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 unsigned int vxlan_net_id;
  56 static struct rtnl_link_ops vxlan_link_ops;
  57
  58 static const u8 all_zeros_mac[ETH_ALEN + 2];
  59
  60 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  61
  62 /* 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         int rc;
 615
 616         f = __vxlan_find_mac(vxlan, mac);
 617         if (f) {
 618                 if (flags & NLM_F_EXCL) {
 619                         netdev_dbg(vxlan->dev,
 620                                    "lost race to create %pM\n", mac);
 621                         return -EEXIST;
 622                 }
 623                 if (f->state != state) {
 624                         f->state = state;
 625                         f->updated = jiffies;
 626                         notify = 1;
 627                 }
 628                 if (f->flags != ndm_flags) {
 629                         f->flags = ndm_flags;
 630                         f->updated = jiffies;
 631                         notify = 1;
 632                 }
 633                 if ((flags & NLM_F_REPLACE)) {
 634                         /* Only change unicasts */
 635                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 636                              is_zero_ether_addr(f->eth_addr))) {
 637                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 638                                                            ifindex);
 639                         } else
 640                                 return -EOPNOTSUPP;
 641                 }
 642                 if ((flags & NLM_F_APPEND) &&
 643                     (is_multicast_ether_addr(f->eth_addr) ||
 644                      is_zero_ether_addr(f->eth_addr))) {
 645                         rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &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                 rc = vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 677                 if (rc < 0) {
 678                         kfree(f);
 679                         return rc;
 680                 }
 681
 682                 ++vxlan->addrcnt;
 683                 hlist_add_head_rcu(&f->hlist,
 684                                    vxlan_fdb_head(vxlan, mac));
 685         }
 686
 687         if (notify) {
 688                 if (rd == NULL)
 689                         rd = first_remote_rtnl(f);
 690                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 691         }
 692
 693         return 0;
 694 }
 695
 696 static void vxlan_fdb_free(struct rcu_head *head)
 697 {
 698         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 699         struct vxlan_rdst *rd, *nd;
 700
 701         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 702                 dst_cache_destroy(&rd->dst_cache);
 703                 kfree(rd);
 704         }
 705         kfree(f);
 706 }
 707
 708 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 709 {
 710         netdev_dbg(vxlan->dev,
 711                     "delete %pM\n", f->eth_addr);
 712
 713         --vxlan->addrcnt;
 714         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 715
 716         hlist_del_rcu(&f->hlist);
 717         call_rcu(&f->rcu, vxlan_fdb_free);
 718 }
 719
 720 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 721                            union vxlan_addr *ip, __be16 *port, __be32 *vni,
 722                            u32 *ifindex)
 723 {
 724         struct net *net = dev_net(vxlan->dev);
 725         int err;
 726
 727         if (tb[NDA_DST]) {
 728                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 729                 if (err)
 730                         return err;
 731         } else {
 732                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 733                 if (remote->sa.sa_family == AF_INET) {
 734                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 735                         ip->sa.sa_family = AF_INET;
 736 #if IS_ENABLED(CONFIG_IPV6)
 737                 } else {
 738                         ip->sin6.sin6_addr = in6addr_any;
 739                         ip->sa.sa_family = AF_INET6;
 740 #endif
 741                 }
 742         }
 743
 744         if (tb[NDA_PORT]) {
 745                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 746                         return -EINVAL;
 747                 *port = nla_get_be16(tb[NDA_PORT]);
 748         } else {
 749                 *port = vxlan->cfg.dst_port;
 750         }
 751
 752         if (tb[NDA_VNI]) {
 753                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 754                         return -EINVAL;
 755                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 756         } else {
 757                 *vni = vxlan->default_dst.remote_vni;
 758         }
 759
 760         if (tb[NDA_IFINDEX]) {
 761                 struct net_device *tdev;
 762
 763                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 764                         return -EINVAL;
 765                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 766                 tdev = __dev_get_by_index(net, *ifindex);
 767                 if (!tdev)
 768                         return -EADDRNOTAVAIL;
 769         } else {
 770                 *ifindex = 0;
 771         }
 772
 773         return 0;
 774 }
 775
 776 /* Add static entry (via netlink) */
 777 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 778                          struct net_device *dev,
 779                          const unsigned char *addr, u16 vid, u16 flags)
 780 {
 781         struct vxlan_dev *vxlan = netdev_priv(dev);
 782         /* struct net *net = dev_net(vxlan->dev); */
 783         union vxlan_addr ip;
 784         __be16 port;
 785         __be32 vni;
 786         u32 ifindex;
 787         int err;
 788
 789         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 790                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 791                         ndm->ndm_state);
 792                 return -EINVAL;
 793         }
 794
 795         if (tb[NDA_DST] == NULL)
 796                 return -EINVAL;
 797
 798         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 799         if (err)
 800                 return err;
 801
 802         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 803                 return -EAFNOSUPPORT;
 804
 805         spin_lock_bh(&vxlan->hash_lock);
 806         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 807                                port, vni, ifindex, ndm->ndm_flags);
 808         spin_unlock_bh(&vxlan->hash_lock);
 809
 810         return err;
 811 }
 812
 813 /* Delete entry (via netlink) */
 814 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 815                             struct net_device *dev,
 816                             const unsigned char *addr, u16 vid)
 817 {
 818         struct vxlan_dev *vxlan = netdev_priv(dev);
 819         struct vxlan_fdb *f;
 820         struct vxlan_rdst *rd = NULL;
 821         union vxlan_addr ip;
 822         __be16 port;
 823         __be32 vni;
 824         u32 ifindex;
 825         int err;
 826
 827         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 828         if (err)
 829                 return err;
 830
 831         err = -ENOENT;
 832
 833         spin_lock_bh(&vxlan->hash_lock);
 834         f = vxlan_find_mac(vxlan, addr);
 835         if (!f)
 836                 goto out;
 837
 838         if (!vxlan_addr_any(&ip)) {
 839                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 840                 if (!rd)
 841                         goto out;
 842         }
 843
 844         err = 0;
 845
 846         /* remove a destination if it's not the only one on the list,
 847          * otherwise destroy the fdb entry
 848          */
 849         if (rd && !list_is_singular(&f->remotes)) {
 850                 list_del_rcu(&rd->list);
 851                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 852                 kfree_rcu(rd, rcu);
 853                 goto out;
 854         }
 855
 856         vxlan_fdb_destroy(vxlan, f);
 857
 858 out:
 859         spin_unlock_bh(&vxlan->hash_lock);
 860
 861         return err;
 862 }
 863
 864 /* Dump forwarding table */
 865 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 866                           struct net_device *dev,
 867                           struct net_device *filter_dev, int *idx)
 868 {
 869         struct vxlan_dev *vxlan = netdev_priv(dev);
 870         unsigned int h;
 871         int err = 0;
 872
 873         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 874                 struct vxlan_fdb *f;
 875
 876                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 877                         struct vxlan_rdst *rd;
 878
 879                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 880                                 if (*idx < cb->args[2])
 881                                         goto skip;
 882
 883                                 err = vxlan_fdb_info(skb, vxlan, f,
 884                                                      NETLINK_CB(cb->skb).portid,
 885                                                      cb->nlh->nlmsg_seq,
 886                                                      RTM_NEWNEIGH,
 887                                                      NLM_F_MULTI, rd);
 888                                 if (err < 0)
 889                                         goto out;
 890 skip:
 891                                 *idx += 1;
 892                         }
 893                 }
 894         }
 895 out:
 896         return err;
 897 }
 898
 899 /* Watch incoming packets to learn mapping between Ethernet address
 900  * and Tunnel endpoint.
 901  * Return true if packet is bogus and should be dropped.
 902  */
 903 static bool vxlan_snoop(struct net_device *dev,
 904                         union vxlan_addr *src_ip, const u8 *src_mac)
 905 {
 906         struct vxlan_dev *vxlan = netdev_priv(dev);
 907         struct vxlan_fdb *f;
 908
 909         f = vxlan_find_mac(vxlan, src_mac);
 910         if (likely(f)) {
 911                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 912
 913                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 914                         return false;
 915
 916                 /* Don't migrate static entries, drop packets */
 917                 if (f->state & NUD_NOARP)
 918                         return true;
 919
 920                 if (net_ratelimit())
 921                         netdev_info(dev,
 922                                     "%pM migrated from %pIS to %pIS\n",
 923                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 924
 925                 rdst->remote_ip = *src_ip;
 926                 f->updated = jiffies;
 927                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 928         } else {
 929                 /* learned new entry */
 930                 spin_lock(&vxlan->hash_lock);
 931
 932                 /* close off race between vxlan_flush and incoming packets */
 933                 if (netif_running(dev))
 934                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 935                                          NUD_REACHABLE,
 936                                          NLM_F_EXCL|NLM_F_CREATE,
 937                                          vxlan->cfg.dst_port,
 938                                          vxlan->default_dst.remote_vni,
 939                                          0, NTF_SELF);
 940                 spin_unlock(&vxlan->hash_lock);
 941         }
 942
 943         return false;
 944 }
 945
 946 /* See if multicast group is already in use by other ID */
 947 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
 948 {
 949         struct vxlan_dev *vxlan;
 950         struct vxlan_sock *sock4;
 951 #if IS_ENABLED(CONFIG_IPV6)
 952         struct vxlan_sock *sock6;
 953 #endif
 954         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
 955
 956         sock4 = rtnl_dereference(dev->vn4_sock);
 957
 958         /* The vxlan_sock is only used by dev, leaving group has
 959          * no effect on other vxlan devices.
 960          */
 961         if (family == AF_INET && sock4 && atomic_read(&sock4->refcnt) == 1)
 962                 return false;
 963 #if IS_ENABLED(CONFIG_IPV6)
 964         sock6 = rtnl_dereference(dev->vn6_sock);
 965         if (family == AF_INET6 && sock6 && atomic_read(&sock6->refcnt) == 1)
 966                 return false;
 967 #endif
 968
 969         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
 970                 if (!netif_running(vxlan->dev) || vxlan == dev)
 971                         continue;
 972
 973                 if (family == AF_INET &&
 974                     rtnl_dereference(vxlan->vn4_sock) != sock4)
 975                         continue;
 976 #if IS_ENABLED(CONFIG_IPV6)
 977                 if (family == AF_INET6 &&
 978                     rtnl_dereference(vxlan->vn6_sock) != sock6)
 979                         continue;
 980 #endif
 981
 982                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
 983                                       &dev->default_dst.remote_ip))
 984                         continue;
 985
 986                 if (vxlan->default_dst.remote_ifindex !=
 987                     dev->default_dst.remote_ifindex)
 988                         continue;
 989
 990                 return true;
 991         }
 992
 993         return false;
 994 }
 995
 996 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
 997 {
 998         struct vxlan_net *vn;
 999
1000         if (!vs)
1001                 return false;
1002         if (!atomic_dec_and_test(&vs->refcnt))
1003                 return false;
1004
1005         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
1006         spin_lock(&vn->sock_lock);
1007         hlist_del_rcu(&vs->hlist);
1008         udp_tunnel_notify_del_rx_port(vs->sock,
1009                                       (vs->flags & VXLAN_F_GPE) ?
1010                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1011                                       UDP_TUNNEL_TYPE_VXLAN);
1012         spin_unlock(&vn->sock_lock);
1013
1014         return true;
1015 }
1016
1017 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1018 {
1019         struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1020 #if IS_ENABLED(CONFIG_IPV6)
1021         struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1022
1023         rcu_assign_pointer(vxlan->vn6_sock, NULL);
1024 #endif
1025
1026         rcu_assign_pointer(vxlan->vn4_sock, NULL);
1027         synchronize_net();
1028
1029         if (__vxlan_sock_release_prep(sock4)) {
1030                 udp_tunnel_sock_release(sock4->sock);
1031                 kfree(sock4);
1032         }
1033
1034 #if IS_ENABLED(CONFIG_IPV6)
1035         if (__vxlan_sock_release_prep(sock6)) {
1036                 udp_tunnel_sock_release(sock6->sock);
1037                 kfree(sock6);
1038         }
1039 #endif
1040 }
1041
1042 /* Update multicast group membership when first VNI on
1043  * multicast address is brought up
1044  */
1045 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1046 {
1047         struct sock *sk;
1048         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1049         int ifindex = vxlan->default_dst.remote_ifindex;
1050         int ret = -EINVAL;
1051
1052         if (ip->sa.sa_family == AF_INET) {
1053                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1054                 struct ip_mreqn mreq = {
1055                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1056                         .imr_ifindex            = ifindex,
1057                 };
1058
1059                 sk = sock4->sock->sk;
1060                 lock_sock(sk);
1061                 ret = ip_mc_join_group(sk, &mreq);
1062                 release_sock(sk);
1063 #if IS_ENABLED(CONFIG_IPV6)
1064         } else {
1065                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1066
1067                 sk = sock6->sock->sk;
1068                 lock_sock(sk);
1069                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1070                                                    &ip->sin6.sin6_addr);
1071                 release_sock(sk);
1072 #endif
1073         }
1074
1075         return ret;
1076 }
1077
1078 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1079 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1080 {
1081         struct sock *sk;
1082         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1083         int ifindex = vxlan->default_dst.remote_ifindex;
1084         int ret = -EINVAL;
1085
1086         if (ip->sa.sa_family == AF_INET) {
1087                 struct vxlan_sock *sock4 = rtnl_dereference(vxlan->vn4_sock);
1088                 struct ip_mreqn mreq = {
1089                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1090                         .imr_ifindex            = ifindex,
1091                 };
1092
1093                 sk = sock4->sock->sk;
1094                 lock_sock(sk);
1095                 ret = ip_mc_leave_group(sk, &mreq);
1096                 release_sock(sk);
1097 #if IS_ENABLED(CONFIG_IPV6)
1098         } else {
1099                 struct vxlan_sock *sock6 = rtnl_dereference(vxlan->vn6_sock);
1100
1101                 sk = sock6->sock->sk;
1102                 lock_sock(sk);
1103                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1104                                                    &ip->sin6.sin6_addr);
1105                 release_sock(sk);
1106 #endif
1107         }
1108
1109         return ret;
1110 }
1111
1112 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1113                           struct sk_buff *skb, u32 vxflags)
1114 {
1115         size_t start, offset;
1116
1117         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1118                 goto out;
1119
1120         start = vxlan_rco_start(unparsed->vx_vni);
1121         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1122
1123         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1124                 return false;
1125
1126         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1127                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1128 out:
1129         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1130         unparsed->vx_vni &= VXLAN_VNI_MASK;
1131         return true;
1132 }
1133
1134 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1135                                 struct sk_buff *skb, u32 vxflags,
1136                                 struct vxlan_metadata *md)
1137 {
1138         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1139         struct metadata_dst *tun_dst;
1140
1141         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1142                 goto out;
1143
1144         md->gbp = ntohs(gbp->policy_id);
1145
1146         tun_dst = (struct metadata_dst *)skb_dst(skb);
1147         if (tun_dst) {
1148                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1149                 tun_dst->u.tun_info.options_len = sizeof(*md);
1150         }
1151         if (gbp->dont_learn)
1152                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1153
1154         if (gbp->policy_applied)
1155                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1156
1157         /* In flow-based mode, GBP is carried in dst_metadata */
1158         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1159                 skb->mark = md->gbp;
1160 out:
1161         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1162 }
1163
1164 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1165                                 __be16 *protocol,
1166                                 struct sk_buff *skb, u32 vxflags)
1167 {
1168         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1169
1170         /* Need to have Next Protocol set for interfaces in GPE mode. */
1171         if (!gpe->np_applied)
1172                 return false;
1173         /* "The initial version is 0. If a receiver does not support the
1174          * version indicated it MUST drop the packet.
1175          */
1176         if (gpe->version != 0)
1177                 return false;
1178         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1179          * processing MUST occur." However, we don't implement OAM
1180          * processing, thus drop the packet.
1181          */
1182         if (gpe->oam_flag)
1183                 return false;
1184
1185         switch (gpe->next_protocol) {
1186         case VXLAN_GPE_NP_IPV4:
1187                 *protocol = htons(ETH_P_IP);
1188                 break;
1189         case VXLAN_GPE_NP_IPV6:
1190                 *protocol = htons(ETH_P_IPV6);
1191                 break;
1192         case VXLAN_GPE_NP_ETHERNET:
1193                 *protocol = htons(ETH_P_TEB);
1194                 break;
1195         default:
1196                 return false;
1197         }
1198
1199         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1200         return true;
1201 }
1202
1203 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1204                           struct vxlan_sock *vs,
1205                           struct sk_buff *skb)
1206 {
1207         union vxlan_addr saddr;
1208
1209         skb_reset_mac_header(skb);
1210         skb->protocol = eth_type_trans(skb, vxlan->dev);
1211         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1212
1213         /* Ignore packet loops (and multicast echo) */
1214         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1215                 return false;
1216
1217         /* Get address from the outer IP header */
1218         if (vxlan_get_sk_family(vs) == AF_INET) {
1219                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1220                 saddr.sa.sa_family = AF_INET;
1221 #if IS_ENABLED(CONFIG_IPV6)
1222         } else {
1223                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1224                 saddr.sa.sa_family = AF_INET6;
1225 #endif
1226         }
1227
1228         if ((vxlan->flags & VXLAN_F_LEARN) &&
1229             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1230                 return false;
1231
1232         return true;
1233 }
1234
1235 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1236                                   struct sk_buff *skb)
1237 {
1238         int err = 0;
1239
1240         if (vxlan_get_sk_family(vs) == AF_INET)
1241                 err = IP_ECN_decapsulate(oiph, skb);
1242 #if IS_ENABLED(CONFIG_IPV6)
1243         else
1244                 err = IP6_ECN_decapsulate(oiph, skb);
1245 #endif
1246
1247         if (unlikely(err) && log_ecn_error) {
1248                 if (vxlan_get_sk_family(vs) == AF_INET)
1249                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1250                                              &((struct iphdr *)oiph)->saddr,
1251                                              ((struct iphdr *)oiph)->tos);
1252                 else
1253                         net_info_ratelimited("non-ECT from %pI6\n",
1254                                              &((struct ipv6hdr *)oiph)->saddr);
1255         }
1256         return err <= 1;
1257 }
1258
1259 /* Callback from net/ipv4/udp.c to receive packets */
1260 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1261 {
1262         struct pcpu_sw_netstats *stats;
1263         struct vxlan_dev *vxlan;
1264         struct vxlan_sock *vs;
1265         struct vxlanhdr unparsed;
1266         struct vxlan_metadata _md;
1267         struct vxlan_metadata *md = &_md;
1268         __be16 protocol = htons(ETH_P_TEB);
1269         bool raw_proto = false;
1270         void *oiph;
1271
1272         /* Need UDP and VXLAN header to be present */
1273         if (!pskb_may_pull(skb, VXLAN_HLEN))
1274                 goto drop;
1275
1276         unparsed = *vxlan_hdr(skb);
1277         /* VNI flag always required to be set */
1278         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1279                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1280                            ntohl(vxlan_hdr(skb)->vx_flags),
1281                            ntohl(vxlan_hdr(skb)->vx_vni));
1282                 /* Return non vxlan pkt */
1283                 goto drop;
1284         }
1285         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1286         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1287
1288         vs = rcu_dereference_sk_user_data(sk);
1289         if (!vs)
1290                 goto drop;
1291
1292         vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1293         if (!vxlan)
1294                 goto drop;
1295
1296         /* For backwards compatibility, only allow reserved fields to be
1297          * used by VXLAN extensions if explicitly requested.
1298          */
1299         if (vs->flags & VXLAN_F_GPE) {
1300                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1301                         goto drop;
1302                 raw_proto = true;
1303         }
1304
1305         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1306                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1307                         goto drop;
1308
1309         if (vxlan_collect_metadata(vs)) {
1310                 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1311                 struct metadata_dst *tun_dst;
1312
1313                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1314                                          key32_to_tunnel_id(vni), sizeof(*md));
1315
1316                 if (!tun_dst)
1317                         goto drop;
1318
1319                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1320
1321                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1322         } else {
1323                 memset(md, 0, sizeof(*md));
1324         }
1325
1326         if (vs->flags & VXLAN_F_REMCSUM_RX)
1327                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1328                         goto drop;
1329         if (vs->flags & VXLAN_F_GBP)
1330                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1331         /* Note that GBP and GPE can never be active together. This is
1332          * ensured in vxlan_dev_configure.
1333          */
1334
1335         if (unparsed.vx_flags || unparsed.vx_vni) {
1336                 /* If there are any unprocessed flags remaining treat
1337                  * this as a malformed packet. This behavior diverges from
1338                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1339                  * in reserved fields are to be ignored. The approach here
1340                  * maintains compatibility with previous stack code, and also
1341                  * is more robust and provides a little more security in
1342                  * adding extensions to VXLAN.
1343                  */
1344                 goto drop;
1345         }
1346
1347         if (!raw_proto) {
1348                 if (!vxlan_set_mac(vxlan, vs, skb))
1349                         goto drop;
1350         } else {
1351                 skb_reset_mac_header(skb);
1352                 skb->dev = vxlan->dev;
1353                 skb->pkt_type = PACKET_HOST;
1354         }
1355
1356         oiph = skb_network_header(skb);
1357         skb_reset_network_header(skb);
1358
1359         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1360                 ++vxlan->dev->stats.rx_frame_errors;
1361                 ++vxlan->dev->stats.rx_errors;
1362                 goto drop;
1363         }
1364
1365         stats = this_cpu_ptr(vxlan->dev->tstats);
1366         u64_stats_update_begin(&stats->syncp);
1367         stats->rx_packets++;
1368         stats->rx_bytes += skb->len;
1369         u64_stats_update_end(&stats->syncp);
1370
1371         gro_cells_receive(&vxlan->gro_cells, skb);
1372         return 0;
1373
1374 drop:
1375         /* Consume bad packet */
1376         kfree_skb(skb);
1377         return 0;
1378 }
1379
1380 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1381 {
1382         struct vxlan_dev *vxlan = netdev_priv(dev);
1383         struct arphdr *parp;
1384         u8 *arpptr, *sha;
1385         __be32 sip, tip;
1386         struct neighbour *n;
1387
1388         if (dev->flags & IFF_NOARP)
1389                 goto out;
1390
1391         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1392                 dev->stats.tx_dropped++;
1393                 goto out;
1394         }
1395         parp = arp_hdr(skb);
1396
1397         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1398              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1399             parp->ar_pro != htons(ETH_P_IP) ||
1400             parp->ar_op != htons(ARPOP_REQUEST) ||
1401             parp->ar_hln != dev->addr_len ||
1402             parp->ar_pln != 4)
1403                 goto out;
1404         arpptr = (u8 *)parp + sizeof(struct arphdr);
1405         sha = arpptr;
1406         arpptr += dev->addr_len;        /* sha */
1407         memcpy(&sip, arpptr, sizeof(sip));
1408         arpptr += sizeof(sip);
1409         arpptr += dev->addr_len;        /* tha */
1410         memcpy(&tip, arpptr, sizeof(tip));
1411
1412         if (ipv4_is_loopback(tip) ||
1413             ipv4_is_multicast(tip))
1414                 goto out;
1415
1416         n = neigh_lookup(&arp_tbl, &tip, dev);
1417
1418         if (n) {
1419                 struct vxlan_fdb *f;
1420                 struct sk_buff  *reply;
1421
1422                 if (!(n->nud_state & NUD_CONNECTED)) {
1423                         neigh_release(n);
1424                         goto out;
1425                 }
1426
1427                 f = vxlan_find_mac(vxlan, n->ha);
1428                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1429                         /* bridge-local neighbor */
1430                         neigh_release(n);
1431                         goto out;
1432                 }
1433
1434                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1435                                 n->ha, sha);
1436
1437                 neigh_release(n);
1438
1439                 if (reply == NULL)
1440                         goto out;
1441
1442                 skb_reset_mac_header(reply);
1443                 __skb_pull(reply, skb_network_offset(reply));
1444                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1445                 reply->pkt_type = PACKET_HOST;
1446
1447                 if (netif_rx_ni(reply) == NET_RX_DROP)
1448                         dev->stats.rx_dropped++;
1449         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1450                 union vxlan_addr ipa = {
1451                         .sin.sin_addr.s_addr = tip,
1452                         .sin.sin_family = AF_INET,
1453                 };
1454
1455                 vxlan_ip_miss(dev, &ipa);
1456         }
1457 out:
1458         consume_skb(skb);
1459         return NETDEV_TX_OK;
1460 }
1461
1462 #if IS_ENABLED(CONFIG_IPV6)
1463 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1464         struct neighbour *n, bool isrouter)
1465 {
1466         struct net_device *dev = request->dev;
1467         struct sk_buff *reply;
1468         struct nd_msg *ns, *na;
1469         struct ipv6hdr *pip6;
1470         u8 *daddr;
1471         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1472         int ns_olen;
1473         int i, len;
1474
1475         if (dev == NULL)
1476                 return NULL;
1477
1478         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1479                 sizeof(*na) + na_olen + dev->needed_tailroom;
1480         reply = alloc_skb(len, GFP_ATOMIC);
1481         if (reply == NULL)
1482                 return NULL;
1483
1484         reply->protocol = htons(ETH_P_IPV6);
1485         reply->dev = dev;
1486         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1487         skb_push(reply, sizeof(struct ethhdr));
1488         skb_reset_mac_header(reply);
1489
1490         ns = (struct nd_msg *)skb_transport_header(request);
1491
1492         daddr = eth_hdr(request)->h_source;
1493         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1494         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1495                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1496                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1497                         break;
1498                 }
1499         }
1500
1501         /* Ethernet header */
1502         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1503         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1504         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1505         reply->protocol = htons(ETH_P_IPV6);
1506
1507         skb_pull(reply, sizeof(struct ethhdr));
1508         skb_reset_network_header(reply);
1509         skb_put(reply, sizeof(struct ipv6hdr));
1510
1511         /* IPv6 header */
1512
1513         pip6 = ipv6_hdr(reply);
1514         memset(pip6, 0, sizeof(struct ipv6hdr));
1515         pip6->version = 6;
1516         pip6->priority = ipv6_hdr(request)->priority;
1517         pip6->nexthdr = IPPROTO_ICMPV6;
1518         pip6->hop_limit = 255;
1519         pip6->daddr = ipv6_hdr(request)->saddr;
1520         pip6->saddr = *(struct in6_addr *)n->primary_key;
1521
1522         skb_pull(reply, sizeof(struct ipv6hdr));
1523         skb_reset_transport_header(reply);
1524
1525         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1526
1527         /* Neighbor Advertisement */
1528         memset(na, 0, sizeof(*na)+na_olen);
1529         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1530         na->icmph.icmp6_router = isrouter;
1531         na->icmph.icmp6_override = 1;
1532         na->icmph.icmp6_solicited = 1;
1533         na->target = ns->target;
1534         ether_addr_copy(&na->opt[2], n->ha);
1535         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1536         na->opt[1] = na_olen >> 3;
1537
1538         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1539                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1540                 csum_partial(na, sizeof(*na)+na_olen, 0));
1541
1542         pip6->payload_len = htons(sizeof(*na)+na_olen);
1543
1544         skb_push(reply, sizeof(struct ipv6hdr));
1545
1546         reply->ip_summed = CHECKSUM_UNNECESSARY;
1547
1548         return reply;
1549 }
1550
1551 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1552 {
1553         struct vxlan_dev *vxlan = netdev_priv(dev);
1554         struct nd_msg *msg;
1555         const struct ipv6hdr *iphdr;
1556         const struct in6_addr *saddr, *daddr;
1557         struct neighbour *n;
1558         struct inet6_dev *in6_dev;
1559
1560         in6_dev = __in6_dev_get(dev);
1561         if (!in6_dev)
1562                 goto out;
1563
1564         iphdr = ipv6_hdr(skb);
1565         saddr = &iphdr->saddr;
1566         daddr = &iphdr->daddr;
1567
1568         msg = (struct nd_msg *)skb_transport_header(skb);
1569         if (msg->icmph.icmp6_code != 0 ||
1570             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1571                 goto out;
1572
1573         if (ipv6_addr_loopback(daddr) ||
1574             ipv6_addr_is_multicast(&msg->target))
1575                 goto out;
1576
1577         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1578
1579         if (n) {
1580                 struct vxlan_fdb *f;
1581                 struct sk_buff *reply;
1582
1583                 if (!(n->nud_state & NUD_CONNECTED)) {
1584                         neigh_release(n);
1585                         goto out;
1586                 }
1587
1588                 f = vxlan_find_mac(vxlan, n->ha);
1589                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1590                         /* bridge-local neighbor */
1591                         neigh_release(n);
1592                         goto out;
1593                 }
1594
1595                 reply = vxlan_na_create(skb, n,
1596                                         !!(f ? f->flags & NTF_ROUTER : 0));
1597
1598                 neigh_release(n);
1599
1600                 if (reply == NULL)
1601                         goto out;
1602
1603                 if (netif_rx_ni(reply) == NET_RX_DROP)
1604                         dev->stats.rx_dropped++;
1605
1606         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1607                 union vxlan_addr ipa = {
1608                         .sin6.sin6_addr = msg->target,
1609                         .sin6.sin6_family = AF_INET6,
1610                 };
1611
1612                 vxlan_ip_miss(dev, &ipa);
1613         }
1614
1615 out:
1616         consume_skb(skb);
1617         return NETDEV_TX_OK;
1618 }
1619 #endif
1620
1621 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1622 {
1623         struct vxlan_dev *vxlan = netdev_priv(dev);
1624         struct neighbour *n;
1625
1626         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1627                 return false;
1628
1629         n = NULL;
1630         switch (ntohs(eth_hdr(skb)->h_proto)) {
1631         case ETH_P_IP:
1632         {
1633                 struct iphdr *pip;
1634
1635                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1636                         return false;
1637                 pip = ip_hdr(skb);
1638                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1639                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1640                         union vxlan_addr ipa = {
1641                                 .sin.sin_addr.s_addr = pip->daddr,
1642                                 .sin.sin_family = AF_INET,
1643                         };
1644
1645                         vxlan_ip_miss(dev, &ipa);
1646                         return false;
1647                 }
1648
1649                 break;
1650         }
1651 #if IS_ENABLED(CONFIG_IPV6)
1652         case ETH_P_IPV6:
1653         {
1654                 struct ipv6hdr *pip6;
1655
1656                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1657                         return false;
1658                 pip6 = ipv6_hdr(skb);
1659                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1660                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1661                         union vxlan_addr ipa = {
1662                                 .sin6.sin6_addr = pip6->daddr,
1663                                 .sin6.sin6_family = AF_INET6,
1664                         };
1665
1666                         vxlan_ip_miss(dev, &ipa);
1667                         return false;
1668                 }
1669
1670                 break;
1671         }
1672 #endif
1673         default:
1674                 return false;
1675         }
1676
1677         if (n) {
1678                 bool diff;
1679
1680                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1681                 if (diff) {
1682                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1683                                 dev->addr_len);
1684                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1685                 }
1686                 neigh_release(n);
1687                 return diff;
1688         }
1689
1690         return false;
1691 }
1692
1693 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1694                                 struct vxlan_metadata *md)
1695 {
1696         struct vxlanhdr_gbp *gbp;
1697
1698         if (!md->gbp)
1699                 return;
1700
1701         gbp = (struct vxlanhdr_gbp *)vxh;
1702         vxh->vx_flags |= VXLAN_HF_GBP;
1703
1704         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1705                 gbp->dont_learn = 1;
1706
1707         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1708                 gbp->policy_applied = 1;
1709
1710         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1711 }
1712
1713 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1714                                __be16 protocol)
1715 {
1716         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1717
1718         gpe->np_applied = 1;
1719
1720         switch (protocol) {
1721         case htons(ETH_P_IP):
1722                 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1723                 return 0;
1724         case htons(ETH_P_IPV6):
1725                 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1726                 return 0;
1727         case htons(ETH_P_TEB):
1728                 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1729                 return 0;
1730         }
1731         return -EPFNOSUPPORT;
1732 }
1733
1734 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1735                            int iphdr_len, __be32 vni,
1736                            struct vxlan_metadata *md, u32 vxflags,
1737                            bool udp_sum)
1738 {
1739         struct vxlanhdr *vxh;
1740         int min_headroom;
1741         int err;
1742         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1743         __be16 inner_protocol = htons(ETH_P_TEB);
1744
1745         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1746             skb->ip_summed == CHECKSUM_PARTIAL) {
1747                 int csum_start = skb_checksum_start_offset(skb);
1748
1749                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1750                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1751                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1752                      skb->csum_offset == offsetof(struct tcphdr, check)))
1753                         type |= SKB_GSO_TUNNEL_REMCSUM;
1754         }
1755
1756         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1757                         + VXLAN_HLEN + iphdr_len;
1758
1759         /* Need space for new headers (invalidates iph ptr) */
1760         err = skb_cow_head(skb, min_headroom);
1761         if (unlikely(err))
1762                 return err;
1763
1764         err = iptunnel_handle_offloads(skb, type);
1765         if (err)
1766                 return err;
1767
1768         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1769         vxh->vx_flags = VXLAN_HF_VNI;
1770         vxh->vx_vni = vxlan_vni_field(vni);
1771
1772         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1773                 unsigned int start;
1774
1775                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1776                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1777                 vxh->vx_flags |= VXLAN_HF_RCO;
1778
1779                 if (!skb_is_gso(skb)) {
1780                         skb->ip_summed = CHECKSUM_NONE;
1781                         skb->encapsulation = 0;
1782                 }
1783         }
1784
1785         if (vxflags & VXLAN_F_GBP)
1786                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1787         if (vxflags & VXLAN_F_GPE) {
1788                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1789                 if (err < 0)
1790                         return err;
1791                 inner_protocol = skb->protocol;
1792         }
1793
1794         skb_set_inner_protocol(skb, inner_protocol);
1795         return 0;
1796 }
1797
1798 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan, struct net_device *dev,
1799                                       struct vxlan_sock *sock4,
1800                                       struct sk_buff *skb, int oif, u8 tos,
1801                                       __be32 daddr, __be32 *saddr,
1802                                       struct dst_cache *dst_cache,
1803                                       const struct ip_tunnel_info *info)
1804 {
1805         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1806         struct rtable *rt = NULL;
1807         struct flowi4 fl4;
1808
1809         if (!sock4)
1810                 return ERR_PTR(-EIO);
1811
1812         if (tos && !info)
1813                 use_cache = false;
1814         if (use_cache) {
1815                 rt = dst_cache_get_ip4(dst_cache, saddr);
1816                 if (rt)
1817                         return rt;
1818         }
1819
1820         memset(&fl4, 0, sizeof(fl4));
1821         fl4.flowi4_oif = oif;
1822         fl4.flowi4_tos = RT_TOS(tos);
1823         fl4.flowi4_mark = skb->mark;
1824         fl4.flowi4_proto = IPPROTO_UDP;
1825         fl4.daddr = daddr;
1826         fl4.saddr = *saddr;
1827
1828         rt = ip_route_output_key(vxlan->net, &fl4);
1829         if (likely(!IS_ERR(rt))) {
1830                 if (rt->dst.dev == dev) {
1831                         netdev_dbg(dev, "circular route to %pI4\n", &daddr);
1832                         ip_rt_put(rt);
1833                         return ERR_PTR(-ELOOP);
1834                 }
1835
1836                 *saddr = fl4.saddr;
1837                 if (use_cache)
1838                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1839         } else {
1840                 netdev_dbg(dev, "no route to %pI4\n", &daddr);
1841                 return ERR_PTR(-ENETUNREACH);
1842         }
1843         return rt;
1844 }
1845
1846 #if IS_ENABLED(CONFIG_IPV6)
1847 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1848                                           struct net_device *dev,
1849                                           struct vxlan_sock *sock6,
1850                                           struct sk_buff *skb, int oif, u8 tos,
1851                                           __be32 label,
1852                                           const struct in6_addr *daddr,
1853                                           struct in6_addr *saddr,
1854                                           struct dst_cache *dst_cache,
1855                                           const struct ip_tunnel_info *info)
1856 {
1857         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1858         struct dst_entry *ndst;
1859         struct flowi6 fl6;
1860         int err;
1861
1862         if (!sock6)
1863                 return ERR_PTR(-EIO);
1864
1865         if (tos && !info)
1866                 use_cache = false;
1867         if (use_cache) {
1868                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1869                 if (ndst)
1870                         return ndst;
1871         }
1872
1873         memset(&fl6, 0, sizeof(fl6));
1874         fl6.flowi6_oif = oif;
1875         fl6.daddr = *daddr;
1876         fl6.saddr = *saddr;
1877         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1878         fl6.flowi6_mark = skb->mark;
1879         fl6.flowi6_proto = IPPROTO_UDP;
1880
1881         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1882                                          sock6->sock->sk,
1883                                          &ndst, &fl6);
1884         if (unlikely(err < 0)) {
1885                 netdev_dbg(dev, "no route to %pI6\n", daddr);
1886                 return ERR_PTR(-ENETUNREACH);
1887         }
1888
1889         if (unlikely(ndst->dev == dev)) {
1890                 netdev_dbg(dev, "circular route to %pI6\n", daddr);
1891                 dst_release(ndst);
1892                 return ERR_PTR(-ELOOP);
1893         }
1894
1895         *saddr = fl6.saddr;
1896         if (use_cache)
1897                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1898         return ndst;
1899 }
1900 #endif
1901
1902 /* Bypass encapsulation if the destination is local */
1903 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1904                                struct vxlan_dev *dst_vxlan)
1905 {
1906         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1907         union vxlan_addr loopback;
1908         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1909         struct net_device *dev = skb->dev;
1910         int len = skb->len;
1911
1912         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1913         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1914         skb->pkt_type = PACKET_HOST;
1915         skb->encapsulation = 0;
1916         skb->dev = dst_vxlan->dev;
1917         __skb_pull(skb, skb_network_offset(skb));
1918
1919         if (remote_ip->sa.sa_family == AF_INET) {
1920                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1921                 loopback.sa.sa_family =  AF_INET;
1922 #if IS_ENABLED(CONFIG_IPV6)
1923         } else {
1924                 loopback.sin6.sin6_addr = in6addr_loopback;
1925                 loopback.sa.sa_family =  AF_INET6;
1926 #endif
1927         }
1928
1929         if (dst_vxlan->flags & VXLAN_F_LEARN)
1930                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1931
1932         u64_stats_update_begin(&tx_stats->syncp);
1933         tx_stats->tx_packets++;
1934         tx_stats->tx_bytes += len;
1935         u64_stats_update_end(&tx_stats->syncp);
1936
1937         if (netif_rx(skb) == NET_RX_SUCCESS) {
1938                 u64_stats_update_begin(&rx_stats->syncp);
1939                 rx_stats->rx_packets++;
1940                 rx_stats->rx_bytes += len;
1941                 u64_stats_update_end(&rx_stats->syncp);
1942         } else {
1943                 dev->stats.rx_dropped++;
1944         }
1945 }
1946
1947 static int encap_bypass_if_local(struct sk_buff *skb, struct net_device *dev,
1948                                  struct vxlan_dev *vxlan, union vxlan_addr *daddr,
1949                                  __be32 dst_port, __be32 vni, struct dst_entry *dst,
1950                                  u32 rt_flags)
1951 {
1952 #if IS_ENABLED(CONFIG_IPV6)
1953         /* IPv6 rt-flags are checked against RTF_LOCAL, but the value of
1954          * RTF_LOCAL is equal to RTCF_LOCAL. So to keep code simple
1955          * we can use RTCF_LOCAL which works for ipv4 and ipv6 route entry.
1956          */
1957         BUILD_BUG_ON(RTCF_LOCAL != RTF_LOCAL);
1958 #endif
1959         /* Bypass encapsulation if the destination is local */
1960         if (rt_flags & RTCF_LOCAL &&
1961             !(rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
1962                 struct vxlan_dev *dst_vxlan;
1963
1964                 dst_release(dst);
1965                 dst_vxlan = vxlan_find_vni(vxlan->net, vni,
1966                                            daddr->sa.sa_family, dst_port,
1967                                            vxlan->flags);
1968                 if (!dst_vxlan) {
1969                         dev->stats.tx_errors++;
1970                         kfree_skb(skb);
1971
1972                         return -ENOENT;
1973                 }
1974                 vxlan_encap_bypass(skb, vxlan, dst_vxlan);
1975                 return 1;
1976         }
1977
1978         return 0;
1979 }
1980
1981 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1982                            struct vxlan_rdst *rdst, bool did_rsc)
1983 {
1984         struct dst_cache *dst_cache;
1985         struct ip_tunnel_info *info;
1986         struct vxlan_dev *vxlan = netdev_priv(dev);
1987         const struct iphdr *old_iph = ip_hdr(skb);
1988         union vxlan_addr *dst;
1989         union vxlan_addr remote_ip, local_ip;
1990         union vxlan_addr *src;
1991         struct vxlan_metadata _md;
1992         struct vxlan_metadata *md = &_md;
1993         __be16 src_port = 0, dst_port;
1994         struct dst_entry *ndst = NULL;
1995         __be32 vni, label;
1996         __u8 tos, ttl;
1997         int err;
1998         u32 flags = vxlan->flags;
1999         bool udp_sum = false;
2000         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
2001
2002         info = skb_tunnel_info(skb);
2003
2004         if (rdst) {
2005                 dst = &rdst->remote_ip;
2006                 if (vxlan_addr_any(dst)) {
2007                         if (did_rsc) {
2008                                 /* short-circuited back to local bridge */
2009                                 vxlan_encap_bypass(skb, vxlan, vxlan);
2010                                 return;
2011                         }
2012                         goto drop;
2013                 }
2014
2015                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
2016                 vni = rdst->remote_vni;
2017                 src = &vxlan->cfg.saddr;
2018                 dst_cache = &rdst->dst_cache;
2019                 md->gbp = skb->mark;
2020                 ttl = vxlan->cfg.ttl;
2021                 if (!ttl && vxlan_addr_multicast(dst))
2022                         ttl = 1;
2023
2024                 tos = vxlan->cfg.tos;
2025                 if (tos == 1)
2026                         tos = ip_tunnel_get_dsfield(old_iph, skb);
2027
2028                 if (dst->sa.sa_family == AF_INET)
2029                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2030                 else
2031                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2032                 label = vxlan->cfg.label;
2033         } else {
2034                 if (!info) {
2035                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
2036                                   dev->name);
2037                         goto drop;
2038                 }
2039                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
2040                 if (remote_ip.sa.sa_family == AF_INET) {
2041                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
2042                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
2043                 } else {
2044                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
2045                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
2046                 }
2047                 dst = &remote_ip;
2048                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
2049                 vni = tunnel_id_to_key32(info->key.tun_id);
2050                 src = &local_ip;
2051                 dst_cache = &info->dst_cache;
2052                 if (info->options_len)
2053                         md = ip_tunnel_info_opts(info);
2054                 ttl = info->key.ttl;
2055                 tos = info->key.tos;
2056                 label = info->key.label;
2057                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
2058         }
2059         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2060                                      vxlan->cfg.port_max, true);
2061
2062         if (dst->sa.sa_family == AF_INET) {
2063                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2064                 struct rtable *rt;
2065                 __be16 df = 0;
2066
2067                 rt = vxlan_get_route(vxlan, dev, sock4, skb,
2068                                      rdst ? rdst->remote_ifindex : 0, tos,
2069                                      dst->sin.sin_addr.s_addr,
2070                                      &src->sin.sin_addr.s_addr,
2071                                      dst_cache, info);
2072                 if (IS_ERR(rt)) {
2073                         err = PTR_ERR(rt);
2074                         goto tx_error;
2075                 }
2076
2077                 /* Bypass encapsulation if the destination is local */
2078                 if (!info) {
2079                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2080                                                     dst_port, vni, &rt->dst,
2081                                                     rt->rt_flags);
2082                         if (err)
2083                                 return;
2084                 } else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT) {
2085                         df = htons(IP_DF);
2086                 }
2087
2088                 ndst = &rt->dst;
2089                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2090                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2091                 err = vxlan_build_skb(skb, ndst, sizeof(struct iphdr),
2092                                       vni, md, flags, udp_sum);
2093                 if (err < 0)
2094                         goto tx_error;
2095
2096                 udp_tunnel_xmit_skb(rt, sock4->sock->sk, skb, src->sin.sin_addr.s_addr,
2097                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2098                                     src_port, dst_port, xnet, !udp_sum);
2099 #if IS_ENABLED(CONFIG_IPV6)
2100         } else {
2101                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2102
2103                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb,
2104                                         rdst ? rdst->remote_ifindex : 0, tos,
2105                                         label, &dst->sin6.sin6_addr,
2106                                         &src->sin6.sin6_addr,
2107                                         dst_cache, info);
2108                 if (IS_ERR(ndst)) {
2109                         err = PTR_ERR(ndst);
2110                         ndst = NULL;
2111                         goto tx_error;
2112                 }
2113
2114                 if (!info) {
2115                         u32 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2116
2117                         err = encap_bypass_if_local(skb, dev, vxlan, dst,
2118                                                     dst_port, vni, ndst,
2119                                                     rt6i_flags);
2120                         if (err)
2121                                 return;
2122                 }
2123
2124                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2125                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2126                 skb_scrub_packet(skb, xnet);
2127                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2128                                       vni, md, flags, udp_sum);
2129                 if (err < 0)
2130                         goto tx_error;
2131
2132                 udp_tunnel6_xmit_skb(ndst, sock6->sock->sk, skb, dev,
2133                                      &src->sin6.sin6_addr,
2134                                      &dst->sin6.sin6_addr, tos, ttl,
2135                                      label, src_port, dst_port, !udp_sum);
2136 #endif
2137         }
2138         return;
2139
2140 drop:
2141         dev->stats.tx_dropped++;
2142         dev_kfree_skb(skb);
2143         return;
2144
2145 tx_error:
2146         if (err == -ELOOP)
2147                 dev->stats.collisions++;
2148         else if (err == -ENETUNREACH)
2149                 dev->stats.tx_carrier_errors++;
2150         dst_release(ndst);
2151         dev->stats.tx_errors++;
2152         kfree_skb(skb);
2153 }
2154
2155 /* Transmit local packets over Vxlan
2156  *
2157  * Outer IP header inherits ECN and DF from inner header.
2158  * Outer UDP destination is the VXLAN assigned port.
2159  *           source port is based on hash of flow
2160  */
2161 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2162 {
2163         struct vxlan_dev *vxlan = netdev_priv(dev);
2164         const struct ip_tunnel_info *info;
2165         struct ethhdr *eth;
2166         bool did_rsc = false;
2167         struct vxlan_rdst *rdst, *fdst = NULL;
2168         struct vxlan_fdb *f;
2169
2170         info = skb_tunnel_info(skb);
2171
2172         skb_reset_mac_header(skb);
2173
2174         if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2175                 if (info && info->mode & IP_TUNNEL_INFO_TX)
2176                         vxlan_xmit_one(skb, dev, NULL, false);
2177                 else
2178                         kfree_skb(skb);
2179                 return NETDEV_TX_OK;
2180         }
2181
2182         if (vxlan->flags & VXLAN_F_PROXY) {
2183                 eth = eth_hdr(skb);
2184                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2185                         return arp_reduce(dev, skb);
2186 #if IS_ENABLED(CONFIG_IPV6)
2187                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2188                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2189                                        + sizeof(struct nd_msg)) &&
2190                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2191                                 struct nd_msg *msg;
2192
2193                                 msg = (struct nd_msg *)skb_transport_header(skb);
2194                                 if (msg->icmph.icmp6_code == 0 &&
2195                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2196                                         return neigh_reduce(dev, skb);
2197                 }
2198 #endif
2199         }
2200
2201         eth = eth_hdr(skb);
2202         f = vxlan_find_mac(vxlan, eth->h_dest);
2203         did_rsc = false;
2204
2205         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2206             (ntohs(eth->h_proto) == ETH_P_IP ||
2207              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2208                 did_rsc = route_shortcircuit(dev, skb);
2209                 if (did_rsc)
2210                         f = vxlan_find_mac(vxlan, eth->h_dest);
2211         }
2212
2213         if (f == NULL) {
2214                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2215                 if (f == NULL) {
2216                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2217                             !is_multicast_ether_addr(eth->h_dest))
2218                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2219
2220                         dev->stats.tx_dropped++;
2221                         kfree_skb(skb);
2222                         return NETDEV_TX_OK;
2223                 }
2224         }
2225
2226         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2227                 struct sk_buff *skb1;
2228
2229                 if (!fdst) {
2230                         fdst = rdst;
2231                         continue;
2232                 }
2233                 skb1 = skb_clone(skb, GFP_ATOMIC);
2234                 if (skb1)
2235                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2236         }
2237
2238         if (fdst)
2239                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2240         else
2241                 kfree_skb(skb);
2242         return NETDEV_TX_OK;
2243 }
2244
2245 /* Walk the forwarding table and purge stale entries */
2246 static void vxlan_cleanup(unsigned long arg)
2247 {
2248         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2249         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2250         unsigned int h;
2251
2252         if (!netif_running(vxlan->dev))
2253                 return;
2254
2255         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2256                 struct hlist_node *p, *n;
2257
2258                 spin_lock_bh(&vxlan->hash_lock);
2259                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2260                         struct vxlan_fdb *f
2261                                 = container_of(p, struct vxlan_fdb, hlist);
2262                         unsigned long timeout;
2263
2264                         if (f->state & NUD_PERMANENT)
2265                                 continue;
2266
2267                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2268                         if (time_before_eq(timeout, jiffies)) {
2269                                 netdev_dbg(vxlan->dev,
2270                                            "garbage collect %pM\n",
2271                                            f->eth_addr);
2272                                 f->state = NUD_STALE;
2273                                 vxlan_fdb_destroy(vxlan, f);
2274                         } else if (time_before(timeout, next_timer))
2275                                 next_timer = timeout;
2276                 }
2277                 spin_unlock_bh(&vxlan->hash_lock);
2278         }
2279
2280         mod_timer(&vxlan->age_timer, next_timer);
2281 }
2282
2283 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2284 {
2285         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2286         __be32 vni = vxlan->default_dst.remote_vni;
2287
2288         spin_lock(&vn->sock_lock);
2289         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2290         spin_unlock(&vn->sock_lock);
2291 }
2292
2293 /* Setup stats when device is created */
2294 static int vxlan_init(struct net_device *dev)
2295 {
2296         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2297         if (!dev->tstats)
2298                 return -ENOMEM;
2299
2300         return 0;
2301 }
2302
2303 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2304 {
2305         struct vxlan_fdb *f;
2306
2307         spin_lock_bh(&vxlan->hash_lock);
2308         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2309         if (f)
2310                 vxlan_fdb_destroy(vxlan, f);
2311         spin_unlock_bh(&vxlan->hash_lock);
2312 }
2313
2314 static void vxlan_uninit(struct net_device *dev)
2315 {
2316         struct vxlan_dev *vxlan = netdev_priv(dev);
2317
2318         vxlan_fdb_delete_default(vxlan);
2319
2320         free_percpu(dev->tstats);
2321 }
2322
2323 /* Start ageing timer and join group when device is brought up */
2324 static int vxlan_open(struct net_device *dev)
2325 {
2326         struct vxlan_dev *vxlan = netdev_priv(dev);
2327         int ret;
2328
2329         ret = vxlan_sock_add(vxlan);
2330         if (ret < 0)
2331                 return ret;
2332
2333         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2334                 ret = vxlan_igmp_join(vxlan);
2335                 if (ret == -EADDRINUSE)
2336                         ret = 0;
2337                 if (ret) {
2338                         vxlan_sock_release(vxlan);
2339                         return ret;
2340                 }
2341         }
2342
2343         if (vxlan->cfg.age_interval)
2344                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2345
2346         return ret;
2347 }
2348
2349 /* Purge the forwarding table */
2350 static void vxlan_flush(struct vxlan_dev *vxlan)
2351 {
2352         unsigned int h;
2353
2354         spin_lock_bh(&vxlan->hash_lock);
2355         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2356                 struct hlist_node *p, *n;
2357                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2358                         struct vxlan_fdb *f
2359                                 = container_of(p, struct vxlan_fdb, hlist);
2360                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2361                         if (!is_zero_ether_addr(f->eth_addr))
2362                                 vxlan_fdb_destroy(vxlan, f);
2363                 }
2364         }
2365         spin_unlock_bh(&vxlan->hash_lock);
2366 }
2367
2368 /* Cleanup timer and forwarding table on shutdown */
2369 static int vxlan_stop(struct net_device *dev)
2370 {
2371         struct vxlan_dev *vxlan = netdev_priv(dev);
2372         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2373         int ret = 0;
2374
2375         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2376             !vxlan_group_used(vn, vxlan))
2377                 ret = vxlan_igmp_leave(vxlan);
2378
2379         del_timer_sync(&vxlan->age_timer);
2380
2381         vxlan_flush(vxlan);
2382         vxlan_sock_release(vxlan);
2383
2384         return ret;
2385 }
2386
2387 /* Stub, nothing needs to be done. */
2388 static void vxlan_set_multicast_list(struct net_device *dev)
2389 {
2390 }
2391
2392 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2393 {
2394         struct vxlan_dev *vxlan = netdev_priv(dev);
2395         struct vxlan_rdst *dst = &vxlan->default_dst;
2396         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2397                                                          dst->remote_ifindex);
2398         bool use_ipv6 = false;
2399
2400         if (dst->remote_ip.sa.sa_family == AF_INET6)
2401                 use_ipv6 = true;
2402
2403         /* This check is different than dev->max_mtu, because it looks at
2404          * the lowerdev->mtu, rather than the static dev->max_mtu
2405          */
2406         if (lowerdev) {
2407                 int max_mtu = lowerdev->mtu -
2408                               (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2409                 if (new_mtu > max_mtu)
2410                         return -EINVAL;
2411         }
2412
2413         dev->mtu = new_mtu;
2414         return 0;
2415 }
2416
2417 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2418 {
2419         struct vxlan_dev *vxlan = netdev_priv(dev);
2420         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2421         __be16 sport, dport;
2422
2423         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2424                                   vxlan->cfg.port_max, true);
2425         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2426
2427         if (ip_tunnel_info_af(info) == AF_INET) {
2428                 struct vxlan_sock *sock4 = rcu_dereference(vxlan->vn4_sock);
2429                 struct rtable *rt;
2430
2431                 rt = vxlan_get_route(vxlan, dev, sock4, skb, 0, info->key.tos,
2432                                      info->key.u.ipv4.dst,
2433                                      &info->key.u.ipv4.src, NULL, info);
2434                 if (IS_ERR(rt))
2435                         return PTR_ERR(rt);
2436                 ip_rt_put(rt);
2437         } else {
2438 #if IS_ENABLED(CONFIG_IPV6)
2439                 struct vxlan_sock *sock6 = rcu_dereference(vxlan->vn6_sock);
2440                 struct dst_entry *ndst;
2441
2442                 ndst = vxlan6_get_route(vxlan, dev, sock6, skb, 0, info->key.tos,
2443                                         info->key.label, &info->key.u.ipv6.dst,
2444                                         &info->key.u.ipv6.src, NULL, info);
2445                 if (IS_ERR(ndst))
2446                         return PTR_ERR(ndst);
2447                 dst_release(ndst);
2448 #else /* !CONFIG_IPV6 */
2449                 return -EPFNOSUPPORT;
2450 #endif
2451         }
2452         info->key.tp_src = sport;
2453         info->key.tp_dst = dport;
2454         return 0;
2455 }
2456
2457 static const struct net_device_ops vxlan_netdev_ether_ops = {
2458         .ndo_init               = vxlan_init,
2459         .ndo_uninit             = vxlan_uninit,
2460         .ndo_open               = vxlan_open,
2461         .ndo_stop               = vxlan_stop,
2462         .ndo_start_xmit         = vxlan_xmit,
2463         .ndo_get_stats64        = ip_tunnel_get_stats64,
2464         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2465         .ndo_change_mtu         = vxlan_change_mtu,
2466         .ndo_validate_addr      = eth_validate_addr,
2467         .ndo_set_mac_address    = eth_mac_addr,
2468         .ndo_fdb_add            = vxlan_fdb_add,
2469         .ndo_fdb_del            = vxlan_fdb_delete,
2470         .ndo_fdb_dump           = vxlan_fdb_dump,
2471         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2472 };
2473
2474 static const struct net_device_ops vxlan_netdev_raw_ops = {
2475         .ndo_init               = vxlan_init,
2476         .ndo_uninit             = vxlan_uninit,
2477         .ndo_open               = vxlan_open,
2478         .ndo_stop               = vxlan_stop,
2479         .ndo_start_xmit         = vxlan_xmit,
2480         .ndo_get_stats64        = ip_tunnel_get_stats64,
2481         .ndo_change_mtu         = vxlan_change_mtu,
2482         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2483 };
2484
2485 /* Info for udev, that this is a virtual tunnel endpoint */
2486 static struct device_type vxlan_type = {
2487         .name = "vxlan",
2488 };
2489
2490 /* Calls the ndo_udp_tunnel_add of the caller in order to
2491  * supply the listening VXLAN udp ports. Callers are expected
2492  * to implement the ndo_udp_tunnel_add.
2493  */
2494 static void vxlan_push_rx_ports(struct net_device *dev)
2495 {
2496         struct vxlan_sock *vs;
2497         struct net *net = dev_net(dev);
2498         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2499         unsigned int i;
2500
2501         spin_lock(&vn->sock_lock);
2502         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2503                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2504                         udp_tunnel_push_rx_port(dev, vs->sock,
2505                                                 (vs->flags & VXLAN_F_GPE) ?
2506                                                 UDP_TUNNEL_TYPE_VXLAN_GPE :
2507                                                 UDP_TUNNEL_TYPE_VXLAN);
2508         }
2509         spin_unlock(&vn->sock_lock);
2510 }
2511
2512 /* Initialize the device structure. */
2513 static void vxlan_setup(struct net_device *dev)
2514 {
2515         struct vxlan_dev *vxlan = netdev_priv(dev);
2516         unsigned int h;
2517
2518         eth_hw_addr_random(dev);
2519         ether_setup(dev);
2520
2521         dev->destructor = free_netdev;
2522         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2523
2524         dev->features   |= NETIF_F_LLTX;
2525         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2526         dev->features   |= NETIF_F_RXCSUM;
2527         dev->features   |= NETIF_F_GSO_SOFTWARE;
2528
2529         dev->vlan_features = dev->features;
2530         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2531         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2532         netif_keep_dst(dev);
2533         dev->priv_flags |= IFF_NO_QUEUE;
2534
2535         INIT_LIST_HEAD(&vxlan->next);
2536         spin_lock_init(&vxlan->hash_lock);
2537
2538         init_timer_deferrable(&vxlan->age_timer);
2539         vxlan->age_timer.function = vxlan_cleanup;
2540         vxlan->age_timer.data = (unsigned long) vxlan;
2541
2542         vxlan->cfg.dst_port = htons(vxlan_port);
2543
2544         vxlan->dev = dev;
2545
2546         gro_cells_init(&vxlan->gro_cells, dev);
2547
2548         for (h = 0; h < FDB_HASH_SIZE; ++h)
2549                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2550 }
2551
2552 static void vxlan_ether_setup(struct net_device *dev)
2553 {
2554         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2555         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2556         dev->netdev_ops = &vxlan_netdev_ether_ops;
2557 }
2558
2559 static void vxlan_raw_setup(struct net_device *dev)
2560 {
2561         dev->header_ops = NULL;
2562         dev->type = ARPHRD_NONE;
2563         dev->hard_header_len = 0;
2564         dev->addr_len = 0;
2565         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2566         dev->netdev_ops = &vxlan_netdev_raw_ops;
2567 }
2568
2569 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2570         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2571         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2572         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2573         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2574         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2575         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2576         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2577         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2578         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2579         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2580         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2581         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2582         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2583         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2584         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2585         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2586         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2587         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2588         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2589         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2590         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2591         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2592         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2593         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2594         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2595         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2596         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2597 };
2598
2599 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2600 {
2601         if (tb[IFLA_ADDRESS]) {
2602                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2603                         pr_debug("invalid link address (not ethernet)\n");
2604                         return -EINVAL;
2605                 }
2606
2607                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2608                         pr_debug("invalid all zero ethernet address\n");
2609                         return -EADDRNOTAVAIL;
2610                 }
2611         }
2612
2613         if (!data)
2614                 return -EINVAL;
2615
2616         if (data[IFLA_VXLAN_ID]) {
2617                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2618                 if (id >= VXLAN_VID_MASK)
2619                         return -ERANGE;
2620         }
2621
2622         if (data[IFLA_VXLAN_PORT_RANGE]) {
2623                 const struct ifla_vxlan_port_range *p
2624                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2625
2626                 if (ntohs(p->high) < ntohs(p->low)) {
2627                         pr_debug("port range %u .. %u not valid\n",
2628                                  ntohs(p->low), ntohs(p->high));
2629                         return -EINVAL;
2630                 }
2631         }
2632
2633         return 0;
2634 }
2635
2636 static void vxlan_get_drvinfo(struct net_device *netdev,
2637                               struct ethtool_drvinfo *drvinfo)
2638 {
2639         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2640         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2641 }
2642
2643 static const struct ethtool_ops vxlan_ethtool_ops = {
2644         .get_drvinfo    = vxlan_get_drvinfo,
2645         .get_link       = ethtool_op_get_link,
2646 };
2647
2648 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2649                                         __be16 port, u32 flags)
2650 {
2651         struct socket *sock;
2652         struct udp_port_cfg udp_conf;
2653         int err;
2654
2655         memset(&udp_conf, 0, sizeof(udp_conf));
2656
2657         if (ipv6) {
2658                 udp_conf.family = AF_INET6;
2659                 udp_conf.use_udp6_rx_checksums =
2660                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2661                 udp_conf.ipv6_v6only = 1;
2662         } else {
2663                 udp_conf.family = AF_INET;
2664         }
2665
2666         udp_conf.local_udp_port = port;
2667
2668         /* Open UDP socket */
2669         err = udp_sock_create(net, &udp_conf, &sock);
2670         if (err < 0)
2671                 return ERR_PTR(err);
2672
2673         return sock;
2674 }
2675
2676 /* Create new listen socket if needed */
2677 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2678                                               __be16 port, u32 flags)
2679 {
2680         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2681         struct vxlan_sock *vs;
2682         struct socket *sock;
2683         unsigned int h;
2684         struct udp_tunnel_sock_cfg tunnel_cfg;
2685
2686         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2687         if (!vs)
2688                 return ERR_PTR(-ENOMEM);
2689
2690         for (h = 0; h < VNI_HASH_SIZE; ++h)
2691                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2692
2693         sock = vxlan_create_sock(net, ipv6, port, flags);
2694         if (IS_ERR(sock)) {
2695                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2696                         PTR_ERR(sock));
2697                 kfree(vs);
2698                 return ERR_CAST(sock);
2699         }
2700
2701         vs->sock = sock;
2702         atomic_set(&vs->refcnt, 1);
2703         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2704
2705         spin_lock(&vn->sock_lock);
2706         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2707         udp_tunnel_notify_add_rx_port(sock,
2708                                       (vs->flags & VXLAN_F_GPE) ?
2709                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2710                                       UDP_TUNNEL_TYPE_VXLAN);
2711         spin_unlock(&vn->sock_lock);
2712
2713         /* Mark socket as an encapsulation socket. */
2714         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2715         tunnel_cfg.sk_user_data = vs;
2716         tunnel_cfg.encap_type = 1;
2717         tunnel_cfg.encap_rcv = vxlan_rcv;
2718         tunnel_cfg.encap_destroy = NULL;
2719         tunnel_cfg.gro_receive = vxlan_gro_receive;
2720         tunnel_cfg.gro_complete = vxlan_gro_complete;
2721
2722         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2723
2724         return vs;
2725 }
2726
2727 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2728 {
2729         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2730         struct vxlan_sock *vs = NULL;
2731
2732         if (!vxlan->cfg.no_share) {
2733                 spin_lock(&vn->sock_lock);
2734                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2735                                      vxlan->cfg.dst_port, vxlan->flags);
2736                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2737                         spin_unlock(&vn->sock_lock);
2738                         return -EBUSY;
2739                 }
2740                 spin_unlock(&vn->sock_lock);
2741         }
2742         if (!vs)
2743                 vs = vxlan_socket_create(vxlan->net, ipv6,
2744                                          vxlan->cfg.dst_port, vxlan->flags);
2745         if (IS_ERR(vs))
2746                 return PTR_ERR(vs);
2747 #if IS_ENABLED(CONFIG_IPV6)
2748         if (ipv6)
2749                 rcu_assign_pointer(vxlan->vn6_sock, vs);
2750         else
2751 #endif
2752                 rcu_assign_pointer(vxlan->vn4_sock, vs);
2753         vxlan_vs_add_dev(vs, vxlan);
2754         return 0;
2755 }
2756
2757 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2758 {
2759         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2760         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2761         int ret = 0;
2762
2763         RCU_INIT_POINTER(vxlan->vn4_sock, NULL);
2764 #if IS_ENABLED(CONFIG_IPV6)
2765         RCU_INIT_POINTER(vxlan->vn6_sock, NULL);
2766         if (ipv6 || metadata)
2767                 ret = __vxlan_sock_add(vxlan, true);
2768 #endif
2769         if (!ret && (!ipv6 || metadata))
2770                 ret = __vxlan_sock_add(vxlan, false);
2771         if (ret < 0)
2772                 vxlan_sock_release(vxlan);
2773         return ret;
2774 }
2775
2776 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2777                                struct vxlan_config *conf)
2778 {
2779         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2780         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2781         struct vxlan_rdst *dst = &vxlan->default_dst;
2782         unsigned short needed_headroom = ETH_HLEN;
2783         int err;
2784         bool use_ipv6 = false;
2785         __be16 default_port = vxlan->cfg.dst_port;
2786         struct net_device *lowerdev = NULL;
2787
2788         if (conf->flags & VXLAN_F_GPE) {
2789                 /* For now, allow GPE only together with COLLECT_METADATA.
2790                  * This can be relaxed later; in such case, the other side
2791                  * of the PtP link will have to be provided.
2792                  */
2793                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2794                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2795                         pr_info("unsupported combination of extensions\n");
2796                         return -EINVAL;
2797                 }
2798
2799                 vxlan_raw_setup(dev);
2800         } else {
2801                 vxlan_ether_setup(dev);
2802         }
2803
2804         /* MTU range: 68 - 65535 */
2805         dev->min_mtu = ETH_MIN_MTU;
2806         dev->max_mtu = ETH_MAX_MTU;
2807
2808         vxlan->net = src_net;
2809
2810         dst->remote_vni = conf->vni;
2811
2812         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2813
2814         /* Unless IPv6 is explicitly requested, assume IPv4 */
2815         if (!dst->remote_ip.sa.sa_family)
2816                 dst->remote_ip.sa.sa_family = AF_INET;
2817
2818         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2819             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2820                 if (!IS_ENABLED(CONFIG_IPV6))
2821                         return -EPFNOSUPPORT;
2822                 use_ipv6 = true;
2823                 vxlan->flags |= VXLAN_F_IPV6;
2824         }
2825
2826         if (conf->label && !use_ipv6) {
2827                 pr_info("label only supported in use with IPv6\n");
2828                 return -EINVAL;
2829         }
2830
2831         if (conf->remote_ifindex) {
2832                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2833                 dst->remote_ifindex = conf->remote_ifindex;
2834
2835                 if (!lowerdev) {
2836                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2837                         return -ENODEV;
2838                 }
2839
2840 #if IS_ENABLED(CONFIG_IPV6)
2841                 if (use_ipv6) {
2842                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2843                         if (idev && idev->cnf.disable_ipv6) {
2844                                 pr_info("IPv6 is disabled via sysctl\n");
2845                                 return -EPERM;
2846                         }
2847                 }
2848 #endif
2849
2850                 if (!conf->mtu)
2851                         dev->mtu = lowerdev->mtu -
2852                                    (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2853
2854                 needed_headroom = lowerdev->hard_header_len;
2855         } else if (vxlan_addr_multicast(&dst->remote_ip)) {
2856                 pr_info("multicast destination requires interface to be specified\n");
2857                 return -EINVAL;
2858         }
2859
2860         if (conf->mtu) {
2861                 int max_mtu = ETH_MAX_MTU;
2862
2863                 if (lowerdev)
2864                         max_mtu = lowerdev->mtu;
2865
2866                 max_mtu -= (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2867
2868                 if (conf->mtu < dev->min_mtu || conf->mtu > dev->max_mtu)
2869                         return -EINVAL;
2870
2871                 dev->mtu = conf->mtu;
2872
2873                 if (conf->mtu > max_mtu)
2874                         dev->mtu = max_mtu;
2875         }
2876
2877         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2878                 needed_headroom += VXLAN6_HEADROOM;
2879         else
2880                 needed_headroom += VXLAN_HEADROOM;
2881         dev->needed_headroom = needed_headroom;
2882
2883         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2884         if (!vxlan->cfg.dst_port) {
2885                 if (conf->flags & VXLAN_F_GPE)
2886                         vxlan->cfg.dst_port = 4790; /* IANA assigned VXLAN-GPE port */
2887                 else
2888                         vxlan->cfg.dst_port = default_port;
2889         }
2890         vxlan->flags |= conf->flags;
2891
2892         if (!vxlan->cfg.age_interval)
2893                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2894
2895         list_for_each_entry(tmp, &vn->vxlan_list, next) {
2896                 if (tmp->cfg.vni == conf->vni &&
2897                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2898                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2899                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2900                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2901                     (vxlan->flags & VXLAN_F_RCV_FLAGS)) {
2902                         pr_info("duplicate VNI %u\n", be32_to_cpu(conf->vni));
2903                         return -EEXIST;
2904                 }
2905         }
2906
2907         dev->ethtool_ops = &vxlan_ethtool_ops;
2908
2909         /* create an fdb entry for a valid default destination */
2910         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2911                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2912                                        &vxlan->default_dst.remote_ip,
2913                                        NUD_REACHABLE|NUD_PERMANENT,
2914                                        NLM_F_EXCL|NLM_F_CREATE,
2915                                        vxlan->cfg.dst_port,
2916                                        vxlan->default_dst.remote_vni,
2917                                        vxlan->default_dst.remote_ifindex,
2918                                        NTF_SELF);
2919                 if (err)
2920                         return err;
2921         }
2922
2923         err = register_netdevice(dev);
2924         if (err) {
2925                 vxlan_fdb_delete_default(vxlan);
2926                 return err;
2927         }
2928
2929         list_add(&vxlan->next, &vn->vxlan_list);
2930
2931         return 0;
2932 }
2933
2934 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2935                          struct nlattr *tb[], struct nlattr *data[])
2936 {
2937         struct vxlan_config conf;
2938
2939         memset(&conf, 0, sizeof(conf));
2940
2941         if (data[IFLA_VXLAN_ID])
2942                 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2943
2944         if (data[IFLA_VXLAN_GROUP]) {
2945                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2946         } else if (data[IFLA_VXLAN_GROUP6]) {
2947                 if (!IS_ENABLED(CONFIG_IPV6))
2948                         return -EPFNOSUPPORT;
2949
2950                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2951                 conf.remote_ip.sa.sa_family = AF_INET6;
2952         }
2953
2954         if (data[IFLA_VXLAN_LOCAL]) {
2955                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2956                 conf.saddr.sa.sa_family = AF_INET;
2957         } else if (data[IFLA_VXLAN_LOCAL6]) {
2958                 if (!IS_ENABLED(CONFIG_IPV6))
2959                         return -EPFNOSUPPORT;
2960
2961                 /* TODO: respect scope id */
2962                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2963                 conf.saddr.sa.sa_family = AF_INET6;
2964         }
2965
2966         if (data[IFLA_VXLAN_LINK])
2967                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2968
2969         if (data[IFLA_VXLAN_TOS])
2970                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2971
2972         if (data[IFLA_VXLAN_TTL])
2973                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2974
2975         if (data[IFLA_VXLAN_LABEL])
2976                 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
2977                              IPV6_FLOWLABEL_MASK;
2978
2979         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2980                 conf.flags |= VXLAN_F_LEARN;
2981
2982         if (data[IFLA_VXLAN_AGEING])
2983                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2984
2985         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2986                 conf.flags |= VXLAN_F_PROXY;
2987
2988         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2989                 conf.flags |= VXLAN_F_RSC;
2990
2991         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2992                 conf.flags |= VXLAN_F_L2MISS;
2993
2994         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2995                 conf.flags |= VXLAN_F_L3MISS;
2996
2997         if (data[IFLA_VXLAN_LIMIT])
2998                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2999
3000         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
3001             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
3002                 conf.flags |= VXLAN_F_COLLECT_METADATA;
3003
3004         if (data[IFLA_VXLAN_PORT_RANGE]) {
3005                 const struct ifla_vxlan_port_range *p
3006                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
3007                 conf.port_min = ntohs(p->low);
3008                 conf.port_max = ntohs(p->high);
3009         }
3010
3011         if (data[IFLA_VXLAN_PORT])
3012                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
3013
3014         if (data[IFLA_VXLAN_UDP_CSUM] &&
3015             !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
3016                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
3017
3018         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
3019             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
3020                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
3021
3022         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3023             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3024                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3025
3026         if (data[IFLA_VXLAN_REMCSUM_TX] &&
3027             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3028                 conf.flags |= VXLAN_F_REMCSUM_TX;
3029
3030         if (data[IFLA_VXLAN_REMCSUM_RX] &&
3031             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3032                 conf.flags |= VXLAN_F_REMCSUM_RX;
3033
3034         if (data[IFLA_VXLAN_GBP])
3035                 conf.flags |= VXLAN_F_GBP;
3036
3037         if (data[IFLA_VXLAN_GPE])
3038                 conf.flags |= VXLAN_F_GPE;
3039
3040         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3041                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3042
3043         if (tb[IFLA_MTU])
3044                 conf.mtu = nla_get_u32(tb[IFLA_MTU]);
3045
3046         return vxlan_dev_configure(src_net, dev, &conf);
3047 }
3048
3049 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3050 {
3051         struct vxlan_dev *vxlan = netdev_priv(dev);
3052         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3053
3054         spin_lock(&vn->sock_lock);
3055         if (!hlist_unhashed(&vxlan->hlist))
3056                 hlist_del_rcu(&vxlan->hlist);
3057         spin_unlock(&vn->sock_lock);
3058
3059         gro_cells_destroy(&vxlan->gro_cells);
3060         list_del(&vxlan->next);
3061         unregister_netdevice_queue(dev, head);
3062 }
3063
3064 static size_t vxlan_get_size(const struct net_device *dev)
3065 {
3066
3067         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3068                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3069                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3070                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3071                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3072                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3073                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3074                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3075                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3076                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3077                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3078                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3079                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3080                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3081                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3082                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3083                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3084                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3085                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3086                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3087                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3088                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3089                 0;
3090 }
3091
3092 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3093 {
3094         const struct vxlan_dev *vxlan = netdev_priv(dev);
3095         const struct vxlan_rdst *dst = &vxlan->default_dst;
3096         struct ifla_vxlan_port_range ports = {
3097                 .low =  htons(vxlan->cfg.port_min),
3098                 .high = htons(vxlan->cfg.port_max),
3099         };
3100
3101         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3102                 goto nla_put_failure;
3103
3104         if (!vxlan_addr_any(&dst->remote_ip)) {
3105                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3106                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3107                                             dst->remote_ip.sin.sin_addr.s_addr))
3108                                 goto nla_put_failure;
3109 #if IS_ENABLED(CONFIG_IPV6)
3110                 } else {
3111                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3112                                              &dst->remote_ip.sin6.sin6_addr))
3113                                 goto nla_put_failure;
3114 #endif
3115                 }
3116         }
3117
3118         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3119                 goto nla_put_failure;
3120
3121         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3122                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3123                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3124                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3125                                 goto nla_put_failure;
3126 #if IS_ENABLED(CONFIG_IPV6)
3127                 } else {
3128                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3129                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3130                                 goto nla_put_failure;
3131 #endif
3132                 }
3133         }
3134
3135         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3136             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3137             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3138             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3139                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3140             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3141                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3142             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3143             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3144                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3145             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3146                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3147             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3148                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3149             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3150             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3151             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3152             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3153                         !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3154             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3155                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3156             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3157                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3158             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3159                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3160             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3161                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3162                 goto nla_put_failure;
3163
3164         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3165                 goto nla_put_failure;
3166
3167         if (vxlan->flags & VXLAN_F_GBP &&
3168             nla_put_flag(skb, IFLA_VXLAN_GBP))
3169                 goto nla_put_failure;
3170
3171         if (vxlan->flags & VXLAN_F_GPE &&
3172             nla_put_flag(skb, IFLA_VXLAN_GPE))
3173                 goto nla_put_failure;
3174
3175         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3176             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3177                 goto nla_put_failure;
3178
3179         return 0;
3180
3181 nla_put_failure:
3182         return -EMSGSIZE;
3183 }
3184
3185 static struct net *vxlan_get_link_net(const struct net_device *dev)
3186 {
3187         struct vxlan_dev *vxlan = netdev_priv(dev);
3188
3189         return vxlan->net;
3190 }
3191
3192 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3193         .kind           = "vxlan",
3194         .maxtype        = IFLA_VXLAN_MAX,
3195         .policy         = vxlan_policy,
3196         .priv_size      = sizeof(struct vxlan_dev),
3197         .setup          = vxlan_setup,
3198         .validate       = vxlan_validate,
3199         .newlink        = vxlan_newlink,
3200         .dellink        = vxlan_dellink,
3201         .get_size       = vxlan_get_size,
3202         .fill_info      = vxlan_fill_info,
3203         .get_link_net   = vxlan_get_link_net,
3204 };
3205
3206 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3207                                     u8 name_assign_type,
3208                                     struct vxlan_config *conf)
3209 {
3210         struct nlattr *tb[IFLA_MAX + 1];
3211         struct net_device *dev;
3212         int err;
3213
3214         memset(&tb, 0, sizeof(tb));
3215
3216         dev = rtnl_create_link(net, name, name_assign_type,
3217                                &vxlan_link_ops, tb);
3218         if (IS_ERR(dev))
3219                 return dev;
3220
3221         err = vxlan_dev_configure(net, dev, conf);
3222         if (err < 0) {
3223                 free_netdev(dev);
3224                 return ERR_PTR(err);
3225         }
3226
3227         err = rtnl_configure_link(dev, NULL);
3228         if (err < 0) {
3229                 LIST_HEAD(list_kill);
3230
3231                 vxlan_dellink(dev, &list_kill);
3232                 unregister_netdevice_many(&list_kill);
3233                 return ERR_PTR(err);
3234         }
3235
3236         return dev;
3237 }
3238 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3239
3240 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3241                                              struct net_device *dev)
3242 {
3243         struct vxlan_dev *vxlan, *next;
3244         LIST_HEAD(list_kill);
3245
3246         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3247                 struct vxlan_rdst *dst = &vxlan->default_dst;
3248
3249                 /* In case we created vxlan device with carrier
3250                  * and we loose the carrier due to module unload
3251                  * we also need to remove vxlan device. In other
3252                  * cases, it's not necessary and remote_ifindex
3253                  * is 0 here, so no matches.
3254                  */
3255                 if (dst->remote_ifindex == dev->ifindex)
3256                         vxlan_dellink(vxlan->dev, &list_kill);
3257         }
3258
3259         unregister_netdevice_many(&list_kill);
3260 }
3261
3262 static int vxlan_netdevice_event(struct notifier_block *unused,
3263                                  unsigned long event, void *ptr)
3264 {
3265         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3266         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3267
3268         if (event == NETDEV_UNREGISTER)
3269                 vxlan_handle_lowerdev_unregister(vn, dev);
3270         else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3271                 vxlan_push_rx_ports(dev);
3272
3273         return NOTIFY_DONE;
3274 }
3275
3276 static struct notifier_block vxlan_notifier_block __read_mostly = {
3277         .notifier_call = vxlan_netdevice_event,
3278 };
3279
3280 static __net_init int vxlan_init_net(struct net *net)
3281 {
3282         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3283         unsigned int h;
3284
3285         INIT_LIST_HEAD(&vn->vxlan_list);
3286         spin_lock_init(&vn->sock_lock);
3287
3288         for (h = 0; h < PORT_HASH_SIZE; ++h)
3289                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3290
3291         return 0;
3292 }
3293
3294 static void __net_exit vxlan_exit_net(struct net *net)
3295 {
3296         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3297         struct vxlan_dev *vxlan, *next;
3298         struct net_device *dev, *aux;
3299         LIST_HEAD(list);
3300
3301         rtnl_lock();
3302         for_each_netdev_safe(net, dev, aux)
3303                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3304                         unregister_netdevice_queue(dev, &list);
3305
3306         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3307                 /* If vxlan->dev is in the same netns, it has already been added
3308                  * to the list by the previous loop.
3309                  */
3310                 if (!net_eq(dev_net(vxlan->dev), net)) {
3311                         gro_cells_destroy(&vxlan->gro_cells);
3312                         unregister_netdevice_queue(vxlan->dev, &list);
3313                 }
3314         }
3315
3316         unregister_netdevice_many(&list);
3317         rtnl_unlock();
3318 }
3319
3320 static struct pernet_operations vxlan_net_ops = {
3321         .init = vxlan_init_net,
3322         .exit = vxlan_exit_net,
3323         .id   = &vxlan_net_id,
3324         .size = sizeof(struct vxlan_net),
3325 };
3326
3327 static int __init vxlan_init_module(void)
3328 {
3329         int rc;
3330
3331         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3332
3333         rc = register_pernet_subsys(&vxlan_net_ops);
3334         if (rc)
3335                 goto out1;
3336
3337         rc = register_netdevice_notifier(&vxlan_notifier_block);
3338         if (rc)
3339                 goto out2;
3340
3341         rc = rtnl_link_register(&vxlan_link_ops);
3342         if (rc)
3343                 goto out3;
3344
3345         return 0;
3346 out3:
3347         unregister_netdevice_notifier(&vxlan_notifier_block);
3348 out2:
3349         unregister_pernet_subsys(&vxlan_net_ops);
3350 out1:
3351         return rc;
3352 }
3353 late_initcall(vxlan_init_module);
3354
3355 static void __exit vxlan_cleanup_module(void)
3356 {
3357         rtnl_link_unregister(&vxlan_link_ops);
3358         unregister_netdevice_notifier(&vxlan_notifier_block);
3359         unregister_pernet_subsys(&vxlan_net_ops);
3360         /* rcu_barrier() is called by netns */
3361 }
3362 module_exit(vxlan_cleanup_module);
3363
3364 MODULE_LICENSE("GPL");
3365 MODULE_VERSION(VXLAN_VERSION);
3366 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3367 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3368 MODULE_ALIAS_RTNL_LINK("vxlan");