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         struct vxlan_sock *sock6 = NULL;
 948         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
 949
 950         sock4 = rtnl_dereference(dev->vn4_sock);
 951
 952         /* The vxlan_sock is only used by dev, leaving group has
 953          * no effect on other vxlan devices.
 954          */
 955         if (family == AF_INET && sock4 && atomic_read(&sock4->refcnt) == 1)
 956                 return false;
 957 #if IS_ENABLED(CONFIG_IPV6)
 958         sock6 = rtnl_dereference(dev->vn6_sock);
 959         if (family == AF_INET6 && sock6 && atomic_read(&sock6->refcnt) == 1)
 960                 return false;
 961 #endif
 962
 963         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
 964                 if (!netif_running(vxlan->dev) || vxlan == dev)
 965                         continue;
 966
 967                 if (family == AF_INET &&
 968                     rtnl_dereference(vxlan->vn4_sock) != sock4)
 969                         continue;
 970 #if IS_ENABLED(CONFIG_IPV6)
 971                 if (family == AF_INET6 &&
 972                     rtnl_dereference(vxlan->vn6_sock) != sock6)
 973                         continue;
 974 #endif
 975
 976                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
 977                                       &dev->default_dst.remote_ip))
 978                         continue;
 979
 980                 if (vxlan->default_dst.remote_ifindex !=
 981                     dev->default_dst.remote_ifindex)
 982                         continue;
 983
 984                 return true;
 985         }
 986
 987         return false;
 988 }
 989
 990 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
 991 {
 992         struct vxlan_net *vn;
 993
 994         if (!vs)
 995                 return false;
 996         if (!atomic_dec_and_test(&vs->refcnt))
 997                 return false;
 998
 999         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1000         spin_lock(&vn->sock_lock);
1001         hlist_del_rcu(&vs->hlist);
1002         udp_tunnel_notify_del_rx_port(vs->sock,
1003                                       (vs->flags & VXLAN_F_GPE) ?
1004                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1005                                       UDP_TUNNEL_TYPE_VXLAN);
1006         spin_unlock(&vn->sock_lock);
1007
1008         return true;
1009 }
1010
1011 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1012 {
1013         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1014 #if IS_ENABLED(CONFIG_IPV6)
1015         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1016
1017         rcu_assign_pointer(vxlan->vn6_sock, NULL);
1018 #endif
1019
1020         rcu_assign_pointer(vxlan->vn4_sock, NULL);
1021         synchronize_net();
1022
1023         if (__vxlan_sock_release_prep(sock4)) {
1024                 udp_tunnel_sock_release(sock4->sock);
1025                 kfree(sock4);
1026         }
1027
1028 #if IS_ENABLED(CONFIG_IPV6)
1029         if (__vxlan_sock_release_prep(sock6)) {
1030                 udp_tunnel_sock_release(sock6->sock);
1031                 kfree(sock6);
1032         }
1033 #endif
1034 }
1035
1036 /* Update multicast group membership when first VNI on
1037  * multicast address is brought up
1038  */
1039 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1040 {
1041         struct sock *sk;
1042         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1043         int ifindex = vxlan->default_dst.remote_ifindex;
1044         int ret = -EINVAL;
1045
1046         if (ip->sa.sa_family == AF_INET) {
1047                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1048                 struct ip_mreqn mreq = {
1049                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1050                         .imr_ifindex            = ifindex,
1051                 };
1052
1053                 sk = sock4->sock->sk;
1054                 lock_sock(sk);
1055                 ret = ip_mc_join_group(sk, &mreq);
1056                 release_sock(sk);
1057 #if IS_ENABLED(CONFIG_IPV6)
1058         } else {
1059                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1060
1061                 sk = sock6->sock->sk;
1062                 lock_sock(sk);
1063                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1064                                                    &ip->sin6.sin6_addr);
1065                 release_sock(sk);
1066 #endif
1067         }
1068
1069         return ret;
1070 }
1071
1072 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1073 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1074 {
1075         struct sock *sk;
1076         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1077         int ifindex = vxlan->default_dst.remote_ifindex;
1078         int ret = -EINVAL;
1079
1080         if (ip->sa.sa_family == AF_INET) {
1081                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1082                 struct ip_mreqn mreq = {
1083                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1084                         .imr_ifindex            = ifindex,
1085                 };
1086
1087                 sk = sock4->sock->sk;
1088                 lock_sock(sk);
1089                 ret = ip_mc_leave_group(sk, &mreq);
1090                 release_sock(sk);
1091 #if IS_ENABLED(CONFIG_IPV6)
1092         } else {
1093                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1094
1095                 sk = sock6->sock->sk;
1096                 lock_sock(sk);
1097                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1098                                                    &ip->sin6.sin6_addr);
1099                 release_sock(sk);
1100 #endif
1101         }
1102
1103         return ret;
1104 }
1105
1106 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1107                           struct sk_buff *skb, u32 vxflags)
1108 {
1109         size_t start, offset;
1110
1111         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1112                 goto out;
1113
1114         start = vxlan_rco_start(unparsed->vx_vni);
1115         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1116
1117         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1118                 return false;
1119
1120         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1121                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1122 out:
1123         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1124         unparsed->vx_vni &= VXLAN_VNI_MASK;
1125         return true;
1126 }
1127
1128 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1129                                 struct sk_buff *skb, u32 vxflags,
1130                                 struct vxlan_metadata *md)
1131 {
1132         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1133         struct metadata_dst *tun_dst;
1134
1135         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1136                 goto out;
1137
1138         md->gbp = ntohs(gbp->policy_id);
1139
1140         tun_dst = (struct metadata_dst *)skb_dst(skb);
1141         if (tun_dst) {
1142                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1143                 tun_dst->u.tun_info.options_len = sizeof(*md);
1144         }
1145         if (gbp->dont_learn)
1146                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1147
1148         if (gbp->policy_applied)
1149                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1150
1151         /* In flow-based mode, GBP is carried in dst_metadata */
1152         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1153                 skb->mark = md->gbp;
1154 out:
1155         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1156 }
1157
1158 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1159                                 __be16 *protocol,
1160                                 struct sk_buff *skb, u32 vxflags)
1161 {
1162         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1163
1164         /* Need to have Next Protocol set for interfaces in GPE mode. */
1165         if (!gpe->np_applied)
1166                 return false;
1167         /* "The initial version is 0. If a receiver does not support the
1168          * version indicated it MUST drop the packet.
1169          */
1170         if (gpe->version != 0)
1171                 return false;
1172         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1173          * processing MUST occur." However, we don't implement OAM
1174          * processing, thus drop the packet.
1175          */
1176         if (gpe->oam_flag)
1177                 return false;
1178
1179         switch (gpe->next_protocol) {
1180         case VXLAN_GPE_NP_IPV4:
1181                 *protocol = htons(ETH_P_IP);
1182                 break;
1183         case VXLAN_GPE_NP_IPV6:
1184                 *protocol = htons(ETH_P_IPV6);
1185                 break;
1186         case VXLAN_GPE_NP_ETHERNET:
1187                 *protocol = htons(ETH_P_TEB);
1188                 break;
1189         default:
1190                 return false;
1191         }
1192
1193         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1194         return true;
1195 }
1196
1197 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1198                           struct vxlan_sock *vs,
1199                           struct sk_buff *skb)
1200 {
1201         union vxlan_addr saddr;
1202
1203         skb_reset_mac_header(skb);
1204         skb->protocol = eth_type_trans(skb, vxlan->dev);
1205         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1206
1207         /* Ignore packet loops (and multicast echo) */
1208         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1209                 return false;
1210
1211         /* Get address from the outer IP header */
1212         if (vxlan_get_sk_family(vs) == AF_INET) {
1213                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1214                 saddr.sa.sa_family = AF_INET;
1215 #if IS_ENABLED(CONFIG_IPV6)
1216         } else {
1217                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1218                 saddr.sa.sa_family = AF_INET6;
1219 #endif
1220         }
1221
1222         if ((vxlan->flags & VXLAN_F_LEARN) &&
1223             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1224                 return false;
1225
1226         return true;
1227 }
1228
1229 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1230                                   struct sk_buff *skb)
1231 {
1232         int err = 0;
1233
1234         if (vxlan_get_sk_family(vs) == AF_INET)
1235                 err = IP_ECN_decapsulate(oiph, skb);
1236 #if IS_ENABLED(CONFIG_IPV6)
1237         else
1238                 err = IP6_ECN_decapsulate(oiph, skb);
1239 #endif
1240
1241         if (unlikely(err) && log_ecn_error) {
1242                 if (vxlan_get_sk_family(vs) == AF_INET)
1243                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1244                                              &((struct iphdr *)oiph)->saddr,
1245                                              ((struct iphdr *)oiph)->tos);
1246                 else
1247                         net_info_ratelimited("non-ECT from %pI6\n",
1248                                              &((struct ipv6hdr *)oiph)->saddr);
1249         }
1250         return err <= 1;
1251 }
1252
1253 /* Callback from net/ipv4/udp.c to receive packets */
1254 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1255 {
1256         struct pcpu_sw_netstats *stats;
1257         struct vxlan_dev *vxlan;
1258         struct vxlan_sock *vs;
1259         struct vxlanhdr unparsed;
1260         struct vxlan_metadata _md;
1261         struct vxlan_metadata *md = &_md;
1262         __be16 protocol = htons(ETH_P_TEB);
1263         bool raw_proto = false;
1264         void *oiph;
1265
1266         /* Need UDP and VXLAN header to be present */
1267         if (!pskb_may_pull(skb, VXLAN_HLEN))
1268                 goto drop;
1269
1270         unparsed = *vxlan_hdr(skb);
1271         /* VNI flag always required to be set */
1272         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1273                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1274                            ntohl(vxlan_hdr(skb)->vx_flags),
1275                            ntohl(vxlan_hdr(skb)->vx_vni));
1276                 /* Return non vxlan pkt */
1277                 goto drop;
1278         }
1279         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1280         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1281
1282         vs = rcu_dereference_sk_user_data(sk);
1283         if (!vs)
1284                 goto drop;
1285
1286         vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1287         if (!vxlan)
1288                 goto drop;
1289
1290         /* For backwards compatibility, only allow reserved fields to be
1291          * used by VXLAN extensions if explicitly requested.
1292          */
1293         if (vs->flags & VXLAN_F_GPE) {
1294                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1295                         goto drop;
1296                 raw_proto = true;
1297         }
1298
1299         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1300                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1301                         goto drop;
1302
1303         if (vxlan_collect_metadata(vs)) {
1304                 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1305                 struct metadata_dst *tun_dst;
1306
1307                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1308                                          key32_to_tunnel_id(vni), sizeof(*md));
1309
1310                 if (!tun_dst)
1311                         goto drop;
1312
1313                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1314
1315                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1316         } else {
1317                 memset(md, 0, sizeof(*md));
1318         }
1319
1320         if (vs->flags & VXLAN_F_REMCSUM_RX)
1321                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1322                         goto drop;
1323         if (vs->flags & VXLAN_F_GBP)
1324                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1325         /* Note that GBP and GPE can never be active together. This is
1326          * ensured in vxlan_dev_configure.
1327          */
1328
1329         if (unparsed.vx_flags || unparsed.vx_vni) {
1330                 /* If there are any unprocessed flags remaining treat
1331                  * this as a malformed packet. This behavior diverges from
1332                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1333                  * in reserved fields are to be ignored. The approach here
1334                  * maintains compatibility with previous stack code, and also
1335                  * is more robust and provides a little more security in
1336                  * adding extensions to VXLAN.
1337                  */
1338                 goto drop;
1339         }
1340
1341         if (!raw_proto) {
1342                 if (!vxlan_set_mac(vxlan, vs, skb))
1343                         goto drop;
1344         } else {
1345                 skb_reset_mac_header(skb);
1346                 skb->dev = vxlan->dev;
1347                 skb->pkt_type = PACKET_HOST;
1348         }
1349
1350         oiph = skb_network_header(skb);
1351         skb_reset_network_header(skb);
1352
1353         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1354                 ++vxlan->dev->stats.rx_frame_errors;
1355                 ++vxlan->dev->stats.rx_errors;
1356                 goto drop;
1357         }
1358
1359         stats = this_cpu_ptr(vxlan->dev->tstats);
1360         u64_stats_update_begin(&stats->syncp);
1361         stats->rx_packets++;
1362         stats->rx_bytes += skb->len;
1363         u64_stats_update_end(&stats->syncp);
1364
1365         gro_cells_receive(&vxlan->gro_cells, skb);
1366         return 0;
1367
1368 drop:
1369         /* Consume bad packet */
1370         kfree_skb(skb);
1371         return 0;
1372 }
1373
1374 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1375 {
1376         struct vxlan_dev *vxlan = netdev_priv(dev);
1377         struct arphdr *parp;
1378         u8 *arpptr, *sha;
1379         __be32 sip, tip;
1380         struct neighbour *n;
1381
1382         if (dev->flags & IFF_NOARP)
1383                 goto out;
1384
1385         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1386                 dev->stats.tx_dropped++;
1387                 goto out;
1388         }
1389         parp = arp_hdr(skb);
1390
1391         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1392              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1393             parp->ar_pro != htons(ETH_P_IP) ||
1394             parp->ar_op != htons(ARPOP_REQUEST) ||
1395             parp->ar_hln != dev->addr_len ||
1396             parp->ar_pln != 4)
1397                 goto out;
1398         arpptr = (u8 *)parp + sizeof(struct arphdr);
1399         sha = arpptr;
1400         arpptr += dev->addr_len;        /* sha */
1401         memcpy(&sip, arpptr, sizeof(sip));
1402         arpptr += sizeof(sip);
1403         arpptr += dev->addr_len;        /* tha */
1404         memcpy(&tip, arpptr, sizeof(tip));
1405
1406         if (ipv4_is_loopback(tip) ||
1407             ipv4_is_multicast(tip))
1408                 goto out;
1409
1410         n = neigh_lookup(&arp_tbl, &tip, dev);
1411
1412         if (n) {
1413                 struct vxlan_fdb *f;
1414                 struct sk_buff  *reply;
1415
1416                 if (!(n->nud_state & NUD_CONNECTED)) {
1417                         neigh_release(n);
1418                         goto out;
1419                 }
1420
1421                 f = vxlan_find_mac(vxlan, n->ha);
1422                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1423                         /* bridge-local neighbor */
1424                         neigh_release(n);
1425                         goto out;
1426                 }
1427
1428                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1429                                 n->ha, sha);
1430
1431                 neigh_release(n);
1432
1433                 if (reply == NULL)
1434                         goto out;
1435
1436                 skb_reset_mac_header(reply);
1437                 __skb_pull(reply, skb_network_offset(reply));
1438                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1439                 reply->pkt_type = PACKET_HOST;
1440
1441                 if (netif_rx_ni(reply) == NET_RX_DROP)
1442                         dev->stats.rx_dropped++;
1443         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1444                 union vxlan_addr ipa = {
1445                         .sin.sin_addr.s_addr = tip,
1446                         .sin.sin_family = AF_INET,
1447                 };
1448
1449                 vxlan_ip_miss(dev, &ipa);
1450         }
1451 out:
1452         consume_skb(skb);
1453         return NETDEV_TX_OK;
1454 }
1455
1456 #if IS_ENABLED(CONFIG_IPV6)
1457 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1458         struct neighbour *n, bool isrouter)
1459 {
1460         struct net_device *dev = request->dev;
1461         struct sk_buff *reply;
1462         struct nd_msg *ns, *na;
1463         struct ipv6hdr *pip6;
1464         u8 *daddr;
1465         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1466         int ns_olen;
1467         int i, len;
1468
1469         if (dev == NULL)
1470                 return NULL;
1471
1472         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1473                 sizeof(*na) + na_olen + dev->needed_tailroom;
1474         reply = alloc_skb(len, GFP_ATOMIC);
1475         if (reply == NULL)
1476                 return NULL;
1477
1478         reply->protocol = htons(ETH_P_IPV6);
1479         reply->dev = dev;
1480         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1481         skb_push(reply, sizeof(struct ethhdr));
1482         skb_reset_mac_header(reply);
1483
1484         ns = (struct nd_msg *)skb_transport_header(request);
1485
1486         daddr = eth_hdr(request)->h_source;
1487         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1488         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1489                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1490                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1491                         break;
1492                 }
1493         }
1494
1495         /* Ethernet header */
1496         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1497         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1498         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1499         reply->protocol = htons(ETH_P_IPV6);
1500
1501         skb_pull(reply, sizeof(struct ethhdr));
1502         skb_reset_network_header(reply);
1503         skb_put(reply, sizeof(struct ipv6hdr));
1504
1505         /* IPv6 header */
1506
1507         pip6 = ipv6_hdr(reply);
1508         memset(pip6, 0, sizeof(struct ipv6hdr));
1509         pip6->version = 6;
1510         pip6->priority = ipv6_hdr(request)->priority;
1511         pip6->nexthdr = IPPROTO_ICMPV6;
1512         pip6->hop_limit = 255;
1513         pip6->daddr = ipv6_hdr(request)->saddr;
1514         pip6->saddr = *(struct in6_addr *)n->primary_key;
1515
1516         skb_pull(reply, sizeof(struct ipv6hdr));
1517         skb_reset_transport_header(reply);
1518
1519         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1520
1521         /* Neighbor Advertisement */
1522         memset(na, 0, sizeof(*na)+na_olen);
1523         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1524         na->icmph.icmp6_router = isrouter;
1525         na->icmph.icmp6_override = 1;
1526         na->icmph.icmp6_solicited = 1;
1527         na->target = ns->target;
1528         ether_addr_copy(&na->opt[2], n->ha);
1529         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1530         na->opt[1] = na_olen >> 3;
1531
1532         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1533                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1534                 csum_partial(na, sizeof(*na)+na_olen, 0));
1535
1536         pip6->payload_len = htons(sizeof(*na)+na_olen);
1537
1538         skb_push(reply, sizeof(struct ipv6hdr));
1539
1540         reply->ip_summed = CHECKSUM_UNNECESSARY;
1541
1542         return reply;
1543 }
1544
1545 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1546 {
1547         struct vxlan_dev *vxlan = netdev_priv(dev);
1548         struct nd_msg *msg;
1549         const struct ipv6hdr *iphdr;
1550         const struct in6_addr *saddr, *daddr;
1551         struct neighbour *n;
1552         struct inet6_dev *in6_dev;
1553
1554         in6_dev = __in6_dev_get(dev);
1555         if (!in6_dev)
1556                 goto out;
1557
1558         iphdr = ipv6_hdr(skb);
1559         saddr = &iphdr->saddr;
1560         daddr = &iphdr->daddr;
1561
1562         msg = (struct nd_msg *)skb_transport_header(skb);
1563         if (msg->icmph.icmp6_code != 0 ||
1564             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1565                 goto out;
1566
1567         if (ipv6_addr_loopback(daddr) ||
1568             ipv6_addr_is_multicast(&msg->target))
1569                 goto out;
1570
1571         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1572
1573         if (n) {
1574                 struct vxlan_fdb *f;
1575                 struct sk_buff *reply;
1576
1577                 if (!(n->nud_state & NUD_CONNECTED)) {
1578                         neigh_release(n);
1579                         goto out;
1580                 }
1581
1582                 f = vxlan_find_mac(vxlan, n->ha);
1583                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1584                         /* bridge-local neighbor */
1585                         neigh_release(n);
1586                         goto out;
1587                 }
1588
1589                 reply = vxlan_na_create(skb, n,
1590                                         !!(f ? f->flags & NTF_ROUTER : 0));
1591
1592                 neigh_release(n);
1593
1594                 if (reply == NULL)
1595                         goto out;
1596
1597                 if (netif_rx_ni(reply) == NET_RX_DROP)
1598                         dev->stats.rx_dropped++;
1599
1600         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1601                 union vxlan_addr ipa = {
1602                         .sin6.sin6_addr = msg->target,
1603                         .sin6.sin6_family = AF_INET6,
1604                 };
1605
1606                 vxlan_ip_miss(dev, &ipa);
1607         }
1608
1609 out:
1610         consume_skb(skb);
1611         return NETDEV_TX_OK;
1612 }
1613 #endif
1614
1615 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1616 {
1617         struct vxlan_dev *vxlan = netdev_priv(dev);
1618         struct neighbour *n;
1619
1620         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1621                 return false;
1622
1623         n = NULL;
1624         switch (ntohs(eth_hdr(skb)->h_proto)) {
1625         case ETH_P_IP:
1626         {
1627                 struct iphdr *pip;
1628
1629                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1630                         return false;
1631                 pip = ip_hdr(skb);
1632                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1633                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1634                         union vxlan_addr ipa = {
1635                                 .sin.sin_addr.s_addr = pip->daddr,
1636                                 .sin.sin_family = AF_INET,
1637                         };
1638
1639                         vxlan_ip_miss(dev, &ipa);
1640                         return false;
1641                 }
1642
1643                 break;
1644         }
1645 #if IS_ENABLED(CONFIG_IPV6)
1646         case ETH_P_IPV6:
1647         {
1648                 struct ipv6hdr *pip6;
1649
1650                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1651                         return false;
1652                 pip6 = ipv6_hdr(skb);
1653                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1654                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1655                         union vxlan_addr ipa = {
1656                                 .sin6.sin6_addr = pip6->daddr,
1657                                 .sin6.sin6_family = AF_INET6,
1658                         };
1659
1660                         vxlan_ip_miss(dev, &ipa);
1661                         return false;
1662                 }
1663
1664                 break;
1665         }
1666 #endif
1667         default:
1668                 return false;
1669         }
1670
1671         if (n) {
1672                 bool diff;
1673
1674                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1675                 if (diff) {
1676                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1677                                 dev->addr_len);
1678                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1679                 }
1680                 neigh_release(n);
1681                 return diff;
1682         }
1683
1684         return false;
1685 }
1686
1687 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1688                                 struct vxlan_metadata *md)
1689 {
1690         struct vxlanhdr_gbp *gbp;
1691
1692         if (!md->gbp)
1693                 return;
1694
1695         gbp = (struct vxlanhdr_gbp *)vxh;
1696         vxh->vx_flags |= VXLAN_HF_GBP;
1697
1698         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1699                 gbp->dont_learn = 1;
1700
1701         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1702                 gbp->policy_applied = 1;
1703
1704         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1705 }
1706
1707 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1708                                __be16 protocol)
1709 {
1710         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1711
1712         gpe->np_applied = 1;
1713
1714         switch (protocol) {
1715         case htons(ETH_P_IP):
1716                 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1717                 return 0;
1718         case htons(ETH_P_IPV6):
1719                 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1720                 return 0;
1721         case htons(ETH_P_TEB):
1722                 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1723                 return 0;
1724         }
1725         return -EPFNOSUPPORT;
1726 }
1727
1728 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1729                            int iphdr_len, __be32 vni,
1730                            struct vxlan_metadata *md, u32 vxflags,
1731                            bool udp_sum)
1732 {
1733         struct vxlanhdr *vxh;
1734         int min_headroom;
1735         int err;
1736         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1737         __be16 inner_protocol = htons(ETH_P_TEB);
1738
1739         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1740             skb->ip_summed == CHECKSUM_PARTIAL) {
1741                 int csum_start = skb_checksum_start_offset(skb);
1742
1743                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1744                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1745                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1746                      skb->csum_offset == offsetof(struct tcphdr, check)))
1747                         type |= SKB_GSO_TUNNEL_REMCSUM;
1748         }
1749
1750         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1751                         + VXLAN_HLEN + iphdr_len
1752                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1753
1754         /* Need space for new headers (invalidates iph ptr) */
1755         err = skb_cow_head(skb, min_headroom);
1756         if (unlikely(err))
1757                 goto out_free;
1758
1759         skb = vlan_hwaccel_push_inside(skb);
1760         if (WARN_ON(!skb))
1761                 return -ENOMEM;
1762
1763         err = iptunnel_handle_offloads(skb, type);
1764         if (err)
1765                 goto out_free;
1766
1767         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1768         vxh->vx_flags = VXLAN_HF_VNI;
1769         vxh->vx_vni = vxlan_vni_field(vni);
1770
1771         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1772                 unsigned int start;
1773
1774                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1775                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1776                 vxh->vx_flags |= VXLAN_HF_RCO;
1777
1778                 if (!skb_is_gso(skb)) {
1779                         skb->ip_summed = CHECKSUM_NONE;
1780                         skb->encapsulation = 0;
1781                 }
1782         }
1783
1784         if (vxflags & VXLAN_F_GBP)
1785                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1786         if (vxflags & VXLAN_F_GPE) {
1787                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1788                 if (err < 0)
1789                         goto out_free;
1790                 inner_protocol = skb->protocol;
1791         }
1792
1793         skb_set_inner_protocol(skb, inner_protocol);
1794         return 0;
1795
1796 out_free:
1797         kfree_skb(skb);
1798         return err;
1799 }
1800
1801 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1802                                       struct sk_buff *skb, int oif, u8 tos,
1803                                       __be32 daddr, __be32 *saddr,
1804                                       struct dst_cache *dst_cache,
1805                                       const struct ip_tunnel_info *info)
1806 {
1807         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1808         struct rtable *rt = NULL;
1809         struct flowi4 fl4;
1810
1811         if (tos && !info)
1812                 use_cache = false;
1813         if (use_cache) {
1814                 rt = dst_cache_get_ip4(dst_cache, saddr);
1815                 if (rt)
1816                         return rt;
1817         }
1818
1819         memset(&fl4, 0, sizeof(fl4));
1820         fl4.flowi4_oif = oif;
1821         fl4.flowi4_tos = RT_TOS(tos);
1822         fl4.flowi4_mark = skb->mark;
1823         fl4.flowi4_proto = IPPROTO_UDP;
1824         fl4.daddr = daddr;
1825         fl4.saddr = *saddr;
1826
1827         rt = ip_route_output_key(vxlan->net, &fl4);
1828         if (!IS_ERR(rt)) {
1829                 *saddr = fl4.saddr;
1830                 if (use_cache)
1831                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1832         }
1833         return rt;
1834 }
1835
1836 #if IS_ENABLED(CONFIG_IPV6)
1837 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1838                                           struct sk_buff *skb, int oif, u8 tos,
1839                                           __be32 label,
1840                                           const struct in6_addr *daddr,
1841                                           struct in6_addr *saddr,
1842                                           struct dst_cache *dst_cache,
1843                                           const struct ip_tunnel_info *info)
1844 {
1845         struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
1846         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1847         struct dst_entry *ndst;
1848         struct flowi6 fl6;
1849         int err;
1850
1851         if (!sock6)
1852                 return ERR_PTR(-EIO);
1853
1854         if (tos && !info)
1855                 use_cache = false;
1856         if (use_cache) {
1857                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1858                 if (ndst)
1859                         return ndst;
1860         }
1861
1862         memset(&fl6, 0, sizeof(fl6));
1863         fl6.flowi6_oif = oif;
1864         fl6.daddr = *daddr;
1865         fl6.saddr = *saddr;
1866         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1867         fl6.flowi6_mark = skb->mark;
1868         fl6.flowi6_proto = IPPROTO_UDP;
1869
1870         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1871                                          sock6->sock->sk,
1872                                          &ndst, &fl6);
1873         if (err < 0)
1874                 return ERR_PTR(err);
1875
1876         *saddr = fl6.saddr;
1877         if (use_cache)
1878                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1879         return ndst;
1880 }
1881 #endif
1882
1883 /* Bypass encapsulation if the destination is local */
1884 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1885                                struct vxlan_dev *dst_vxlan)
1886 {
1887         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1888         union vxlan_addr loopback;
1889         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1890         struct net_device *dev = skb->dev;
1891         int len = skb->len;
1892
1893         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1894         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1895         skb->pkt_type = PACKET_HOST;
1896         skb->encapsulation = 0;
1897         skb->dev = dst_vxlan->dev;
1898         __skb_pull(skb, skb_network_offset(skb));
1899
1900         if (remote_ip->sa.sa_family == AF_INET) {
1901                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1902                 loopback.sa.sa_family =  AF_INET;
1903 #if IS_ENABLED(CONFIG_IPV6)
1904         } else {
1905                 loopback.sin6.sin6_addr = in6addr_loopback;
1906                 loopback.sa.sa_family =  AF_INET6;
1907 #endif
1908         }
1909
1910         if (dst_vxlan->flags & VXLAN_F_LEARN)
1911                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1912
1913         u64_stats_update_begin(&tx_stats->syncp);
1914         tx_stats->tx_packets++;
1915         tx_stats->tx_bytes += len;
1916         u64_stats_update_end(&tx_stats->syncp);
1917
1918         if (netif_rx(skb) == NET_RX_SUCCESS) {
1919                 u64_stats_update_begin(&rx_stats->syncp);
1920                 rx_stats->rx_packets++;
1921                 rx_stats->rx_bytes += len;
1922                 u64_stats_update_end(&rx_stats->syncp);
1923         } else {
1924                 dev->stats.rx_dropped++;
1925         }
1926 }
1927
1928 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1929                            struct vxlan_rdst *rdst, bool did_rsc)
1930 {
1931         struct dst_cache *dst_cache;
1932         struct ip_tunnel_info *info;
1933         struct vxlan_dev *vxlan = netdev_priv(dev);
1934         struct sock *sk;
1935         struct rtable *rt = NULL;
1936         const struct iphdr *old_iph;
1937         union vxlan_addr *dst;
1938         union vxlan_addr remote_ip, local_ip;
1939         union vxlan_addr *src;
1940         struct vxlan_metadata _md;
1941         struct vxlan_metadata *md = &_md;
1942         __be16 src_port = 0, dst_port;
1943         __be32 vni, label;
1944         __be16 df = 0;
1945         __u8 tos, ttl;
1946         int err;
1947         u32 flags = vxlan->flags;
1948         bool udp_sum = false;
1949         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1950
1951         info = skb_tunnel_info(skb);
1952
1953         if (rdst) {
1954                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1955                 vni = rdst->remote_vni;
1956                 dst = &rdst->remote_ip;
1957                 src = &vxlan->cfg.saddr;
1958                 dst_cache = &rdst->dst_cache;
1959         } else {
1960                 if (!info) {
1961                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1962                                   dev->name);
1963                         goto drop;
1964                 }
1965                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1966                 vni = tunnel_id_to_key32(info->key.tun_id);
1967                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1968                 if (remote_ip.sa.sa_family == AF_INET) {
1969                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1970                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
1971                 } else {
1972                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1973                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
1974                 }
1975                 dst = &remote_ip;
1976                 src = &local_ip;
1977                 dst_cache = &info->dst_cache;
1978         }
1979
1980         if (vxlan_addr_any(dst)) {
1981                 if (did_rsc) {
1982                         /* short-circuited back to local bridge */
1983                         vxlan_encap_bypass(skb, vxlan, vxlan);
1984                         return;
1985                 }
1986                 goto drop;
1987         }
1988
1989         old_iph = ip_hdr(skb);
1990
1991         ttl = vxlan->cfg.ttl;
1992         if (!ttl && vxlan_addr_multicast(dst))
1993                 ttl = 1;
1994
1995         tos = vxlan->cfg.tos;
1996         if (tos == 1)
1997                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1998
1999         label = vxlan->cfg.label;
2000         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2001                                      vxlan->cfg.port_max, true);
2002
2003         if (info) {
2004                 ttl = info->key.ttl;
2005                 tos = info->key.tos;
2006                 label = info->key.label;
2007                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2008
2009                 if (info->options_len)
2010                         md = ip_tunnel_info_opts(info);
2011         } else {
2012                 md->gbp = skb->mark;
2013         }
2014
2015         if (dst->sa.sa_family == AF_INET) {
2016                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2017
2018                 if (!sock4)
2019                         goto drop;
2020                 sk = sock4->sock->sk;
2021
2022                 rt = vxlan_get_route(vxlan, skb,
2023                                      rdst ? rdst->remote_ifindex : 0, tos,
2024                                      dst->sin.sin_addr.s_addr,
2025                                      &src->sin.sin_addr.s_addr,
2026                                      dst_cache, info);
2027                 if (IS_ERR(rt)) {
2028                         netdev_dbg(dev, "no route to %pI4\n",
2029                                    &dst->sin.sin_addr.s_addr);
2030                         dev->stats.tx_carrier_errors++;
2031                         goto tx_error;
2032                 }
2033
2034                 if (rt->dst.dev == dev) {
2035                         netdev_dbg(dev, "circular route to %pI4\n",
2036                                    &dst->sin.sin_addr.s_addr);
2037                         dev->stats.collisions++;
2038                         goto rt_tx_error;
2039                 }
2040
2041                 /* Bypass encapsulation if the destination is local */
2042                 if (!info && rt->rt_flags & RTCF_LOCAL &&
2043                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2044                         struct vxlan_dev *dst_vxlan;
2045
2046                         ip_rt_put(rt);
2047                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2048                                                    dst->sa.sa_family, dst_port,
2049                                                    vxlan->flags);
2050                         if (!dst_vxlan)
2051                                 goto tx_error;
2052                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2053                         return;
2054                 }
2055
2056                 if (!info)
2057                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2058                 else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2059                         df = htons(IP_DF);
2060
2061                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2062                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2063                 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2064                                       vni, md, flags, udp_sum);
2065                 if (err < 0)
2066                         goto xmit_tx_error;
2067
2068                 udp_tunnel_xmit_skb(rt, sk, skb, src->sin.sin_addr.s_addr,
2069                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2070                                     src_port, dst_port, xnet, !udp_sum);
2071 #if IS_ENABLED(CONFIG_IPV6)
2072         } else {
2073                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2074                 struct dst_entry *ndst;
2075                 u32 rt6i_flags;
2076
2077                 if (!sock6)
2078                         goto drop;
2079                 sk = sock6->sock->sk;
2080
2081                 ndst = vxlan6_get_route(vxlan, skb,
2082                                         rdst ? rdst->remote_ifindex : 0, tos,
2083                                         label, &dst->sin6.sin6_addr,
2084                                         &src->sin6.sin6_addr,
2085                                         dst_cache, info);
2086                 if (IS_ERR(ndst)) {
2087                         netdev_dbg(dev, "no route to %pI6\n",
2088                                    &dst->sin6.sin6_addr);
2089                         dev->stats.tx_carrier_errors++;
2090                         goto tx_error;
2091                 }
2092
2093                 if (ndst->dev == dev) {
2094                         netdev_dbg(dev, "circular route to %pI6\n",
2095                                    &dst->sin6.sin6_addr);
2096                         dst_release(ndst);
2097                         dev->stats.collisions++;
2098                         goto tx_error;
2099                 }
2100
2101                 /* Bypass encapsulation if the destination is local */
2102                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2103                 if (!info && rt6i_flags & RTF_LOCAL &&
2104                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2105                         struct vxlan_dev *dst_vxlan;
2106
2107                         dst_release(ndst);
2108                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2109                                                    dst->sa.sa_family, dst_port,
2110                                                    vxlan->flags);
2111                         if (!dst_vxlan)
2112                                 goto tx_error;
2113                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2114                         return;
2115                 }
2116
2117                 if (!info)
2118                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2119
2120                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2121                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2122                 skb_scrub_packet(skb, xnet);
2123                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2124                                       vni, md, flags, udp_sum);
2125                 if (err < 0) {
2126                         dst_release(ndst);
2127                         dev->stats.tx_errors++;
2128                         return;
2129                 }
2130                 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2131                                      &src->sin6.sin6_addr,
2132                                      &dst->sin6.sin6_addr, tos, ttl,
2133                                      label, src_port, dst_port, !udp_sum);
2134 #endif
2135         }
2136
2137         return;
2138
2139 drop:
2140         dev->stats.tx_dropped++;
2141         goto tx_free;
2142
2143 xmit_tx_error:
2144         /* skb is already freed. */
2145         skb = NULL;
2146 rt_tx_error:
2147         ip_rt_put(rt);
2148 tx_error:
2149         dev->stats.tx_errors++;
2150 tx_free:
2151         dev_kfree_skb(skb);
2152 }
2153
2154 /* Transmit local packets over Vxlan
2155  *
2156  * Outer IP header inherits ECN and DF from inner header.
2157  * Outer UDP destination is the VXLAN assigned port.
2158  *           source port is based on hash of flow
2159  */
2160 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2161 {
2162         struct vxlan_dev *vxlan = netdev_priv(dev);
2163         const struct ip_tunnel_info *info;
2164         struct ethhdr *eth;
2165         bool did_rsc = false;
2166         struct vxlan_rdst *rdst, *fdst = NULL;
2167         struct vxlan_fdb *f;
2168
2169         info = skb_tunnel_info(skb);
2170
2171         skb_reset_mac_header(skb);
2172
2173         if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2174                 if (info && info->mode & IP_TUNNEL_INFO_TX)
2175                         vxlan_xmit_one(skb, dev, NULL, false);
2176                 else
2177                         kfree_skb(skb);
2178                 return NETDEV_TX_OK;
2179         }
2180
2181         if (vxlan->flags & VXLAN_F_PROXY) {
2182                 eth = eth_hdr(skb);
2183                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2184                         return arp_reduce(dev, skb);
2185 #if IS_ENABLED(CONFIG_IPV6)
2186                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2187                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2188                                        + sizeof(struct nd_msg)) &&
2189                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2190                                 struct nd_msg *msg;
2191
2192                                 msg = (struct nd_msg *)skb_transport_header(skb);
2193                                 if (msg->icmph.icmp6_code == 0 &&
2194                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2195                                         return neigh_reduce(dev, skb);
2196                 }
2197 #endif
2198         }
2199
2200         eth = eth_hdr(skb);
2201         f = vxlan_find_mac(vxlan, eth->h_dest);
2202         did_rsc = false;
2203
2204         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2205             (ntohs(eth->h_proto) == ETH_P_IP ||
2206              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2207                 did_rsc = route_shortcircuit(dev, skb);
2208                 if (did_rsc)
2209                         f = vxlan_find_mac(vxlan, eth->h_dest);
2210         }
2211
2212         if (f == NULL) {
2213                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2214                 if (f == NULL) {
2215                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2216                             !is_multicast_ether_addr(eth->h_dest))
2217                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2218
2219                         dev->stats.tx_dropped++;
2220                         kfree_skb(skb);
2221                         return NETDEV_TX_OK;
2222                 }
2223         }
2224
2225         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2226                 struct sk_buff *skb1;
2227
2228                 if (!fdst) {
2229                         fdst = rdst;
2230                         continue;
2231                 }
2232                 skb1 = skb_clone(skb, GFP_ATOMIC);
2233                 if (skb1)
2234                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2235         }
2236
2237         if (fdst)
2238                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2239         else
2240                 kfree_skb(skb);
2241         return NETDEV_TX_OK;
2242 }
2243
2244 /* Walk the forwarding table and purge stale entries */
2245 static void vxlan_cleanup(unsigned long arg)
2246 {
2247         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2248         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2249         unsigned int h;
2250
2251         if (!netif_running(vxlan->dev))
2252                 return;
2253
2254         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2255                 struct hlist_node *p, *n;
2256
2257                 spin_lock_bh(&vxlan->hash_lock);
2258                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2259                         struct vxlan_fdb *f
2260                                 = container_of(p, struct vxlan_fdb, hlist);
2261                         unsigned long timeout;
2262
2263                         if (f->state & NUD_PERMANENT)
2264                                 continue;
2265
2266                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2267                         if (time_before_eq(timeout, jiffies)) {
2268                                 netdev_dbg(vxlan->dev,
2269                                            "garbage collect %pM\n",
2270                                            f->eth_addr);
2271                                 f->state = NUD_STALE;
2272                                 vxlan_fdb_destroy(vxlan, f);
2273                         } else if (time_before(timeout, next_timer))
2274                                 next_timer = timeout;
2275                 }
2276                 spin_unlock_bh(&vxlan->hash_lock);
2277         }
2278
2279         mod_timer(&vxlan->age_timer, next_timer);
2280 }
2281
2282 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2283 {
2284         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2285         __be32 vni = vxlan->default_dst.remote_vni;
2286
2287         spin_lock(&vn->sock_lock);
2288         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2289         spin_unlock(&vn->sock_lock);
2290 }
2291
2292 /* Setup stats when device is created */
2293 static int vxlan_init(struct net_device *dev)
2294 {
2295         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2296         if (!dev->tstats)
2297                 return -ENOMEM;
2298
2299         return 0;
2300 }
2301
2302 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2303 {
2304         struct vxlan_fdb *f;
2305
2306         spin_lock_bh(&vxlan->hash_lock);
2307         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2308         if (f)
2309                 vxlan_fdb_destroy(vxlan, f);
2310         spin_unlock_bh(&vxlan->hash_lock);
2311 }
2312
2313 static void vxlan_uninit(struct net_device *dev)
2314 {
2315         struct vxlan_dev *vxlan = netdev_priv(dev);
2316
2317         vxlan_fdb_delete_default(vxlan);
2318
2319         free_percpu(dev->tstats);
2320 }
2321
2322 /* Start ageing timer and join group when device is brought up */
2323 static int vxlan_open(struct net_device *dev)
2324 {
2325         struct vxlan_dev *vxlan = netdev_priv(dev);
2326         int ret;
2327
2328         ret = vxlan_sock_add(vxlan);
2329         if (ret < 0)
2330                 return ret;
2331
2332         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2333                 ret = vxlan_igmp_join(vxlan);
2334                 if (ret == -EADDRINUSE)
2335                         ret = 0;
2336                 if (ret) {
2337                         vxlan_sock_release(vxlan);
2338                         return ret;
2339                 }
2340         }
2341
2342         if (vxlan->cfg.age_interval)
2343                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2344
2345         return ret;
2346 }
2347
2348 /* Purge the forwarding table */
2349 static void vxlan_flush(struct vxlan_dev *vxlan)
2350 {
2351         unsigned int h;
2352
2353         spin_lock_bh(&vxlan->hash_lock);
2354         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2355                 struct hlist_node *p, *n;
2356                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2357                         struct vxlan_fdb *f
2358                                 = container_of(p, struct vxlan_fdb, hlist);
2359                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2360                         if (!is_zero_ether_addr(f->eth_addr))
2361                                 vxlan_fdb_destroy(vxlan, f);
2362                 }
2363         }
2364         spin_unlock_bh(&vxlan->hash_lock);
2365 }
2366
2367 /* Cleanup timer and forwarding table on shutdown */
2368 static int vxlan_stop(struct net_device *dev)
2369 {
2370         struct vxlan_dev *vxlan = netdev_priv(dev);
2371         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2372         int ret = 0;
2373
2374         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2375             !vxlan_group_used(vn, vxlan))
2376                 ret = vxlan_igmp_leave(vxlan);
2377
2378         del_timer_sync(&vxlan->age_timer);
2379
2380         vxlan_flush(vxlan);
2381         vxlan_sock_release(vxlan);
2382
2383         return ret;
2384 }
2385
2386 /* Stub, nothing needs to be done. */
2387 static void vxlan_set_multicast_list(struct net_device *dev)
2388 {
2389 }
2390
2391 static int __vxlan_change_mtu(struct net_device *dev,
2392                               struct net_device *lowerdev,
2393                               struct vxlan_rdst *dst, int new_mtu, bool strict)
2394 {
2395         int max_mtu = IP_MAX_MTU;
2396
2397         if (lowerdev)
2398                 max_mtu = lowerdev->mtu;
2399
2400         if (dst->remote_ip.sa.sa_family == AF_INET6)
2401                 max_mtu -= VXLAN6_HEADROOM;
2402         else
2403                 max_mtu -= VXLAN_HEADROOM;
2404
2405         if (new_mtu < 68)
2406                 return -EINVAL;
2407
2408         if (new_mtu > max_mtu) {
2409                 if (strict)
2410                         return -EINVAL;
2411
2412                 new_mtu = max_mtu;
2413         }
2414
2415         dev->mtu = new_mtu;
2416         return 0;
2417 }
2418
2419 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2420 {
2421         struct vxlan_dev *vxlan = netdev_priv(dev);
2422         struct vxlan_rdst *dst = &vxlan->default_dst;
2423         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2424                                                          dst->remote_ifindex);
2425         return __vxlan_change_mtu(dev, lowerdev, dst, new_mtu, true);
2426 }
2427
2428 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2429 {
2430         struct vxlan_dev *vxlan = netdev_priv(dev);
2431         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2432         __be16 sport, dport;
2433
2434         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2435                                   vxlan->cfg.port_max, true);
2436         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2437
2438         if (ip_tunnel_info_af(info) == AF_INET) {
2439                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2440                 struct rtable *rt;
2441
2442                 if (!sock4)
2443                         return -EINVAL;
2444                 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2445                                      info->key.u.ipv4.dst,
2446                                      &info->key.u.ipv4.src, NULL, info);
2447                 if (IS_ERR(rt))
2448                         return PTR_ERR(rt);
2449                 ip_rt_put(rt);
2450         } else {
2451 #if IS_ENABLED(CONFIG_IPV6)
2452                 struct dst_entry *ndst;
2453
2454                 ndst = vxlan6_get_route(vxlan, skb, 0, info->key.tos,
2455                                         info->key.label, &info->key.u.ipv6.dst,
2456                                         &info->key.u.ipv6.src, NULL, info);
2457                 if (IS_ERR(ndst))
2458                         return PTR_ERR(ndst);
2459                 dst_release(ndst);
2460 #else /* !CONFIG_IPV6 */
2461                 return -EPFNOSUPPORT;
2462 #endif
2463         }
2464         info->key.tp_src = sport;
2465         info->key.tp_dst = dport;
2466         return 0;
2467 }
2468
2469 static const struct net_device_ops vxlan_netdev_ether_ops = {
2470         .ndo_init               = vxlan_init,
2471         .ndo_uninit             = vxlan_uninit,
2472         .ndo_open               = vxlan_open,
2473         .ndo_stop               = vxlan_stop,
2474         .ndo_start_xmit         = vxlan_xmit,
2475         .ndo_get_stats64        = ip_tunnel_get_stats64,
2476         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2477         .ndo_change_mtu         = vxlan_change_mtu,
2478         .ndo_validate_addr      = eth_validate_addr,
2479         .ndo_set_mac_address    = eth_mac_addr,
2480         .ndo_fdb_add            = vxlan_fdb_add,
2481         .ndo_fdb_del            = vxlan_fdb_delete,
2482         .ndo_fdb_dump           = vxlan_fdb_dump,
2483         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2484 };
2485
2486 static const struct net_device_ops vxlan_netdev_raw_ops = {
2487         .ndo_init               = vxlan_init,
2488         .ndo_uninit             = vxlan_uninit,
2489         .ndo_open               = vxlan_open,
2490         .ndo_stop               = vxlan_stop,
2491         .ndo_start_xmit         = vxlan_xmit,
2492         .ndo_get_stats64        = ip_tunnel_get_stats64,
2493         .ndo_change_mtu         = vxlan_change_mtu,
2494         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2495 };
2496
2497 /* Info for udev, that this is a virtual tunnel endpoint */
2498 static struct device_type vxlan_type = {
2499         .name = "vxlan",
2500 };
2501
2502 /* Calls the ndo_udp_tunnel_add of the caller in order to
2503  * supply the listening VXLAN udp ports. Callers are expected
2504  * to implement the ndo_udp_tunnel_add.
2505  */
2506 static void vxlan_push_rx_ports(struct net_device *dev)
2507 {
2508         struct vxlan_sock *vs;
2509         struct net *net = dev_net(dev);
2510         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2511         unsigned int i;
2512
2513         spin_lock(&vn->sock_lock);
2514         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2515                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2516                         udp_tunnel_push_rx_port(dev, vs->sock,
2517                                                 (vs->flags & VXLAN_F_GPE) ?
2518                                                 UDP_TUNNEL_TYPE_VXLAN_GPE :
2519                                                 UDP_TUNNEL_TYPE_VXLAN);
2520         }
2521         spin_unlock(&vn->sock_lock);
2522 }
2523
2524 /* Initialize the device structure. */
2525 static void vxlan_setup(struct net_device *dev)
2526 {
2527         struct vxlan_dev *vxlan = netdev_priv(dev);
2528         unsigned int h;
2529
2530         eth_hw_addr_random(dev);
2531         ether_setup(dev);
2532
2533         dev->destructor = free_netdev;
2534         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2535
2536         dev->features   |= NETIF_F_LLTX;
2537         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2538         dev->features   |= NETIF_F_RXCSUM;
2539         dev->features   |= NETIF_F_GSO_SOFTWARE;
2540
2541         dev->vlan_features = dev->features;
2542         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2543         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2544         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2545         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2546         netif_keep_dst(dev);
2547         dev->priv_flags |= IFF_NO_QUEUE;
2548
2549         INIT_LIST_HEAD(&vxlan->next);
2550         spin_lock_init(&vxlan->hash_lock);
2551
2552         init_timer_deferrable(&vxlan->age_timer);
2553         vxlan->age_timer.function = vxlan_cleanup;
2554         vxlan->age_timer.data = (unsigned long) vxlan;
2555
2556         vxlan->cfg.dst_port = htons(vxlan_port);
2557
2558         vxlan->dev = dev;
2559
2560         gro_cells_init(&vxlan->gro_cells, dev);
2561
2562         for (h = 0; h < FDB_HASH_SIZE; ++h)
2563                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2564 }
2565
2566 static void vxlan_ether_setup(struct net_device *dev)
2567 {
2568         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2569         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2570         dev->netdev_ops = &vxlan_netdev_ether_ops;
2571 }
2572
2573 static void vxlan_raw_setup(struct net_device *dev)
2574 {
2575         dev->header_ops = NULL;
2576         dev->type = ARPHRD_NONE;
2577         dev->hard_header_len = 0;
2578         dev->addr_len = 0;
2579         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2580         dev->netdev_ops = &vxlan_netdev_raw_ops;
2581 }
2582
2583 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2584         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2585         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2586         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2587         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2588         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2589         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2590         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2591         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2592         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2593         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2594         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2595         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2596         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2597         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2598         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2599         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2600         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2601         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2602         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2603         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2604         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2605         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2606         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2607         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2608         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2609         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2610         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2611 };
2612
2613 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2614 {
2615         if (tb[IFLA_ADDRESS]) {
2616                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2617                         pr_debug("invalid link address (not ethernet)\n");
2618                         return -EINVAL;
2619                 }
2620
2621                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2622                         pr_debug("invalid all zero ethernet address\n");
2623                         return -EADDRNOTAVAIL;
2624                 }
2625         }
2626
2627         if (!data)
2628                 return -EINVAL;
2629
2630         if (data[IFLA_VXLAN_ID]) {
2631                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2632                 if (id >= VXLAN_VID_MASK)
2633                         return -ERANGE;
2634         }
2635
2636         if (data[IFLA_VXLAN_PORT_RANGE]) {
2637                 const struct ifla_vxlan_port_range *p
2638                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2639
2640                 if (ntohs(p->high) < ntohs(p->low)) {
2641                         pr_debug("port range %u .. %u not valid\n",
2642                                  ntohs(p->low), ntohs(p->high));
2643                         return -EINVAL;
2644                 }
2645         }
2646
2647         return 0;
2648 }
2649
2650 static void vxlan_get_drvinfo(struct net_device *netdev,
2651                               struct ethtool_drvinfo *drvinfo)
2652 {
2653         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2654         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2655 }
2656
2657 static const struct ethtool_ops vxlan_ethtool_ops = {
2658         .get_drvinfo    = vxlan_get_drvinfo,
2659         .get_link       = ethtool_op_get_link,
2660 };
2661
2662 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2663                                         __be16 port, u32 flags)
2664 {
2665         struct socket *sock;
2666         struct udp_port_cfg udp_conf;
2667         int err;
2668
2669         memset(&udp_conf, 0, sizeof(udp_conf));
2670
2671         if (ipv6) {
2672                 udp_conf.family = AF_INET6;
2673                 udp_conf.use_udp6_rx_checksums =
2674                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2675                 udp_conf.ipv6_v6only = 1;
2676         } else {
2677                 udp_conf.family = AF_INET;
2678         }
2679
2680         udp_conf.local_udp_port = port;
2681
2682         /* Open UDP socket */
2683         err = udp_sock_create(net, &udp_conf, &sock);
2684         if (err < 0)
2685                 return ERR_PTR(err);
2686
2687         return sock;
2688 }
2689
2690 /* Create new listen socket if needed */
2691 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2692                                               __be16 port, u32 flags)
2693 {
2694         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2695         struct vxlan_sock *vs;
2696         struct socket *sock;
2697         unsigned int h;
2698         struct udp_tunnel_sock_cfg tunnel_cfg;
2699
2700         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2701         if (!vs)
2702                 return ERR_PTR(-ENOMEM);
2703
2704         for (h = 0; h < VNI_HASH_SIZE; ++h)
2705                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2706
2707         sock = vxlan_create_sock(net, ipv6, port, flags);
2708         if (IS_ERR(sock)) {
2709                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2710                         PTR_ERR(sock));
2711                 kfree(vs);
2712                 return ERR_CAST(sock);
2713         }
2714
2715         vs->sock = sock;
2716         atomic_set(&vs->refcnt, 1);
2717         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2718
2719         spin_lock(&vn->sock_lock);
2720         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2721         udp_tunnel_notify_add_rx_port(sock,
2722                                       (vs->flags & VXLAN_F_GPE) ?
2723                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2724                                       UDP_TUNNEL_TYPE_VXLAN);
2725         spin_unlock(&vn->sock_lock);
2726
2727         /* Mark socket as an encapsulation socket. */
2728         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2729         tunnel_cfg.sk_user_data = vs;
2730         tunnel_cfg.encap_type = 1;
2731         tunnel_cfg.encap_rcv = vxlan_rcv;
2732         tunnel_cfg.encap_destroy = NULL;
2733         tunnel_cfg.gro_receive = vxlan_gro_receive;
2734         tunnel_cfg.gro_complete = vxlan_gro_complete;
2735
2736         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2737
2738         return vs;
2739 }
2740
2741 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2742 {
2743         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2744         struct vxlan_sock *vs = NULL;
2745
2746         if (!vxlan->cfg.no_share) {
2747                 spin_lock(&vn->sock_lock);
2748                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2749                                      vxlan->cfg.dst_port, vxlan->flags);
2750                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2751                         spin_unlock(&vn->sock_lock);
2752                         return -EBUSY;
2753                 }
2754                 spin_unlock(&vn->sock_lock);
2755         }
2756         if (!vs)
2757                 vs = vxlan_socket_create(vxlan->net, ipv6,
2758                                          vxlan->cfg.dst_port, vxlan->flags);
2759         if (IS_ERR(vs))
2760                 return PTR_ERR(vs);
2761 #if IS_ENABLED(CONFIG_IPV6)
2762         if (ipv6)
2763                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2764         else
2765 #endif
2766                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2767         vxlan_vs_add_dev(vs, vxlan);
2768         return 0;
2769 }
2770
2771 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2772 {
2773         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2774         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2775         int ret = 0;
2776
2777         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2778 #if IS_ENABLED(CONFIG_IPV6)
2779         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2780         if (ipv6 || metadata)
2781                 ret = __vxlan_sock_add(vxlan, true);
2782 #endif
2783         if (!ret && (!ipv6 || metadata))
2784                 ret = __vxlan_sock_add(vxlan, false);
2785         if (ret < 0)
2786                 vxlan_sock_release(vxlan);
2787         return ret;
2788 }
2789
2790 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2791                                struct vxlan_config *conf)
2792 {
2793         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2794         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2795         struct vxlan_rdst *dst = &vxlan->default_dst;
2796         unsigned short needed_headroom = ETH_HLEN;
2797         int err;
2798         bool use_ipv6 = false;
2799         __be16 default_port = vxlan->cfg.dst_port;
2800         struct net_device *lowerdev = NULL;
2801
2802         if (conf->flags & VXLAN_F_GPE) {
2803                 /* For now, allow GPE only together with COLLECT_METADATA.
2804                  * This can be relaxed later; in such case, the other side
2805                  * of the PtP link will have to be provided.
2806                  */
2807                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2808                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2809                         pr_info("unsupported combination of extensions\n");
2810                         return -EINVAL;
2811                 }
2812
2813                 vxlan_raw_setup(dev);
2814         } else {
2815                 vxlan_ether_setup(dev);
2816         }
2817
2818         vxlan->net = src_net;
2819
2820         dst->remote_vni = conf->vni;
2821
2822         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2823
2824         /* Unless IPv6 is explicitly requested, assume IPv4 */
2825         if (!dst->remote_ip.sa.sa_family)
2826                 dst->remote_ip.sa.sa_family = AF_INET;
2827
2828         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2829             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2830                 if (!IS_ENABLED(CONFIG_IPV6))
2831                         return -EPFNOSUPPORT;
2832                 use_ipv6 = true;
2833                 vxlan->flags |= VXLAN_F_IPV6;
2834         }
2835
2836         if (conf->label && !use_ipv6) {
2837                 pr_info("label only supported in use with IPv6\n");
2838                 return -EINVAL;
2839         }
2840
2841         if (conf->remote_ifindex) {
2842                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2843                 dst->remote_ifindex = conf->remote_ifindex;
2844
2845                 if (!lowerdev) {
2846                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2847                         return -ENODEV;
2848                 }
2849
2850 #if IS_ENABLED(CONFIG_IPV6)
2851                 if (use_ipv6) {
2852                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2853                         if (idev && idev->cnf.disable_ipv6) {
2854                                 pr_info("IPv6 is disabled via sysctl\n");
2855                                 return -EPERM;
2856                         }
2857                 }
2858 #endif
2859
2860                 if (!conf->mtu)
2861                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2862
2863                 needed_headroom = lowerdev->hard_header_len;
2864         } else if (vxlan_addr_multicast(&dst->remote_ip)) {
2865                 pr_info("multicast destination requires interface to be specified\n");
2866                 return -EINVAL;
2867         }
2868
2869         if (conf->mtu) {
2870                 err = __vxlan_change_mtu(dev, lowerdev, dst, conf->mtu, false);
2871                 if (err)
2872                         return err;
2873         }
2874
2875         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2876                 needed_headroom += VXLAN6_HEADROOM;
2877         else
2878                 needed_headroom += VXLAN_HEADROOM;
2879         dev->needed_headroom = needed_headroom;
2880
2881         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2882         if (!vxlan->cfg.dst_port) {
2883                 if (conf->flags & VXLAN_F_GPE)
2884                         vxlan->cfg.dst_port = 4790; /* IANA assigned VXLAN-GPE port */
2885                 else
2886                         vxlan->cfg.dst_port = default_port;
2887         }
2888         vxlan->flags |= conf->flags;
2889
2890         if (!vxlan->cfg.age_interval)
2891                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2892
2893         list_for_each_entry(tmp, &vn->vxlan_list, next) {
2894                 if (tmp->cfg.vni == conf->vni &&
2895                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2896                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2897                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2898                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2899                     (vxlan->flags & VXLAN_F_RCV_FLAGS)) {
2900                         pr_info("duplicate VNI %u\n", be32_to_cpu(conf->vni));
2901                         return -EEXIST;
2902                 }
2903         }
2904
2905         dev->ethtool_ops = &vxlan_ethtool_ops;
2906
2907         /* create an fdb entry for a valid default destination */
2908         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2909                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2910                                        &vxlan->default_dst.remote_ip,
2911                                        NUD_REACHABLE|NUD_PERMANENT,
2912                                        NLM_F_EXCL|NLM_F_CREATE,
2913                                        vxlan->cfg.dst_port,
2914                                        vxlan->default_dst.remote_vni,
2915                                        vxlan->default_dst.remote_ifindex,
2916                                        NTF_SELF);
2917                 if (err)
2918                         return err;
2919         }
2920
2921         err = register_netdevice(dev);
2922         if (err) {
2923                 vxlan_fdb_delete_default(vxlan);
2924                 return err;
2925         }
2926
2927         list_add(&vxlan->next, &vn->vxlan_list);
2928
2929         return 0;
2930 }
2931
2932 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2933                          struct nlattr *tb[], struct nlattr *data[])
2934 {
2935         struct vxlan_config conf;
2936
2937         memset(&conf, 0, sizeof(conf));
2938
2939         if (data[IFLA_VXLAN_ID])
2940                 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2941
2942         if (data[IFLA_VXLAN_GROUP]) {
2943                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2944         } else if (data[IFLA_VXLAN_GROUP6]) {
2945                 if (!IS_ENABLED(CONFIG_IPV6))
2946                         return -EPFNOSUPPORT;
2947
2948                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2949                 conf.remote_ip.sa.sa_family = AF_INET6;
2950         }
2951
2952         if (data[IFLA_VXLAN_LOCAL]) {
2953                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2954                 conf.saddr.sa.sa_family = AF_INET;
2955         } else if (data[IFLA_VXLAN_LOCAL6]) {
2956                 if (!IS_ENABLED(CONFIG_IPV6))
2957                         return -EPFNOSUPPORT;
2958
2959                 /* TODO: respect scope id */
2960                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2961                 conf.saddr.sa.sa_family = AF_INET6;
2962         }
2963
2964         if (data[IFLA_VXLAN_LINK])
2965                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2966
2967         if (data[IFLA_VXLAN_TOS])
2968                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2969
2970         if (data[IFLA_VXLAN_TTL])
2971                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2972
2973         if (data[IFLA_VXLAN_LABEL])
2974                 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
2975                              IPV6_FLOWLABEL_MASK;
2976
2977         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2978                 conf.flags |= VXLAN_F_LEARN;
2979
2980         if (data[IFLA_VXLAN_AGEING])
2981                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2982
2983         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2984                 conf.flags |= VXLAN_F_PROXY;
2985
2986         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2987                 conf.flags |= VXLAN_F_RSC;
2988
2989         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2990                 conf.flags |= VXLAN_F_L2MISS;
2991
2992         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2993                 conf.flags |= VXLAN_F_L3MISS;
2994
2995         if (data[IFLA_VXLAN_LIMIT])
2996                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2997
2998         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2999             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3000                 conf.flags |= VXLAN_F_COLLECT_METADATA;
3001
3002         if (data[IFLA_VXLAN_PORT_RANGE]) {
3003                 const struct ifla_vxlan_port_range *p
3004                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3005                 conf.port_min = ntohs(p->low);
3006                 conf.port_max = ntohs(p->high);
3007         }
3008
3009         if (data[IFLA_VXLAN_PORT])
3010                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3011
3012         if (data[IFLA_VXLAN_UDP_CSUM] &&
3013             !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3014                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3015
3016         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
3017             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3018                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3019
3020         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3021             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3022                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3023
3024         if (data[IFLA_VXLAN_REMCSUM_TX] &&
3025             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3026                 conf.flags |= VXLAN_F_REMCSUM_TX;
3027
3028         if (data[IFLA_VXLAN_REMCSUM_RX] &&
3029             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3030                 conf.flags |= VXLAN_F_REMCSUM_RX;
3031
3032         if (data[IFLA_VXLAN_GBP])
3033                 conf.flags |= VXLAN_F_GBP;
3034
3035         if (data[IFLA_VXLAN_GPE])
3036                 conf.flags |= VXLAN_F_GPE;
3037
3038         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3039                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3040
3041         if (tb[IFLA_MTU])
3042                 conf.mtu = nla_get_u32(tb[IFLA_MTU]);
3043
3044         return vxlan_dev_configure(src_net, dev, &conf);
3045 }
3046
3047 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3048 {
3049         struct vxlan_dev *vxlan = netdev_priv(dev);
3050         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3051
3052         spin_lock(&vn->sock_lock);
3053         if (!hlist_unhashed(&vxlan->hlist))
3054                 hlist_del_rcu(&vxlan->hlist);
3055         spin_unlock(&vn->sock_lock);
3056
3057         gro_cells_destroy(&vxlan->gro_cells);
3058         list_del(&vxlan->next);
3059         unregister_netdevice_queue(dev, head);
3060 }
3061
3062 static size_t vxlan_get_size(const struct net_device *dev)
3063 {
3064
3065         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3066                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3067                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3068                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3069                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3070                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3071                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3072                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3073                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3074                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3075                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3076                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3077                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3078                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3079                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3080                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3081                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3082                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3083                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3084                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3085                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3086                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3087                 0;
3088 }
3089
3090 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3091 {
3092         const struct vxlan_dev *vxlan = netdev_priv(dev);
3093         const struct vxlan_rdst *dst = &vxlan->default_dst;
3094         struct ifla_vxlan_port_range ports = {
3095                 .low =  htons(vxlan->cfg.port_min),
3096                 .high = htons(vxlan->cfg.port_max),
3097         };
3098
3099         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3100                 goto nla_put_failure;
3101
3102         if (!vxlan_addr_any(&dst->remote_ip)) {
3103                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3104                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3105                                             dst->remote_ip.sin.sin_addr.s_addr))
3106                                 goto nla_put_failure;
3107 #if IS_ENABLED(CONFIG_IPV6)
3108                 } else {
3109                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3110                                              &dst->remote_ip.sin6.sin6_addr))
3111                                 goto nla_put_failure;
3112 #endif
3113                 }
3114         }
3115
3116         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3117                 goto nla_put_failure;
3118
3119         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3120                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3121                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3122                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3123                                 goto nla_put_failure;
3124 #if IS_ENABLED(CONFIG_IPV6)
3125                 } else {
3126                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3127                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3128                                 goto nla_put_failure;
3129 #endif
3130                 }
3131         }
3132
3133         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3134             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3135             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3136             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3137                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3138             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3139                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3140             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3141             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3142                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3143             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3144                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3145             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3146                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3147             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3148             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3149             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3150             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3151                         !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3152             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3153                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3154             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3155                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3156             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3157                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3158             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3159                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3160                 goto nla_put_failure;
3161
3162         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3163                 goto nla_put_failure;
3164
3165         if (vxlan->flags & VXLAN_F_GBP &&
3166             nla_put_flag(skb, IFLA_VXLAN_GBP))
3167                 goto nla_put_failure;
3168
3169         if (vxlan->flags & VXLAN_F_GPE &&
3170             nla_put_flag(skb, IFLA_VXLAN_GPE))
3171                 goto nla_put_failure;
3172
3173         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3174             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3175                 goto nla_put_failure;
3176
3177         return 0;
3178
3179 nla_put_failure:
3180         return -EMSGSIZE;
3181 }
3182
3183 static struct net *vxlan_get_link_net(const struct net_device *dev)
3184 {
3185         struct vxlan_dev *vxlan = netdev_priv(dev);
3186
3187         return vxlan->net;
3188 }
3189
3190 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3191         .kind           = "vxlan",
3192         .maxtype        = IFLA_VXLAN_MAX,
3193         .policy         = vxlan_policy,
3194         .priv_size      = sizeof(struct vxlan_dev),
3195         .setup          = vxlan_setup,
3196         .validate       = vxlan_validate,
3197         .newlink        = vxlan_newlink,
3198         .dellink        = vxlan_dellink,
3199         .get_size       = vxlan_get_size,
3200         .fill_info      = vxlan_fill_info,
3201         .get_link_net   = vxlan_get_link_net,
3202 };
3203
3204 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3205                                     u8 name_assign_type,
3206                                     struct vxlan_config *conf)
3207 {
3208         struct nlattr *tb[IFLA_MAX + 1];
3209         struct net_device *dev;
3210         int err;
3211
3212         memset(&tb, 0, sizeof(tb));
3213
3214         dev = rtnl_create_link(net, name, name_assign_type,
3215                                &vxlan_link_ops, tb);
3216         if (IS_ERR(dev))
3217                 return dev;
3218
3219         err = vxlan_dev_configure(net, dev, conf);
3220         if (err < 0) {
3221                 free_netdev(dev);
3222                 return ERR_PTR(err);
3223         }
3224
3225         err = rtnl_configure_link(dev, NULL);
3226         if (err < 0) {
3227                 LIST_HEAD(list_kill);
3228
3229                 vxlan_dellink(dev, &list_kill);
3230                 unregister_netdevice_many(&list_kill);
3231                 return ERR_PTR(err);
3232         }
3233
3234         return dev;
3235 }
3236 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3237
3238 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3239                                              struct net_device *dev)
3240 {
3241         struct vxlan_dev *vxlan, *next;
3242         LIST_HEAD(list_kill);
3243
3244         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3245                 struct vxlan_rdst *dst = &vxlan->default_dst;
3246
3247                 /* In case we created vxlan device with carrier
3248                  * and we loose the carrier due to module unload
3249                  * we also need to remove vxlan device. In other
3250                  * cases, it's not necessary and remote_ifindex
3251                  * is 0 here, so no matches.
3252                  */
3253                 if (dst->remote_ifindex == dev->ifindex)
3254                         vxlan_dellink(vxlan->dev, &list_kill);
3255         }
3256
3257         unregister_netdevice_many(&list_kill);
3258 }
3259
3260 static int vxlan_netdevice_event(struct notifier_block *unused,
3261                                  unsigned long event, void *ptr)
3262 {
3263         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3264         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3265
3266         if (event == NETDEV_UNREGISTER)
3267                 vxlan_handle_lowerdev_unregister(vn, dev);
3268         else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3269                 vxlan_push_rx_ports(dev);
3270
3271         return NOTIFY_DONE;
3272 }
3273
3274 static struct notifier_block vxlan_notifier_block __read_mostly = {
3275         .notifier_call = vxlan_netdevice_event,
3276 };
3277
3278 static __net_init int vxlan_init_net(struct net *net)
3279 {
3280         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3281         unsigned int h;
3282
3283         INIT_LIST_HEAD(&vn->vxlan_list);
3284         spin_lock_init(&vn->sock_lock);
3285
3286         for (h = 0; h < PORT_HASH_SIZE; ++h)
3287                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3288
3289         return 0;
3290 }
3291
3292 static void __net_exit vxlan_exit_net(struct net *net)
3293 {
3294         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3295         struct vxlan_dev *vxlan, *next;
3296         struct net_device *dev, *aux;
3297         LIST_HEAD(list);
3298
3299         rtnl_lock();
3300         for_each_netdev_safe(net, dev, aux)
3301                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3302                         unregister_netdevice_queue(dev, &list);
3303
3304         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3305                 /* If vxlan->dev is in the same netns, it has already been added
3306                  * to the list by the previous loop.
3307                  */
3308                 if (!net_eq(dev_net(vxlan->dev), net)) {
3309                         gro_cells_destroy(&vxlan->gro_cells);
3310                         unregister_netdevice_queue(vxlan->dev, &list);
3311                 }
3312         }
3313
3314         unregister_netdevice_many(&list);
3315         rtnl_unlock();
3316 }
3317
3318 static struct pernet_operations vxlan_net_ops = {
3319         .init = vxlan_init_net,
3320         .exit = vxlan_exit_net,
3321         .id   = &vxlan_net_id,
3322         .size = sizeof(struct vxlan_net),
3323 };
3324
3325 static int __init vxlan_init_module(void)
3326 {
3327         int rc;
3328
3329         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3330
3331         rc = register_pernet_subsys(&vxlan_net_ops);
3332         if (rc)
3333                 goto out1;
3334
3335         rc = register_netdevice_notifier(&vxlan_notifier_block);
3336         if (rc)
3337                 goto out2;
3338
3339         rc = rtnl_link_register(&vxlan_link_ops);
3340         if (rc)
3341                 goto out3;
3342
3343         return 0;
3344 out3:
3345         unregister_netdevice_notifier(&vxlan_notifier_block);
3346 out2:
3347         unregister_pernet_subsys(&vxlan_net_ops);
3348 out1:
3349         return rc;
3350 }
3351 late_initcall(vxlan_init_module);
3352
3353 static void __exit vxlan_cleanup_module(void)
3354 {
3355         rtnl_link_unregister(&vxlan_link_ops);
3356         unregister_netdevice_notifier(&vxlan_notifier_block);
3357         unregister_pernet_subsys(&vxlan_net_ops);
3358         /* rcu_barrier() is called by netns */
3359 }
3360 module_exit(vxlan_cleanup_module);
3361
3362 MODULE_LICENSE("GPL");
3363 MODULE_VERSION(VXLAN_VERSION);
3364 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3365 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3366 MODULE_ALIAS_RTNL_LINK("vxlan");