QUEUE;
2550
2551         INIT_LIST_HEAD(&vxlan->next);
2552         spin_lock_init(&vxlan->hash_lock);
2553
2554         init_timer_deferrable(&vxlan->age_timer);
2555         vxlan->age_timer.function = vxlan_cleanup;
2556         vxlan->age_timer.data = (unsigned long) vxlan;
2557
2558         vxlan->cfg.dst_port = htons(vxlan_port);
2559
2560         vxlan->dev = dev;
2561
2562         gro_cells_init(&vxlan->gro_cells, dev);
2563
2564         for (h = 0; h < FDB_HASH_SIZE; ++h)
2565                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2566 }
2567
2568 static void vxlan_ether_setup(struct net_device *dev)
2569 {
2570         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2571         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2572         dev->netdev_ops = &vxlan_netdev_ether_ops;
2573 }
2574
2575 static void vxlan_raw_setup(struct net_device *dev)
2576 {
2577         dev->header_ops = NULL;
2578         dev->type = ARPHRD_NONE;
2579         dev->hard_header_len = 0;
2580         dev->addr_len = 0;
2581         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2582         dev->netdev_ops = &vxlan_netdev_raw_ops;
2583 }
2584
2585 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2586         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2587         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2588         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2589         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2590         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2591         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2592         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2593         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2594         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2595         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2596         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2597         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2598         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2599         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2600         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2601         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2602         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2603         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2604         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2605         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2606         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2607         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2608         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2609         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2610         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2611         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2612         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2613 };
2614
2615 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2616 {
2617         if (tb[IFLA_ADDRESS]) {
2618                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2619                         pr_debug("invalid link address (not ethernet)\n");
2620                         return -EINVAL;
2621                 }
2622
2623                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2624                         pr_debug("invalid all zero ethernet address\n");
2625                         return -EADDRNOTAVAIL;
2626                 }
2627         }
2628
2629         if (!data)
2630                 return -EINVAL;
2631
2632         if (data[IFLA_VXLAN_ID]) {
2633                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2634                 if (id >= VXLAN_VID_MASK)
2635                         return -ERANGE;
2636         }
2637
2638         if (data[IFLA_VXLAN_PORT_RANGE]) {
2639                 const struct ifla_vxlan_port_range *p
2640                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2641
2642                 if (ntohs(p->high) < ntohs(p->low)) {
2643                         pr_debug("port range %u .. %u not valid\n",
2644                                  ntohs(p->low), ntohs(p->high));
2645                         return -EINVAL;
2646                 }
2647         }
2648
2649         return 0;
2650 }
2651
2652 static void vxlan_get_drvinfo(struct net_device *netdev,
2653                               struct ethtool_drvinfo *drvinfo)
2654 {
2655         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2656         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2657 }
2658
2659 static const struct ethtool_ops vxlan_ethtool_ops = {
2660         .get_drvinfo    = vxlan_get_drvinfo,
2661         .get_link       = ethtool_op_get_link,
2662 };
2663
2664 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2665                                         __be16 port, u32 flags)
2666 {
2667         struct socket *sock;
2668         struct udp_port_cfg udp_conf;
2669         int err;
2670
2671         memset(&udp_conf, 0, sizeof(udp_conf));
2672
2673         if (ipv6) {
2674                 udp_conf.family = AF_INET6;
2675                 udp_conf.use_udp6_rx_checksums =
2676                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2677                 udp_conf.ipv6_v6only = 1;
2678         } else {
2679                 udp_conf.family = AF_INET;
2680         }
2681
2682         udp_conf.local_udp_port = port;
2683
2684         /* Open UDP socket */
2685         err = udp_sock_create(net, &udp_conf, &sock);
2686         if (err < 0)
2687                 return ERR_PTR(err);
2688
2689         return sock;
2690 }
2691
2692 /* Create new listen socket if needed */
2693 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2694                                               __be16 port, u32 flags)
2695 {
2696         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2697         struct vxlan_sock *vs;
2698         struct socket *sock;
2699         unsigned int h;
2700         struct udp_tunnel_sock_cfg tunnel_cfg;
2701
2702         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2703         if (!vs)
2704                 return ERR_PTR(-ENOMEM);
2705
2706         for (h = 0; h < VNI_HASH_SIZE; ++h)
2707                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2708
2709         sock = vxlan_create_sock(net, ipv6, port, flags);
2710         if (IS_ERR(sock)) {
2711                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2712                         PTR_ERR(sock));
2713                 kfree(vs);
2714                 return ERR_CAST(sock);
2715         }
2716
2717         vs->sock = sock;
2718         atomic_set(&vs->refcnt, 1);
2719         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2720
2721         spin_lock(&vn->sock_lock);
2722         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2723         udp_tunnel_notify_add_rx_port(sock,
2724                                       (vs->flags & VXLAN_F_GPE) ?
2725                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2726                                       UDP_TUNNEL_TYPE_VXLAN);
2727         spin_unlock(&vn->sock_lock);
2728
2729         /* Mark socket as an encapsulation socket. */
2730         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2731         tunnel_cfg.sk_user_data = vs;
2732         tunnel_cfg.encap_type = 1;
2733         tunnel_cfg.encap_rcv = vxlan_rcv;
2734         tunnel_cfg.encap_destroy = NULL;
2735         tunnel_cfg.gro_receive = vxlan_gro_receive;
2736         tunnel_cfg.gro_complete = vxlan_gro_complete;
2737
2738         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2739
2740         return vs;
2741 }
2742
2743 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2744 {
2745         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2746         struct vxlan_sock *vs = NULL;
2747
2748         if (!vxlan->cfg.no_share) {
2749                 spin_lock(&vn->sock_lock);
2750                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2751                                      vxlan->cfg.dst_port, vxlan->flags);
2752                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2753                         spin_unlock(&vn->sock_lock);
2754                         return -EBUSY;
2755                 }
2756                 spin_unlock(&vn->sock_lock);
2757         }
2758         if (!vs)
2759                 vs = vxlan_socket_create(vxlan->net, ipv6,
2760                                          vxlan->cfg.dst_port, vxlan->flags);
2761         if (IS_ERR(vs))
2762                 return PTR_ERR(vs);
2763 #if IS_ENABLED(CONFIG_IPV6)
2764         if (ipv6)
2765                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2766         else
2767 #endif
2768                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2769         vxlan_vs_add_dev(vs, vxlan);
2770         return 0;
2771 }
2772
2773 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2774 {
2775         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2776         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2777         int ret = 0;
2778
2779         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2780 #if IS_ENABLED(CONFIG_IPV6)
2781         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2782         if (ipv6 || metadata)
2783                 ret = __vxlan_sock_add(vxlan, true);
2784 #endif
2785         if (!ret && (!ipv6 || metadata))
2786                 ret = __vxlan_sock_add(vxlan, false);
2787         if (ret < 0)
2788                 vxlan_sock_release(vxlan);
2789         return ret;
2790 }
2791
2792 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2793                                struct vxlan_config *conf)
2794 {
2795         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2796         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2797         struct vxlan_rdst *dst = &vxlan->default_dst;
2798         unsigned short needed_headroom = ETH_HLEN;
2799         int err;
2800         bool use_ipv6 = false;
2801         __be16 default_port = vxlan->cfg.dst_port;
2802         struct net_device *lowerdev = NULL;
2803
2804         if (conf->flags & VXLAN_F_GPE) {
2805                 /* For now, allow GPE only together with COLLECT_METADATA.
2806                  * This can be relaxed later; in such case, the other side
2807                  * of the PtP link will have to be provided.
2808                  */
2809                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2810                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2811                         pr_info("unsupported combination of extensions\n");
2812                         return -EINVAL;
2813                 }
2814
2815                 vxlan_raw_setup(dev);
2816         } else {
2817                 vxlan_ether_setup(dev);
2818         }
2819
2820         vxlan->net = src_net;
2821
2822         dst->remote_vni = conf->vni;
2823
2824         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2825
2826         /* Unless IPv6 is explicitly requested, assume IPv4 */
2827         if (!dst->remote_ip.sa.sa_family)
2828                 dst->remote_ip.sa.sa_family = AF_INET;
2829
2830         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2831             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2832                 if (!IS_ENABLED(CONFIG_IPV6))
2833                         return -EPFNOSUPPORT;
2834                 use_ipv6 = true;
2835                 vxlan->flags |= VXLAN_F_IPV6;
2836         }
2837
2838         if (conf->label && !use_ipv6) {
2839                 pr_info("label only supported in use with IPv6\n");
2840                 return -EINVAL;
2841         }
2842
2843         if (conf->remote_ifindex) {
2844                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2845                 dst->remote_ifindex = conf->remote_ifindex;
2846
2847                 if (!lowerdev) {
2848                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2849                         return -ENODEV;
2850                 }
2851
2852 #if IS_ENABLED(CONFIG_IPV6)
2853                 if (use_ipv6) {
2854                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2855                         if (idev && idev->cnf.disable_ipv6) {
2856                                 pr_info("IPv6 is disabled via sysctl\n");
2857                                 return -EPERM;
2858                         }
2859                 }
2860 #endif
2861
2862                 if (!conf->mtu)
2863                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2864
2865                 needed_headroom = lowerdev->hard_header_len;
2866         } else if (vxlan_addr_multicast(&dst->remote_ip)) {
2867                 pr_info("multicast destination requires interface to be specified\n");
2868                 return -EINVAL;
2869         }
2870
2871         if (conf->mtu) {
2872                 err = __vxlan_change_mtu(dev, lowerdev, dst, conf->mtu, false);
2873                 if (err)
2874                         return err;
2875         }
2876
2877         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2878                 needed_headroom += VXLAN6_HEADROOM;
2879         else
2880                 needed_headroom += VXLAN_HEADROOM;
2881         dev->needed_headroom = needed_headroom;
2882
2883         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2884         if (!vxlan->cfg.dst_port) {
2885                 if (conf->flags & VXLAN_F_GPE)
2886                         vxlan->cfg.dst_port = 4790; /* IANA assigned VXLAN-GPE port */
2887                 else
2888                         vxlan->cfg.dst_port = default_port;
2889         }
2890         vxlan->flags |= conf->flags;
2891
2892         if (!vxlan->cfg.age_interval)
2893                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2894
2895         list_for_each_entry(tmp, &vn->vxlan_list, next) {
2896                 if (tmp->cfg.vni == conf->vni &&
2897                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2898                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2899                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2900                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2901                     (vxlan->flags & VXLAN_F_RCV_FLAGS)) {
2902                         pr_info("duplicate VNI %u\n", be32_to_cpu(conf->vni));
2903                         return -EEXIST;
2904                 }
2905         }
2906
2907         dev->ethtool_ops = &vxlan_ethtool_ops;
2908
2909         /* create an fdb entry for a valid default destination */
2910         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2911                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2912                                        &vxlan->default_dst.remote_ip,
2913                                        NUD_REACHABLE|NUD_PERMANENT,
2914                                        NLM_F_EXCL|NLM_F_CREATE,
2915                                        vxlan->cfg.dst_port,
2916                                        vxlan->default_dst.remote_vni,
2917                                        vxlan->default_dst.remote_ifindex,
2918                                        NTF_SELF);
2919                 if (err)
2920                         return err;
2921         }
2922
2923         err = register_netdevice(dev);
2924         if (err) {
2925                 vxlan_fdb_delete_default(vxlan);
2926                 return err;
2927         }
2928
2929         list_add(&vxlan->next, &vn->vxlan_list);
2930
2931         return 0;
2932 }
2933
2934 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2935                          struct nlattr *tb[], struct nlattr *data[])
2936 {
2937         struct vxlan_config conf;
2938
2939         memset(&conf, 0, sizeof(conf));
2940
2941         if (data[IFLA_VXLAN_ID])
2942                 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2943
2944         if (data[IFLA_VXLAN_GROUP]) {
2945                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2946         } else if (data[IFLA_VXLAN_GROUP6]) {
2947                 if (!IS_ENABLED(CONFIG_IPV6))
2948                         return -EPFNOSUPPORT;
2949
2950                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2951                 conf.remote_ip.sa.sa_family = AF_INET6;
2952         }
2953
2954         if (data[IFLA_VXLAN_LOCAL]) {
2955                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2956                 conf.saddr.sa.sa_family = AF_INET;
2957         } else if (data[IFLA_VXLAN_LOCAL6]) {
2958                 if (!IS_ENABLED(CONFIG_IPV6))
2959                         return -EPFNOSUPPORT;
2960
2961                 /* TODO: respect scope id */
2962                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2963                 conf.saddr.sa.sa_family = AF_INET6;
2964         }
2965
2966         if (data[IFLA_VXLAN_LINK])
2967                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2968
2969         if (data[IFLA_VXLAN_TOS])
2970                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2971
2972         if (data[IFLA_VXLAN_TTL])
2973                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2974
2975         if (data[IFLA_VXLAN_LABEL])
2976                 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
2977                              IPV6_FLOWLABEL_MASK;
2978
2979         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2980                 conf.flags |= VXLAN_F_LEARN;
2981
2982         if (data[IFLA_VXLAN_AGEING])
2983                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2984
2985         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2986                 conf.flags |= VXLAN_F_PROXY;
2987
2988         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2989                 conf.flags |= VXLAN_F_RSC;
2990
2991         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2992                 conf.flags |= VXLAN_F_L2MISS;
2993
2994         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2995                 conf.flags |= VXLAN_F_L3MISS;
2996
2997         if (data[IFLA_VXLAN_LIMIT])
2998                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2999
3000         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
3001             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3002                 conf.flags |= VXLAN_F_COLLECT_METADATA;
3003
3004         if (data[IFLA_VXLAN_PORT_RANGE]) {
3005                 const struct ifla_vxlan_port_range *p
3006                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3007                 conf.port_min = ntohs(p->low);
3008                 conf.port_max = ntohs(p->high);
3009         }
3010
3011         if (data[IFLA_VXLAN_PORT])
3012                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3013
3014         if (data[IFLA_VXLAN_UDP_CSUM] &&
3015             !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3016                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3017
3018         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
3019             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3020                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3021
3022         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3023             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3024                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3025
3026         if (data[IFLA_VXLAN_REMCSUM_TX] &&
3027             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3028                 conf.flags |= VXLAN_F_REMCSUM_TX;
3029
3030         if (data[IFLA_VXLAN_REMCSUM_RX] &&
3031             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3032                 conf.flags |= VXLAN_F_REMCSUM_RX;
3033
3034         if (data[IFLA_VXLAN_GBP])
3035                 conf.flags |= VXLAN_F_GBP;
3036
3037         if (data[IFLA_VXLAN_GPE])
3038                 conf.flags |= VXLAN_F_GPE;
3039
3040         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3041                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3042
3043         if (tb[IFLA_MTU])
3044                 conf.mtu = nla_get_u32(tb[IFLA_MTU]);
3045
3046         return vxlan_dev_configure(src_net, dev, &conf);
3047 }
3048
3049 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3050 {
3051         struct vxlan_dev *vxlan = netdev_priv(dev);
3052         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3053
3054         spin_lock(&vn->sock_lock);
3055         if (!hlist_unhashed(&vxlan->hlist))
3056                 hlist_del_rcu(&vxlan->hlist);
3057         spin_unlock(&vn->sock_lock);
3058
3059         gro_cells_destroy(&vxlan->gro_cells);
3060         list_del(&vxlan->next);
3061         unregister_netdevice_queue(dev, head);
3062 }
3063
3064 static size_t vxlan_get_size(const struct net_device *dev)
3065 {
3066
3067         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3068                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3069                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3070                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3071                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3072                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3073                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3074                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3075                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3076                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3077                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3078                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3079                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3080                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3081                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3082                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3083                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3084                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3085                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3086                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3087                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3088                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3089                 0;
3090 }
3091
3092 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3093 {
3094         const struct vxlan_dev *vxlan = netdev_priv(dev);
3095         const struct vxlan_rdst *dst = &vxlan->default_dst;
3096         struct ifla_vxlan_port_range ports = {
3097                 .low =  htons(vxlan->cfg.port_min),
3098                 .high = htons(vxlan->cfg.port_max),
3099         };
3100
3101         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3102                 goto nla_put_failure;
3103
3104         if (!vxlan_addr_any(&dst->remote_ip)) {
3105                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3106                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3107                                             dst->remote_ip.sin.sin_addr.s_addr))
3108                                 goto nla_put_failure;
3109 #if IS_ENABLED(CONFIG_IPV6)
3110                 } else {
3111                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3112                                              &dst->remote_ip.sin6.sin6_addr))
3113                                 goto nla_put_failure;
3114 #endif
3115                 }
3116         }
3117
3118         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3119                 goto nla_put_failure;
3120
3121         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3122                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3123                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3124                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3125                                 goto nla_put_failure;
3126 #if IS_ENABLED(CONFIG_IPV6)
3127                 } else {
3128                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3129                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3130                                 goto nla_put_failure;
3131 #endif
3132                 }
3133         }
3134
3135         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3136             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3137             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3138             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3139                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3140             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3141                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3142             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3143             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3144                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3145             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3146                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3147             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3148                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3149             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3150             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3151             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3152             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3153                         !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3154             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3155                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3156             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3157                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3158             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3159                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3160             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3161                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3162                 goto nla_put_failure;
3163
3164         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3165                 goto nla_put_failure;
3166
3167         if (vxlan->flags & VXLAN_F_GBP &&
3168             nla_put_flag(skb, IFLA_VXLAN_GBP))
3169                 goto nla_put_failure;
3170
3171         if (vxlan->flags & VXLAN_F_GPE &&
3172             nla_put_flag(skb, IFLA_VXLAN_GPE))
3173                 goto nla_put_failure;
3174
3175         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3176             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3177                 goto nla_put_failure;
3178
3179         return 0;
3180
3181 nla_put_failure:
3182         return -EMSGSIZE;
3183 }
3184
3185 static struct net *vxlan_get_link_net(const struct net_device *dev)
3186 {
3187         struct vxlan_dev *vxlan = netdev_priv(dev);
3188
3189         return vxlan->net;
3190 }
3191
3192 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3193         .kind           = "vxlan",
3194         .maxtype        = IFLA_VXLAN_MAX,
3195         .policy         = vxlan_policy,
3196         .priv_size      = sizeof(struct vxlan_dev),
3197         .setup          = vxlan_setup,
3198         .validate       = vxlan_validate,
3199         .newlink        = vxlan_newlink,
3200         .dellink        = vxlan_dellink,
3201         .get_size       = vxlan_get_size,
3202         .fill_info      = vxlan_fill_info,
3203         .get_link_net   = vxlan_get_link_net,
3204 };
3205
3206 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3207                                     u8 name_assign_type,
3208                                     struct vxlan_config *conf)
3209 {
3210         struct nlattr *tb[IFLA_MAX + 1];
3211         struct net_device *dev;
3212         int err;
3213
3214         memset(&tb, 0, sizeof(tb));
3215
3216         dev = rtnl_create_link(net, name, name_assign_type,
3217                                &vxlan_link_ops, tb);
3218         if (IS_ERR(dev))
3219                 return dev;
3220
3221         err = vxlan_dev_configure(net, dev, conf);
3222         if (err < 0) {
3223                 free_netdev(dev);
3224                 return ERR_PTR(err);
3225         }
3226
3227         err = rtnl_configure_link(dev, NULL);
3228         if (err < 0) {
3229                 LIST_HEAD(list_kill);
3230
3231                 vxlan_dellink(dev, &list_kill);
3232                 unregister_netdevice_many(&list_kill);
3233                 return ERR_PTR(err);
3234         }
3235
3236         return dev;
3237 }
3238 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3239
3240 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3241                                              struct net_device *dev)
3242 {
3243         struct vxlan_dev *vxlan, *next;
3244         LIST_HEAD(list_kill);
3245
3246         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3247                 struct vxlan_rdst *dst = &vxlan->default_dst;
3248
3249                 /* In case we created vxlan device with carrier
3250                  * and we loose the carrier due to module unload
3251                  * we also need to remove vxlan device. In other
3252                  * cases, it's not necessary and remote_ifindex
3253                  * is 0 here, so no matches.
3254                  */
3255                 if (dst->remote_ifindex == dev->ifindex)
3256                         vxlan_dellink(vxlan->dev, &list_kill);
3257         }
3258
3259         unregister_netdevice_many(&list_kill);
3260 }
3261
3262 static int vxlan_netdevice_event(struct notifier_block *unused,
3263                                  unsigned long event, void *ptr)
3264 {
3265         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3266         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3267
3268         if (event == NETDEV_UNREGISTER)
3269                 vxlan_handle_lowerdev_unregister(vn, dev);
3270         else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3271                 vxlan_push_rx_ports(dev);
3272
3273         return NOTIFY_DONE;
3274 }
3275
3276 static struct notifier_block vxlan_notifier_block __read_mostly = {
3277         .notifier_call = vxlan_netdevice_event,
3278 };
3279
3280 static __net_init int vxlan_init_net(struct net *net)
3281 {
3282         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3283         unsigned int h;
3284
3285         INIT_LIST_HEAD(&vn->vxlan_list);
3286         spin_lock_init(&vn->sock_lock);
3287
3288         for (h = 0; h < PORT_HASH_SIZE; ++h)
3289                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3290
3291         return 0;
3292 }
3293
3294 static void __net_exit vxlan_exit_net(struct net *net)
3295 {
3296         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3297         struct vxlan_dev *vxlan, *next;
3298         struct net_device *dev, *aux;
3299         LIST_HEAD(list);
3300
3301         rtnl_lock();
3302         for_each_netdev_safe(net, dev, aux)
3303                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3304                         unregister_netdevice_queue(dev, &list);
3305
3306         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3307                 /* If vxlan->dev is in the same netns, it has already been added
3308                  * to the list by the previous loop.
3309                  */
3310                 if (!net_eq(dev_net(vxlan->dev), net)) {
3311                         gro_cells_destroy(&vxlan->gro_cells);
3312                         unregister_netdevice_queue(vxlan->dev, &list);
3313                 }
3314         }
3315
3316         unregister_netdevice_many(&list);
3317         rtnl_unlock();
3318 }
3319
3320 static struct pernet_operations vxlan_net_ops = {
3321         .init = vxlan_init_net,
3322         .exit = vxlan_exit_net,
3323         .id   = &vxlan_net_id,
3324         .size = sizeof(struct vxlan_net),
3325 };
3326
3327 static int __init vxlan_init_module(void)
3328 {
3329         int rc;
3330
3331         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3332
3333         rc = register_pernet_subsys(&vxlan_net_ops);
3334         if (rc)
3335                 goto out1;
3336
3337         rc = register_netdevice_notifier(&vxlan_notifier_block);
3338         if (rc)
3339                 goto out2;
3340
3341         rc = rtnl_link_register(&vxlan_link_ops);
3342         if (rc)
3343                 goto out3;
3344
3345         return 0;
3346 out3:
3347         unregister_netdevice_notifier(&vxlan_notifier_block);
3348 out2:
3349         unregister_pernet_subsys(&vxlan_net_ops);
3350 out1:
3351         return rc;
3352 }
3353 late_initcall(vxlan_init_module);
3354
3355 static void __exit vxlan_cleanup_module(void)
3356 {
3357         rtnl_link_unregister(&vxlan_link_ops);
3358         unregister_netdevice_notifier(&vxlan_notifier_block);
3359         unregister_pernet_subsys(&vxlan_net_ops);
3360         /* rcu_barrier() is called by netns */
3361 }
3362 module_exit(vxlan_cleanup_module);
3363
3364 MODULE_LICENSE("GPL");
3365 MODULE_VERSION(VXLAN_VERSION);
3366 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3367 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3368 MODULE_ALIAS_RTNL_LINK("vxlan");