drivers/net/vxlan.c

   1 /*
   2  * VXLAN: Virtual eXtensible Local Area Network
   3  *
   4  * Copyright (c) 2012-2013 Vyatta Inc.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  12
  13 #include <linux/kernel.h>
  14 #include <linux/module.h>
  15 #include <linux/errno.h>
  16 #include <linux/slab.h>
  17 #include <linux/udp.h>
  18 #include <linux/igmp.h>
  19 #include <linux/if_ether.h>
  20 #include <linux/ethtool.h>
  21 #include <net/arp.h>
  22 #include <net/ndisc.h>
  23 #include <net/ip.h>
  24 #include <net/icmp.h>
  25 #include <net/rtnetlink.h>
  26 #include <net/inet_ecn.h>
  27 #include <net/net_namespace.h>
  28 #include <net/netns/generic.h>
  29 #include <net/vxlan.h>
  30
  31 #if IS_ENABLED(CONFIG_IPV6)
  32 #include <net/ip6_tunnel.h>
  33 #include <net/ip6_checksum.h>
  34 #endif
  35
  36 #define VXLAN_VERSION   "0.1"
  37
  38 #define PORT_HASH_BITS  8
  39 #define PORT_HASH_SIZE  (1<<PORT_HASH_BITS)
  40 #define FDB_AGE_DEFAULT 300 /* 5 min */
  41 #define FDB_AGE_INTERVAL (10 * HZ)      /* rescan interval */
  42
  43 /* UDP port for VXLAN traffic.
  44  * The IANA assigned port is 4789, but the Linux default is 8472
  45  * for compatibility with early adopters.
  46  */
  47 static unsigned short vxlan_port __read_mostly = 8472;
  48 module_param_named(udp_port, vxlan_port, ushort, 0444);
  49 MODULE_PARM_DESC(udp_port, "Destination UDP port");
  50
  51 static bool log_ecn_error = true;
  52 module_param(log_ecn_error, bool, 0644);
  53 MODULE_PARM_DESC(log_ecn_error, "Log packets received with corrupted ECN");
  54
  55 static int vxlan_net_id;
  56 static struct rtnl_link_ops vxlan_link_ops;
  57
  58 static const u8 all_zeros_mac[ETH_ALEN + 2];
  59
  60 static int vxlan_sock_add(struct vxlan_dev *vxlan);
  61
  62 /* per-network namespace private data for this module */
  63 struct vxlan_net {
  64         struct list_head  vxlan_list;
  65         struct hlist_head sock_list[PORT_HASH_SIZE];
  66         spinlock_t        sock_lock;
  67 };
  68
  69 /* Forwarding table entry */
  70 struct vxlan_fdb {
  71         struct hlist_node hlist;        /* linked list of entries */
  72         struct rcu_head   rcu;
  73         unsigned long     updated;      /* jiffies */
  74         unsigned long     used;
  75         struct list_head  remotes;
  76         u8                eth_addr[ETH_ALEN];
  77         u16               state;        /* see ndm_state */
  78         u8                flags;        /* see ndm_flags */
  79 };
  80
  81 /* salt for hash table */
  82 static u32 vxlan_salt __read_mostly;
  83
  84 static inline bool vxlan_collect_metadata(struct vxlan_sock *vs)
  85 {
  86         return vs->flags & VXLAN_F_COLLECT_METADATA ||
  87                ip_tunnel_collect_metadata();
  88 }
  89
  90 #if IS_ENABLED(CONFIG_IPV6)
  91 static inline
  92 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
  93 {
  94         if (a->sa.sa_family != b->sa.sa_family)
  95                 return false;
  96         if (a->sa.sa_family == AF_INET6)
  97                 return ipv6_addr_equal(&a->sin6.sin6_addr, &b->sin6.sin6_addr);
  98         else
  99                 return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 100 }
 101
 102 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 103 {
 104         if (ipa->sa.sa_family == AF_INET6)
 105                 return ipv6_addr_any(&ipa->sin6.sin6_addr);
 106         else
 107                 return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 108 }
 109
 110 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 111 {
 112         if (ipa->sa.sa_family == AF_INET6)
 113                 return ipv6_addr_is_multicast(&ipa->sin6.sin6_addr);
 114         else
 115                 return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 116 }
 117
 118 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 119 {
 120         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 121                 ip->sin6.sin6_addr = nla_get_in6_addr(nla);
 122                 ip->sa.sa_family = AF_INET6;
 123                 return 0;
 124         } else if (nla_len(nla) >= sizeof(__be32)) {
 125                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 126                 ip->sa.sa_family = AF_INET;
 127                 return 0;
 128         } else {
 129                 return -EAFNOSUPPORT;
 130         }
 131 }
 132
 133 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 134                               const union vxlan_addr *ip)
 135 {
 136         if (ip->sa.sa_family == AF_INET6)
 137                 return nla_put_in6_addr(skb, attr, &ip->sin6.sin6_addr);
 138         else
 139                 return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 140 }
 141
 142 #else /* !CONFIG_IPV6 */
 143
 144 static inline
 145 bool vxlan_addr_equal(const union vxlan_addr *a, const union vxlan_addr *b)
 146 {
 147         return a->sin.sin_addr.s_addr == b->sin.sin_addr.s_addr;
 148 }
 149
 150 static inline bool vxlan_addr_any(const union vxlan_addr *ipa)
 151 {
 152         return ipa->sin.sin_addr.s_addr == htonl(INADDR_ANY);
 153 }
 154
 155 static inline bool vxlan_addr_multicast(const union vxlan_addr *ipa)
 156 {
 157         return IN_MULTICAST(ntohl(ipa->sin.sin_addr.s_addr));
 158 }
 159
 160 static int vxlan_nla_get_addr(union vxlan_addr *ip, struct nlattr *nla)
 161 {
 162         if (nla_len(nla) >= sizeof(struct in6_addr)) {
 163                 return -EAFNOSUPPORT;
 164         } else if (nla_len(nla) >= sizeof(__be32)) {
 165                 ip->sin.sin_addr.s_addr = nla_get_in_addr(nla);
 166                 ip->sa.sa_family = AF_INET;
 167                 return 0;
 168         } else {
 169                 return -EAFNOSUPPORT;
 170         }
 171 }
 172
 173 static int vxlan_nla_put_addr(struct sk_buff *skb, int attr,
 174                               const union vxlan_addr *ip)
 175 {
 176         return nla_put_in_addr(skb, attr, ip->sin.sin_addr.s_addr);
 177 }
 178 #endif
 179
 180 /* Virtual Network hash table head */
 181 static inline struct hlist_head *vni_head(struct vxlan_sock *vs, __be32 vni)
 182 {
 183         return &vs->vni_list[hash_32((__force u32)vni, VNI_HASH_BITS)];
 184 }
 185
 186 /* Socket hash table head */
 187 static inline struct hlist_head *vs_head(struct net *net, __be16 port)
 188 {
 189         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
 190
 191         return &vn->sock_list[hash_32(ntohs(port), PORT_HASH_BITS)];
 192 }
 193
 194 /* First remote destination for a forwarding entry.
 195  * Guaranteed to be non-NULL because remotes are never deleted.
 196  */
 197 static inline struct vxlan_rdst *first_remote_rcu(struct vxlan_fdb *fdb)
 198 {
 199         return list_entry_rcu(fdb->remotes.next, struct vxlan_rdst, list);
 200 }
 201
 202 static inline struct vxlan_rdst *first_remote_rtnl(struct vxlan_fdb *fdb)
 203 {
 204         return list_first_entry(&fdb->remotes, struct vxlan_rdst, list);
 205 }
 206
 207 /* Find VXLAN socket based on network namespace, address family and UDP port
 208  * and enabled unshareable flags.
 209  */
 210 static struct vxlan_sock *vxlan_find_sock(struct net *net, sa_family_t family,
 211                                           __be16 port, u32 flags)
 212 {
 213         struct vxlan_sock *vs;
 214
 215         flags &= VXLAN_F_RCV_FLAGS;
 216
 217         hlist_for_each_entry_rcu(vs, vs_head(net, port), hlist) {
 218                 if (inet_sk(vs->sock->sk)->inet_sport == port &&
 219                     vxlan_get_sk_family(vs) == family &&
 220                     vs->flags == flags)
 221                         return vs;
 222         }
 223         return NULL;
 224 }
 225
 226 static struct vxlan_dev *vxlan_vs_find_vni(struct vxlan_sock *vs, __be32 vni)
 227 {
 228         struct vxlan_dev *vxlan;
 229
 230         /* For flow based devices, map all packets to VNI 0 */
 231         if (vs->flags & VXLAN_F_COLLECT_METADATA)
 232                 vni = 0;
 233
 234         hlist_for_each_entry_rcu(vxlan, vni_head(vs, vni), hlist) {
 235                 if (vxlan->default_dst.remote_vni == vni)
 236                         return vxlan;
 237         }
 238
 239         return NULL;
 240 }
 241
 242 /* Look up VNI in a per net namespace table */
 243 static struct vxlan_dev *vxlan_find_vni(struct net *net, __be32 vni,
 244                                         sa_family_t family, __be16 port,
 245                                         u32 flags)
 246 {
 247         struct vxlan_sock *vs;
 248
 249         vs = vxlan_find_sock(net, family, port, flags);
 250         if (!vs)
 251                 return NULL;
 252
 253         return vxlan_vs_find_vni(vs, vni);
 254 }
 255
 256 /* Fill in neighbour message in skbuff. */
 257 static int vxlan_fdb_info(struct sk_buff *skb, struct vxlan_dev *vxlan,
 258                           const struct vxlan_fdb *fdb,
 259                           u32 portid, u32 seq, int type, unsigned int flags,
 260                           const struct vxlan_rdst *rdst)
 261 {
 262         unsigned long now = jiffies;
 263         struct nda_cacheinfo ci;
 264         struct nlmsghdr *nlh;
 265         struct ndmsg *ndm;
 266         bool send_ip, send_eth;
 267
 268         nlh = nlmsg_put(skb, portid, seq, type, sizeof(*ndm), flags);
 269         if (nlh == NULL)
 270                 return -EMSGSIZE;
 271
 272         ndm = nlmsg_data(nlh);
 273         memset(ndm, 0, sizeof(*ndm));
 274
 275         send_eth = send_ip = true;
 276
 277         if (type == RTM_GETNEIGH) {
 278                 ndm->ndm_family = AF_INET;
 279                 send_ip = !vxlan_addr_any(&rdst->remote_ip);
 280                 send_eth = !is_zero_ether_addr(fdb->eth_addr);
 281         } else
 282                 ndm->ndm_family = AF_BRIDGE;
 283         ndm->ndm_state = fdb->state;
 284         ndm->ndm_ifindex = vxlan->dev->ifindex;
 285         ndm->ndm_flags = fdb->flags;
 286         ndm->ndm_type = RTN_UNICAST;
 287
 288         if (!net_eq(dev_net(vxlan->dev), vxlan->net) &&
 289             nla_put_s32(skb, NDA_LINK_NETNSID,
 290                         peernet2id(dev_net(vxlan->dev), vxlan->net)))
 291                 goto nla_put_failure;
 292
 293         if (send_eth && nla_put(skb, NDA_LLADDR, ETH_ALEN, &fdb->eth_addr))
 294                 goto nla_put_failure;
 295
 296         if (send_ip && vxlan_nla_put_addr(skb, NDA_DST, &rdst->remote_ip))
 297                 goto nla_put_failure;
 298
 299         if (rdst->remote_port && rdst->remote_port != vxlan->cfg.dst_port &&
 300             nla_put_be16(skb, NDA_PORT, rdst->remote_port))
 301                 goto nla_put_failure;
 302         if (rdst->remote_vni != vxlan->default_dst.remote_vni &&
 303             nla_put_u32(skb, NDA_VNI, be32_to_cpu(rdst->remote_vni)))
 304                 goto nla_put_failure;
 305         if (rdst->remote_ifindex &&
 306             nla_put_u32(skb, NDA_IFINDEX, rdst->remote_ifindex))
 307                 goto nla_put_failure;
 308
 309         ci.ndm_used      = jiffies_to_clock_t(now - fdb->used);
 310         ci.ndm_confirmed = 0;
 311         ci.ndm_updated   = jiffies_to_clock_t(now - fdb->updated);
 312         ci.ndm_refcnt    = 0;
 313
 314         if (nla_put(skb, NDA_CACHEINFO, sizeof(ci), &ci))
 315                 goto nla_put_failure;
 316
 317         nlmsg_end(skb, nlh);
 318         return 0;
 319
 320 nla_put_failure:
 321         nlmsg_cancel(skb, nlh);
 322         return -EMSGSIZE;
 323 }
 324
 325 static inline size_t vxlan_nlmsg_size(void)
 326 {
 327         return NLMSG_ALIGN(sizeof(struct ndmsg))
 328                 + nla_total_size(ETH_ALEN) /* NDA_LLADDR */
 329                 + nla_total_size(sizeof(struct in6_addr)) /* NDA_DST */
 330                 + nla_total_size(sizeof(__be16)) /* NDA_PORT */
 331                 + nla_total_size(sizeof(__be32)) /* NDA_VNI */
 332                 + nla_total_size(sizeof(__u32)) /* NDA_IFINDEX */
 333                 + nla_total_size(sizeof(__s32)) /* NDA_LINK_NETNSID */
 334                 + nla_total_size(sizeof(struct nda_cacheinfo));
 335 }
 336
 337 static void vxlan_fdb_notify(struct vxlan_dev *vxlan, struct vxlan_fdb *fdb,
 338                              struct vxlan_rdst *rd, int type)
 339 {
 340         struct net *net = dev_net(vxlan->dev);
 341         struct sk_buff *skb;
 342         int err = -ENOBUFS;
 343
 344         skb = nlmsg_new(vxlan_nlmsg_size(), GFP_ATOMIC);
 345         if (skb == NULL)
 346                 goto errout;
 347
 348         err = vxlan_fdb_info(skb, vxlan, fdb, 0, 0, type, 0, rd);
 349         if (err < 0) {
 350                 /* -EMSGSIZE implies BUG in vxlan_nlmsg_size() */
 351                 WARN_ON(err == -EMSGSIZE);
 352                 kfree_skb(skb);
 353                 goto errout;
 354         }
 355
 356         rtnl_notify(skb, net, 0, RTNLGRP_NEIGH, NULL, GFP_ATOMIC);
 357         return;
 358 errout:
 359         if (err < 0)
 360                 rtnl_set_sk_err(net, RTNLGRP_NEIGH, err);
 361 }
 362
 363 static void vxlan_ip_miss(struct net_device *dev, union vxlan_addr *ipa)
 364 {
 365         struct vxlan_dev *vxlan = netdev_priv(dev);
 366         struct vxlan_fdb f = {
 367                 .state = NUD_STALE,
 368         };
 369         struct vxlan_rdst remote = {
 370                 .remote_ip = *ipa, /* goes to NDA_DST */
 371                 .remote_vni = cpu_to_be32(VXLAN_N_VID),
 372         };
 373
 374         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 375 }
 376
 377 static void vxlan_fdb_miss(struct vxlan_dev *vxlan, const u8 eth_addr[ETH_ALEN])
 378 {
 379         struct vxlan_fdb f = {
 380                 .state = NUD_STALE,
 381         };
 382         struct vxlan_rdst remote = { };
 383
 384         memcpy(f.eth_addr, eth_addr, ETH_ALEN);
 385
 386         vxlan_fdb_notify(vxlan, &f, &remote, RTM_GETNEIGH);
 387 }
 388
 389 /* Hash Ethernet address */
 390 static u32 eth_hash(const unsigned char *addr)
 391 {
 392         u64 value = get_unaligned((u64 *)addr);
 393
 394         /* only want 6 bytes */
 395 #ifdef __BIG_ENDIAN
 396         value >>= 16;
 397 #else
 398         value <<= 16;
 399 #endif
 400         return hash_64(value, FDB_HASH_BITS);
 401 }
 402
 403 /* Hash chain to use given mac address */
 404 static inline struct hlist_head *vxlan_fdb_head(struct vxlan_dev *vxlan,
 405                                                 const u8 *mac)
 406 {
 407         return &vxlan->fdb_head[eth_hash(mac)];
 408 }
 409
 410 /* Look up Ethernet address in forwarding table */
 411 static struct vxlan_fdb *__vxlan_find_mac(struct vxlan_dev *vxlan,
 412                                         const u8 *mac)
 413 {
 414         struct hlist_head *head = vxlan_fdb_head(vxlan, mac);
 415         struct vxlan_fdb *f;
 416
 417         hlist_for_each_entry_rcu(f, head, hlist) {
 418                 if (ether_addr_equal(mac, f->eth_addr))
 419                         return f;
 420         }
 421
 422         return NULL;
 423 }
 424
 425 static struct vxlan_fdb *vxlan_find_mac(struct vxlan_dev *vxlan,
 426                                         const u8 *mac)
 427 {
 428         struct vxlan_fdb *f;
 429
 430         f = __vxlan_find_mac(vxlan, mac);
 431         if (f)
 432                 f->used = jiffies;
 433
 434         return f;
 435 }
 436
 437 /* caller should hold vxlan->hash_lock */
 438 static struct vxlan_rdst *vxlan_fdb_find_rdst(struct vxlan_fdb *f,
 439                                               union vxlan_addr *ip, __be16 port,
 440                                               __be32 vni, __u32 ifindex)
 441 {
 442         struct vxlan_rdst *rd;
 443
 444         list_for_each_entry(rd, &f->remotes, list) {
 445                 if (vxlan_addr_equal(&rd->remote_ip, ip) &&
 446                     rd->remote_port == port &&
 447                     rd->remote_vni == vni &&
 448                     rd->remote_ifindex == ifindex)
 449                         return rd;
 450         }
 451
 452         return NULL;
 453 }
 454
 455 /* Replace destination of unicast mac */
 456 static int vxlan_fdb_replace(struct vxlan_fdb *f,
 457                              union vxlan_addr *ip, __be16 port, __be32 vni,
 458                              __u32 ifindex)
 459 {
 460         struct vxlan_rdst *rd;
 461
 462         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 463         if (rd)
 464                 return 0;
 465
 466         rd = list_first_entry_or_null(&f->remotes, struct vxlan_rdst, list);
 467         if (!rd)
 468                 return 0;
 469
 470         dst_cache_reset(&rd->dst_cache);
 471         rd->remote_ip = *ip;
 472         rd->remote_port = port;
 473         rd->remote_vni = vni;
 474         rd->remote_ifindex = ifindex;
 475         return 1;
 476 }
 477
 478 /* Add/update destinations for multicast */
 479 static int vxlan_fdb_append(struct vxlan_fdb *f,
 480                             union vxlan_addr *ip, __be16 port, __be32 vni,
 481                             __u32 ifindex, struct vxlan_rdst **rdp)
 482 {
 483         struct vxlan_rdst *rd;
 484
 485         rd = vxlan_fdb_find_rdst(f, ip, port, vni, ifindex);
 486         if (rd)
 487                 return 0;
 488
 489         rd = kmalloc(sizeof(*rd), GFP_ATOMIC);
 490         if (rd == NULL)
 491                 return -ENOBUFS;
 492
 493         if (dst_cache_init(&rd->dst_cache, GFP_ATOMIC)) {
 494                 kfree(rd);
 495                 return -ENOBUFS;
 496         }
 497
 498         rd->remote_ip = *ip;
 499         rd->remote_port = port;
 500         rd->remote_vni = vni;
 501         rd->remote_ifindex = ifindex;
 502
 503         list_add_tail_rcu(&rd->list, &f->remotes);
 504
 505         *rdp = rd;
 506         return 1;
 507 }
 508
 509 static struct vxlanhdr *vxlan_gro_remcsum(struct sk_buff *skb,
 510                                           unsigned int off,
 511                                           struct vxlanhdr *vh, size_t hdrlen,
 512                                           __be32 vni_field,
 513                                           struct gro_remcsum *grc,
 514                                           bool nopartial)
 515 {
 516         size_t start, offset;
 517
 518         if (skb->remcsum_offload)
 519                 return vh;
 520
 521         if (!NAPI_GRO_CB(skb)->csum_valid)
 522                 return NULL;
 523
 524         start = vxlan_rco_start(vni_field);
 525         offset = start + vxlan_rco_offset(vni_field);
 526
 527         vh = skb_gro_remcsum_process(skb, (void *)vh, off, hdrlen,
 528                                      start, offset, grc, nopartial);
 529
 530         skb->remcsum_offload = 1;
 531
 532         return vh;
 533 }
 534
 535 static struct sk_buff **vxlan_gro_receive(struct sock *sk,
 536                                           struct sk_buff **head,
 537                                           struct sk_buff *skb)
 538 {
 539         struct sk_buff *p, **pp = NULL;
 540         struct vxlanhdr *vh, *vh2;
 541         unsigned int hlen, off_vx;
 542         int flush = 1;
 543         struct vxlan_sock *vs = rcu_dereference_sk_user_data(sk);
 544         __be32 flags;
 545         struct gro_remcsum grc;
 546
 547         skb_gro_remcsum_init(&grc);
 548
 549         off_vx = skb_gro_offset(skb);
 550         hlen = off_vx + sizeof(*vh);
 551         vh   = skb_gro_header_fast(skb, off_vx);
 552         if (skb_gro_header_hard(skb, hlen)) {
 553                 vh = skb_gro_header_slow(skb, hlen, off_vx);
 554                 if (unlikely(!vh))
 555                         goto out;
 556         }
 557
 558         skb_gro_postpull_rcsum(skb, vh, sizeof(struct vxlanhdr));
 559
 560         flags = vh->vx_flags;
 561
 562         if ((flags & VXLAN_HF_RCO) && (vs->flags & VXLAN_F_REMCSUM_RX)) {
 563                 vh = vxlan_gro_remcsum(skb, off_vx, vh, sizeof(struct vxlanhdr),
 564                                        vh->vx_vni, &grc,
 565                                        !!(vs->flags &
 566                                           VXLAN_F_REMCSUM_NOPARTIAL));
 567
 568                 if (!vh)
 569                         goto out;
 570         }
 571
 572         skb_gro_pull(skb, sizeof(struct vxlanhdr)); /* pull vxlan header */
 573
 574         for (p = *head; p; p = p->next) {
 575                 if (!NAPI_GRO_CB(p)->same_flow)
 576                         continue;
 577
 578                 vh2 = (struct vxlanhdr *)(p->data + off_vx);
 579                 if (vh->vx_flags != vh2->vx_flags ||
 580                     vh->vx_vni != vh2->vx_vni) {
 581                         NAPI_GRO_CB(p)->same_flow = 0;
 582                         continue;
 583                 }
 584         }
 585
 586         pp = call_gro_receive(eth_gro_receive, head, skb);
 587         flush = 0;
 588
 589 out:
 590         skb_gro_remcsum_cleanup(skb, &grc);
 591         NAPI_GRO_CB(skb)->flush |= flush;
 592
 593         return pp;
 594 }
 595
 596 static int vxlan_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
 597 {
 598         /* Sets 'skb->inner_mac_header' since we are always called with
 599          * 'skb->encapsulation' set.
 600          */
 601         return eth_gro_complete(skb, nhoff + sizeof(struct vxlanhdr));
 602 }
 603
 604 /* Add new entry to forwarding table -- assumes lock held */
 605 static int vxlan_fdb_create(struct vxlan_dev *vxlan,
 606                             const u8 *mac, union vxlan_addr *ip,
 607                             __u16 state, __u16 flags,
 608                             __be16 port, __be32 vni, __u32 ifindex,
 609                             __u8 ndm_flags)
 610 {
 611         struct vxlan_rdst *rd = NULL;
 612         struct vxlan_fdb *f;
 613         int notify = 0;
 614
 615         f = __vxlan_find_mac(vxlan, mac);
 616         if (f) {
 617                 if (flags & NLM_F_EXCL) {
 618                         netdev_dbg(vxlan->dev,
 619                                    "lost race to create %pM\n", mac);
 620                         return -EEXIST;
 621                 }
 622                 if (f->state != state) {
 623                         f->state = state;
 624                         f->updated = jiffies;
 625                         notify = 1;
 626                 }
 627                 if (f->flags != ndm_flags) {
 628                         f->flags = ndm_flags;
 629                         f->updated = jiffies;
 630                         notify = 1;
 631                 }
 632                 if ((flags & NLM_F_REPLACE)) {
 633                         /* Only change unicasts */
 634                         if (!(is_multicast_ether_addr(f->eth_addr) ||
 635                              is_zero_ether_addr(f->eth_addr))) {
 636                                 notify |= vxlan_fdb_replace(f, ip, port, vni,
 637                                                            ifindex);
 638                         } else
 639                                 return -EOPNOTSUPP;
 640                 }
 641                 if ((flags & NLM_F_APPEND) &&
 642                     (is_multicast_ether_addr(f->eth_addr) ||
 643                      is_zero_ether_addr(f->eth_addr))) {
 644                         int rc = vxlan_fdb_append(f, ip, port, vni, ifindex,
 645                                                   &rd);
 646
 647                         if (rc < 0)
 648                                 return rc;
 649                         notify |= rc;
 650                 }
 651         } else {
 652                 if (!(flags & NLM_F_CREATE))
 653                         return -ENOENT;
 654
 655                 if (vxlan->cfg.addrmax &&
 656                     vxlan->addrcnt >= vxlan->cfg.addrmax)
 657                         return -ENOSPC;
 658
 659                 /* Disallow replace to add a multicast entry */
 660                 if ((flags & NLM_F_REPLACE) &&
 661                     (is_multicast_ether_addr(mac) || is_zero_ether_addr(mac)))
 662                         return -EOPNOTSUPP;
 663
 664                 netdev_dbg(vxlan->dev, "add %pM -> %pIS\n", mac, ip);
 665                 f = kmalloc(sizeof(*f), GFP_ATOMIC);
 666                 if (!f)
 667                         return -ENOMEM;
 668
 669                 notify = 1;
 670                 f->state = state;
 671                 f->flags = ndm_flags;
 672                 f->updated = f->used = jiffies;
 673                 INIT_LIST_HEAD(&f->remotes);
 674                 memcpy(f->eth_addr, mac, ETH_ALEN);
 675
 676                 vxlan_fdb_append(f, ip, port, vni, ifindex, &rd);
 677
 678                 ++vxlan->addrcnt;
 679                 hlist_add_head_rcu(&f->hlist,
 680                                    vxlan_fdb_head(vxlan, mac));
 681         }
 682
 683         if (notify) {
 684                 if (rd == NULL)
 685                         rd = first_remote_rtnl(f);
 686                 vxlan_fdb_notify(vxlan, f, rd, RTM_NEWNEIGH);
 687         }
 688
 689         return 0;
 690 }
 691
 692 static void vxlan_fdb_free(struct rcu_head *head)
 693 {
 694         struct vxlan_fdb *f = container_of(head, struct vxlan_fdb, rcu);
 695         struct vxlan_rdst *rd, *nd;
 696
 697         list_for_each_entry_safe(rd, nd, &f->remotes, list) {
 698                 dst_cache_destroy(&rd->dst_cache);
 699                 kfree(rd);
 700         }
 701         kfree(f);
 702 }
 703
 704 static void vxlan_fdb_destroy(struct vxlan_dev *vxlan, struct vxlan_fdb *f)
 705 {
 706         netdev_dbg(vxlan->dev,
 707                     "delete %pM\n", f->eth_addr);
 708
 709         --vxlan->addrcnt;
 710         vxlan_fdb_notify(vxlan, f, first_remote_rtnl(f), RTM_DELNEIGH);
 711
 712         hlist_del_rcu(&f->hlist);
 713         call_rcu(&f->rcu, vxlan_fdb_free);
 714 }
 715
 716 static int vxlan_fdb_parse(struct nlattr *tb[], struct vxlan_dev *vxlan,
 717                            union vxlan_addr *ip, __be16 *port, __be32 *vni,
 718                            u32 *ifindex)
 719 {
 720         struct net *net = dev_net(vxlan->dev);
 721         int err;
 722
 723         if (tb[NDA_DST]) {
 724                 err = vxlan_nla_get_addr(ip, tb[NDA_DST]);
 725                 if (err)
 726                         return err;
 727         } else {
 728                 union vxlan_addr *remote = &vxlan->default_dst.remote_ip;
 729                 if (remote->sa.sa_family == AF_INET) {
 730                         ip->sin.sin_addr.s_addr = htonl(INADDR_ANY);
 731                         ip->sa.sa_family = AF_INET;
 732 #if IS_ENABLED(CONFIG_IPV6)
 733                 } else {
 734                         ip->sin6.sin6_addr = in6addr_any;
 735                         ip->sa.sa_family = AF_INET6;
 736 #endif
 737                 }
 738         }
 739
 740         if (tb[NDA_PORT]) {
 741                 if (nla_len(tb[NDA_PORT]) != sizeof(__be16))
 742                         return -EINVAL;
 743                 *port = nla_get_be16(tb[NDA_PORT]);
 744         } else {
 745                 *port = vxlan->cfg.dst_port;
 746         }
 747
 748         if (tb[NDA_VNI]) {
 749                 if (nla_len(tb[NDA_VNI]) != sizeof(u32))
 750                         return -EINVAL;
 751                 *vni = cpu_to_be32(nla_get_u32(tb[NDA_VNI]));
 752         } else {
 753                 *vni = vxlan->default_dst.remote_vni;
 754         }
 755
 756         if (tb[NDA_IFINDEX]) {
 757                 struct net_device *tdev;
 758
 759                 if (nla_len(tb[NDA_IFINDEX]) != sizeof(u32))
 760                         return -EINVAL;
 761                 *ifindex = nla_get_u32(tb[NDA_IFINDEX]);
 762                 tdev = __dev_get_by_index(net, *ifindex);
 763                 if (!tdev)
 764                         return -EADDRNOTAVAIL;
 765         } else {
 766                 *ifindex = 0;
 767         }
 768
 769         return 0;
 770 }
 771
 772 /* Add static entry (via netlink) */
 773 static int vxlan_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
 774                          struct net_device *dev,
 775                          const unsigned char *addr, u16 vid, u16 flags)
 776 {
 777         struct vxlan_dev *vxlan = netdev_priv(dev);
 778         /* struct net *net = dev_net(vxlan->dev); */
 779         union vxlan_addr ip;
 780         __be16 port;
 781         __be32 vni;
 782         u32 ifindex;
 783         int err;
 784
 785         if (!(ndm->ndm_state & (NUD_PERMANENT|NUD_REACHABLE))) {
 786                 pr_info("RTM_NEWNEIGH with invalid state %#x\n",
 787                         ndm->ndm_state);
 788                 return -EINVAL;
 789         }
 790
 791         if (tb[NDA_DST] == NULL)
 792                 return -EINVAL;
 793
 794         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 795         if (err)
 796                 return err;
 797
 798         if (vxlan->default_dst.remote_ip.sa.sa_family != ip.sa.sa_family)
 799                 return -EAFNOSUPPORT;
 800
 801         spin_lock_bh(&vxlan->hash_lock);
 802         err = vxlan_fdb_create(vxlan, addr, &ip, ndm->ndm_state, flags,
 803                                port, vni, ifindex, ndm->ndm_flags);
 804         spin_unlock_bh(&vxlan->hash_lock);
 805
 806         return err;
 807 }
 808
 809 /* Delete entry (via netlink) */
 810 static int vxlan_fdb_delete(struct ndmsg *ndm, struct nlattr *tb[],
 811                             struct net_device *dev,
 812                             const unsigned char *addr, u16 vid)
 813 {
 814         struct vxlan_dev *vxlan = netdev_priv(dev);
 815         struct vxlan_fdb *f;
 816         struct vxlan_rdst *rd = NULL;
 817         union vxlan_addr ip;
 818         __be16 port;
 819         __be32 vni;
 820         u32 ifindex;
 821         int err;
 822
 823         err = vxlan_fdb_parse(tb, vxlan, &ip, &port, &vni, &ifindex);
 824         if (err)
 825                 return err;
 826
 827         err = -ENOENT;
 828
 829         spin_lock_bh(&vxlan->hash_lock);
 830         f = vxlan_find_mac(vxlan, addr);
 831         if (!f)
 832                 goto out;
 833
 834         if (!vxlan_addr_any(&ip)) {
 835                 rd = vxlan_fdb_find_rdst(f, &ip, port, vni, ifindex);
 836                 if (!rd)
 837                         goto out;
 838         }
 839
 840         err = 0;
 841
 842         /* remove a destination if it's not the only one on the list,
 843          * otherwise destroy the fdb entry
 844          */
 845         if (rd && !list_is_singular(&f->remotes)) {
 846                 list_del_rcu(&rd->list);
 847                 vxlan_fdb_notify(vxlan, f, rd, RTM_DELNEIGH);
 848                 kfree_rcu(rd, rcu);
 849                 goto out;
 850         }
 851
 852         vxlan_fdb_destroy(vxlan, f);
 853
 854 out:
 855         spin_unlock_bh(&vxlan->hash_lock);
 856
 857         return err;
 858 }
 859
 860 /* Dump forwarding table */
 861 static int vxlan_fdb_dump(struct sk_buff *skb, struct netlink_callback *cb,
 862                           struct net_device *dev,
 863                           struct net_device *filter_dev, int *idx)
 864 {
 865         struct vxlan_dev *vxlan = netdev_priv(dev);
 866         unsigned int h;
 867         int err = 0;
 868
 869         for (h = 0; h < FDB_HASH_SIZE; ++h) {
 870                 struct vxlan_fdb *f;
 871
 872                 hlist_for_each_entry_rcu(f, &vxlan->fdb_head[h], hlist) {
 873                         struct vxlan_rdst *rd;
 874
 875                         list_for_each_entry_rcu(rd, &f->remotes, list) {
 876                                 if (*idx < cb->args[2])
 877                                         goto skip;
 878
 879                                 err = vxlan_fdb_info(skb, vxlan, f,
 880                                                      NETLINK_CB(cb->skb).portid,
 881                                                      cb->nlh->nlmsg_seq,
 882                                                      RTM_NEWNEIGH,
 883                                                      NLM_F_MULTI, rd);
 884                                 if (err < 0)
 885                                         goto out;
 886 skip:
 887                                 *idx += 1;
 888                         }
 889                 }
 890         }
 891 out:
 892         return err;
 893 }
 894
 895 /* Watch incoming packets to learn mapping between Ethernet address
 896  * and Tunnel endpoint.
 897  * Return true if packet is bogus and should be dropped.
 898  */
 899 static bool vxlan_snoop(struct net_device *dev,
 900                         union vxlan_addr *src_ip, const u8 *src_mac)
 901 {
 902         struct vxlan_dev *vxlan = netdev_priv(dev);
 903         struct vxlan_fdb *f;
 904
 905         f = vxlan_find_mac(vxlan, src_mac);
 906         if (likely(f)) {
 907                 struct vxlan_rdst *rdst = first_remote_rcu(f);
 908
 909                 if (likely(vxlan_addr_equal(&rdst->remote_ip, src_ip)))
 910                         return false;
 911
 912                 /* Don't migrate static entries, drop packets */
 913                 if (f->state & NUD_NOARP)
 914                         return true;
 915
 916                 if (net_ratelimit())
 917                         netdev_info(dev,
 918                                     "%pM migrated from %pIS to %pIS\n",
 919                                     src_mac, &rdst->remote_ip.sa, &src_ip->sa);
 920
 921                 rdst->remote_ip = *src_ip;
 922                 f->updated = jiffies;
 923                 vxlan_fdb_notify(vxlan, f, rdst, RTM_NEWNEIGH);
 924         } else {
 925                 /* learned new entry */
 926                 spin_lock(&vxlan->hash_lock);
 927
 928                 /* close off race between vxlan_flush and incoming packets */
 929                 if (netif_running(dev))
 930                         vxlan_fdb_create(vxlan, src_mac, src_ip,
 931                                          NUD_REACHABLE,
 932                                          NLM_F_EXCL|NLM_F_CREATE,
 933                                          vxlan->cfg.dst_port,
 934                                          vxlan->default_dst.remote_vni,
 935                                          0, NTF_SELF);
 936                 spin_unlock(&vxlan->hash_lock);
 937         }
 938
 939         return false;
 940 }
 941
 942 /* See if multicast group is already in use by other ID */
 943 static bool vxlan_group_used(struct vxlan_net *vn, struct vxlan_dev *dev)
 944 {
 945         struct vxlan_dev *vxlan;
 946         unsigned short family = dev->default_dst.remote_ip.sa.sa_family;
 947
 948         /* The vxlan_sock is only used by dev, leaving group has
 949          * no effect on other vxlan devices.
 950          */
 951         if (family == AF_INET && dev->vn4_sock &&
 952             atomic_read(&dev->vn4_sock->refcnt) == 1)
 953                 return false;
 954 #if IS_ENABLED(CONFIG_IPV6)
 955         if (family == AF_INET6 && dev->vn6_sock &&
 956             atomic_read(&dev->vn6_sock->refcnt) == 1)
 957                 return false;
 958 #endif
 959
 960         list_for_each_entry(vxlan, &vn->vxlan_list, next) {
 961                 if (!netif_running(vxlan->dev) || vxlan == dev)
 962                         continue;
 963
 964                 if (family == AF_INET && vxlan->vn4_sock != dev->vn4_sock)
 965                         continue;
 966 #if IS_ENABLED(CONFIG_IPV6)
 967                 if (family == AF_INET6 && vxlan->vn6_sock != dev->vn6_sock)
 968                         continue;
 969 #endif
 970
 971                 if (!vxlan_addr_equal(&vxlan->default_dst.remote_ip,
 972                                       &dev->default_dst.remote_ip))
 973                         continue;
 974
 975                 if (vxlan->default_dst.remote_ifindex !=
 976                     dev->default_dst.remote_ifindex)
 977                         continue;
 978
 979                 return true;
 980         }
 981
 982         return false;
 983 }
 984
 985 static bool __vxlan_sock_release_prep(struct vxlan_sock *vs)
 986 {
 987         struct vxlan_net *vn;
 988
 989         if (!vs)
 990                 return false;
 991         if (!atomic_dec_and_test(&vs->refcnt))
 992                 return false;
 993
 994         vn = net_generic(sock_net(vs->sock->sk), vxlan_net_id);
 995         spin_lock(&vn->sock_lock);
 996         hlist_del_rcu(&vs->hlist);
 997         udp_tunnel_notify_del_rx_port(vs->sock,
 998                                       (vs->flags & VXLAN_F_GPE) ?
 999                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
1000                                       UDP_TUNNEL_TYPE_VXLAN);
1001         spin_unlock(&vn->sock_lock);
1002
1003         return true;
1004 }
1005
1006 static void vxlan_sock_release(struct vxlan_dev *vxlan)
1007 {
1008         bool ipv4 = __vxlan_sock_release_prep(vxlan->vn4_sock);
1009 #if IS_ENABLED(CONFIG_IPV6)
1010         bool ipv6 = __vxlan_sock_release_prep(vxlan->vn6_sock);
1011 #endif
1012
1013         synchronize_net();
1014
1015         if (ipv4) {
1016                 udp_tunnel_sock_release(vxlan->vn4_sock->sock);
1017                 kfree(vxlan->vn4_sock);
1018         }
1019
1020 #if IS_ENABLED(CONFIG_IPV6)
1021         if (ipv6) {
1022                 udp_tunnel_sock_release(vxlan->vn6_sock->sock);
1023                 kfree(vxlan->vn6_sock);
1024         }
1025 #endif
1026 }
1027
1028 /* Update multicast group membership when first VNI on
1029  * multicast address is brought up
1030  */
1031 static int vxlan_igmp_join(struct vxlan_dev *vxlan)
1032 {
1033         struct sock *sk;
1034         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1035         int ifindex = vxlan->default_dst.remote_ifindex;
1036         int ret = -EINVAL;
1037
1038         if (ip->sa.sa_family == AF_INET) {
1039                 struct ip_mreqn mreq = {
1040                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1041                         .imr_ifindex            = ifindex,
1042                 };
1043
1044                 sk = vxlan->vn4_sock->sock->sk;
1045                 lock_sock(sk);
1046                 ret = ip_mc_join_group(sk, &mreq);
1047                 release_sock(sk);
1048 #if IS_ENABLED(CONFIG_IPV6)
1049         } else {
1050                 sk = vxlan->vn6_sock->sock->sk;
1051                 lock_sock(sk);
1052                 ret = ipv6_stub->ipv6_sock_mc_join(sk, ifindex,
1053                                                    &ip->sin6.sin6_addr);
1054                 release_sock(sk);
1055 #endif
1056         }
1057
1058         return ret;
1059 }
1060
1061 /* Inverse of vxlan_igmp_join when last VNI is brought down */
1062 static int vxlan_igmp_leave(struct vxlan_dev *vxlan)
1063 {
1064         struct sock *sk;
1065         union vxlan_addr *ip = &vxlan->default_dst.remote_ip;
1066         int ifindex = vxlan->default_dst.remote_ifindex;
1067         int ret = -EINVAL;
1068
1069         if (ip->sa.sa_family == AF_INET) {
1070                 struct ip_mreqn mreq = {
1071                         .imr_multiaddr.s_addr   = ip->sin.sin_addr.s_addr,
1072                         .imr_ifindex            = ifindex,
1073                 };
1074
1075                 sk = vxlan->vn4_sock->sock->sk;
1076                 lock_sock(sk);
1077                 ret = ip_mc_leave_group(sk, &mreq);
1078                 release_sock(sk);
1079 #if IS_ENABLED(CONFIG_IPV6)
1080         } else {
1081                 sk = vxlan->vn6_sock->sock->sk;
1082                 lock_sock(sk);
1083                 ret = ipv6_stub->ipv6_sock_mc_drop(sk, ifindex,
1084                                                    &ip->sin6.sin6_addr);
1085                 release_sock(sk);
1086 #endif
1087         }
1088
1089         return ret;
1090 }
1091
1092 static bool vxlan_remcsum(struct vxlanhdr *unparsed,
1093                           struct sk_buff *skb, u32 vxflags)
1094 {
1095         size_t start, offset;
1096
1097         if (!(unparsed->vx_flags & VXLAN_HF_RCO) || skb->remcsum_offload)
1098                 goto out;
1099
1100         start = vxlan_rco_start(unparsed->vx_vni);
1101         offset = start + vxlan_rco_offset(unparsed->vx_vni);
1102
1103         if (!pskb_may_pull(skb, offset + sizeof(u16)))
1104                 return false;
1105
1106         skb_remcsum_process(skb, (void *)(vxlan_hdr(skb) + 1), start, offset,
1107                             !!(vxflags & VXLAN_F_REMCSUM_NOPARTIAL));
1108 out:
1109         unparsed->vx_flags &= ~VXLAN_HF_RCO;
1110         unparsed->vx_vni &= VXLAN_VNI_MASK;
1111         return true;
1112 }
1113
1114 static void vxlan_parse_gbp_hdr(struct vxlanhdr *unparsed,
1115                                 struct sk_buff *skb, u32 vxflags,
1116                                 struct vxlan_metadata *md)
1117 {
1118         struct vxlanhdr_gbp *gbp = (struct vxlanhdr_gbp *)unparsed;
1119         struct metadata_dst *tun_dst;
1120
1121         if (!(unparsed->vx_flags & VXLAN_HF_GBP))
1122                 goto out;
1123
1124         md->gbp = ntohs(gbp->policy_id);
1125
1126         tun_dst = (struct metadata_dst *)skb_dst(skb);
1127         if (tun_dst) {
1128                 tun_dst->u.tun_info.key.tun_flags |= TUNNEL_VXLAN_OPT;
1129                 tun_dst->u.tun_info.options_len = sizeof(*md);
1130         }
1131         if (gbp->dont_learn)
1132                 md->gbp |= VXLAN_GBP_DONT_LEARN;
1133
1134         if (gbp->policy_applied)
1135                 md->gbp |= VXLAN_GBP_POLICY_APPLIED;
1136
1137         /* In flow-based mode, GBP is carried in dst_metadata */
1138         if (!(vxflags & VXLAN_F_COLLECT_METADATA))
1139                 skb->mark = md->gbp;
1140 out:
1141         unparsed->vx_flags &= ~VXLAN_GBP_USED_BITS;
1142 }
1143
1144 static bool vxlan_parse_gpe_hdr(struct vxlanhdr *unparsed,
1145                                 __be16 *protocol,
1146                                 struct sk_buff *skb, u32 vxflags)
1147 {
1148         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)unparsed;
1149
1150         /* Need to have Next Protocol set for interfaces in GPE mode. */
1151         if (!gpe->np_applied)
1152                 return false;
1153         /* "The initial version is 0. If a receiver does not support the
1154          * version indicated it MUST drop the packet.
1155          */
1156         if (gpe->version != 0)
1157                 return false;
1158         /* "When the O bit is set to 1, the packet is an OAM packet and OAM
1159          * processing MUST occur." However, we don't implement OAM
1160          * processing, thus drop the packet.
1161          */
1162         if (gpe->oam_flag)
1163                 return false;
1164
1165         switch (gpe->next_protocol) {
1166         case VXLAN_GPE_NP_IPV4:
1167                 *protocol = htons(ETH_P_IP);
1168                 break;
1169         case VXLAN_GPE_NP_IPV6:
1170                 *protocol = htons(ETH_P_IPV6);
1171                 break;
1172         case VXLAN_GPE_NP_ETHERNET:
1173                 *protocol = htons(ETH_P_TEB);
1174                 break;
1175         default:
1176                 return false;
1177         }
1178
1179         unparsed->vx_flags &= ~VXLAN_GPE_USED_BITS;
1180         return true;
1181 }
1182
1183 static bool vxlan_set_mac(struct vxlan_dev *vxlan,
1184                           struct vxlan_sock *vs,
1185                           struct sk_buff *skb)
1186 {
1187         union vxlan_addr saddr;
1188
1189         skb_reset_mac_header(skb);
1190         skb->protocol = eth_type_trans(skb, vxlan->dev);
1191         skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1192
1193         /* Ignore packet loops (and multicast echo) */
1194         if (ether_addr_equal(eth_hdr(skb)->h_source, vxlan->dev->dev_addr))
1195                 return false;
1196
1197         /* Get address from the outer IP header */
1198         if (vxlan_get_sk_family(vs) == AF_INET) {
1199                 saddr.sin.sin_addr.s_addr = ip_hdr(skb)->saddr;
1200                 saddr.sa.sa_family = AF_INET;
1201 #if IS_ENABLED(CONFIG_IPV6)
1202         } else {
1203                 saddr.sin6.sin6_addr = ipv6_hdr(skb)->saddr;
1204                 saddr.sa.sa_family = AF_INET6;
1205 #endif
1206         }
1207
1208         if ((vxlan->flags & VXLAN_F_LEARN) &&
1209             vxlan_snoop(skb->dev, &saddr, eth_hdr(skb)->h_source))
1210                 return false;
1211
1212         return true;
1213 }
1214
1215 static bool vxlan_ecn_decapsulate(struct vxlan_sock *vs, void *oiph,
1216                                   struct sk_buff *skb)
1217 {
1218         int err = 0;
1219
1220         if (vxlan_get_sk_family(vs) == AF_INET)
1221                 err = IP_ECN_decapsulate(oiph, skb);
1222 #if IS_ENABLED(CONFIG_IPV6)
1223         else
1224                 err = IP6_ECN_decapsulate(oiph, skb);
1225 #endif
1226
1227         if (unlikely(err) && log_ecn_error) {
1228                 if (vxlan_get_sk_family(vs) == AF_INET)
1229                         net_info_ratelimited("non-ECT from %pI4 with TOS=%#x\n",
1230                                              &((struct iphdr *)oiph)->saddr,
1231                                              ((struct iphdr *)oiph)->tos);
1232                 else
1233                         net_info_ratelimited("non-ECT from %pI6\n",
1234                                              &((struct ipv6hdr *)oiph)->saddr);
1235         }
1236         return err <= 1;
1237 }
1238
1239 /* Callback from net/ipv4/udp.c to receive packets */
1240 static int vxlan_rcv(struct sock *sk, struct sk_buff *skb)
1241 {
1242         struct pcpu_sw_netstats *stats;
1243         struct vxlan_dev *vxlan;
1244         struct vxlan_sock *vs;
1245         struct vxlanhdr unparsed;
1246         struct vxlan_metadata _md;
1247         struct vxlan_metadata *md = &_md;
1248         __be16 protocol = htons(ETH_P_TEB);
1249         bool raw_proto = false;
1250         void *oiph;
1251
1252         /* Need UDP and VXLAN header to be present */
1253         if (!pskb_may_pull(skb, VXLAN_HLEN))
1254                 goto drop;
1255
1256         unparsed = *vxlan_hdr(skb);
1257         /* VNI flag always required to be set */
1258         if (!(unparsed.vx_flags & VXLAN_HF_VNI)) {
1259                 netdev_dbg(skb->dev, "invalid vxlan flags=%#x vni=%#x\n",
1260                            ntohl(vxlan_hdr(skb)->vx_flags),
1261                            ntohl(vxlan_hdr(skb)->vx_vni));
1262                 /* Return non vxlan pkt */
1263                 goto drop;
1264         }
1265         unparsed.vx_flags &= ~VXLAN_HF_VNI;
1266         unparsed.vx_vni &= ~VXLAN_VNI_MASK;
1267
1268         vs = rcu_dereference_sk_user_data(sk);
1269         if (!vs)
1270                 goto drop;
1271
1272         vxlan = vxlan_vs_find_vni(vs, vxlan_vni(vxlan_hdr(skb)->vx_vni));
1273         if (!vxlan)
1274                 goto drop;
1275
1276         /* For backwards compatibility, only allow reserved fields to be
1277          * used by VXLAN extensions if explicitly requested.
1278          */
1279         if (vs->flags & VXLAN_F_GPE) {
1280                 if (!vxlan_parse_gpe_hdr(&unparsed, &protocol, skb, vs->flags))
1281                         goto drop;
1282                 raw_proto = true;
1283         }
1284
1285         if (__iptunnel_pull_header(skb, VXLAN_HLEN, protocol, raw_proto,
1286                                    !net_eq(vxlan->net, dev_net(vxlan->dev))))
1287                         goto drop;
1288
1289         if (vxlan_collect_metadata(vs)) {
1290                 __be32 vni = vxlan_vni(vxlan_hdr(skb)->vx_vni);
1291                 struct metadata_dst *tun_dst;
1292
1293                 tun_dst = udp_tun_rx_dst(skb, vxlan_get_sk_family(vs), TUNNEL_KEY,
1294                                          key32_to_tunnel_id(vni), sizeof(*md));
1295
1296                 if (!tun_dst)
1297                         goto drop;
1298
1299                 md = ip_tunnel_info_opts(&tun_dst->u.tun_info);
1300
1301                 skb_dst_set(skb, (struct dst_entry *)tun_dst);
1302         } else {
1303                 memset(md, 0, sizeof(*md));
1304         }
1305
1306         if (vs->flags & VXLAN_F_REMCSUM_RX)
1307                 if (!vxlan_remcsum(&unparsed, skb, vs->flags))
1308                         goto drop;
1309         if (vs->flags & VXLAN_F_GBP)
1310                 vxlan_parse_gbp_hdr(&unparsed, skb, vs->flags, md);
1311         /* Note that GBP and GPE can never be active together. This is
1312          * ensured in vxlan_dev_configure.
1313          */
1314
1315         if (unparsed.vx_flags || unparsed.vx_vni) {
1316                 /* If there are any unprocessed flags remaining treat
1317                  * this as a malformed packet. This behavior diverges from
1318                  * VXLAN RFC (RFC7348) which stipulates that bits in reserved
1319                  * in reserved fields are to be ignored. The approach here
1320                  * maintains compatibility with previous stack code, and also
1321                  * is more robust and provides a little more security in
1322                  * adding extensions to VXLAN.
1323                  */
1324                 goto drop;
1325         }
1326
1327         if (!raw_proto) {
1328                 if (!vxlan_set_mac(vxlan, vs, skb))
1329                         goto drop;
1330         } else {
1331                 skb_reset_mac_header(skb);
1332                 skb->dev = vxlan->dev;
1333                 skb->pkt_type = PACKET_HOST;
1334         }
1335
1336         oiph = skb_network_header(skb);
1337         skb_reset_network_header(skb);
1338
1339         if (!vxlan_ecn_decapsulate(vs, oiph, skb)) {
1340                 ++vxlan->dev->stats.rx_frame_errors;
1341                 ++vxlan->dev->stats.rx_errors;
1342                 goto drop;
1343         }
1344
1345         stats = this_cpu_ptr(vxlan->dev->tstats);
1346         u64_stats_update_begin(&stats->syncp);
1347         stats->rx_packets++;
1348         stats->rx_bytes += skb->len;
1349         u64_stats_update_end(&stats->syncp);
1350
1351         gro_cells_receive(&vxlan->gro_cells, skb);
1352         return 0;
1353
1354 drop:
1355         /* Consume bad packet */
1356         kfree_skb(skb);
1357         return 0;
1358 }
1359
1360 static int arp_reduce(struct net_device *dev, struct sk_buff *skb)
1361 {
1362         struct vxlan_dev *vxlan = netdev_priv(dev);
1363         struct arphdr *parp;
1364         u8 *arpptr, *sha;
1365         __be32 sip, tip;
1366         struct neighbour *n;
1367
1368         if (dev->flags & IFF_NOARP)
1369                 goto out;
1370
1371         if (!pskb_may_pull(skb, arp_hdr_len(dev))) {
1372                 dev->stats.tx_dropped++;
1373                 goto out;
1374         }
1375         parp = arp_hdr(skb);
1376
1377         if ((parp->ar_hrd != htons(ARPHRD_ETHER) &&
1378              parp->ar_hrd != htons(ARPHRD_IEEE802)) ||
1379             parp->ar_pro != htons(ETH_P_IP) ||
1380             parp->ar_op != htons(ARPOP_REQUEST) ||
1381             parp->ar_hln != dev->addr_len ||
1382             parp->ar_pln != 4)
1383                 goto out;
1384         arpptr = (u8 *)parp + sizeof(struct arphdr);
1385         sha = arpptr;
1386         arpptr += dev->addr_len;        /* sha */
1387         memcpy(&sip, arpptr, sizeof(sip));
1388         arpptr += sizeof(sip);
1389         arpptr += dev->addr_len;        /* tha */
1390         memcpy(&tip, arpptr, sizeof(tip));
1391
1392         if (ipv4_is_loopback(tip) ||
1393             ipv4_is_multicast(tip))
1394                 goto out;
1395
1396         n = neigh_lookup(&arp_tbl, &tip, dev);
1397
1398         if (n) {
1399                 struct vxlan_fdb *f;
1400                 struct sk_buff  *reply;
1401
1402                 if (!(n->nud_state & NUD_CONNECTED)) {
1403                         neigh_release(n);
1404                         goto out;
1405                 }
1406
1407                 f = vxlan_find_mac(vxlan, n->ha);
1408                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1409                         /* bridge-local neighbor */
1410                         neigh_release(n);
1411                         goto out;
1412                 }
1413
1414                 reply = arp_create(ARPOP_REPLY, ETH_P_ARP, sip, dev, tip, sha,
1415                                 n->ha, sha);
1416
1417                 neigh_release(n);
1418
1419                 if (reply == NULL)
1420                         goto out;
1421
1422                 skb_reset_mac_header(reply);
1423                 __skb_pull(reply, skb_network_offset(reply));
1424                 reply->ip_summed = CHECKSUM_UNNECESSARY;
1425                 reply->pkt_type = PACKET_HOST;
1426
1427                 if (netif_rx_ni(reply) == NET_RX_DROP)
1428                         dev->stats.rx_dropped++;
1429         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1430                 union vxlan_addr ipa = {
1431                         .sin.sin_addr.s_addr = tip,
1432                         .sin.sin_family = AF_INET,
1433                 };
1434
1435                 vxlan_ip_miss(dev, &ipa);
1436         }
1437 out:
1438         consume_skb(skb);
1439         return NETDEV_TX_OK;
1440 }
1441
1442 #if IS_ENABLED(CONFIG_IPV6)
1443 static struct sk_buff *vxlan_na_create(struct sk_buff *request,
1444         struct neighbour *n, bool isrouter)
1445 {
1446         struct net_device *dev = request->dev;
1447         struct sk_buff *reply;
1448         struct nd_msg *ns, *na;
1449         struct ipv6hdr *pip6;
1450         u8 *daddr;
1451         int na_olen = 8; /* opt hdr + ETH_ALEN for target */
1452         int ns_olen;
1453         int i, len;
1454
1455         if (dev == NULL)
1456                 return NULL;
1457
1458         len = LL_RESERVED_SPACE(dev) + sizeof(struct ipv6hdr) +
1459                 sizeof(*na) + na_olen + dev->needed_tailroom;
1460         reply = alloc_skb(len, GFP_ATOMIC);
1461         if (reply == NULL)
1462                 return NULL;
1463
1464         reply->protocol = htons(ETH_P_IPV6);
1465         reply->dev = dev;
1466         skb_reserve(reply, LL_RESERVED_SPACE(request->dev));
1467         skb_push(reply, sizeof(struct ethhdr));
1468         skb_reset_mac_header(reply);
1469
1470         ns = (struct nd_msg *)skb_transport_header(request);
1471
1472         daddr = eth_hdr(request)->h_source;
1473         ns_olen = request->len - skb_transport_offset(request) - sizeof(*ns);
1474         for (i = 0; i < ns_olen-1; i += (ns->opt[i+1]<<3)) {
1475                 if (ns->opt[i] == ND_OPT_SOURCE_LL_ADDR) {
1476                         daddr = ns->opt + i + sizeof(struct nd_opt_hdr);
1477                         break;
1478                 }
1479         }
1480
1481         /* Ethernet header */
1482         ether_addr_copy(eth_hdr(reply)->h_dest, daddr);
1483         ether_addr_copy(eth_hdr(reply)->h_source, n->ha);
1484         eth_hdr(reply)->h_proto = htons(ETH_P_IPV6);
1485         reply->protocol = htons(ETH_P_IPV6);
1486
1487         skb_pull(reply, sizeof(struct ethhdr));
1488         skb_reset_network_header(reply);
1489         skb_put(reply, sizeof(struct ipv6hdr));
1490
1491         /* IPv6 header */
1492
1493         pip6 = ipv6_hdr(reply);
1494         memset(pip6, 0, sizeof(struct ipv6hdr));
1495         pip6->version = 6;
1496         pip6->priority = ipv6_hdr(request)->priority;
1497         pip6->nexthdr = IPPROTO_ICMPV6;
1498         pip6->hop_limit = 255;
1499         pip6->daddr = ipv6_hdr(request)->saddr;
1500         pip6->saddr = *(struct in6_addr *)n->primary_key;
1501
1502         skb_pull(reply, sizeof(struct ipv6hdr));
1503         skb_reset_transport_header(reply);
1504
1505         na = (struct nd_msg *)skb_put(reply, sizeof(*na) + na_olen);
1506
1507         /* Neighbor Advertisement */
1508         memset(na, 0, sizeof(*na)+na_olen);
1509         na->icmph.icmp6_type = NDISC_NEIGHBOUR_ADVERTISEMENT;
1510         na->icmph.icmp6_router = isrouter;
1511         na->icmph.icmp6_override = 1;
1512         na->icmph.icmp6_solicited = 1;
1513         na->target = ns->target;
1514         ether_addr_copy(&na->opt[2], n->ha);
1515         na->opt[0] = ND_OPT_TARGET_LL_ADDR;
1516         na->opt[1] = na_olen >> 3;
1517
1518         na->icmph.icmp6_cksum = csum_ipv6_magic(&pip6->saddr,
1519                 &pip6->daddr, sizeof(*na)+na_olen, IPPROTO_ICMPV6,
1520                 csum_partial(na, sizeof(*na)+na_olen, 0));
1521
1522         pip6->payload_len = htons(sizeof(*na)+na_olen);
1523
1524         skb_push(reply, sizeof(struct ipv6hdr));
1525
1526         reply->ip_summed = CHECKSUM_UNNECESSARY;
1527
1528         return reply;
1529 }
1530
1531 static int neigh_reduce(struct net_device *dev, struct sk_buff *skb)
1532 {
1533         struct vxlan_dev *vxlan = netdev_priv(dev);
1534         struct nd_msg *msg;
1535         const struct ipv6hdr *iphdr;
1536         const struct in6_addr *saddr, *daddr;
1537         struct neighbour *n;
1538         struct inet6_dev *in6_dev;
1539
1540         in6_dev = __in6_dev_get(dev);
1541         if (!in6_dev)
1542                 goto out;
1543
1544         iphdr = ipv6_hdr(skb);
1545         saddr = &iphdr->saddr;
1546         daddr = &iphdr->daddr;
1547
1548         msg = (struct nd_msg *)skb_transport_header(skb);
1549         if (msg->icmph.icmp6_code != 0 ||
1550             msg->icmph.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION)
1551                 goto out;
1552
1553         if (ipv6_addr_loopback(daddr) ||
1554             ipv6_addr_is_multicast(&msg->target))
1555                 goto out;
1556
1557         n = neigh_lookup(ipv6_stub->nd_tbl, &msg->target, dev);
1558
1559         if (n) {
1560                 struct vxlan_fdb *f;
1561                 struct sk_buff *reply;
1562
1563                 if (!(n->nud_state & NUD_CONNECTED)) {
1564                         neigh_release(n);
1565                         goto out;
1566                 }
1567
1568                 f = vxlan_find_mac(vxlan, n->ha);
1569                 if (f && vxlan_addr_any(&(first_remote_rcu(f)->remote_ip))) {
1570                         /* bridge-local neighbor */
1571                         neigh_release(n);
1572                         goto out;
1573                 }
1574
1575                 reply = vxlan_na_create(skb, n,
1576                                         !!(f ? f->flags & NTF_ROUTER : 0));
1577
1578                 neigh_release(n);
1579
1580                 if (reply == NULL)
1581                         goto out;
1582
1583                 if (netif_rx_ni(reply) == NET_RX_DROP)
1584                         dev->stats.rx_dropped++;
1585
1586         } else if (vxlan->flags & VXLAN_F_L3MISS) {
1587                 union vxlan_addr ipa = {
1588                         .sin6.sin6_addr = msg->target,
1589                         .sin6.sin6_family = AF_INET6,
1590                 };
1591
1592                 vxlan_ip_miss(dev, &ipa);
1593         }
1594
1595 out:
1596         consume_skb(skb);
1597         return NETDEV_TX_OK;
1598 }
1599 #endif
1600
1601 static bool route_shortcircuit(struct net_device *dev, struct sk_buff *skb)
1602 {
1603         struct vxlan_dev *vxlan = netdev_priv(dev);
1604         struct neighbour *n;
1605
1606         if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
1607                 return false;
1608
1609         n = NULL;
1610         switch (ntohs(eth_hdr(skb)->h_proto)) {
1611         case ETH_P_IP:
1612         {
1613                 struct iphdr *pip;
1614
1615                 if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1616                         return false;
1617                 pip = ip_hdr(skb);
1618                 n = neigh_lookup(&arp_tbl, &pip->daddr, dev);
1619                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1620                         union vxlan_addr ipa = {
1621                                 .sin.sin_addr.s_addr = pip->daddr,
1622                                 .sin.sin_family = AF_INET,
1623                         };
1624
1625                         vxlan_ip_miss(dev, &ipa);
1626                         return false;
1627                 }
1628
1629                 break;
1630         }
1631 #if IS_ENABLED(CONFIG_IPV6)
1632         case ETH_P_IPV6:
1633         {
1634                 struct ipv6hdr *pip6;
1635
1636                 if (!pskb_may_pull(skb, sizeof(struct ipv6hdr)))
1637                         return false;
1638                 pip6 = ipv6_hdr(skb);
1639                 n = neigh_lookup(ipv6_stub->nd_tbl, &pip6->daddr, dev);
1640                 if (!n && (vxlan->flags & VXLAN_F_L3MISS)) {
1641                         union vxlan_addr ipa = {
1642                                 .sin6.sin6_addr = pip6->daddr,
1643                                 .sin6.sin6_family = AF_INET6,
1644                         };
1645
1646                         vxlan_ip_miss(dev, &ipa);
1647                         return false;
1648                 }
1649
1650                 break;
1651         }
1652 #endif
1653         default:
1654                 return false;
1655         }
1656
1657         if (n) {
1658                 bool diff;
1659
1660                 diff = !ether_addr_equal(eth_hdr(skb)->h_dest, n->ha);
1661                 if (diff) {
1662                         memcpy(eth_hdr(skb)->h_source, eth_hdr(skb)->h_dest,
1663                                 dev->addr_len);
1664                         memcpy(eth_hdr(skb)->h_dest, n->ha, dev->addr_len);
1665                 }
1666                 neigh_release(n);
1667                 return diff;
1668         }
1669
1670         return false;
1671 }
1672
1673 static void vxlan_build_gbp_hdr(struct vxlanhdr *vxh, u32 vxflags,
1674                                 struct vxlan_metadata *md)
1675 {
1676         struct vxlanhdr_gbp *gbp;
1677
1678         if (!md->gbp)
1679                 return;
1680
1681         gbp = (struct vxlanhdr_gbp *)vxh;
1682         vxh->vx_flags |= VXLAN_HF_GBP;
1683
1684         if (md->gbp & VXLAN_GBP_DONT_LEARN)
1685                 gbp->dont_learn = 1;
1686
1687         if (md->gbp & VXLAN_GBP_POLICY_APPLIED)
1688                 gbp->policy_applied = 1;
1689
1690         gbp->policy_id = htons(md->gbp & VXLAN_GBP_ID_MASK);
1691 }
1692
1693 static int vxlan_build_gpe_hdr(struct vxlanhdr *vxh, u32 vxflags,
1694                                __be16 protocol)
1695 {
1696         struct vxlanhdr_gpe *gpe = (struct vxlanhdr_gpe *)vxh;
1697
1698         gpe->np_applied = 1;
1699
1700         switch (protocol) {
1701         case htons(ETH_P_IP):
1702                 gpe->next_protocol = VXLAN_GPE_NP_IPV4;
1703                 return 0;
1704         case htons(ETH_P_IPV6):
1705                 gpe->next_protocol = VXLAN_GPE_NP_IPV6;
1706                 return 0;
1707         case htons(ETH_P_TEB):
1708                 gpe->next_protocol = VXLAN_GPE_NP_ETHERNET;
1709                 return 0;
1710         }
1711         return -EPFNOSUPPORT;
1712 }
1713
1714 static int vxlan_build_skb(struct sk_buff *skb, struct dst_entry *dst,
1715                            int iphdr_len, __be32 vni,
1716                            struct vxlan_metadata *md, u32 vxflags,
1717                            bool udp_sum)
1718 {
1719         struct vxlanhdr *vxh;
1720         int min_headroom;
1721         int err;
1722         int type = udp_sum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
1723         __be16 inner_protocol = htons(ETH_P_TEB);
1724
1725         if ((vxflags & VXLAN_F_REMCSUM_TX) &&
1726             skb->ip_summed == CHECKSUM_PARTIAL) {
1727                 int csum_start = skb_checksum_start_offset(skb);
1728
1729                 if (csum_start <= VXLAN_MAX_REMCSUM_START &&
1730                     !(csum_start & VXLAN_RCO_SHIFT_MASK) &&
1731                     (skb->csum_offset == offsetof(struct udphdr, check) ||
1732                      skb->csum_offset == offsetof(struct tcphdr, check)))
1733                         type |= SKB_GSO_TUNNEL_REMCSUM;
1734         }
1735
1736         min_headroom = LL_RESERVED_SPACE(dst->dev) + dst->header_len
1737                         + VXLAN_HLEN + iphdr_len
1738                         + (skb_vlan_tag_present(skb) ? VLAN_HLEN : 0);
1739
1740         /* Need space for new headers (invalidates iph ptr) */
1741         err = skb_cow_head(skb, min_headroom);
1742         if (unlikely(err))
1743                 goto out_free;
1744
1745         skb = vlan_hwaccel_push_inside(skb);
1746         if (WARN_ON(!skb))
1747                 return -ENOMEM;
1748
1749         err = iptunnel_handle_offloads(skb, type);
1750         if (err)
1751                 goto out_free;
1752
1753         vxh = (struct vxlanhdr *) __skb_push(skb, sizeof(*vxh));
1754         vxh->vx_flags = VXLAN_HF_VNI;
1755         vxh->vx_vni = vxlan_vni_field(vni);
1756
1757         if (type & SKB_GSO_TUNNEL_REMCSUM) {
1758                 unsigned int start;
1759
1760                 start = skb_checksum_start_offset(skb) - sizeof(struct vxlanhdr);
1761                 vxh->vx_vni |= vxlan_compute_rco(start, skb->csum_offset);
1762                 vxh->vx_flags |= VXLAN_HF_RCO;
1763
1764                 if (!skb_is_gso(skb)) {
1765                         skb->ip_summed = CHECKSUM_NONE;
1766                         skb->encapsulation = 0;
1767                 }
1768         }
1769
1770         if (vxflags & VXLAN_F_GBP)
1771                 vxlan_build_gbp_hdr(vxh, vxflags, md);
1772         if (vxflags & VXLAN_F_GPE) {
1773                 err = vxlan_build_gpe_hdr(vxh, vxflags, skb->protocol);
1774                 if (err < 0)
1775                         goto out_free;
1776                 inner_protocol = skb->protocol;
1777         }
1778
1779         skb_set_inner_protocol(skb, inner_protocol);
1780         return 0;
1781
1782 out_free:
1783         kfree_skb(skb);
1784         return err;
1785 }
1786
1787 static struct rtable *vxlan_get_route(struct vxlan_dev *vxlan,
1788                                       struct sk_buff *skb, int oif, u8 tos,
1789                                       __be32 daddr, __be32 *saddr,
1790                                       struct dst_cache *dst_cache,
1791                                       const struct ip_tunnel_info *info)
1792 {
1793         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1794         struct rtable *rt = NULL;
1795         struct flowi4 fl4;
1796
1797         if (tos && !info)
1798                 use_cache = false;
1799         if (use_cache) {
1800                 rt = dst_cache_get_ip4(dst_cache, saddr);
1801                 if (rt)
1802                         return rt;
1803         }
1804
1805         memset(&fl4, 0, sizeof(fl4));
1806         fl4.flowi4_oif = oif;
1807         fl4.flowi4_tos = RT_TOS(tos);
1808         fl4.flowi4_mark = skb->mark;
1809         fl4.flowi4_proto = IPPROTO_UDP;
1810         fl4.daddr = daddr;
1811         fl4.saddr = *saddr;
1812
1813         rt = ip_route_output_key(vxlan->net, &fl4);
1814         if (!IS_ERR(rt)) {
1815                 *saddr = fl4.saddr;
1816                 if (use_cache)
1817                         dst_cache_set_ip4(dst_cache, &rt->dst, fl4.saddr);
1818         }
1819         return rt;
1820 }
1821
1822 #if IS_ENABLED(CONFIG_IPV6)
1823 static struct dst_entry *vxlan6_get_route(struct vxlan_dev *vxlan,
1824                                           struct sk_buff *skb, int oif, u8 tos,
1825                                           __be32 label,
1826                                           const struct in6_addr *daddr,
1827                                           struct in6_addr *saddr,
1828                                           struct dst_cache *dst_cache,
1829                                           const struct ip_tunnel_info *info)
1830 {
1831         bool use_cache = ip_tunnel_dst_cache_usable(skb, info);
1832         struct dst_entry *ndst;
1833         struct flowi6 fl6;
1834         int err;
1835
1836         if (tos && !info)
1837                 use_cache = false;
1838         if (use_cache) {
1839                 ndst = dst_cache_get_ip6(dst_cache, saddr);
1840                 if (ndst)
1841                         return ndst;
1842         }
1843
1844         memset(&fl6, 0, sizeof(fl6));
1845         fl6.flowi6_oif = oif;
1846         fl6.daddr = *daddr;
1847         fl6.saddr = *saddr;
1848         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tos), label);
1849         fl6.flowi6_mark = skb->mark;
1850         fl6.flowi6_proto = IPPROTO_UDP;
1851
1852         err = ipv6_stub->ipv6_dst_lookup(vxlan->net,
1853                                          vxlan->vn6_sock->sock->sk,
1854                                          &ndst, &fl6);
1855         if (err < 0)
1856                 return ERR_PTR(err);
1857
1858         *saddr = fl6.saddr;
1859         if (use_cache)
1860                 dst_cache_set_ip6(dst_cache, ndst, saddr);
1861         return ndst;
1862 }
1863 #endif
1864
1865 /* Bypass encapsulation if the destination is local */
1866 static void vxlan_encap_bypass(struct sk_buff *skb, struct vxlan_dev *src_vxlan,
1867                                struct vxlan_dev *dst_vxlan)
1868 {
1869         struct pcpu_sw_netstats *tx_stats, *rx_stats;
1870         union vxlan_addr loopback;
1871         union vxlan_addr *remote_ip = &dst_vxlan->default_dst.remote_ip;
1872         struct net_device *dev = skb->dev;
1873         int len = skb->len;
1874
1875         tx_stats = this_cpu_ptr(src_vxlan->dev->tstats);
1876         rx_stats = this_cpu_ptr(dst_vxlan->dev->tstats);
1877         skb->pkt_type = PACKET_HOST;
1878         skb->encapsulation = 0;
1879         skb->dev = dst_vxlan->dev;
1880         __skb_pull(skb, skb_network_offset(skb));
1881
1882         if (remote_ip->sa.sa_family == AF_INET) {
1883                 loopback.sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1884                 loopback.sa.sa_family =  AF_INET;
1885 #if IS_ENABLED(CONFIG_IPV6)
1886         } else {
1887                 loopback.sin6.sin6_addr = in6addr_loopback;
1888                 loopback.sa.sa_family =  AF_INET6;
1889 #endif
1890         }
1891
1892         if (dst_vxlan->flags & VXLAN_F_LEARN)
1893                 vxlan_snoop(skb->dev, &loopback, eth_hdr(skb)->h_source);
1894
1895         u64_stats_update_begin(&tx_stats->syncp);
1896         tx_stats->tx_packets++;
1897         tx_stats->tx_bytes += len;
1898         u64_stats_update_end(&tx_stats->syncp);
1899
1900         if (netif_rx(skb) == NET_RX_SUCCESS) {
1901                 u64_stats_update_begin(&rx_stats->syncp);
1902                 rx_stats->rx_packets++;
1903                 rx_stats->rx_bytes += len;
1904                 u64_stats_update_end(&rx_stats->syncp);
1905         } else {
1906                 dev->stats.rx_dropped++;
1907         }
1908 }
1909
1910 static void vxlan_xmit_one(struct sk_buff *skb, struct net_device *dev,
1911                            struct vxlan_rdst *rdst, bool did_rsc)
1912 {
1913         struct dst_cache *dst_cache;
1914         struct ip_tunnel_info *info;
1915         struct vxlan_dev *vxlan = netdev_priv(dev);
1916         struct sock *sk;
1917         struct rtable *rt = NULL;
1918         const struct iphdr *old_iph;
1919         union vxlan_addr *dst;
1920         union vxlan_addr remote_ip, local_ip;
1921         union vxlan_addr *src;
1922         struct vxlan_metadata _md;
1923         struct vxlan_metadata *md = &_md;
1924         __be16 src_port = 0, dst_port;
1925         __be32 vni, label;
1926         __be16 df = 0;
1927         __u8 tos, ttl;
1928         int err;
1929         u32 flags = vxlan->flags;
1930         bool udp_sum = false;
1931         bool xnet = !net_eq(vxlan->net, dev_net(vxlan->dev));
1932
1933         info = skb_tunnel_info(skb);
1934
1935         if (rdst) {
1936                 dst_port = rdst->remote_port ? rdst->remote_port : vxlan->cfg.dst_port;
1937                 vni = rdst->remote_vni;
1938                 dst = &rdst->remote_ip;
1939                 src = &vxlan->cfg.saddr;
1940                 dst_cache = &rdst->dst_cache;
1941         } else {
1942                 if (!info) {
1943                         WARN_ONCE(1, "%s: Missing encapsulation instructions\n",
1944                                   dev->name);
1945                         goto drop;
1946                 }
1947                 dst_port = info->key.tp_dst ? : vxlan->cfg.dst_port;
1948                 vni = tunnel_id_to_key32(info->key.tun_id);
1949                 remote_ip.sa.sa_family = ip_tunnel_info_af(info);
1950                 if (remote_ip.sa.sa_family == AF_INET) {
1951                         remote_ip.sin.sin_addr.s_addr = info->key.u.ipv4.dst;
1952                         local_ip.sin.sin_addr.s_addr = info->key.u.ipv4.src;
1953                 } else {
1954                         remote_ip.sin6.sin6_addr = info->key.u.ipv6.dst;
1955                         local_ip.sin6.sin6_addr = info->key.u.ipv6.src;
1956                 }
1957                 dst = &remote_ip;
1958                 src = &local_ip;
1959                 dst_cache = &info->dst_cache;
1960         }
1961
1962         if (vxlan_addr_any(dst)) {
1963                 if (did_rsc) {
1964                         /* short-circuited back to local bridge */
1965                         vxlan_encap_bypass(skb, vxlan, vxlan);
1966                         return;
1967                 }
1968                 goto drop;
1969         }
1970
1971         old_iph = ip_hdr(skb);
1972
1973         ttl = vxlan->cfg.ttl;
1974         if (!ttl && vxlan_addr_multicast(dst))
1975                 ttl = 1;
1976
1977         tos = vxlan->cfg.tos;
1978         if (tos == 1)
1979                 tos = ip_tunnel_get_dsfield(old_iph, skb);
1980
1981         label = vxlan->cfg.label;
1982         src_port = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
1983                                      vxlan->cfg.port_max, true);
1984
1985         if (info) {
1986                 ttl = info->key.ttl;
1987                 tos = info->key.tos;
1988                 label = info->key.label;
1989                 udp_sum = !!(info->key.tun_flags & TUNNEL_CSUM);
1990
1991                 if (info->options_len)
1992                         md = ip_tunnel_info_opts(info);
1993         } else {
1994                 md->gbp = skb->mark;
1995         }
1996
1997         if (dst->sa.sa_family == AF_INET) {
1998                 if (!vxlan->vn4_sock)
1999                         goto drop;
2000                 sk = vxlan->vn4_sock->sock->sk;
2001
2002                 rt = vxlan_get_route(vxlan, skb,
2003                                      rdst ? rdst->remote_ifindex : 0, tos,
2004                                      dst->sin.sin_addr.s_addr,
2005                                      &src->sin.sin_addr.s_addr,
2006                                      dst_cache, info);
2007                 if (IS_ERR(rt)) {
2008                         netdev_dbg(dev, "no route to %pI4\n",
2009                                    &dst->sin.sin_addr.s_addr);
2010                         dev->stats.tx_carrier_errors++;
2011                         goto tx_error;
2012                 }
2013
2014                 if (rt->dst.dev == dev) {
2015                         netdev_dbg(dev, "circular route to %pI4\n",
2016                                    &dst->sin.sin_addr.s_addr);
2017                         dev->stats.collisions++;
2018                         goto rt_tx_error;
2019                 }
2020
2021                 /* Bypass encapsulation if the destination is local */
2022                 if (!info && rt->rt_flags & RTCF_LOCAL &&
2023                     !(rt->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2024                         struct vxlan_dev *dst_vxlan;
2025
2026                         ip_rt_put(rt);
2027                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2028                                                    dst->sa.sa_family, dst_port,
2029                                                    vxlan->flags);
2030                         if (!dst_vxlan)
2031                                 goto tx_error;
2032                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2033                         return;
2034                 }
2035
2036                 if (!info)
2037                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM_TX);
2038                 else if (info->key.tun_flags & TUNNEL_DONT_FRAGMENT)
2039                         df = htons(IP_DF);
2040
2041                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2042                 ttl = ttl ? : ip4_dst_hoplimit(&rt->dst);
2043                 err = vxlan_build_skb(skb, &rt->dst, sizeof(struct iphdr),
2044                                       vni, md, flags, udp_sum);
2045                 if (err < 0)
2046                         goto xmit_tx_error;
2047
2048                 udp_tunnel_xmit_skb(rt, sk, skb, src->sin.sin_addr.s_addr,
2049                                     dst->sin.sin_addr.s_addr, tos, ttl, df,
2050                                     src_port, dst_port, xnet, !udp_sum);
2051 #if IS_ENABLED(CONFIG_IPV6)
2052         } else {
2053                 struct dst_entry *ndst;
2054                 u32 rt6i_flags;
2055
2056                 if (!vxlan->vn6_sock)
2057                         goto drop;
2058                 sk = vxlan->vn6_sock->sock->sk;
2059
2060                 ndst = vxlan6_get_route(vxlan, skb,
2061                                         rdst ? rdst->remote_ifindex : 0, tos,
2062                                         label, &dst->sin6.sin6_addr,
2063                                         &src->sin6.sin6_addr,
2064                                         dst_cache, info);
2065                 if (IS_ERR(ndst)) {
2066                         netdev_dbg(dev, "no route to %pI6\n",
2067                                    &dst->sin6.sin6_addr);
2068                         dev->stats.tx_carrier_errors++;
2069                         goto tx_error;
2070                 }
2071
2072                 if (ndst->dev == dev) {
2073                         netdev_dbg(dev, "circular route to %pI6\n",
2074                                    &dst->sin6.sin6_addr);
2075                         dst_release(ndst);
2076                         dev->stats.collisions++;
2077                         goto tx_error;
2078                 }
2079
2080                 /* Bypass encapsulation if the destination is local */
2081                 rt6i_flags = ((struct rt6_info *)ndst)->rt6i_flags;
2082                 if (!info && rt6i_flags & RTF_LOCAL &&
2083                     !(rt6i_flags & (RTCF_BROADCAST | RTCF_MULTICAST))) {
2084                         struct vxlan_dev *dst_vxlan;
2085
2086                         dst_release(ndst);
2087                         dst_vxlan = vxlan_find_vni(vxlan->net, vni,
2088                                                    dst->sa.sa_family, dst_port,
2089                                                    vxlan->flags);
2090                         if (!dst_vxlan)
2091                                 goto tx_error;
2092                         vxlan_encap_bypass(skb, vxlan, dst_vxlan);
2093                         return;
2094                 }
2095
2096                 if (!info)
2097                         udp_sum = !(flags & VXLAN_F_UDP_ZERO_CSUM6_TX);
2098
2099                 tos = ip_tunnel_ecn_encap(tos, old_iph, skb);
2100                 ttl = ttl ? : ip6_dst_hoplimit(ndst);
2101                 skb_scrub_packet(skb, xnet);
2102                 err = vxlan_build_skb(skb, ndst, sizeof(struct ipv6hdr),
2103                                       vni, md, flags, udp_sum);
2104                 if (err < 0) {
2105                         dst_release(ndst);
2106                         dev->stats.tx_errors++;
2107                         return;
2108                 }
2109                 udp_tunnel6_xmit_skb(ndst, sk, skb, dev,
2110                                      &src->sin6.sin6_addr,
2111                                      &dst->sin6.sin6_addr, tos, ttl,
2112                                      label, src_port, dst_port, !udp_sum);
2113 #endif
2114         }
2115
2116         return;
2117
2118 drop:
2119         dev->stats.tx_dropped++;
2120         goto tx_free;
2121
2122 xmit_tx_error:
2123         /* skb is already freed. */
2124         skb = NULL;
2125 rt_tx_error:
2126         ip_rt_put(rt);
2127 tx_error:
2128         dev->stats.tx_errors++;
2129 tx_free:
2130         dev_kfree_skb(skb);
2131 }
2132
2133 /* Transmit local packets over Vxlan
2134  *
2135  * Outer IP header inherits ECN and DF from inner header.
2136  * Outer UDP destination is the VXLAN assigned port.
2137  *           source port is based on hash of flow
2138  */
2139 static netdev_tx_t vxlan_xmit(struct sk_buff *skb, struct net_device *dev)
2140 {
2141         struct vxlan_dev *vxlan = netdev_priv(dev);
2142         const struct ip_tunnel_info *info;
2143         struct ethhdr *eth;
2144         bool did_rsc = false;
2145         struct vxlan_rdst *rdst, *fdst = NULL;
2146         struct vxlan_fdb *f;
2147
2148         info = skb_tunnel_info(skb);
2149
2150         skb_reset_mac_header(skb);
2151
2152         if (vxlan->flags & VXLAN_F_COLLECT_METADATA) {
2153                 if (info && info->mode & IP_TUNNEL_INFO_TX)
2154                         vxlan_xmit_one(skb, dev, NULL, false);
2155                 else
2156                         kfree_skb(skb);
2157                 return NETDEV_TX_OK;
2158         }
2159
2160         if (vxlan->flags & VXLAN_F_PROXY) {
2161                 eth = eth_hdr(skb);
2162                 if (ntohs(eth->h_proto) == ETH_P_ARP)
2163                         return arp_reduce(dev, skb);
2164 #if IS_ENABLED(CONFIG_IPV6)
2165                 else if (ntohs(eth->h_proto) == ETH_P_IPV6 &&
2166                          pskb_may_pull(skb, sizeof(struct ipv6hdr)
2167                                        + sizeof(struct nd_msg)) &&
2168                          ipv6_hdr(skb)->nexthdr == IPPROTO_ICMPV6) {
2169                                 struct nd_msg *msg;
2170
2171                                 msg = (struct nd_msg *)skb_transport_header(skb);
2172                                 if (msg->icmph.icmp6_code == 0 &&
2173                                     msg->icmph.icmp6_type == NDISC_NEIGHBOUR_SOLICITATION)
2174                                         return neigh_reduce(dev, skb);
2175                 }
2176 #endif
2177         }
2178
2179         eth = eth_hdr(skb);
2180         f = vxlan_find_mac(vxlan, eth->h_dest);
2181         did_rsc = false;
2182
2183         if (f && (f->flags & NTF_ROUTER) && (vxlan->flags & VXLAN_F_RSC) &&
2184             (ntohs(eth->h_proto) == ETH_P_IP ||
2185              ntohs(eth->h_proto) == ETH_P_IPV6)) {
2186                 did_rsc = route_shortcircuit(dev, skb);
2187                 if (did_rsc)
2188                         f = vxlan_find_mac(vxlan, eth->h_dest);
2189         }
2190
2191         if (f == NULL) {
2192                 f = vxlan_find_mac(vxlan, all_zeros_mac);
2193                 if (f == NULL) {
2194                         if ((vxlan->flags & VXLAN_F_L2MISS) &&
2195                             !is_multicast_ether_addr(eth->h_dest))
2196                                 vxlan_fdb_miss(vxlan, eth->h_dest);
2197
2198                         dev->stats.tx_dropped++;
2199                         kfree_skb(skb);
2200                         return NETDEV_TX_OK;
2201                 }
2202         }
2203
2204         list_for_each_entry_rcu(rdst, &f->remotes, list) {
2205                 struct sk_buff *skb1;
2206
2207                 if (!fdst) {
2208                         fdst = rdst;
2209                         continue;
2210                 }
2211                 skb1 = skb_clone(skb, GFP_ATOMIC);
2212                 if (skb1)
2213                         vxlan_xmit_one(skb1, dev, rdst, did_rsc);
2214         }
2215
2216         if (fdst)
2217                 vxlan_xmit_one(skb, dev, fdst, did_rsc);
2218         else
2219                 kfree_skb(skb);
2220         return NETDEV_TX_OK;
2221 }
2222
2223 /* Walk the forwarding table and purge stale entries */
2224 static void vxlan_cleanup(unsigned long arg)
2225 {
2226         struct vxlan_dev *vxlan = (struct vxlan_dev *) arg;
2227         unsigned long next_timer = jiffies + FDB_AGE_INTERVAL;
2228         unsigned int h;
2229
2230         if (!netif_running(vxlan->dev))
2231                 return;
2232
2233         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2234                 struct hlist_node *p, *n;
2235
2236                 spin_lock_bh(&vxlan->hash_lock);
2237                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2238                         struct vxlan_fdb *f
2239                                 = container_of(p, struct vxlan_fdb, hlist);
2240                         unsigned long timeout;
2241
2242                         if (f->state & NUD_PERMANENT)
2243                                 continue;
2244
2245                         timeout = f->used + vxlan->cfg.age_interval * HZ;
2246                         if (time_before_eq(timeout, jiffies)) {
2247                                 netdev_dbg(vxlan->dev,
2248                                            "garbage collect %pM\n",
2249                                            f->eth_addr);
2250                                 f->state = NUD_STALE;
2251                                 vxlan_fdb_destroy(vxlan, f);
2252                         } else if (time_before(timeout, next_timer))
2253                                 next_timer = timeout;
2254                 }
2255                 spin_unlock_bh(&vxlan->hash_lock);
2256         }
2257
2258         mod_timer(&vxlan->age_timer, next_timer);
2259 }
2260
2261 static void vxlan_vs_add_dev(struct vxlan_sock *vs, struct vxlan_dev *vxlan)
2262 {
2263         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2264         __be32 vni = vxlan->default_dst.remote_vni;
2265
2266         spin_lock(&vn->sock_lock);
2267         hlist_add_head_rcu(&vxlan->hlist, vni_head(vs, vni));
2268         spin_unlock(&vn->sock_lock);
2269 }
2270
2271 /* Setup stats when device is created */
2272 static int vxlan_init(struct net_device *dev)
2273 {
2274         dev->tstats = netdev_alloc_pcpu_stats(struct pcpu_sw_netstats);
2275         if (!dev->tstats)
2276                 return -ENOMEM;
2277
2278         return 0;
2279 }
2280
2281 static void vxlan_fdb_delete_default(struct vxlan_dev *vxlan)
2282 {
2283         struct vxlan_fdb *f;
2284
2285         spin_lock_bh(&vxlan->hash_lock);
2286         f = __vxlan_find_mac(vxlan, all_zeros_mac);
2287         if (f)
2288                 vxlan_fdb_destroy(vxlan, f);
2289         spin_unlock_bh(&vxlan->hash_lock);
2290 }
2291
2292 static void vxlan_uninit(struct net_device *dev)
2293 {
2294         struct vxlan_dev *vxlan = netdev_priv(dev);
2295
2296         vxlan_fdb_delete_default(vxlan);
2297
2298         free_percpu(dev->tstats);
2299 }
2300
2301 /* Start ageing timer and join group when device is brought up */
2302 static int vxlan_open(struct net_device *dev)
2303 {
2304         struct vxlan_dev *vxlan = netdev_priv(dev);
2305         int ret;
2306
2307         ret = vxlan_sock_add(vxlan);
2308         if (ret < 0)
2309                 return ret;
2310
2311         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip)) {
2312                 ret = vxlan_igmp_join(vxlan);
2313                 if (ret == -EADDRINUSE)
2314                         ret = 0;
2315                 if (ret) {
2316                         vxlan_sock_release(vxlan);
2317                         return ret;
2318                 }
2319         }
2320
2321         if (vxlan->cfg.age_interval)
2322                 mod_timer(&vxlan->age_timer, jiffies + FDB_AGE_INTERVAL);
2323
2324         return ret;
2325 }
2326
2327 /* Purge the forwarding table */
2328 static void vxlan_flush(struct vxlan_dev *vxlan)
2329 {
2330         unsigned int h;
2331
2332         spin_lock_bh(&vxlan->hash_lock);
2333         for (h = 0; h < FDB_HASH_SIZE; ++h) {
2334                 struct hlist_node *p, *n;
2335                 hlist_for_each_safe(p, n, &vxlan->fdb_head[h]) {
2336                         struct vxlan_fdb *f
2337                                 = container_of(p, struct vxlan_fdb, hlist);
2338                         /* the all_zeros_mac entry is deleted at vxlan_uninit */
2339                         if (!is_zero_ether_addr(f->eth_addr))
2340                                 vxlan_fdb_destroy(vxlan, f);
2341                 }
2342         }
2343         spin_unlock_bh(&vxlan->hash_lock);
2344 }
2345
2346 /* Cleanup timer and forwarding table on shutdown */
2347 static int vxlan_stop(struct net_device *dev)
2348 {
2349         struct vxlan_dev *vxlan = netdev_priv(dev);
2350         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2351         int ret = 0;
2352
2353         if (vxlan_addr_multicast(&vxlan->default_dst.remote_ip) &&
2354             !vxlan_group_used(vn, vxlan))
2355                 ret = vxlan_igmp_leave(vxlan);
2356
2357         del_timer_sync(&vxlan->age_timer);
2358
2359         vxlan_flush(vxlan);
2360         vxlan_sock_release(vxlan);
2361
2362         return ret;
2363 }
2364
2365 /* Stub, nothing needs to be done. */
2366 static void vxlan_set_multicast_list(struct net_device *dev)
2367 {
2368 }
2369
2370 static int __vxlan_change_mtu(struct net_device *dev,
2371                               struct net_device *lowerdev,
2372                               struct vxlan_rdst *dst, int new_mtu, bool strict)
2373 {
2374         int max_mtu = IP_MAX_MTU;
2375
2376         if (lowerdev)
2377                 max_mtu = lowerdev->mtu;
2378
2379         if (dst->remote_ip.sa.sa_family == AF_INET6)
2380                 max_mtu -= VXLAN6_HEADROOM;
2381         else
2382                 max_mtu -= VXLAN_HEADROOM;
2383
2384         if (new_mtu < 68)
2385                 return -EINVAL;
2386
2387         if (new_mtu > max_mtu) {
2388                 if (strict)
2389                         return -EINVAL;
2390
2391                 new_mtu = max_mtu;
2392         }
2393
2394         dev->mtu = new_mtu;
2395         return 0;
2396 }
2397
2398 static int vxlan_change_mtu(struct net_device *dev, int new_mtu)
2399 {
2400         struct vxlan_dev *vxlan = netdev_priv(dev);
2401         struct vxlan_rdst *dst = &vxlan->default_dst;
2402         struct net_device *lowerdev = __dev_get_by_index(vxlan->net,
2403                                                          dst->remote_ifindex);
2404         return __vxlan_change_mtu(dev, lowerdev, dst, new_mtu, true);
2405 }
2406
2407 static int vxlan_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
2408 {
2409         struct vxlan_dev *vxlan = netdev_priv(dev);
2410         struct ip_tunnel_info *info = skb_tunnel_info(skb);
2411         __be16 sport, dport;
2412
2413         sport = udp_flow_src_port(dev_net(dev), skb, vxlan->cfg.port_min,
2414                                   vxlan->cfg.port_max, true);
2415         dport = info->key.tp_dst ? : vxlan->cfg.dst_port;
2416
2417         if (ip_tunnel_info_af(info) == AF_INET) {
2418                 struct rtable *rt;
2419
2420                 if (!vxlan->vn4_sock)
2421                         return -EINVAL;
2422                 rt = vxlan_get_route(vxlan, skb, 0, info->key.tos,
2423                                      info->key.u.ipv4.dst,
2424                                      &info->key.u.ipv4.src, NULL, info);
2425                 if (IS_ERR(rt))
2426                         return PTR_ERR(rt);
2427                 ip_rt_put(rt);
2428         } else {
2429 #if IS_ENABLED(CONFIG_IPV6)
2430                 struct dst_entry *ndst;
2431
2432                 if (!vxlan->vn6_sock)
2433                         return -EINVAL;
2434                 ndst = vxlan6_get_route(vxlan, skb, 0, info->key.tos,
2435                                         info->key.label, &info->key.u.ipv6.dst,
2436                                         &info->key.u.ipv6.src, NULL, info);
2437                 if (IS_ERR(ndst))
2438                         return PTR_ERR(ndst);
2439                 dst_release(ndst);
2440 #else /* !CONFIG_IPV6 */
2441                 return -EPFNOSUPPORT;
2442 #endif
2443         }
2444         info->key.tp_src = sport;
2445         info->key.tp_dst = dport;
2446         return 0;
2447 }
2448
2449 static const struct net_device_ops vxlan_netdev_ether_ops = {
2450         .ndo_init               = vxlan_init,
2451         .ndo_uninit             = vxlan_uninit,
2452         .ndo_open               = vxlan_open,
2453         .ndo_stop               = vxlan_stop,
2454         .ndo_start_xmit         = vxlan_xmit,
2455         .ndo_get_stats64        = ip_tunnel_get_stats64,
2456         .ndo_set_rx_mode        = vxlan_set_multicast_list,
2457         .ndo_change_mtu         = vxlan_change_mtu,
2458         .ndo_validate_addr      = eth_validate_addr,
2459         .ndo_set_mac_address    = eth_mac_addr,
2460         .ndo_fdb_add            = vxlan_fdb_add,
2461         .ndo_fdb_del            = vxlan_fdb_delete,
2462         .ndo_fdb_dump           = vxlan_fdb_dump,
2463         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2464 };
2465
2466 static const struct net_device_ops vxlan_netdev_raw_ops = {
2467         .ndo_init               = vxlan_init,
2468         .ndo_uninit             = vxlan_uninit,
2469         .ndo_open               = vxlan_open,
2470         .ndo_stop               = vxlan_stop,
2471         .ndo_start_xmit         = vxlan_xmit,
2472         .ndo_get_stats64        = ip_tunnel_get_stats64,
2473         .ndo_change_mtu         = vxlan_change_mtu,
2474         .ndo_fill_metadata_dst  = vxlan_fill_metadata_dst,
2475 };
2476
2477 /* Info for udev, that this is a virtual tunnel endpoint */
2478 static struct device_type vxlan_type = {
2479         .name = "vxlan",
2480 };
2481
2482 /* Calls the ndo_udp_tunnel_add of the caller in order to
2483  * supply the listening VXLAN udp ports. Callers are expected
2484  * to implement the ndo_udp_tunnel_add.
2485  */
2486 static void vxlan_push_rx_ports(struct net_device *dev)
2487 {
2488         struct vxlan_sock *vs;
2489         struct net *net = dev_net(dev);
2490         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2491         unsigned int i;
2492
2493         spin_lock(&vn->sock_lock);
2494         for (i = 0; i < PORT_HASH_SIZE; ++i) {
2495                 hlist_for_each_entry_rcu(vs, &vn->sock_list[i], hlist)
2496                         udp_tunnel_push_rx_port(dev, vs->sock,
2497                                                 (vs->flags & VXLAN_F_GPE) ?
2498                                                 UDP_TUNNEL_TYPE_VXLAN_GPE :
2499                                                 UDP_TUNNEL_TYPE_VXLAN);
2500         }
2501         spin_unlock(&vn->sock_lock);
2502 }
2503
2504 /* Initialize the device structure. */
2505 static void vxlan_setup(struct net_device *dev)
2506 {
2507         struct vxlan_dev *vxlan = netdev_priv(dev);
2508         unsigned int h;
2509
2510         eth_hw_addr_random(dev);
2511         ether_setup(dev);
2512
2513         dev->destructor = free_netdev;
2514         SET_NETDEV_DEVTYPE(dev, &vxlan_type);
2515
2516         dev->features   |= NETIF_F_LLTX;
2517         dev->features   |= NETIF_F_SG | NETIF_F_HW_CSUM;
2518         dev->features   |= NETIF_F_RXCSUM;
2519         dev->features   |= NETIF_F_GSO_SOFTWARE;
2520
2521         dev->vlan_features = dev->features;
2522         dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2523         dev->hw_features |= NETIF_F_SG | NETIF_F_HW_CSUM | NETIF_F_RXCSUM;
2524         dev->hw_features |= NETIF_F_GSO_SOFTWARE;
2525         dev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
2526         netif_keep_dst(dev);
2527         dev->priv_flags |= IFF_NO_QUEUE;
2528
2529         INIT_LIST_HEAD(&vxlan->next);
2530         spin_lock_init(&vxlan->hash_lock);
2531
2532         init_timer_deferrable(&vxlan->age_timer);
2533         vxlan->age_timer.function = vxlan_cleanup;
2534         vxlan->age_timer.data = (unsigned long) vxlan;
2535
2536         vxlan->cfg.dst_port = htons(vxlan_port);
2537
2538         vxlan->dev = dev;
2539
2540         gro_cells_init(&vxlan->gro_cells, dev);
2541
2542         for (h = 0; h < FDB_HASH_SIZE; ++h)
2543                 INIT_HLIST_HEAD(&vxlan->fdb_head[h]);
2544 }
2545
2546 static void vxlan_ether_setup(struct net_device *dev)
2547 {
2548         dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2549         dev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2550         dev->netdev_ops = &vxlan_netdev_ether_ops;
2551 }
2552
2553 static void vxlan_raw_setup(struct net_device *dev)
2554 {
2555         dev->header_ops = NULL;
2556         dev->type = ARPHRD_NONE;
2557         dev->hard_header_len = 0;
2558         dev->addr_len = 0;
2559         dev->flags = IFF_POINTOPOINT | IFF_NOARP | IFF_MULTICAST;
2560         dev->netdev_ops = &vxlan_netdev_raw_ops;
2561 }
2562
2563 static const struct nla_policy vxlan_policy[IFLA_VXLAN_MAX + 1] = {
2564         [IFLA_VXLAN_ID]         = { .type = NLA_U32 },
2565         [IFLA_VXLAN_GROUP]      = { .len = FIELD_SIZEOF(struct iphdr, daddr) },
2566         [IFLA_VXLAN_GROUP6]     = { .len = sizeof(struct in6_addr) },
2567         [IFLA_VXLAN_LINK]       = { .type = NLA_U32 },
2568         [IFLA_VXLAN_LOCAL]      = { .len = FIELD_SIZEOF(struct iphdr, saddr) },
2569         [IFLA_VXLAN_LOCAL6]     = { .len = sizeof(struct in6_addr) },
2570         [IFLA_VXLAN_TOS]        = { .type = NLA_U8 },
2571         [IFLA_VXLAN_TTL]        = { .type = NLA_U8 },
2572         [IFLA_VXLAN_LABEL]      = { .type = NLA_U32 },
2573         [IFLA_VXLAN_LEARNING]   = { .type = NLA_U8 },
2574         [IFLA_VXLAN_AGEING]     = { .type = NLA_U32 },
2575         [IFLA_VXLAN_LIMIT]      = { .type = NLA_U32 },
2576         [IFLA_VXLAN_PORT_RANGE] = { .len  = sizeof(struct ifla_vxlan_port_range) },
2577         [IFLA_VXLAN_PROXY]      = { .type = NLA_U8 },
2578         [IFLA_VXLAN_RSC]        = { .type = NLA_U8 },
2579         [IFLA_VXLAN_L2MISS]     = { .type = NLA_U8 },
2580         [IFLA_VXLAN_L3MISS]     = { .type = NLA_U8 },
2581         [IFLA_VXLAN_COLLECT_METADATA]   = { .type = NLA_U8 },
2582         [IFLA_VXLAN_PORT]       = { .type = NLA_U16 },
2583         [IFLA_VXLAN_UDP_CSUM]   = { .type = NLA_U8 },
2584         [IFLA_VXLAN_UDP_ZERO_CSUM6_TX]  = { .type = NLA_U8 },
2585         [IFLA_VXLAN_UDP_ZERO_CSUM6_RX]  = { .type = NLA_U8 },
2586         [IFLA_VXLAN_REMCSUM_TX] = { .type = NLA_U8 },
2587         [IFLA_VXLAN_REMCSUM_RX] = { .type = NLA_U8 },
2588         [IFLA_VXLAN_GBP]        = { .type = NLA_FLAG, },
2589         [IFLA_VXLAN_GPE]        = { .type = NLA_FLAG, },
2590         [IFLA_VXLAN_REMCSUM_NOPARTIAL]  = { .type = NLA_FLAG },
2591 };
2592
2593 static int vxlan_validate(struct nlattr *tb[], struct nlattr *data[])
2594 {
2595         if (tb[IFLA_ADDRESS]) {
2596                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) {
2597                         pr_debug("invalid link address (not ethernet)\n");
2598                         return -EINVAL;
2599                 }
2600
2601                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) {
2602                         pr_debug("invalid all zero ethernet address\n");
2603                         return -EADDRNOTAVAIL;
2604                 }
2605         }
2606
2607         if (!data)
2608                 return -EINVAL;
2609
2610         if (data[IFLA_VXLAN_ID]) {
2611                 __u32 id = nla_get_u32(data[IFLA_VXLAN_ID]);
2612                 if (id >= VXLAN_VID_MASK)
2613                         return -ERANGE;
2614         }
2615
2616         if (data[IFLA_VXLAN_PORT_RANGE]) {
2617                 const struct ifla_vxlan_port_range *p
2618                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2619
2620                 if (ntohs(p->high) < ntohs(p->low)) {
2621                         pr_debug("port range %u .. %u not valid\n",
2622                                  ntohs(p->low), ntohs(p->high));
2623                         return -EINVAL;
2624                 }
2625         }
2626
2627         return 0;
2628 }
2629
2630 static void vxlan_get_drvinfo(struct net_device *netdev,
2631                               struct ethtool_drvinfo *drvinfo)
2632 {
2633         strlcpy(drvinfo->version, VXLAN_VERSION, sizeof(drvinfo->version));
2634         strlcpy(drvinfo->driver, "vxlan", sizeof(drvinfo->driver));
2635 }
2636
2637 static const struct ethtool_ops vxlan_ethtool_ops = {
2638         .get_drvinfo    = vxlan_get_drvinfo,
2639         .get_link       = ethtool_op_get_link,
2640 };
2641
2642 static struct socket *vxlan_create_sock(struct net *net, bool ipv6,
2643                                         __be16 port, u32 flags)
2644 {
2645         struct socket *sock;
2646         struct udp_port_cfg udp_conf;
2647         int err;
2648
2649         memset(&udp_conf, 0, sizeof(udp_conf));
2650
2651         if (ipv6) {
2652                 udp_conf.family = AF_INET6;
2653                 udp_conf.use_udp6_rx_checksums =
2654                     !(flags & VXLAN_F_UDP_ZERO_CSUM6_RX);
2655                 udp_conf.ipv6_v6only = 1;
2656         } else {
2657                 udp_conf.family = AF_INET;
2658         }
2659
2660         udp_conf.local_udp_port = port;
2661
2662         /* Open UDP socket */
2663         err = udp_sock_create(net, &udp_conf, &sock);
2664         if (err < 0)
2665                 return ERR_PTR(err);
2666
2667         return sock;
2668 }
2669
2670 /* Create new listen socket if needed */
2671 static struct vxlan_sock *vxlan_socket_create(struct net *net, bool ipv6,
2672                                               __be16 port, u32 flags)
2673 {
2674         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2675         struct vxlan_sock *vs;
2676         struct socket *sock;
2677         unsigned int h;
2678         struct udp_tunnel_sock_cfg tunnel_cfg;
2679
2680         vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2681         if (!vs)
2682                 return ERR_PTR(-ENOMEM);
2683
2684         for (h = 0; h < VNI_HASH_SIZE; ++h)
2685                 INIT_HLIST_HEAD(&vs->vni_list[h]);
2686
2687         sock = vxlan_create_sock(net, ipv6, port, flags);
2688         if (IS_ERR(sock)) {
2689                 pr_info("Cannot bind port %d, err=%ld\n", ntohs(port),
2690                         PTR_ERR(sock));
2691                 kfree(vs);
2692                 return ERR_CAST(sock);
2693         }
2694
2695         vs->sock = sock;
2696         atomic_set(&vs->refcnt, 1);
2697         vs->flags = (flags & VXLAN_F_RCV_FLAGS);
2698
2699         spin_lock(&vn->sock_lock);
2700         hlist_add_head_rcu(&vs->hlist, vs_head(net, port));
2701         udp_tunnel_notify_add_rx_port(sock,
2702                                       (vs->flags & VXLAN_F_GPE) ?
2703                                       UDP_TUNNEL_TYPE_VXLAN_GPE :
2704                                       UDP_TUNNEL_TYPE_VXLAN);
2705         spin_unlock(&vn->sock_lock);
2706
2707         /* Mark socket as an encapsulation socket. */
2708         memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
2709         tunnel_cfg.sk_user_data = vs;
2710         tunnel_cfg.encap_type = 1;
2711         tunnel_cfg.encap_rcv = vxlan_rcv;
2712         tunnel_cfg.encap_destroy = NULL;
2713         tunnel_cfg.gro_receive = vxlan_gro_receive;
2714         tunnel_cfg.gro_complete = vxlan_gro_complete;
2715
2716         setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
2717
2718         return vs;
2719 }
2720
2721 static int __vxlan_sock_add(struct vxlan_dev *vxlan, bool ipv6)
2722 {
2723         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
2724         struct vxlan_sock *vs = NULL;
2725
2726         if (!vxlan->cfg.no_share) {
2727                 spin_lock(&vn->sock_lock);
2728                 vs = vxlan_find_sock(vxlan->net, ipv6 ? AF_INET6 : AF_INET,
2729                                      vxlan->cfg.dst_port, vxlan->flags);
2730                 if (vs && !atomic_add_unless(&vs->refcnt, 1, 0)) {
2731                         spin_unlock(&vn->sock_lock);
2732                         return -EBUSY;
2733                 }
2734                 spin_unlock(&vn->sock_lock);
2735         }
2736         if (!vs)
2737                 vs = vxlan_socket_create(vxlan->net, ipv6,
2738                                          vxlan->cfg.dst_port, vxlan->flags);
2739         if (IS_ERR(vs))
2740                 return PTR_ERR(vs);
2741 #if IS_ENABLED(CONFIG_IPV6)
2742         if (ipv6)
2743                 vxlan->vn6_sock = vs;
2744         else
2745 #endif
2746                 vxlan->vn4_sock = vs;
2747         vxlan_vs_add_dev(vs, vxlan);
2748         return 0;
2749 }
2750
2751 static int vxlan_sock_add(struct vxlan_dev *vxlan)
2752 {
2753         bool ipv6 = vxlan->flags & VXLAN_F_IPV6;
2754         bool metadata = vxlan->flags & VXLAN_F_COLLECT_METADATA;
2755         int ret = 0;
2756
2757         vxlan->vn4_sock = NULL;
2758 #if IS_ENABLED(CONFIG_IPV6)
2759         vxlan->vn6_sock = NULL;
2760         if (ipv6 || metadata)
2761                 ret = __vxlan_sock_add(vxlan, true);
2762 #endif
2763         if (!ret && (!ipv6 || metadata))
2764                 ret = __vxlan_sock_add(vxlan, false);
2765         if (ret < 0)
2766                 vxlan_sock_release(vxlan);
2767         return ret;
2768 }
2769
2770 static int vxlan_dev_configure(struct net *src_net, struct net_device *dev,
2771                                struct vxlan_config *conf)
2772 {
2773         struct vxlan_net *vn = net_generic(src_net, vxlan_net_id);
2774         struct vxlan_dev *vxlan = netdev_priv(dev), *tmp;
2775         struct vxlan_rdst *dst = &vxlan->default_dst;
2776         unsigned short needed_headroom = ETH_HLEN;
2777         int err;
2778         bool use_ipv6 = false;
2779         __be16 default_port = vxlan->cfg.dst_port;
2780         struct net_device *lowerdev = NULL;
2781
2782         if (conf->flags & VXLAN_F_GPE) {
2783                 /* For now, allow GPE only together with COLLECT_METADATA.
2784                  * This can be relaxed later; in such case, the other side
2785                  * of the PtP link will have to be provided.
2786                  */
2787                 if ((conf->flags & ~VXLAN_F_ALLOWED_GPE) ||
2788                     !(conf->flags & VXLAN_F_COLLECT_METADATA)) {
2789                         pr_info("unsupported combination of extensions\n");
2790                         return -EINVAL;
2791                 }
2792
2793                 vxlan_raw_setup(dev);
2794         } else {
2795                 vxlan_ether_setup(dev);
2796         }
2797
2798         vxlan->net = src_net;
2799
2800         dst->remote_vni = conf->vni;
2801
2802         memcpy(&dst->remote_ip, &conf->remote_ip, sizeof(conf->remote_ip));
2803
2804         /* Unless IPv6 is explicitly requested, assume IPv4 */
2805         if (!dst->remote_ip.sa.sa_family)
2806                 dst->remote_ip.sa.sa_family = AF_INET;
2807
2808         if (dst->remote_ip.sa.sa_family == AF_INET6 ||
2809             vxlan->cfg.saddr.sa.sa_family == AF_INET6) {
2810                 if (!IS_ENABLED(CONFIG_IPV6))
2811                         return -EPFNOSUPPORT;
2812                 use_ipv6 = true;
2813                 vxlan->flags |= VXLAN_F_IPV6;
2814         }
2815
2816         if (conf->label && !use_ipv6) {
2817                 pr_info("label only supported in use with IPv6\n");
2818                 return -EINVAL;
2819         }
2820
2821         if (conf->remote_ifindex) {
2822                 lowerdev = __dev_get_by_index(src_net, conf->remote_ifindex);
2823                 dst->remote_ifindex = conf->remote_ifindex;
2824
2825                 if (!lowerdev) {
2826                         pr_info("ifindex %d does not exist\n", dst->remote_ifindex);
2827                         return -ENODEV;
2828                 }
2829
2830 #if IS_ENABLED(CONFIG_IPV6)
2831                 if (use_ipv6) {
2832                         struct inet6_dev *idev = __in6_dev_get(lowerdev);
2833                         if (idev && idev->cnf.disable_ipv6) {
2834                                 pr_info("IPv6 is disabled via sysctl\n");
2835                                 return -EPERM;
2836                         }
2837                 }
2838 #endif
2839
2840                 if (!conf->mtu)
2841                         dev->mtu = lowerdev->mtu - (use_ipv6 ? VXLAN6_HEADROOM : VXLAN_HEADROOM);
2842
2843                 needed_headroom = lowerdev->hard_header_len;
2844         } else if (vxlan_addr_multicast(&dst->remote_ip)) {
2845                 pr_info("multicast destination requires interface to be specified\n");
2846                 return -EINVAL;
2847         }
2848
2849         if (conf->mtu) {
2850                 err = __vxlan_change_mtu(dev, lowerdev, dst, conf->mtu, false);
2851                 if (err)
2852                         return err;
2853         }
2854
2855         if (use_ipv6 || conf->flags & VXLAN_F_COLLECT_METADATA)
2856                 needed_headroom += VXLAN6_HEADROOM;
2857         else
2858                 needed_headroom += VXLAN_HEADROOM;
2859         dev->needed_headroom = needed_headroom;
2860
2861         memcpy(&vxlan->cfg, conf, sizeof(*conf));
2862         if (!vxlan->cfg.dst_port) {
2863                 if (conf->flags & VXLAN_F_GPE)
2864                         vxlan->cfg.dst_port = 4790; /* IANA assigned VXLAN-GPE port */
2865                 else
2866                         vxlan->cfg.dst_port = default_port;
2867         }
2868         vxlan->flags |= conf->flags;
2869
2870         if (!vxlan->cfg.age_interval)
2871                 vxlan->cfg.age_interval = FDB_AGE_DEFAULT;
2872
2873         list_for_each_entry(tmp, &vn->vxlan_list, next) {
2874                 if (tmp->cfg.vni == conf->vni &&
2875                     (tmp->default_dst.remote_ip.sa.sa_family == AF_INET6 ||
2876                      tmp->cfg.saddr.sa.sa_family == AF_INET6) == use_ipv6 &&
2877                     tmp->cfg.dst_port == vxlan->cfg.dst_port &&
2878                     (tmp->flags & VXLAN_F_RCV_FLAGS) ==
2879                     (vxlan->flags & VXLAN_F_RCV_FLAGS)) {
2880                         pr_info("duplicate VNI %u\n", be32_to_cpu(conf->vni));
2881                         return -EEXIST;
2882                 }
2883         }
2884
2885         dev->ethtool_ops = &vxlan_ethtool_ops;
2886
2887         /* create an fdb entry for a valid default destination */
2888         if (!vxlan_addr_any(&vxlan->default_dst.remote_ip)) {
2889                 err = vxlan_fdb_create(vxlan, all_zeros_mac,
2890                                        &vxlan->default_dst.remote_ip,
2891                                        NUD_REACHABLE|NUD_PERMANENT,
2892                                        NLM_F_EXCL|NLM_F_CREATE,
2893                                        vxlan->cfg.dst_port,
2894                                        vxlan->default_dst.remote_vni,
2895                                        vxlan->default_dst.remote_ifindex,
2896                                        NTF_SELF);
2897                 if (err)
2898                         return err;
2899         }
2900
2901         err = register_netdevice(dev);
2902         if (err) {
2903                 vxlan_fdb_delete_default(vxlan);
2904                 return err;
2905         }
2906
2907         list_add(&vxlan->next, &vn->vxlan_list);
2908
2909         return 0;
2910 }
2911
2912 static int vxlan_newlink(struct net *src_net, struct net_device *dev,
2913                          struct nlattr *tb[], struct nlattr *data[])
2914 {
2915         struct vxlan_config conf;
2916
2917         memset(&conf, 0, sizeof(conf));
2918
2919         if (data[IFLA_VXLAN_ID])
2920                 conf.vni = cpu_to_be32(nla_get_u32(data[IFLA_VXLAN_ID]));
2921
2922         if (data[IFLA_VXLAN_GROUP]) {
2923                 conf.remote_ip.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_GROUP]);
2924         } else if (data[IFLA_VXLAN_GROUP6]) {
2925                 if (!IS_ENABLED(CONFIG_IPV6))
2926                         return -EPFNOSUPPORT;
2927
2928                 conf.remote_ip.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_GROUP6]);
2929                 conf.remote_ip.sa.sa_family = AF_INET6;
2930         }
2931
2932         if (data[IFLA_VXLAN_LOCAL]) {
2933                 conf.saddr.sin.sin_addr.s_addr = nla_get_in_addr(data[IFLA_VXLAN_LOCAL]);
2934                 conf.saddr.sa.sa_family = AF_INET;
2935         } else if (data[IFLA_VXLAN_LOCAL6]) {
2936                 if (!IS_ENABLED(CONFIG_IPV6))
2937                         return -EPFNOSUPPORT;
2938
2939                 /* TODO: respect scope id */
2940                 conf.saddr.sin6.sin6_addr = nla_get_in6_addr(data[IFLA_VXLAN_LOCAL6]);
2941                 conf.saddr.sa.sa_family = AF_INET6;
2942         }
2943
2944         if (data[IFLA_VXLAN_LINK])
2945                 conf.remote_ifindex = nla_get_u32(data[IFLA_VXLAN_LINK]);
2946
2947         if (data[IFLA_VXLAN_TOS])
2948                 conf.tos  = nla_get_u8(data[IFLA_VXLAN_TOS]);
2949
2950         if (data[IFLA_VXLAN_TTL])
2951                 conf.ttl = nla_get_u8(data[IFLA_VXLAN_TTL]);
2952
2953         if (data[IFLA_VXLAN_LABEL])
2954                 conf.label = nla_get_be32(data[IFLA_VXLAN_LABEL]) &
2955                              IPV6_FLOWLABEL_MASK;
2956
2957         if (!data[IFLA_VXLAN_LEARNING] || nla_get_u8(data[IFLA_VXLAN_LEARNING]))
2958                 conf.flags |= VXLAN_F_LEARN;
2959
2960         if (data[IFLA_VXLAN_AGEING])
2961                 conf.age_interval = nla_get_u32(data[IFLA_VXLAN_AGEING]);
2962
2963         if (data[IFLA_VXLAN_PROXY] && nla_get_u8(data[IFLA_VXLAN_PROXY]))
2964                 conf.flags |= VXLAN_F_PROXY;
2965
2966         if (data[IFLA_VXLAN_RSC] && nla_get_u8(data[IFLA_VXLAN_RSC]))
2967                 conf.flags |= VXLAN_F_RSC;
2968
2969         if (data[IFLA_VXLAN_L2MISS] && nla_get_u8(data[IFLA_VXLAN_L2MISS]))
2970                 conf.flags |= VXLAN_F_L2MISS;
2971
2972         if (data[IFLA_VXLAN_L3MISS] && nla_get_u8(data[IFLA_VXLAN_L3MISS]))
2973                 conf.flags |= VXLAN_F_L3MISS;
2974
2975         if (data[IFLA_VXLAN_LIMIT])
2976                 conf.addrmax = nla_get_u32(data[IFLA_VXLAN_LIMIT]);
2977
2978         if (data[IFLA_VXLAN_COLLECT_METADATA] &&
2979             nla_get_u8(data[IFLA_VXLAN_COLLECT_METADATA]))
2980                 conf.flags |= VXLAN_F_COLLECT_METADATA;
2981
2982         if (data[IFLA_VXLAN_PORT_RANGE]) {
2983                 const struct ifla_vxlan_port_range *p
2984                         = nla_data(data[IFLA_VXLAN_PORT_RANGE]);
2985                 conf.port_min = ntohs(p->low);
2986                 conf.port_max = ntohs(p->high);
2987         }
2988
2989         if (data[IFLA_VXLAN_PORT])
2990                 conf.dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2991
2992         if (data[IFLA_VXLAN_UDP_CSUM] &&
2993             !nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
2994                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM_TX;
2995
2996         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
2997             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
2998                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_TX;
2999
3000         if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
3001             nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
3002                 conf.flags |= VXLAN_F_UDP_ZERO_CSUM6_RX;
3003
3004         if (data[IFLA_VXLAN_REMCSUM_TX] &&
3005             nla_get_u8(data[IFLA_VXLAN_REMCSUM_TX]))
3006                 conf.flags |= VXLAN_F_REMCSUM_TX;
3007
3008         if (data[IFLA_VXLAN_REMCSUM_RX] &&
3009             nla_get_u8(data[IFLA_VXLAN_REMCSUM_RX]))
3010                 conf.flags |= VXLAN_F_REMCSUM_RX;
3011
3012         if (data[IFLA_VXLAN_GBP])
3013                 conf.flags |= VXLAN_F_GBP;
3014
3015         if (data[IFLA_VXLAN_GPE])
3016                 conf.flags |= VXLAN_F_GPE;
3017
3018         if (data[IFLA_VXLAN_REMCSUM_NOPARTIAL])
3019                 conf.flags |= VXLAN_F_REMCSUM_NOPARTIAL;
3020
3021         if (tb[IFLA_MTU])
3022                 conf.mtu = nla_get_u32(tb[IFLA_MTU]);
3023
3024         return vxlan_dev_configure(src_net, dev, &conf);
3025 }
3026
3027 static void vxlan_dellink(struct net_device *dev, struct list_head *head)
3028 {
3029         struct vxlan_dev *vxlan = netdev_priv(dev);
3030         struct vxlan_net *vn = net_generic(vxlan->net, vxlan_net_id);
3031
3032         spin_lock(&vn->sock_lock);
3033         if (!hlist_unhashed(&vxlan->hlist))
3034                 hlist_del_rcu(&vxlan->hlist);
3035         spin_unlock(&vn->sock_lock);
3036
3037         gro_cells_destroy(&vxlan->gro_cells);
3038         list_del(&vxlan->next);
3039         unregister_netdevice_queue(dev, head);
3040 }
3041
3042 static size_t vxlan_get_size(const struct net_device *dev)
3043 {
3044
3045         return nla_total_size(sizeof(__u32)) +  /* IFLA_VXLAN_ID */
3046                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_GROUP{6} */
3047                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LINK */
3048                 nla_total_size(sizeof(struct in6_addr)) + /* IFLA_VXLAN_LOCAL{6} */
3049                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TTL */
3050                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_TOS */
3051                 nla_total_size(sizeof(__be32)) + /* IFLA_VXLAN_LABEL */
3052                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_LEARNING */
3053                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_PROXY */
3054                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_RSC */
3055                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L2MISS */
3056                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_L3MISS */
3057                 nla_total_size(sizeof(__u8)) +  /* IFLA_VXLAN_COLLECT_METADATA */
3058                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
3059                 nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
3060                 nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
3061                 nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
3062                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
3063                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
3064                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
3065                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_TX */
3066                 nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_REMCSUM_RX */
3067                 0;
3068 }
3069
3070 static int vxlan_fill_info(struct sk_buff *skb, const struct net_device *dev)
3071 {
3072         const struct vxlan_dev *vxlan = netdev_priv(dev);
3073         const struct vxlan_rdst *dst = &vxlan->default_dst;
3074         struct ifla_vxlan_port_range ports = {
3075                 .low =  htons(vxlan->cfg.port_min),
3076                 .high = htons(vxlan->cfg.port_max),
3077         };
3078
3079         if (nla_put_u32(skb, IFLA_VXLAN_ID, be32_to_cpu(dst->remote_vni)))
3080                 goto nla_put_failure;
3081
3082         if (!vxlan_addr_any(&dst->remote_ip)) {
3083                 if (dst->remote_ip.sa.sa_family == AF_INET) {
3084                         if (nla_put_in_addr(skb, IFLA_VXLAN_GROUP,
3085                                             dst->remote_ip.sin.sin_addr.s_addr))
3086                                 goto nla_put_failure;
3087 #if IS_ENABLED(CONFIG_IPV6)
3088                 } else {
3089                         if (nla_put_in6_addr(skb, IFLA_VXLAN_GROUP6,
3090                                              &dst->remote_ip.sin6.sin6_addr))
3091                                 goto nla_put_failure;
3092 #endif
3093                 }
3094         }
3095
3096         if (dst->remote_ifindex && nla_put_u32(skb, IFLA_VXLAN_LINK, dst->remote_ifindex))
3097                 goto nla_put_failure;
3098
3099         if (!vxlan_addr_any(&vxlan->cfg.saddr)) {
3100                 if (vxlan->cfg.saddr.sa.sa_family == AF_INET) {
3101                         if (nla_put_in_addr(skb, IFLA_VXLAN_LOCAL,
3102                                             vxlan->cfg.saddr.sin.sin_addr.s_addr))
3103                                 goto nla_put_failure;
3104 #if IS_ENABLED(CONFIG_IPV6)
3105                 } else {
3106                         if (nla_put_in6_addr(skb, IFLA_VXLAN_LOCAL6,
3107                                              &vxlan->cfg.saddr.sin6.sin6_addr))
3108                                 goto nla_put_failure;
3109 #endif
3110                 }
3111         }
3112
3113         if (nla_put_u8(skb, IFLA_VXLAN_TTL, vxlan->cfg.ttl) ||
3114             nla_put_u8(skb, IFLA_VXLAN_TOS, vxlan->cfg.tos) ||
3115             nla_put_be32(skb, IFLA_VXLAN_LABEL, vxlan->cfg.label) ||
3116             nla_put_u8(skb, IFLA_VXLAN_LEARNING,
3117                         !!(vxlan->flags & VXLAN_F_LEARN)) ||
3118             nla_put_u8(skb, IFLA_VXLAN_PROXY,
3119                         !!(vxlan->flags & VXLAN_F_PROXY)) ||
3120             nla_put_u8(skb, IFLA_VXLAN_RSC, !!(vxlan->flags & VXLAN_F_RSC)) ||
3121             nla_put_u8(skb, IFLA_VXLAN_L2MISS,
3122                         !!(vxlan->flags & VXLAN_F_L2MISS)) ||
3123             nla_put_u8(skb, IFLA_VXLAN_L3MISS,
3124                         !!(vxlan->flags & VXLAN_F_L3MISS)) ||
3125             nla_put_u8(skb, IFLA_VXLAN_COLLECT_METADATA,
3126                        !!(vxlan->flags & VXLAN_F_COLLECT_METADATA)) ||
3127             nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->cfg.age_interval) ||
3128             nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->cfg.addrmax) ||
3129             nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->cfg.dst_port) ||
3130             nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
3131                         !(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM_TX)) ||
3132             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
3133                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) ||
3134             nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
3135                         !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)) ||
3136             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_TX,
3137                         !!(vxlan->flags & VXLAN_F_REMCSUM_TX)) ||
3138             nla_put_u8(skb, IFLA_VXLAN_REMCSUM_RX,
3139                         !!(vxlan->flags & VXLAN_F_REMCSUM_RX)))
3140                 goto nla_put_failure;
3141
3142         if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))
3143                 goto nla_put_failure;
3144
3145         if (vxlan->flags & VXLAN_F_GBP &&
3146             nla_put_flag(skb, IFLA_VXLAN_GBP))
3147                 goto nla_put_failure;
3148
3149         if (vxlan->flags & VXLAN_F_GPE &&
3150             nla_put_flag(skb, IFLA_VXLAN_GPE))
3151                 goto nla_put_failure;
3152
3153         if (vxlan->flags & VXLAN_F_REMCSUM_NOPARTIAL &&
3154             nla_put_flag(skb, IFLA_VXLAN_REMCSUM_NOPARTIAL))
3155                 goto nla_put_failure;
3156
3157         return 0;
3158
3159 nla_put_failure:
3160         return -EMSGSIZE;
3161 }
3162
3163 static struct net *vxlan_get_link_net(const struct net_device *dev)
3164 {
3165         struct vxlan_dev *vxlan = netdev_priv(dev);
3166
3167         return vxlan->net;
3168 }
3169
3170 static struct rtnl_link_ops vxlan_link_ops __read_mostly = {
3171         .kind           = "vxlan",
3172         .maxtype        = IFLA_VXLAN_MAX,
3173         .policy         = vxlan_policy,
3174         .priv_size      = sizeof(struct vxlan_dev),
3175         .setup          = vxlan_setup,
3176         .validate       = vxlan_validate,
3177         .newlink        = vxlan_newlink,
3178         .dellink        = vxlan_dellink,
3179         .get_size       = vxlan_get_size,
3180         .fill_info      = vxlan_fill_info,
3181         .get_link_net   = vxlan_get_link_net,
3182 };
3183
3184 struct net_device *vxlan_dev_create(struct net *net, const char *name,
3185                                     u8 name_assign_type,
3186                                     struct vxlan_config *conf)
3187 {
3188         struct nlattr *tb[IFLA_MAX + 1];
3189         struct net_device *dev;
3190         int err;
3191
3192         memset(&tb, 0, sizeof(tb));
3193
3194         dev = rtnl_create_link(net, name, name_assign_type,
3195                                &vxlan_link_ops, tb);
3196         if (IS_ERR(dev))
3197                 return dev;
3198
3199         err = vxlan_dev_configure(net, dev, conf);
3200         if (err < 0) {
3201                 free_netdev(dev);
3202                 return ERR_PTR(err);
3203         }
3204
3205         err = rtnl_configure_link(dev, NULL);
3206         if (err < 0) {
3207                 LIST_HEAD(list_kill);
3208
3209                 vxlan_dellink(dev, &list_kill);
3210                 unregister_netdevice_many(&list_kill);
3211                 return ERR_PTR(err);
3212         }
3213
3214         return dev;
3215 }
3216 EXPORT_SYMBOL_GPL(vxlan_dev_create);
3217
3218 static void vxlan_handle_lowerdev_unregister(struct vxlan_net *vn,
3219                                              struct net_device *dev)
3220 {
3221         struct vxlan_dev *vxlan, *next;
3222         LIST_HEAD(list_kill);
3223
3224         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3225                 struct vxlan_rdst *dst = &vxlan->default_dst;
3226
3227                 /* In case we created vxlan device with carrier
3228                  * and we loose the carrier due to module unload
3229                  * we also need to remove vxlan device. In other
3230                  * cases, it's not necessary and remote_ifindex
3231                  * is 0 here, so no matches.
3232                  */
3233                 if (dst->remote_ifindex == dev->ifindex)
3234                         vxlan_dellink(vxlan->dev, &list_kill);
3235         }
3236
3237         unregister_netdevice_many(&list_kill);
3238 }
3239
3240 static int vxlan_netdevice_event(struct notifier_block *unused,
3241                                  unsigned long event, void *ptr)
3242 {
3243         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3244         struct vxlan_net *vn = net_generic(dev_net(dev), vxlan_net_id);
3245
3246         if (event == NETDEV_UNREGISTER)
3247                 vxlan_handle_lowerdev_unregister(vn, dev);
3248         else if (event == NETDEV_UDP_TUNNEL_PUSH_INFO)
3249                 vxlan_push_rx_ports(dev);
3250
3251         return NOTIFY_DONE;
3252 }
3253
3254 static struct notifier_block vxlan_notifier_block __read_mostly = {
3255         .notifier_call = vxlan_netdevice_event,
3256 };
3257
3258 static __net_init int vxlan_init_net(struct net *net)
3259 {
3260         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3261         unsigned int h;
3262
3263         INIT_LIST_HEAD(&vn->vxlan_list);
3264         spin_lock_init(&vn->sock_lock);
3265
3266         for (h = 0; h < PORT_HASH_SIZE; ++h)
3267                 INIT_HLIST_HEAD(&vn->sock_list[h]);
3268
3269         return 0;
3270 }
3271
3272 static void __net_exit vxlan_exit_net(struct net *net)
3273 {
3274         struct vxlan_net *vn = net_generic(net, vxlan_net_id);
3275         struct vxlan_dev *vxlan, *next;
3276         struct net_device *dev, *aux;
3277         LIST_HEAD(list);
3278
3279         rtnl_lock();
3280         for_each_netdev_safe(net, dev, aux)
3281                 if (dev->rtnl_link_ops == &vxlan_link_ops)
3282                         unregister_netdevice_queue(dev, &list);
3283
3284         list_for_each_entry_safe(vxlan, next, &vn->vxlan_list, next) {
3285                 /* If vxlan->dev is in the same netns, it has already been added
3286                  * to the list by the previous loop.
3287                  */
3288                 if (!net_eq(dev_net(vxlan->dev), net)) {
3289                         gro_cells_destroy(&vxlan->gro_cells);
3290                         unregister_netdevice_queue(vxlan->dev, &list);
3291                 }
3292         }
3293
3294         unregister_netdevice_many(&list);
3295         rtnl_unlock();
3296 }
3297
3298 static struct pernet_operations vxlan_net_ops = {
3299         .init = vxlan_init_net,
3300         .exit = vxlan_exit_net,
3301         .id   = &vxlan_net_id,
3302         .size = sizeof(struct vxlan_net),
3303 };
3304
3305 static int __init vxlan_init_module(void)
3306 {
3307         int rc;
3308
3309         get_random_bytes(&vxlan_salt, sizeof(vxlan_salt));
3310
3311         rc = register_pernet_subsys(&vxlan_net_ops);
3312         if (rc)
3313                 goto out1;
3314
3315         rc = register_netdevice_notifier(&vxlan_notifier_block);
3316         if (rc)
3317                 goto out2;
3318
3319         rc = rtnl_link_register(&vxlan_link_ops);
3320         if (rc)
3321                 goto out3;
3322
3323         return 0;
3324 out3:
3325         unregister_netdevice_notifier(&vxlan_notifier_block);
3326 out2:
3327         unregister_pernet_subsys(&vxlan_net_ops);
3328 out1:
3329         return rc;
3330 }
3331 late_initcall(vxlan_init_module);
3332
3333 static void __exit vxlan_cleanup_module(void)
3334 {
3335         rtnl_link_unregister(&vxlan_link_ops);
3336         unregister_netdevice_notifier(&vxlan_notifier_block);
3337         unregister_pernet_subsys(&vxlan_net_ops);
3338         /* rcu_barrier() is called by netns */
3339 }
3340 module_exit(vxlan_cleanup_module);
3341
3342 MODULE_LICENSE("GPL");
3343 MODULE_VERSION(VXLAN_VERSION);
3344 MODULE_AUTHOR("Stephen Hemminger <stephen@networkplumber.org>");
3345 MODULE_DESCRIPTION("Driver for VXLAN encapsulated traffic");
3346 MODULE_ALIAS_RTNL_LINK("vxlan");